University of Southern California Dissertations and Theses

Hyperspectral phasor for multiplexed fluorescence microscopy and autofluorescence-based pathologic diagnosis

(USC Thesis Other)

Hyperspectral phasor for multiplexed fluorescence microscopy and autofluorescence-based pathologic diagnosis

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content

Hyperspectral Phasor
for Multiplexed Fluorescence Microscopy and Autofluorescence-based Pathologic Diagnosis

by

Pu Wang

A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BIOMEDICAL ENGINEERING)

December 2022

Copyright 2022 Pu Wang
ii

Dedication

I would like to dedicate the thesis to my parents, who would always be there and support their
son for every endeavor in his life. I would also like to dedicate this thesis to my loving wife who
gave birth to our lovely son Lennon while the thesis was being written.

iii

Acknowledgements

I was a master student who was unsure about my next step, finding a job or pursuing a Ph.D.
degree, back in 2017. That was when I stumbled across a volunteer student worker job posting
from Sarah Madaan, who was a Ph.D. student with Dr. Scott Fraser and Dr. Thai Truong at that
time. She was looking for a student to help her with a side project and I submitted my resume
just to take my chance as I did not have any experience with microscopy at that time. I was
lucky enough to get the chance to work in the lab as a volunteer student. Instead of working
directly with Sarah, I was assigned to a project that was led by Dr. Francesco Cutrale, who later
became my Ph.D. mentor and a life-long friend of mine.
First and foremost, I would like to give my special thanks to my PI and mentor Dr. Scott Fraser
for giving me the opportunity of my Ph.D. endeavor. I am grateful that he provides me with
such a great platform and lots of opportunities to expose myself to the most advanced
technologies in the field. Scott has been giving my full trust and freedom to explore new ideas.
When I have doubt in myself, I always remind myself of what he said to me, ‘If you think it’s
going to fail, let it fail fast’. In short, he gave me so much encouragement and helped me
through many obstacles along the way.
I would also like to thank my mentor Dr. Francesco Cutrale who has taught me so much not
only his professional knowledge but also the positive attitude during difficult times. We worked
so closely on every project, and I am always grateful for his attendance to lots of details even
though he was occupied by so many projects. Francesco set a great example to me of how much
potential a person can squeeze out of himself by working hard and smartly.
iv

Of course, it was team efforts for the projects that I was involved in during and prior my Ph.D. I
would like to thank my collaborator Dr. David Warburton and Dr. Gianluca Turcatel from
Children’s Hospital Los Angeles (CHLA) for working together on an amazing project in 2017
and 2018. My colleagues Kevin Keomanee-Dizon, Thai Truong and Matthew Jones have been a
tremendous help for my Ph.D. main project, and I would like to give my thanks to them. I would
also thank Dr. Keyue Shen who gave me full trust and allowed me to experimentally build a new
SHy-Cam system for their microscope and provided funding for the project cost.
I am grateful to Dr. Andrea Armani, Dr. Cristina Zavaleta, Dr. Jennifer Treweek for serving my
committee for my Ph.D. qualifying exam and thesis defense and providing invaluable
suggestions on the direction of my research. My graduate affairs advisors, William Yang and
Mischalgrace Diasanta, have always been extremely patient and helping with every major step
throughout my Ph.D. journey. I want to thank them for lots of support happening in the
background. Also, I would like to express my special thanks to Jona Cura, Georgian Keller,
Claire Cato, Socorro Aguirre and all the lab members who provided me with love and support.
Over the past few years during my Ph.D., I received financial support from USC Annenberg
Fellowship, United States Department of Defense (PR150666), Chan Zuckerberg Initiative
(CZI), and USC Alfred E. Mann Institute (AMI). I want to specially thank Winn Hong, the
deputy executive director of AMI, who generously provided me with lots of support and
resources.
Finally, I want to thank my parents, who always give one hundred percent of support to their son
to pursue his dream. I want to thank my wife, Xiaoxi, who has been always by my side giving
me emotional support. Thank you for bringing us such a lovely son and reminding me how life
can be much more meaningful with him. I also want to say thank you to my old friend, my old
v

partner Tianlu Tang for witnessing my growth and being part of my life during the most
important years.

vi

Table of Contents
Dedication ....................................................................................................................................... ii
Acknowledgements ........................................................................................................................ iii
List of Tables ................................................................................................................................. ix
List of Figures ................................................................................................................................. x
Abstract ........................................................................................................................................ xii
Chapter 1 Introduction .................................................................................................................... 1
1.1 Light microscopy: magnifying power that visualizes the invisible.................................. 1
1.1.1 Optical Magnification ............................................................................................... 1
1.1.2 Resolution ................................................................................................................. 3
1.1.3 The essentials of infinity-corrected light microscopes ............................................. 5
1.2 Fluorescence: lighting up biological processes ................................................................ 7
1.2.1 Principle of fluorescence........................................................................................... 7
1.2.2 Autofluorescence (endogenous/native fluorescence) ............................................... 9
1.2.3 Photobleaching ........................................................................................................ 10
1.2.4 Fluorescence microscope instrumentation .............................................................. 12
1.2.5 Multiplexed/multi-color fluorescence imaging....................................................... 16
1.3 Summary ........................................................................................................................ 19
Chapter 2 SHy-Cam: a snapshot hyperspectral phasor camera for fast, multi-color
fluorescence microscopy ......................................................................................... 21
2.1 Introduction .................................................................................................................... 21
2.2 Results ............................................................................................................................ 25
2.2.1 Accessing spectral information with faster single snapshot acquisition ................. 25
2.2.2 SHy-Cam is a cost-effective and versatile add-on device for existing
microscopes ............................................................................................................ 26
2.2.3 SHy-Cam imaging validation tests ......................................................................... 27
2.3 Discussion ...................................................................................................................... 33
2.4 Methods .......................................................................................................................... 36
2.4.1 Spectral compression with hyperspectral phasor .................................................... 36
2.4.2 Sinusoidal dichroic mirror ...................................................................................... 38
2.4.3 Optical design and setup ......................................................................................... 38
vii

2.4.4 Opto-mechanics ...................................................................................................... 40
2.4.5 Image acquisition .................................................................................................... 41
2.4.6 Image pre-processing .............................................................................................. 42
2.4.7 Image analysis ......................................................................................................... 45
2.4.8 Photon-efficiency and -throughput estimation........................................................ 47
2.4.9 Sample preparation ................................................................................................. 48
2.4.10 Zebrafish lines ......................................................................................................... 50
2.5 Code availability ............................................................................................................ 51
2.6 Publication information .................................................................................................. 51
Chapter 3 Hyperspectral phasor autofluorescence characterization of unstained Barrett’s
esophagus dysplasia and adenocarcinoma unstained label-free biopsies ............... 53
3.1 Introduction .................................................................................................................... 53
3.2 Methods .......................................................................................................................... 55
3.2.1 Singleshot Hyperspectral Phasor Digital Slide Scanner (SHy-DSS) ...................... 55
3.2.2 SHy-DSS image acquisition and processing pipeline ............................................. 59
3.2.3 Biopsy slide preparation and standard H&E imaging............................................. 61
3.2.4 Autofluorescence characterization with SHy-DSS image ...................................... 62
3.3 Results ............................................................................................................................ 63
3.3.1 High-grade dysplasia (HGD) in Barrett’s esophagus (BE) ..................................... 63
3.3.2 Esophageal adenocarcinoma at different depths of invasion .................................. 64
3.4 Discussion ...................................................................................................................... 68
3.5 Data Availability ............................................................................................................ 70
3.6 Publication information .................................................................................................. 70
Chapter 4 Conclusions ............................................................................................................... 72
4.1 Summary of findings ...................................................................................................... 72
4.2 Future applications ......................................................................................................... 74
4.2.1 Multiplexed stem cell imaging................................................................................ 74
4.2.2 Autofluorescence-based endoscopic esophageal adenocarcinoma diagnosis ......... 75
4.2.3 Behavior study based on zebrafish embryo whole-brain functional imaging......... 78
4.2.4 Beyond biomedical: hyperspectral phasor handheld fruit quality inspection
camera ..................................................................................................................... 78
References ................................................................................................................................... 82
Appendices ................................................................................................................................... 91
viii

A Part list and CAD diagram of SHy-Cam-SPIM .................................................................. 91

ix

List of Tables
Table 1 SHy-Cam photon efficiency of 5 commonly used fluorescence. .................................. 48

Table A SHy-Cam part list ........................................................................................................... 83

x

List of Figures
Fig. 1.1 Achieving angular magnification by moving subject closer to the eye.. .......................... 2
Fig. 1.2 Air disk and Rayleigh criterion for light microscope theoretical resolution. ................... 4
Fig. 1.3 Comparison of infinity-corrected microscopes and finite corrected microscopes. .......... 5
Fig. 1.4 Jablonski diagram showing the process of fluorescence. . .............................................. 8
Fig. 1.5 Emission spectrum of nicotinamide adenine dinucleotide (NADH) and other three
commonly used fluorophores.. ......................................................................................... 9
Fig. 1.6 Diagram of the essential components of an infinity-correct epi-fluorescence
microscope. ..................................................................................................................... 13
Fig. 1.7 Basic epi-fluorescence microscopes lack optical sectioning capability. ......................... 14
Fig. 1.8 Confocal laser scanning microscopes (CLSM) achieve optical sectioning by
selectively detecting fluorescence from a single plane. ................................................. 15
Fig. 1.9 Selective plane illumination microscopes (SPIMs) achieve optical sectioning by
selectively exciting a thin layer of the thick sample. ..................................................... 16
Fig. 1.10 Hyperspectral phasor forms a 2-D compressed representation of high-dimension
hyperspectral fluorescence data.. .................................................................................. 19
Fig. 2.1 SHy-Cam approach overview. ........................................................................................ 23
Fig. 2.2 SHy-Cam spectral unmixing on standard and fixed samples. ........................................ 28
Fig. 2.3 SPIM-SHy-Cam tiled volumetric in-vivo imaging. ....................................................... 31
Fig. 2.4 SPIM-SHy-Cam dynamic in-vivo imaging.. .................................................................. 33
Fig. 2.5 SHy-Cam 3D model and optical diagram of one channel. ............................................. 40
Fig. 2.6 Image pre-processing pipeline.. ...................................................................................... 44
Fig. 2.7 Spectral analysis on spectral phasor plane. .................................................................... 46
Fig. 3.1 Single snapshot hyperspectral phasor camera (SHy-DSS) system overview. ................ 56
Fig. 3.2 The 3D rendering and picture of SHy-DSS. ................................................................... 58
xi

Fig. 3.3 SHy-DSS image acquisition and processing pipeline. ................................................... 61
Fig. 3.4 Autofluorescence characterization of high-grade dysplasia (HGD) in esophagus
epithelium. ..................................................................................................................... 64
Fig. 3.5 Anatomy of esophagus cross section. ........................................................................... 65
Fig. 3.6 Autofluorescence characterization of esophageal adenocarcinoma. .............................. 67
Fig. 4.1 SHy-Cam for cancer stem cell imaging. ......................................................................... 75
Fig. 4.2 Diagram of SHy-ES. ....................................................................................................... 77
Fig. 4.3 SHy-PC concept and workflow. ..................................................................................... 81
Fig. A 1 The CAD model of SHy-CAM adapted to hSPIM in Chapter 2 .................................... 92

xii

Abstract
This thesis summarizes my efforts in defining a new imaging method for faster, more efficient
multi-color fluorescence imaging with a lower instrumentation cost. Chapter 1 provides the
readers with a brief introduction of light microscopes, fluorescence and existing fluorescence
imaging methods and their limitations, specifically for multi-color imaging applications.
Chapter 2 presents the efforts to build an optical imaging module for a selective plane
illumination microscope (SPIM) to achieve fast fluorescence multiplexing. The system was
named Single-Shot Hyperspectral Phasor Camera (SHy-Cam) and validated with in-vivo imaging
experiments with transgenic multi-color zebrafish embryos. Chapter 3 extends the concept of
SHy-Cam from biological imaging applications to medical applications of diagnostic imaging
with label-free biopsy slides based-on tissue autofluorescence. Chapter 4 summarizes the
research findings and further discusses the potential future applications of SHy-Cam.

1

Chapter 1
Introduction
1.1 Light microscopy: magnifying power that visualizes the invisible
1.1.1 Optical Magnification
There are two types of “magnifications” in our daily life, one affects the physical size while the
other affects the apparent size, without changes in the physical size. The famous Abraham
Lincoln statue in Lincoln Memorial is a good example of the former case and it is a physically
scaled version of the president Lincoln. Obviously, one cannot ask the local newspaper office to
print their newspaper ten times larger so that the news can be read more clearly. One possibility
is to somehow make the text appear larger than its actual printed size to get a better reading
experience. This approach refers to the second type of magnification. The magnification we will
talk about in this thesis is only related to the second type as one of the most important tasks of
light microscopy is truthfully present a magnified version of the object so that the observer can
2

have an objective understanding of the object with as little disturbance to its shape as possible.
This type of magnification is achieved utilizing optics.
Magnification in optics is further divided into two types: linear magnification and angular
magnification. One example of linear magnification is projecting the 35mm films onto a large
movie screen so that even the audience sitting 100 feet away from the screen in the last row can
enjoy the moving pictures and understand the stories unfolding in the movie. Linear
magnification represents a proportional enlargement (or shrinking) in size comparing the image
with the actual object and is normally represented in a two-dimension plane. In microscopy
settings, the optics inside of the microscope projects a magnified version of the 2-D sample plane
onto another 2-D plane called image plane, where an imager is placed to capture the magnified
spatial information. The magnification that will be mentioned multiple times in the thesis
throughout the next two chapters is linear magnification. The other type is called angular
magnification which naturally determines how our eyes perceive the size of an object. The
example in Fig 1.1 represents the simplest and everyday example of the angular magnification
which happens naturally with the optics that everyone has, our eye lenses. Angular
magnification is a very important concept when describing microscopes with an eyepiece for
Fig. 0.1 Achieving angular magnification by moving subject closer to the eye. The observer perceives the enlargement of
the object by simply moving it closer to the eye or move him/herself close to the subject. The nodal ray which is the light ray
that pass through the nodal point of the eye lens without bending determines the ray angle 𝛼 , 𝜃 with respect to the optical axis
and ultimately determines where the light is focused on the retina. Figure cited from [112].
3

directly viewing with our eyes instead of capturing linearly magnified images of the object with
a sensor as well.
1.1.2 Resolution
Just like magnification, resolution is another most essential concept and specification of light
microscopes. Resolution defines the minimum distance at which two distinct points on the
object or specimen being observed can still be distinguished from each other. Without
considering the real-world imperfection of the optics and alignment of a light microscope
system, resolution is governed by a consequence of the wave nature of light called
diffraction [1]. The image of an infinitesimal point in the specimen through a light microscope is
not a point with a defined edge due to diffraction. Instead, the image appears as a slightly
blurred spot surrounded with multiple ring-like patterns with decreasing intensity called an Air
disk (Fig. 1.2).
4

The theoretical resolution d
𝑅 (Fig. 1.2b) of a light microscope is directly related to the effective
numerical aperture (NA) (Eq. 1.2) of the objective lens and the wavelength of the light by Eq.
1.3 and the distance is called Rayleigh criterion [2]. d
𝑅 is the resolution limit on the sample
plane. The theoretical resolution limit on the sensor plane D
𝑅 can be estimated with the total
magnification of the microscope M by Eq. 1.4. D
𝑅 is a very important parameter to determine
the pixel size of the camera used to capture the magnified microscope image. To match the
camera to the light microscope resolution, the pixel size needs to be equal or smaller than half of
D
𝑅 in order to satisfy Nyquist-Shannon sampling theorem.
d =
1.22𝜆 𝑁𝐴
(1.1)
Fig. 0.2 Air disk and Rayleigh criterion for light microscope theoretical resolution. (a) 𝛼 is the half light acceptance
angle of the microscope objective lens. d on the Airy disk represents the diameter of the center airy pattern which contains
84% of total light intensity and can be calculated by Eq. 1.1. (b) As two infinitesimal points come closer, their Airy discs
starts to merge and overlay with each other. When the distance of the two centers is less than 𝑑 𝑅 called Rayleigh critetion, we
define two two points are no longer distinguishable and this number can be used to estimate the theoretical resolution limit of
a light microscope. Note 𝑑 𝑅 represents the distance in the sample/object plane. Figure adapted from [113].
5

𝑁𝐴 = 𝑛 ∙ sin (𝛼 ) (1.2)
d
𝑅 =
0.61𝜆 𝑁𝐴
(1.3)
D
𝑅 = 𝑀 ∙ d
𝑅 (1.4)
1.1.3 The essentials of infinity-corrected light microscopes
Light microscopes are optical systems that can manipulate and bend the light coming from the
sample in a certain way to achieve linear or angular magnification for taking detailed
photographs or directly viewing the object in great details. They have been the most important
tools for scientists in the life science field as most of the underlying biological processes are in
such a small scale that will not be seen without the help of light microscopes. There are many
different variants and configurations of light microscopes for different applications. The one that
is most related to the topic of this thesis is infinity-corrected fluorescence microscope. We will
introduce fluorescence in the next sub-chapter.
Infinity-corrected microscopes (Fig. 1.3a) use an objective lens to collect light from the
sample placed at a finite distance called working distance and collimate the light from the back
aperture into a space called infinity space. Then, a positive lens called the tube lens refocuses
the collimated light on two an imager which can be a single point detector or a 2-D detector such
Fig. 0.3 Comparison of infinity-corrected microscopes and finite corrected microscopes. (a) The essential components of
an infinity-corrected microscope detection path. The space between the objective lens and tube lens is extendable and inserting
optical filters has minimum effect to the operation of the microscope. (b) The essential components of a finite-corrected
microscope which has a fixed length between the objective lens and the tube lens.
6

as charge-coupled device (CCD) or complementary metal-oxide semiconductor (CMOS). The
naming of ‘infinity space’ comes from the fact that the light rays coming from the same point on
the sample are collimated focused to a plane that is located at an infinity distance. The
collimation of the light rays in infinity space provides infinity-corrected microscopes with two
desired advantages over its counterpart, finite-corrected microscopes (Fig. 1.3b) which has
diverging and converging light between the objective lens and tube lens. The first is the length
of infinity space will not affect the total magnification of the microscope. The finite corrected
microscopes, on the other hand, have a fixed distance between the objective lens and tube lens to
achieve the designed magnification. This provides the flexibility to change the infinity space
length to accommodate design requirements for specific applications. The second advantage is
the infinity corrected microscopes do not need to be refocused when inserting optical filters in
the infinity space which make it an ideal configuration for fluorescence imaging applications
which sometimes require switching multiple sets of optical filters with different thicknesses. The
total linear magnification can be calculated from the
effective focal length of the objective lens 𝑓 𝑜 and the tube lens 𝑓 𝑡 (Eq. 1.5). In practice, the
magnification is already specified with the objective lens if used with its corresponding tube lens
which is distinct for each manufacture and Eq. 1.1 is normally used to calculate the effective
focal length of the objective lens. We can gather more useful information regarding the back
aperture size of the objective lens combining the information of its focal length and working
distance, which determines the diameter of the marginal rays in the infinity space. Then we
know the upper limit of the infinity space length given the certain size constraints of the tube
lens and opto-mechanics along downstream of the optical train. This is an important design
criterion and consideration for the work presented in Chapter 2 and 3.
7

𝑀 =
𝑓 𝑡 𝑓 𝑜 (1.5)
1.2 Fluorescence: lighting up biological processes
1.2.1 Principle of fluorescence
In the previous sub-chapter, we briefly introduced the concept of magnification and resolution in
the context of light microscopy. In this sub-chapter, we will look at another important aspect of
light microscopy imaging which is the contrast, more specifically, the contrast generated with
fluorescence.
Fluorescence is a phenomenon that happens in our daily life. Have you ever wondered why the
ink coming out of a highlighter looks so bright? The deep blue and ultraviolet (UV) region of
sunlight or other white light source can cause the fluorescent dye inside of the highlighter ink to
emit photons, which is the process called fluoresce. The same process is largely used in today’s
medical and biological researches and provides the contrast for visualizing complex biological
processes in living samples. Traditional light microscopes rely on a staining process to create
transmission contrast for viewing otherwise nearly transparent cell structures. The sample must
be fixed before the staining process; thus, it is not feasible for studying the on-going biological
processes happening in living samples. Fluorescent probes can bind to specific proteins or
biological structures with high specificity in live samples, which provides a clear three-
dimensional internal structure of the sample. Genetically modified animals expressing multiple
fluorescent proteins [3,4] provide invaluable animal models for developmental studies that
traditional light microscopy cannot do.
8

Fluorescence describes the phenomenon where a fluorescent molecule is driven to the excited
state from the ground state 𝑆 0
(Fig. 1.4) by the absorption of photons at a specific wavelength
and relaxes to the ground again through the emission of photons of a longer wavelength without
a change in electron spin. As shown in the simplified Jablonski diagram, fluorescence typically
involves the first singlet excited state 𝑆 1
(Fig. 1.4). The molecules that are excited to higher
states 𝑆 𝑛 ,(𝑛 >1)
first relax to the lowest vibrational level of the first excited state 𝑆 1
by dissipating
energy through non-radiative processes, including internal conversion followed by vibrational
relaxation [5]. The whole process can be visualized with the famous Jablonski diagram (Fig.
1.4). The ‘red-shift’ phenomenon of longer re-emitted fluorescence photons compared to the
excitation photons is called the Stokes shift [7] which is due to the photon energy loss during the
non-radiative relaxation.
Fig. 0.4 Jablonski diagram showing the process of fluorescence. After being excited to the excited state, the fluorescence
molecules relax to the lowest-energy excited state with non-radiative transition in a short time, ranging from 10
−15
to 10
−9

seconds. This short time is called fluorescence lifetime which is an important property for different fluorescence molecules or
same fluorescence molecules in different micro-environments. The molecules then take another 10
−9
to 10
−7
seconds to relax
to the ground state with the emission of photons with longer wavelength than the originally absorbed light. Image cited
from [115].
9

1.2.2 Autofluorescence (endogenous/native fluorescence)
In addition to fluorescent markers such as fluorophores and fluorescent proteins, fluorescence
process can also happen within biological structures such as lysosomes and mitochondria, which
is called autofluorescence [6]. In fluorescence microscopy, autofluorescence is generally
undesirable as it can interfere the detection of other fluorescent labels. To make things worse,
autofluorescent molecules normally have a much wider emission spectrum compared with
fluorophores (Fig. 1.5) making it a challenging task to remove the autofluorescence cross talk
from the fluorescence of interests. One commonly used computational method for cleaning
autofluorescence cross talk in multispectral or hyperspectral fluorescence microscope images is
spectral linear unmixing [7,8]. It uses the known spectra of the autofluorescence and
fluorescence signals as the reference and attempts to estimate the relative contribution of each
signal from the mixed signals. Another computational method called hyperspectral phasor [9–
12] has been gaining more popularity in recent years for its robustness in low signal-to-noise
conditions and its intuitive use. Chapter 2 presents an example of combining the two methods
for an effective unmixing of autofluorescence with multiple fluorescent signals. In medical
applications, autofluorescence is a rapidly emerging diagnostic imaging contrast [13–17]. The
Fig. 0.5 Emission spectrum of nicotinamide adenine dinucleotide (NADH) and other three commonly used fluorophores.
NADH (autofluorescence) has significantly wider emission spectrum and can cover the emission spectrum of multiple
fluorophores that emit in different wavelength ranges.
10

spectral and lifetime properties of tissue autofluorescence correlate closely with tissue
metabolism. Chapter 4 explores autofluorescence as a potential diagnostic contrast and metric of
esophageal adenocarcinoma.
1.2.3 Photobleaching
Fluorescent molecules have limited lifespans in terms of the capability to fluoresce. This is one
of the challenges in fluorescence microscopy requires careful avoidance of overexposure of the
sample to the excitation light source which would render the sample unusable for imaging
purposes. Photobleaching is the phenomenon where the fluorescent molecules start to
permanently lose the ability to fluoresce, leading to the fading of the fluorescent signals [18,19].
When the fluorescent molecules are excited from the ground state to the excited state, they may
interact with molecular oxygen and undergo permanent covalent modifications before
fluorescence emission [20,21]. The total number of fluoresce cycles between exciting and
emission is dependent on the structure of the fluorescent molecules and the surrounding
environment. In fluorescence microscopy applications, photobleaching is one of the most
concerning factors that researchers must take into consideration during the planning of the
imaging experiments. There several commonly used ways to reduce photobleaching:
• Reduction of the excitation power
Every fluorescent molecule has a certain wavelength band for most efficient and effective
excitation and it is normally quantified with quantum yield [22]. By wisely choosing the
excitation source to match the fluorescence excitation peak can help reduce the excitation
power needed for achieving a sufficient signal-to-noise ratio (SNR). In addition, with the
advancement of high sensitivity scientific CMOS cameras with significantly lower read
noise, we get the opportunity to achieve good imaging results with lower excitation power.
11

In multiplexed or multi-color fluorescence imaging, choosing the right spectral sampling
strategies which allows more photons to reach the detector will also help reduce the
excitation power and minimize photobleaching.
• Minimizing exposure time and frequency
In addition to using a reduced excitation power, minimizing the time required for the
excitation and the frequency of exposing the sample to the light source will help minimize
photobleaching in a same way. Similarly, choosing high sensitivity imagers and spectral
sampling strategies help minimize the exposure time. The exposure frequency can be
minimized by using snapshot imaging methods instead of imaging with multiple exposures
which is normally found in multi-color fluorescence imaging applications with multiple
fluorescence filter sets.
• Using gentler excitation
When determining whether an excitation method is gentle or not, there are two aspects to
consider: the excitation volume and the photobleaching caused with a certain power within a
certain duration of the experiment. The most basic configuration of fluorescence is called
epifluorescence microscope. It uses the same objective lens to send the excitation light to
and collect the fluorescence emission light from the sample. It excites the entire depth of the
sample and causes more photobleaching. In contrast, the more complex laser scanning
confocal microscope, two-photon microscope and selective plane illumination microscope
can selectively excite a small volume of the sample without unnecessary excitation to other
portions of the sample, which help reduce photobleaching. Two-photon excitation has
proven to be a gentler excitation source compared to traditional single-photon excitation not
12

only due to the confined excitation volume but also to the relatively lower energy load of
longer wavelengths [23,24] which reduces the overall harm to the biological samples.
1.2.4 Fluorescence microscope instrumentation
Previously, we briefly talked about infinity-correct microscopes and the advantages. We also
touched on the topic of imaging contrast with fluorescence. In this sub-chapter, we combine
both concept and introduce the basic instrumentation of fluorescence microscopes in infinity-
corrected configuration.
When talking about light microscope, people tend to refer to the traditional bright-field
microscopes [27–29] which use the transmitted light as the imaging contrast and arguably are the
most common type of microscopes. Besides the imaging contrast difference between
fluorescence and transmitted light, there are a few important aspects of fluorescence microscopes
that distinguishes them from bright-field microscope, including:
Fluorescence emission and excitation separation
In order to increase fluorescence imaging specificity, it is very important to selectively detect the
emission signal and reject the excitation light and prevent light contamination. Fig. 1.6 shows
the diagram of the infinity-corrected epi-fluorescence microscope which has the most ‘basic’
form factor among all fluorescence microscopes. Comparing Fig. 1.6 with Fig. 1.3a, epi-
fluorescence microscope adds a critical component called filter cube into its infinity space. Epi-
fluorescence microscope is normally equipped with a white light source with a broad spectrum
such as xenon arc, mercury vapor lamp or LED. An optical excitation filter inside of the filter
cube allows a certain wavelength range of light to pass through and excite the fluorescence. A
45-degree positioned long-pass dichroic mirror reflects the filtered excitation light to the back
aperture of the objective lens for sample excitation and allows the returning fluorescence
13

emission with longer wavelength to be transmitted through. The emission light then gets filtered
with emission filter before being converged by the tube lens and focused on the imager.
3-D volumetric imaging capabilities
Basic epi-fluorescence microscopes are only capable of imaging sliced samples with the
thickness up to 10𝜇𝑚 due to the lack of optical sectioning capability to resolve the depth
information. The excitation light coming out of an epi-fluorescence microscope illuminates the
whole sample (Fig. 1.7) depth, so it does not have the capability of specifically imaging a certain
plane thus it is only suitable for imaging thin samples. However, biological processes extend to
the 3-D volume of live samples with the thickness beyond the imaging capability of epi-
fluorescence microscopes. Several complex fluorescence microscope configurations were
invented to overcome the 2-D barrier.
Fig. 0.6 Diagram of the essential components of an infinity-correct epi-fluorescence microscope. Imaging contrast is the
main difference between fluorescence microscopes and traditional transmission or reflection-based light microscopes. A filter
cube is placed in the infinity space to selectively reflect the excitation light to the sample and allows the emitted fluorescence
light to transmit through the cube for detection by the sensor. The filter cube contains three optical filters, a short-pass or band-
pass EX (excitation) filter, a dichroic mirror and a long-pass or band-pass EM (emission) filter.
14

Confocal laser scanning microscopes (CLSMs) are the most used fluorescence microscope in
biological imaging applications that provide diffraction limited performance with optical
sectioning capability [38]. CLSMs use a small spatial pin hole placed at the conjugate image
plane of the object plane to block out-of-focus light and achieve high resolution and contrast
imaging of the single plane (Fig. 1.8). A pair of scan mirrors By capturing multiple 2-D images
at different depths of the sample, a volumetric 3-D image can be reconstructed from those 2-D
slices.
Fig. 0.7 Basic epi-fluorescence microscopes lack optical sectioning capability. As the excitation light penetrate the thick
sample, sample area above and below the focal plane in Z direction gets excited and emits fluorescence as well. Not only the
fluorescence from the point where the excitation light is focused gets collected by the objective lens and detected by the sensor,
the fluorescence from the rest of excited area (detection volume) gets detected as well. As a result, the image captured by the
sensor cannot properly resolve the spatial information of a certain plane in Z direction. Thus, 3-D volumetric imaging is not
achievable using basic epifluorescence microscopes without the optical sectioning capability.
15

Selective plane illumination microscopes (SPIMs) or light sheet fluorescence microscopes
(LSFMs) describe a type of widefield microscopes which have excellent optical sectioning
capabilities. There are several configurations of SPIMs based on the direction of excitation and
detection. The most commonly seen configuration is the orthogonally posed excitation path and
detection path (Fig. 1.9), in which one or two illumination/excitation objective lenses focus a
thin light sheet at a cross section of the sample and a detection objective lens is positioned at 90-
Fig. 0.8 Confocal laser scanning microscopes (CLSM) achieve optical sectioning by selectively detecting fluorescence
from a single plane [115]. Similar with the basic epi-fluorescence microscopes, CLSMs also excite the fluorescence from a
thick sample beyond the focal plane in Z direction. CLSMs are equipped with a small pinhole at a conjugate plan of the focal
plane which physically blocks the fluorescence light from other Z planes (out-of-focus fluorescence). By selectively detecting
the fluorescence coming from the focal plane only, CLSMs are capable of high-specificity optical sectioning. 3-D volumetric
imaging can be achieved by reconstructing from multiple imagines captured at the different depths in Z direction. At each focal
plan. two scan mirrors (only one is shown in the figure) sweep the focused excitation laser beam across the focal plane while the
sensor, normally a photomultiplier tube (PMT), collecting the fluorescence point by point to form a 2-D image.
16

degree angle collects the fluorescence emission. Compared with CLSMs and two-photon
microscopes, SPIMs can achieve faster imaging speed as it does not require point scanning
mechanism for capturing imagery. It is also more sensitive and efficient in terms of photon
throughput without the pinhole rejecting photons to achieve optical sectioning. The added
sensitivity reduces the amount of laser power and acquisition time required to detect the same
fluorescence signal compared with CLSMs, which ultimately reduces the photobleaching. In
conclusion, SPIM is overall a faster and gentler fluorescence microscope configuration with
great optical sectional capability.
1.2.5 Multiplexed/multi-color fluorescence imaging
Multiplexed or multicolor fluorescence imaging is important for studying and gaining better of
complex biological processes by observing multiple fluorescence-labeled entities with different
colors among the same sample. Normally, each fluorescence has a broad emission spectrum and
spectral crosstalk can happen when using multiple fluorescence. The most important
consideration when conducting multiplexed fluorescence imaging is minimizing spectral
Fig. 0.9 Selective plane illumination microscopes (SPIMs) achieve optical sectioning by selectively exciting a thin layer of
the thick sample. SPIMs have a totally different excitation strategy with epi-fluorescence or confocal laser scanning
microscopes which use the same objective lens to deliver and the excitation light and collect fluorescence light. Instead, SPIMs
are equipped with one or two additional objective lenses along with other optics to generate a thin sheet of light which is
normally orthogonally opposed with the detection objective lens. By selectively illuminating and exciting a very small among of
Z depths, the detector can achieve high-specificity optical sectioning.
17

crosstalk and improves signal specificity from each fluorescence. There are several commonly
used multiplexed fluorescence imaging modalities.
Sequential filter switching acquisition
As we already briefly discussed about epi-fluorescence microscope earlier, a filter cube
consisting of an excitation filter, a dichroic mirror, and an emission filter (Fig. 1.6) is essential
for selectively exciting and detecting a certain fluorescence of interest. A sequential imaging is
conducted with each dedicated filter cube during multiplexed imaging.
Sequential filter switching method has the advantage of simple system configuration and can
be used for both widefield and point scanning microscopes. However, this filter switching
approach physically slows the imaging speed and limits its application in speed demanding in-
vivo imaging where biological processes happen in real time. In addition, sample is exposed to
the excitation light source multiple times, which inevitably induces more photodamage and
causes more photobleaching. This approach also requires the experiments to be designed
carefully beforehand in terms of which fluorophores to be used and their corresponding
excitation light sources and filter sets to minimize crosstalk. A narrower bandwidth is normally
preferred to achieve better spectral specificity and reduce crosstalk however this would decrease
the detection photon throughput and reduce the sensitivity.
Hyperspectral acquisition
Hyperspectral acquisition samples fluorescence emission with a fine spectral resolution in a
single exposure, which theoretically increases the imaging speed by overcoming the physical
limitation of sequential filter switching mechanism required for achieving multi-color imaging.
In addition, the added spectral resolution helps fluorescence unmixing results and minimizes
18

signal cross talk combined with computational analysis method such as spectral linear
unmixing [7,8].
Hyperspectral detectors utilize a diffraction grating to spatially separate the emission light into
multiple spectral channels by the angular dispersion. In point scanning configuration such as
CLSMs, an array of photomultipliers (PMTs) are used to collect photons from each channel.
The SPIM counterpart is known as image mapping spectrometry (IMS) [15,16], which disperses
different spectral channels spatially onto a 2-D camera sensor. Hyperspectral acquisition
samples the entire spectrum which is normally the visible range where most of the fluorescence
emits. [7,8] As previously mentioned, hyperspectral acquisition can potentially achieve faster
imaging speed without the need of physical sequential filter switching. However, hyperspectral
imaging has a major limitation, which is the low photo throughput and sensitivity. It achieves
high spectral resolution by dividing and sampling the visible spectrum with many color channels.
Therefore, each color channel has a limited bandwidth and bottlenecks the amount of
fluorescence photons to pass through and get detected by the sensor. In in-vivo fluorescence
imaging applications with photon budget, the reduced sensitivity inevitably requires increasing
excitation power or exposure time to compensate for the low SNR, which ultimately reduces the
imaging speed and increases photobleaching.
Hyperspectral phasor encoding/sampling
Hyperspectral phasor [9–12] was previously introduced as a computational hyperspectral
fluorescence data analysis method. It compresses the high-dimension hyperspectral fluorescence
data with the first-harmonic Fourier coefficients (Fig. 1.10). The previous study demonstrates
that the 2-D hyperspectral phasor contains the spectral information to achieve clean unmixing of
seven fluorescent signals with spatial and spectral overlapping [12]. This reveals hyperspectral
19

fluorescence data’s spectral resolution redundancy in fluorescence imaging applications.
Researchers from Gratton lab at the University of California, Irvine conducted a multi-color
fluorescence imaging experiment with two sinusoidal color filters, one with a harmonic of cosine
transmittance profile the other with a harmonic of sine transmittance profile [30,31], which
essentially acquired hyperspectral phasor optically. Fluorescence multiplexing was achieved
based on the experiment data. More importantly, this acquisition method has much higher
sensitivity compared with traditional hyperspectral acquisition using a 32-channel spectral
detector as the sinusoidal channels has much wider bandwidth, which allows more fluorescence
photons to be detected. This thesis will present a novel multiplexed fluorescence imaging
strategy in light of the previously mentioned hyperspectral phasor acquisition method with over a
10-fold improvement of imaging speed.
1.3 Summary
In Chapter 1, we briefly touched on the concept of magnification in the setting of microscopy,
fluorescence, basics of fluorescence instrumentation and multiplexed fluorescence microscopy.
Fig. 0.10 Hyperspectral phasor forms a 2-D compressed representation of high-dimension hyperspectral fluorescence
data. Each pixel of the hyperspectral data is represented as a high-dimension vector (𝐼 1
(𝜆 ) and 𝐼 2
(𝜆 )) in the spectral space.
Hyperspectral phasor calculates the normalized first harmonic Fourier coefficients of the spectral vector which are numerically
located on a 2-D unit circle called phasor plane. Each point on the phasor plane corresponds to a unique spectral vector in the
original spectral space.
20

The information provided in this chapter shall provide readers outside of the microscopy field
with enough background to dive into some technical details that this thesis is going to unfold in
the next two chapters. If readers have interests in further exploring fluorescence microscopy
from different technical and applicational angles, I recommend the online resources that are
available from Leica, Nikon, Olympus, and Zeiss website [32–35].

21

Chapter 2
SHy-Cam: a snapshot hyperspectral phasor camera for fast, multi-
color fluorescence microscopy
2.1 Introduction
Hyperspectral fluorescence imaging has been gaining popularity in life sciences because of its
ability to improve the selective detection of signals in specimens with significant background,
and to simultaneously detect multiple analytes, which increases multiplexing capability [36–40].
This advancement in fluorescence imaging is enabled by extending the image acquisition
dimension into the spectral domain. Hyperspectral detection is available in commercial confocal
laser-scanning microscopes (CLSM), separating distinct spectral bands to different detectors or
to different elements of detector arrays. However, inefficiency in the light path of CLSM
devices, combined with the low efficiency of the detectors employed, detect a small minority of
the collected fluorescence light. Furthermore, point-scanning imaging tools are not ideal for
multiplexed live imaging of light-sensitive samples because of their relatively slow imaging
speeds, and their use of an epi-illumination path that exposes the entire depth of the specimen to
the exciting light. Thus, the improved label selectivity of hyperspectral CLSMs is achieved at a
significant cost in speed and photo-toxicity.
22

Faster and gentler wide field imaging has been offered by approaches based on the principle of
Selective Plane Illumination Microscopy (SPIM) [41,42]. SPIM typically uncouples excitation
and detection paths by using orthogonally oriented objectives to provide separate fluorescence
excitation and detection pathways. The thin sheet of exciting light preferentially illuminates the
focal plane, minimizing unnecessary excitation of fluorophores above and below, reducing
photo-bleaching and phototoxicity. The widefield detection permits the use of efficient (>90%
quantum efficiency) and fast sCMOS detectors, further improving the photon budget. Higher
imaging efficiency makes it feasible to perform large volumetric imaging at high spatial and
temporal resolution for extended periods of time [43,44]. However, live multiplexing of
fluorescent signals in SPIM has been limited by the complexity of acquiring multi-spectral
datasets (3D: x, y, wavelength) efficiently on a 2-D camera sensor. SPIM systems usually
acquire multiple fluorescent channels sequentially using band-pass filters, which reject a
significant fraction of the fluorescence emission, reducing the photon efficiency and performance
of the system. Furthermore, this approach limits temporal resolution, because of the need for
multiple images and the time required for filter changes. Multiple exposures increase the total
illumination energy load on the sample, increasing photo-toxicity. Approaches employing
spectral separation of the signals [41,42] have done so at a cost to the imaging speed.
Single exposure (snapshot) spectral acquisitions either using multiple cameras [45] or a single
camera coupled with an image splitter [46,47], as well as Image Mapping Spectrometry
(IMS) [38,40] can help decrease the temporal and photo-toxicity costs of multi-label SPIM.
23

IMS and similar widefield snapshot methods separate the emission light into a large number of
color channels by dispersive optics, so that both spectral and spatial information can be re-
Fig. 2.1 SHy-Cam approach overview. (a) Theoretical (solid line) and real (dashed line) transmission (tr) and reflection (re)
profiles of the two custom sinusoidal dichroic mirrors (𝐷𝑀
𝑠𝑖𝑛 : Sine, 𝐷𝑀
𝑐𝑜𝑠 : Cosine dichroic mirrors) in the range (400nm-
700nm). In this schematic representation, a fluorescently labelled zebrafish embryo provides signal input for the optical setup.
(b) Fluorescence emission is divided and encoded into four spectrally correlated channels using DMsin and DMcos, before being
projected onto a single sCMOS sensor. A zoom-in view of the sensor illustrates this optical subdivision of the sensor. (𝐹 : filter
set, 𝑅𝐿
1−2
: relay lenses, 𝐹𝑆 : field stop, 𝐵𝑆 : 50/50 beamsplitter, 𝐹𝑀
1−3
: folding mirrors, 𝐷𝑀
𝑠 : Sine dichroic mirror, 𝐷𝑀
𝑐 : Cosine
dichroic mirror, 𝐺𝑀
1−4
: gimbal mirrors, 𝑇𝐿
1−4
: tube lenses). (c) Preprocessing (image registration and stitching) is performed on
the four encoded channels to perform tiling of a large field of view. (d) Multiplexing can be performed through any of multiple
phasor approaches, separating the collected fluorescence into independent signals (CH1-CH5) and assembling them in a final
volumetric dataset (Merge).
24

assigned to different areas of the camera sensor acquiring a spectral cube (x,y,wavelength). The
low photon efficiency of the dispersive optics (normally below 60%), combined with many
narrow spectral bands [48], result in as little as ~1% of the emitted fluorescence reaching the
detector in each channel. Given the limited signal intensities obtained from in-vivo fluorescence
imaging, achieving sufficient signal-to-noise ratio (SNR) with the low sensitivity of standard
snapshot hyperspectral approaches requires that exposure time and/or laser power be increased.
These factors hamper temporal resolution and cause increases in photo-toxicity. For example,
the fastest spectral fluorescence imaging method, SPIM IMS, is reported to require 250
milliseconds (ms) for snapshot acquisition, achieving 4 spectral datasets per second [40]. Such
exposure times are too long for many dynamic in-vivo imaging applications which benefit from
video of faster framerates. A more efficient spectral sampling and compressing strategy that
offers higher photon efficiency is key to achieving the needed sensitivity and speed.
Spectral phasor (SP) has proven to be an efficient and robust method for compressing and
analyzing hyperspectral fluorescence datasets [9–12]. SP performs a Fourier transform on the
original high-dimension hyperspectral vectors and uses the first harmonic coefficients, the so-
called 2-D phasor coefficients, to offer a compressed spectral representation (Methods). It
simplifies the interpretation and processing of high-dimension hyperspectral data and preserves
most of the spectral information. The Gratton lab has integrated SP into the image acquisition
step (Phasor-based Hyperspectral Snapshot Microscopy) [30,31] by collective three images:
optically sampling fluorescent emission spectra with sequentially collective two images through
distinct sinusoidal optical filters followed by another with no filters for data normalization.
Compared with traditional hyperspectral sampling with many narrow uniform color channels,
each sinusoidal filter allows many more emitted photons to reach the detector (~40% efficient,
25

for commonly used fluorescent labels), which increases SNR while maintaining reduced laser
power during sensitive in-vivo imaging. This filter-based hyperspectral phasor method can
properly and quantitatively separate more than four fluorescence signatures from multi-labeled
samples [30,31].
Any filter changing approach, including the filter-based hyperspectral phasor approach,
requires multiple exposures with intervening filter changes, limiting the temporal resolution ~3
spectral datasets per second [31]. Photo-toxicity is increased due to the additional illumination
required for multiple exposures. In addition, even with increased photon throughput efficiency,
the absorption-based sinusoidal filters reject more than half of the fluorescent photons,
sacrificing spectral information [49]. The method presented here, Singleshot Hyperspectral
Phasor Camera (SHy-Cam) removes the need for multiple exposures, thus harvesting the
advantages of the efficient hyperspectral phasor sampling strategy with higher temporal and
photon efficiency. This improved performance is validated for fluorescent imaging of live tissue
in low-signal and low-SNR conditions.
2.2 Results
2.2.1 Accessing spectral information with faster single snapshot acquisition
Inspired by the capabilities of the Gratton lab’s filter-based hyperspectral phasor
approach [30,31], described in the previous section, we sought to include the spectral
information from the originally discarded reflected fluorescence [49]. In this process we
developed the Single-snapshot Hyperspectral Phasor Camera (SHy-Cam). Shy-Cam is designed
to fit in detection path of wide-field microscopes, and to offer hyperspectral imaging by
projecting spectrally encoded compressed channels onto a single camera sensor with a single -
26

exposure acquisition (Fig 2.1). Hyperspectral phasor compression is achieved by simultaneously
capturing transmitted and reflected fluorescence emission on the camera sensor using two
custom sinusoidal mirrors, folding mirrors, and lenses (Fig 2.1a, b, Methods). SHy-Cam
distinguishes itself from existing work [30,31] in two main aspects: speed (requiring only one
single exposure to capture a spectral dataset) and photon recycling (permitting the majority of the
collected fluorescence signal to be detected).
SHy-Cam easily achieves video-rate detection and faster, by eliminating the need for multiple
exposures and mechanical filter changes. The recycling and collection of reflected photons with
SHy-Cam increase the detected spectral information [49] and provide the flexibility of spectral
analysis. Ratiometric multiplexing such as spectral linear unmixing (LU) [7,8] (Methods) can
be applied to the four-channel data acquired with SHy-Cam, providing an addition to the
standard spectral 2-D phasor plane analysis (Methods) shared with filter-based hyperspectral
phasor approach. Our results show SHy-Cam’s capability of cleanly separate five fluorescence
emissions with spectral LU analysis (Fig 2.3).
2.2.2 SHy-Cam is a cost-effective and versatile add-on device for existing
microscopes
SHy-Cam is designed as an add-on spectral compressing device which can be adapted to most
widefield microscopes equipped with infinity-corrected detection objective lenses. SHy-Cam
captures four spectrally correlated channels simultaneously on the four quadrants of a single
sCMOS camera sensor with a single-snapshot (Fig 2.1, Methods). The single-camera
architecture simplifies the design and reduces the cost compared with a multi-camera setup.
SHy-Cam utilizes readily available optical and opto-mechanical components, combined with a
few custom-made 3D printed parts; this provides the advantages of easy replication and
27

necessary modification for new microscope adaptations. SHy-Cam is cost-effective, as it does
not require multiple specialized detection filters, filter changers or complex dispersing and
spatial mapping optics found in other approaches such as IMS; the cost of the complete list of
parts and 3D printing models is approximately $6,000 (Supplemental Information).
2.2.3 SHy-Cam imaging validation tests
To validate the real-world performance of SHy-Cam, we integrated it into a custom
SPIM [50,51] designed to offer optical sectioning capability and high-speed imaging. Its
multiple single photon laser lines permit simultaneous excitation of multiple fluorophores in the
visible range (400-700nm). Our first implementation with simple 5M-pixel scientific CMOS
(sCMOS) camera (PCO, Gmbh) offered total magnification of 44x, including the magnification
introduced by SHy-Cam itself, and an X-Y field of view (FoV) in object space of 120um
diagonally (Fig. 2.6b-e, Methods), requiring tiled acquisition for imaging larger specimens. The
total magnification can be changed by using different objective lenses, relay lenses or tube lenses
(Supplemental Information) depending on the application.
As a first test of SHy-Cam, we imaged a set of different mixtures of two commonly used
fluorescent dyes, Rhodamine B and Rhodamine 6G (Sigma-Aldrich Corp.) (Fig. 2.2a); The
emission spectrum of each mixture is mapped to a cluster on the 2-D phasor plane showing
reliable and distinct cluster locations for mixtures with different relative mixing ratios (Fig. 2.2b,
c). SHy-CAM could distinguish small spectral changes of only 1.2% in relative concentrations,
easily separating mixtures of ratio 1:12 from those of 1:14, for example (Fig. 2.2c, d). The
mixtures are visualized using SEER encoded colormap [52] with different colors representing
the emission spectra of different relative concentrations (Fig. 2.2d).
28

We tested fluorescent signal multiplexing using geometric segmentation on 2-D phasor plane
Fig. 2.2 SHy-Cam spectral unmixing on standard and fixed samples. (a) Normalized emission spectra of Rhodamine
6G and Rhodamine B in water. (b) Phasor plot demonstrating distinct clusters of pure Rhodamine 6G (R6G) and
Rhodamine B (RB) and of five solution mixtures with different relative concentrations (Rhodamine 6G:Rhodamine B).
(c) Zoom-in view of phasor plot clusters in (b) visualized using the SEER phasor map encoding (Red circle indicates the
SEER map center for the morphed angle rendering of the spectral information). (d) SEER pseudo-color bars representing
each phasor-cluster, with a continuous color map (panel-bottom) showing the distinction of relative concentrations of the
solutions. The emission profiles (400nm-700nm) and peak emission wavelengths of each mixture are shown to the right
of the corresponding pseudo-color bars; as Rhodamine B relative concentration increases, the peak and shape of emission
profile resembles pure Rhodmine B. (e) Normalized emission intensities of six 1𝜇𝑚 fluorescent beads acquired with
SHy-Cam. (f) Collected phasor plots with circular selections used for unmixing. (g) Unmixed and color coded 6-channel
volume of fluorescent beads. (h-j) Images of a triple-labeled zebrafish embryo collected with conventional and Shy-Cam
microscopy. Green - membrane Tg(krt4:GFP), yellow - neutrophil Tg(lyz:TagRFP), purple - ubiquitous cell nucleus
ubi:H2B-iRFP670. (h) Image acquired with Zeiss LSM 880 with sequential laser excitation for each fluorescent label,
rendered as a maximum intensity projection (MIP) of a coronal plane of a 4-dpf fixed zebrafish embryo (inset, k). (i)
same region acquired with Zeiss LSM 880 using three-laser simutaneous excitation in 32-channel hyperspectral mode,
rendered using Linear Unmixing (LU) for fluorescence separaton. (j) Spectral LU result from SHy-Cam image acquired
from the same embryo, reasonably close to the confocal microscope images in h, i.
29

on SHy-Cam. Immobilized fluorescent beads (Bangs Laboratories, Inc., Sigma-Aldrich Corp.)
with six distinct but highly overlapping spectra (Fig. 2.2e) were imaged with SHy-Cam (Fig.
2.2g). Pixels from the four channels were transformed into the phasor plane (Methods), creating
a look-up table linking pixels and corresponding phasor coefficients which were represented as a
2D-histogram, the 2-D phasor plane. Beads with the similar spectral emission create clusters on
the 2-D phasor plane (Fig. 2.2f). These clusters were selected with six regions of interests
(ROIs) on the phasor plane (Fig. 2.2f). The corresponding pixels were respectively pseudo-
colored permitting the unambiguous visualization of beads within the imaging volume (Fig.
2.2g).
Multiplexing using 2-D phasor plane segmentation is an intuitive and fast analysis method for
spectrally overlapping but spatially sparse fluorescence. However, clean and unambiguous
signal separation can be challenging when fluorescent signals are spectrally and spatially
overlapping. SHy-Cam added two reflected channels allow ratiometric analysis using spectral
LU [7,8]. We tested SHy-Cam in multiplexing spatially and spectrally overlapping fluorescence
of a fixed 3-color zebrafish embryo (Methods) and compared the imaging results with data
acquired on a reference system, a laser scanning confocal microscope (LSM-880, Carl Zeiss,
AG) equipped with a 32-channel spectral detector (Methods). A similar coronal plane of the
fixed sample was acquired on both LSM and Shy-CAM instruments (Fig. 2.2h-j). We acquired a
“ground truth” image on the confocal by sequentially exciting the fluorescent proteins in the
sample, capturing three separate fluorescence emission datasets (Fig. 2.2h). We then acquired a
confocal 32-channel hyperspectral dataset by simultaneously exciting with the three lasers,
followed by spectral LU to separate fluorescence signals (Fig. 2.2i). Following this, we re-
mounted the same fixed embryo for imaging on the SHy-Cam-SPIM using simultaneous
30

excitation of three laser lines, similar to the LSM images, and used the same spectral LU analysis
(Fig. 2.2j, Methods). The SHy-Cam unmixed results resemble very closely the “ground truth”
but were collected 4-fold more quickly than the confocal counterpart. Deeper statistical analysis
in this case is hampered by the fact that the sample needed to be re-mounted in a separate
instrument.
SHy-Cam worked well for in-vivo multiplexed volumetric imaging of transgenic zebrafish
embryos with up to 5 colors. Fig. 2.3 presents results obtained from a 4-day post fertilization
(dpf) transgenic zebrafish embryo, separating the signals arising from 4 spatially and spectrally
overlapping fluorescent proteins from one another and from the tissue-generated
autofluorescence using spectral LU. The number of signals being unmixed is larger than the
number of imaging channels on the sCMOS camera, making the linear unmixing appear to be
under-determined system; however, the unmixing results reveal clean separation, even when the
results are presented at high contrast (Fig. 2.3b-f). This result shows SHy-Cam’s ability to
capture sufficient fluorescence spectral information in challenging imaging conditions. The
acquisition was conducted by simultaneous excitation of by three laser lines, with their powers
adjusted to yield roughly similar emission intensity levels across the five labels in single
exposures of 200 msec. Even including the inevitable time delays required for moving the
specimen for tiling and for Z-stacking, the volumetric image of the embryo trunk area (Fig. 2.3a)
was acquired in under 4 minutes.
31

SHy-Cam’s rapid in-vivo imaging capability is well suited to two distinct imaging challenges in
live zebrafish embryos: cell migration and the beating heart. Cell migration was imaged at a
high spatio-temporal resolution over a large field of view to follow the migration of neutrophils
toward a zebrafish tail-clipping wound (Fig. 2.4a-d). For each time point, a mosaic of volumes
was collected with the co-excitation of 3 laser lines; the tiles were stitched together and analyzed
with spectral LU. The imaging rate of 2.7 volumetric tiles per minute offered sufficient temporal
resolution to unambiguously track neutrophil migration. SHy-Cam’s high detection efficiency
for fluorescent signals, combined with the advantages of light-sheet microscopy, permitted
Fig. 2.3 SPIM-SHy-Cam tiled volumetric in-vivo imaging. (a) Maximum intensity projection image (MIP)
acquired from the trunk area of a 4dpf of a transgenic zebrafish embryo; five labels were unmixed by linear unmixing
from Shy-Cam image data (autofluorescence and four fluorescent proteins). The volumetric image consists of 21 tiles,
each with 62 optical sections, covering a field of view of 602 x 256 x 120𝜇𝑚 . (b-f) The individual unmixed
fluorescence signals displayed for the sub-region (dashed box) in (a): (b) cyan - autofluorescence, (c) green -
membrane Tg(krt4:GFP), (d) magenta - vasculature Tg(kdrl:mCherry), (e) yellow - neutrophil Tg(lyz:TagRFP), (f)
purple - ubiquitous cell nucleus ubi:H2B-iRFP670.
32

continuous and prolonged time-lapse imaging with only 11% photo-bleaching (Fig. 2.4j). This
permitted high resolution imaging with no photo-toxicity to the embryo.
Imaging the beating heart is even more demanding. Shy-Cam was used to image the heart
continuously at 20 frames per second for 10 seconds (Fig. 2.4f-i). This high-speed acquisition
was sufficient to track of fast-moving neutrophils (up to 200um/sec) as they flowed across the
atrial cavity. The excitation laser power was increased to minimize the camera exposure,
without inducing visible photo-damage in the sample. The rapid and continuous imagining
resulted in only 4% photo-bleaching in the 10-second acquisition (Fig. 2.4j). Because each
frame captured all the spectral information there were no movement artifacts, providing striking
multiplexing of fluorescent labels.
We performed high-speed imaging on a zebrafish embryo expressing a different combination
of transgenic fluorescent proteins at 33 frames/sec. The single-plane SPIM image captured a
cross-sectional recording of the movement of the nuclei in the cells of the atrial wall, while the
heart was beating at around 180 beats/min. As with the above two examples, this performance
highlights SHy-Cam’s ability to perform challenging high-speed multiplexed fluorescence
imaging.
33

2.3 Discussion
Our results demonstrate that the Single-shot Hyperspectral Phasor Camera (SHy-Cam) offers
efficient and rapid multiplexed imaging of fluorescent signals. The speed and photon efficiency
Fig. 2.4 SPIM-SHy-Cam dynamic in-vivo imaging. (a-d) Zebrafish tail-clip wound healing (dashed line) acquired as a 2-
hour tiled volumetric timelapse. The transgenic embryos are labeled in their cell membranes (green; Tg(krt4:GFP)), nuclei
(purple; ubi:H2B-iRFP670), and neutrophils (yellow; Tg(lyz:TagRFP)). Neutrophil migration toward the wound can be
observed at high-resolution in 3D in the context of tissue and tracked over time. (i) Zebrafish embryo representing the imaging
positions of (a-d) (square) and (e-h) (circle). The volumetric images are each composed of 6 tiles, each of 21 optical sections
(Video S2), covering a field of view of 300 x 250 x 42𝜇𝑚 . The frames displayed here are taken from a timelapse recording of
60 timepoints acquired quasi-continuously. (e-h) The same embryo’s beating heart (dashed circles) is captured continuously at
20 frames/second for 10 seconds covering an area of 17.1 ∙ 10
3
μm
2
. Neutrophils flowing through the cardiac area can be
tracked (jet dragon-tail). (j, k) Photo-bleaching is estimated as relative intensity of the emission signal from all channels during
the 2-hour tail-clip wound healing timelapse imaging and during the 10-s heart beating timelapse.
34

are significant advantages, especially in applications involving in-vivo time-lapse and volumetric
imaging where minimize acquisition time and photo-bleaching in the sample. The hyperspectral
phasor encoding captures the differences between even similar labels (Fig. 2.2) but removes the
redundancy of traditional uniform hyperspectral sampling strategies, improving not only photon
efficiency, but also processing and data storage needs. SHy-Cam’s single-snapshot acquisition
achieves more than 10-fold faster imaging speed compared with existing filter-based
hyperspectral phasor approach [30,31]. The added spectral information by recycling the
reflected photons [49] enables ratiometric fluorescence multiplexing using spectral LU [7,8].
We designed SHy-Cam focusing on versatility, cost-effectiveness and accessibility. The
versatile modular design was easily adapted to commercial microscopes and to a custom-built
SPIM, showing its potential for application to any camera-based systems. While the SHy-Cam
implementation in this work utilizes a single-camera architecture to remove the necessity of
utilizing and synchronizing multiple expensive sCMOS cameras, it could easily be adapted to
implementations using two or four cameras. In addition, the design can be adapted to enable
multifocal, line-scan or point-scan confocal laser scanning microscopes, using sets or arrays of
silicon-based detectors or photomultiplier tubes (PMTs).
The SHy-Cam data rendering on the 2-D phasor plot offers intuitive and fast display and
interactive analysis of multiplexing data [12,52]. The same data can be analyzed with spectral
linear unmixing, enabling ratiometric fluorescence multiplexing even in the presence of
significant spatial and spectral overlap between multiple fluorophores, or between the labels and
high background fluorescence. SHy-Cam is poised to be a solution for high-speed and -
throughput imaging and multiplexing applications, where improved detection at low-signal and
at low-SNR would considerably shorten experimental pipelines.
35

Besides the in-vivo fluorescence imaging applications, other applications such as multi-color
high-content cellular analyses are poised to benefit from SHy-Cam’s efficiency, streamlined
spectral analysis and visualization using the hyperspectral phasor. Recent advancements in
imaging flow cytometry (IFC) [53–60] have focused on overcoming the common trade-off
between throughput, sensitivity and spatial resolution of traditional IFC [61]. The same
advantages of SHy-Cam demonstrated here for microscopy and SPIM, position it as a powerful
means to improve other quantitative detection and multiplexing applications such as IFC.
A potential limitation of the single-camera SHy-Cam implementation here is the reduced FoV,
resulting from the sub-division of the sensor into four quadrants. Here, we compensated for this
by tiling our acquisition to achieve the same FoV as a non-subdivided sensor. Pilot
implementations have shown that we can achieve the same advantages as the single Shy-Cam
design presented here, but over large FoVs by employing multiple sCMOS cameras or the
growing set of large sCMOS sensors with many more pixels. Of course, for applications where
spatial resolution is not demanding, a lower magnification relay lenses can offer a full the FoV,
with improved SNR, and excellent spectral sensitivity, but at reduced spatial resolution.
The 50:50 beam splitter (Fig. 2.1b), equally dividing the emission signal into two light paths for
simultaneous spectral encoding, might reduce the lowest light level performance of Shy-Cam,
given the electronic noise of each pixel of the camera. For each channel the numbers of photons
reaching the detector are halved. The contribution of this reduction may be less than it first
appears, as no detection filter passes 100% of the light in its band-pass, and by its very nature, a
band-pass filter rejects many photons that Shy-Cam would collect. Shy-Cam’s re-distribution of
light on the sensor, rather than rejection of unwanted spectral ranges, provides a higher
efficiency in total number of photons detected, allowing higher spectral information collection
36

and a faster single snapshot acquisition. With ongoing development of detectors with lower
noise and with larger number of pixels (permitting pixel binning if needed), the low-light
performance of Shy-Cam should meet or exceed that of other approaches.
SHy-Cam is poised as a solution for high-speed and -throughput fluorescent imaging
applications, where improved detection at low-SNR and low signal conditions, combined with
multiplexing, should considerably speed analyses and shorten experimental pipelines.
2.4 Methods
2.4.1 Spectral compression with hyperspectral phasor
Spectral phasor is an established Fourier transform-based analysis method of hyperspectral data
that compresses high dimensional spectral vectors into their normalized first harmonic Fourier
coefficients [9–12] represented by equation (2.1, 2.2). The linear additive property of
fluorescence emission and their wide spectral profile of common fluorescence make it possible
to recover substantial amount of spectral information with the ‘base frequency’ and DC
components of hyperspectral data (Supplemental Information).
𝐺 =
∑ 𝐼 (𝜆 𝑛 )∙cos(
2𝜋 𝑁 𝜆 )∗∆𝜆 𝑁 −1
𝑛 =0
∑ 𝐼 (𝜆 𝑛 )∗∆𝜆 𝑁 −1
𝑛 =0
(2.1)
𝑆 =
∑ 𝐼 (𝜆 𝑛 )∙cos (
2𝜋 𝑁 𝜆 )∗∆𝜆 𝑁 −1
𝑛 =0
∑ 𝐼 (𝜆 𝑛 )∗∆𝜆 𝑁 −1
𝑛 =0
(2.2)
G and S are the real and imaginary coefficients, called phasor coefficients. 𝜆 𝑛 represents the
wavelength of the n-th spectral channel out of a N-channel spectral vector. 𝐼 (𝜆 𝑛 ) denotes the
intensity value of the n-th channel. ∆𝜆 is the wavelength bandwidth of a single channel. The
denominators in equation (2.1) represent the integral of the intensity values across all N channels
and are the normalization factors that eliminate the influence of different intensity levels, which
37

also incorporates the DC component of the spectral vector. Phasor coefficients are visually
presented as a 2-D histogram called phasor plane (Fig. 2b, f). Each (G,S) pair corresponds to a
unique spectral vector in the original hyperspectral image plane.
SHy-Cam performs hyperspectral phasor compression optically during acquisition and directly
get the unnormalized G,S phasor coefficients via transmitting the emission through two custom
sinusoidal dichroic mirrors (Fig. 2.1a, b), similar to filter-based hyperspectral phasor
approach [30,31]. ‘SIN’ and ‘COS’ channels shown in (Fig. 2.1b) are the transmitted channels
that can be used to estimate phasor coefficients with some simple calculation. ‘ASIN’ and
‘ACOS’ channels are the reflected channels. Phasor coefficients calculation from SHy-Cam is
represented with equation (2.3, 2.4). As transmitted and reflected emission are collected at the
same time, the normalization factor, i.e., the intensity value, is readily available by summing
‘SIN’ with ‘ASIN’ or ‘COS’ with ‘ACOS’ (5).
𝐺 =
𝐶 𝑖𝑑𝑒𝑎𝑙 𝐼 (2.3)
𝑆 =
𝑆 𝑖𝑑𝑒𝑎𝑙 𝐼 (2.4)
𝐼 = 𝐶𝑂𝑆 + 𝐴𝐶𝑂𝑆 𝑜𝑟 𝑆𝐼𝑁 + 𝐴𝑆𝐼𝑁 (2.5)
𝐶 𝑖𝑑𝑒𝑎𝑙 =
𝐶𝑂𝑆 −𝑐 𝐶 𝑎 𝐶 (2.6)
𝑆 𝑖𝑑𝑒𝑎𝑙 =
𝑆𝐼𝑁 −𝑐 𝑆 𝑎 𝑆 (2.7)
Since transmittance is physically non-negative, an affine transformation needs to be applied to
‘SIN’ and ‘COS’ channel (6, 7) in order to estimate the Cosine and Sine transformation. 𝑐 𝐶 and
𝑐 𝑆 are respectively the center value of cosine and sine transmittance profiles (Fig. 2.1a). 𝑎 𝐶 and
𝑎 𝑆 are respectively the amplitude of cosine and sine transmittance profiles (Fig. 2.1a). With our
specific set of dichroic mirrors, these parameters are 𝑐 𝐶 =0.52, 𝑐 𝑆 =0.51, 𝑎 𝐶 =0.44 and 𝑎 𝑆 = 0.40
(Supplemental Information).
38

2.4.2 Sinusoidal dichroic mirror
The two sinusoidal dichroic mirrors are the core components of SHy-Cam. The dichroic mirrors
are designed to work in the spectral range of 400nm to 700nm (Fig. 2.1a) as the most commonly
used fluorescent proteins and fluorophores emit in the visible spectrum. They have 45-degree
angle of incidence and random polarization (Chroma, Inc). The spectral transmittance for the
dichroic mirrors closely resembles the shape of a sine and a cosine function, with a maximum
transmittance peaking at 95.8% for sine and at 91.1% for cosine, as per manufacturer provided
specifications.
2.4.3 Optical design and setup
SHy-Cam presented in this work is designed as a spectral compression add-on device (Fig. 2.1b,
Fig. 2.5) to be placed in the infinity space of a custom-built SPIM [50,51]. The SPIM is
equipped with multiple single-photon laser lines (404nm, 450nm, 488nm, 561nm and 640nm)
which are combined and fill the back focal plane of two oppositely positioned excitation
objective lenses (LMPL10XIR, numerical aperture (NA) = 0.25, Olympus). A galvanometer
scanner (6215HB, Cambridge Technology) positioned upstream of the excitation objective
lenses generates fast scanning spherically focused laser beam along the y direction at the sample,
creating a scanned light sheet in the x-y plane (Fig. 2.5b, c) with 2𝜇𝑚 thickness axially. The
scanned light sheet was tuned so that it only illuminates the sample covered by the FoV of SHy-
Cam in y direction (Fig. 2.5c) to minimize laser-induced photo-toxicity. The fluorescence
emission was imaged through a 25x water-immersion detection objective lens (CFI75
Apochromat 25XC W, Nikon) which was positioned upstream of the input port of SHy-Cam.
SHy-Cam itself can be broken down in two three main sections (Fig. 2.1b, Fig. 2.5a):
39

Optical filter set: light at wavelengths outside of 400-700nm were rejected by a short-pass
filter (FF01-715/SP25, Semrock, Inc) and a long-pass filter (FF01-380/SP25, Semrock, Inc).
In addition, we employed a set of notch filters for removing the exciting laser light, centered at
404nm, 488nm, 561nm and 640nm (NF01-488/647-25x5.0; NF03-561E-25, Semrock, Inc.).
Relay lens group: two 2-inch diameter relay lenses 𝑅𝐿
1
, 𝑅𝐿
2
(32-886, 49-291, Thorlabs, Inc.)
and a field stop 𝐹𝑆 (SM1D12C, Thorlabs Inc.). The first relay lens 𝑅𝐿
1
creates an intermediate
image plane at the field stop. The field stop, a ring-actuated iris diaphragm is adjusted to a size
that the final images formed on the sensor do not overlap with each other. The second relay
lens re-collimates the light to infinity space.
Image splitter: after the relay lens group, a 50/50 beam splitter 𝐵𝑆 (BSW10R, Thorlabs, Inc)
equally divides the fluorescence emission light into two paths. Sine dichroic mirror 𝐷𝑀
𝑠 and
cosine dichroic mirror 𝐷𝑀
𝑐 (Chroma, Inc.) along with three mirrors 𝑅𝑀
1−3
(CCM1-E02,
Thorlabs Inc.) create four spectrally compressed and correlated light paths. Four tube lenses
𝑇𝐿
1−4
(32-884, Edmund Optics) with focal length f =175mm focus the light and form the four-
channel final image on a 5MP sCMOS camera sensor (Edge 5.5, PCO, Gmbh). Four mirrors
𝐺𝑀 1−4
(BB1-E02, Thorlabs, Inc.) mounted on gimbal mounts (KC45D, Thorlabs, Inc.) are
used in front of the tube lenses to adjust the incidence angles so the images can be formed on
the correct locations of the sensor.
40

All the optical components used in this work can be found from the part list on GitHub
repository (Code Availability).
2.4.4 Opto-mechanics
SHy-Cam is built based on both off-the-shelf components (Thorlabs, Inc.) and custom-made 3D
printed components with black polylactic acid (PLA) thermal plastic material extruded from an
Fig. 2.5 SHy-Cam 3D model and optical diagram of one channel. (a) The optical train of one channel. o represents a point on
the object plane (on- or off-axis) and i represents its corresponding point on the image plane. 𝐿 𝑜𝑏𝑗 is the microscope detection
objective lens with effective focal length 𝑓 1
. 𝐹 represents a pair of long-pass (FF01-380/SP25, Semrock, Inc) and short-pass
(FF01-715/SP25, Semrock, Inc) filters which reject photons outside of 400-700nm working spectral range. 𝑅𝐿
1
and 𝑅𝐿
2
are a pair
of relay lenses with focal lengths of 𝑓 2
and 𝑓 3
. 𝐵𝐹𝐿 1
and 𝐵𝐹𝐿 2
are the back focal lengths of 𝑅𝐿
1
and 𝑅𝐿
2
which decide the
spacing 𝐷 between the last surface of 𝑅𝐿
1
and the first surface of 𝑅𝐿
2
. 𝐹𝑆 is a field stop placed at the intermediate image plane (at
a distance 𝐵𝐹𝐿 1
after the last surface of 𝑅𝐿
1
) of the relay lens pair. 𝐼𝑆 represents the image splitter portion of SHy-Cam (related to
Figure 2.1 and Methods). 𝑇𝐿 represents a tube lens which converges and focuses the light with an effective focal length of 𝑓 4
.
The SHy-Cam presented in this work uses: 𝑓 1
= 8𝑚𝑚 (CFI75 Apochromat 25XC W, Nikon), 𝑓 2
= 150𝑚𝑚 (AC508-150-A,
Thorlabs, Inc.), 𝑓 3
= 150 𝑚𝑚 (AC508-075-A, Thorlabs, Inc.), 𝑓 4
= 175𝑚𝑚 (32-884, Edmund Optics). The total magnification 𝐷
of SHy-Cam can be estimated via the formula 𝑀 =
𝑓 2
∙𝑓 4
𝑓 1
∙𝑓 3
≈ 44𝑥 . (b) Schematic of SPIM illumination and detetcion in relation to
SHy-Cam. (c) Comparison of light sheet illumination area with the actual field of view of SHy-Cam image. The galvanometer
scanner is tuned so that the generated scanning light sheet has the matching size in y direction with the SHy-Cam field of view to
reduce unnecessary laser dose to the samples that is not being imaged.
41

Ender 5 (Creality 3D Technology Co., Ltd) fused deposition modeling 3D printer. The printing
nozzle temperature was set to 200°C while the print bed was set to 60°C for improved adhesion.
The layer height was set to 0.2𝜇𝑚 . The 3D printer stepper motors were calibrated on all three
axes with rigorous print bed leveling to ensure the dimensional accuracy. All the opto-
mechanical components used in this work can be found from the part list on GitHub repository
(Code Availability).
2.4.5 Image acquisition
SHy-Cam-SPIM image acquisition
In this work, lasers, sample stages and camera were all controlled by an integrated microscope
controlling software called micro-manager [62,63]. The acquisition with SHy-Cam consists of a
hybrid registration image for registering four channels, a series of dark frames for removing the
background signals and the final fluorescence images of interest. A hybrid registration image is
acquired with both bright-field transmitted illumination and fluorescence excitation with the
samples to be examined in place. Bright-field illumination illuminate the whole FoV and the
boundaries confined by the field stop (Fig. 2.5) can be observed clearly. Fluorescence excitation
provides fluorescent contrast and features within the FoV which will help with the registration
step. The dark frames are acquired using the same settings that are used for final fluorescence
images, i.e., the same laser power, same exposure time, same sensor gain and pixel binning.
They are acquired continuously for at least 20 frames. During image pre-processing stage, the
averaged dark frames is subtracted from the final fluorescence images. There is no special
procedure for the acquisition of final fluorescence images with SHy-Cam equipped with single
sCMOS camera (Edge 5.5, PCO, Gmbh). When image tiling is used for covering a larger FoV,
the fixed reference channel used for image registration is used to navigate through the sample in
42

‘Multi-Dimensional Acquisition’ function in Micro-manager with the sample stage controller.
During our test, a proper tile overlapping of 25%-30% was used to achieve good image stitching
results. We also use a 2x2 pixel binning during acquisition to enhance SNR and dynamic range.
Confocal Image acquisition
We did an imaging comparison test of SHy-Cam-SPIM with a laser scanning confocal
microscope (LSM 880, Zeiss) on the same three-color transgenic zebrafish embryo (Fig. 2.2i-k).
We chose the closest combination of laser excitation available on the confocal (488nm, 561nm,
633nm) to match the laser combination used on SHy-Cam-SPIM (488nm, 561nm, 640nm). The
ground truth image (Fig. 2.2i) was acquired sequentially with one laser excitation at a time using
the 32-channel spectral PMT (410nm-696nm with 8.9nm resolution) with 16-bit bit depth. For
each ground truth channel, the intensity value of each pixel was calculated by summing the 32
channels and normalizing the summed value to 16-bit. The hyperspectral image (Fig. 2.2j) was
acquired with simultaneous excitation of three laser lines using the 32-channel spectral PMT
under the same settings. Both images (Fig. 2.2i, j) were acquired using the same 20x objective
lens (Plan-Apochromat 20x/0.8 M27, Zeiss). The pinhole was set to 34𝜇𝑚 in order to match the
axial resolution of the SPIM.
2.4.6 Image pre-processing
Image registration
SHy-Cam data analysis relies on pixel-wise calculation. Proper registration and alignment of
four channels is important and here performed in a three-step process. First, areas on the sensor
that respectively correspond to the four channels (Fig. 2.6a), are defined via regions of interest
(ROIs) (Code Availability) onto a hybrid registration image acquired with SHy-Cam and serve
as reference for later registration. Second, a set of control points from the four channels is
43

located and used for registration, choosing one channel as the fixed reference, three others as the
moving channels to be registered. In this work we use ASIN channel as the reference channel.
We performed the registration by using a deformable warping-based image registration method
and an ImageJ plug-in software called ‘Big Warp’ (Fig. 2.6b-e) [64]. We used the apices of the
hexagonal FoV boundaries defined by the 10-blade field stop 𝐹𝑆 (SM1D12C, Thorlabs Inc.)
along with 2-3 distinct fluorescence features at the center of the FoV as control points. The
registration result can be estimated using the same plug-in software with image overlay
visualization (Fig. 2.6b-e). Finally, we utilize a custom MATLAB script (Code Availability)
based on local weighted mean registration algorithm [65] to complete the image registration.
The script reads the coordinates of rectangular ROI generated in the first step and the control
points exported in second step along with the dark frames and fluorescence images of interest
during acquisition. The registered images are saved as OME.TIFF format [66] with the
formatting order of (x,y,z,c,t). (x,y,z) is the spatial index. c is the channel index (0: ASIN, 1:
ACOS, 2: SIN, 3: COS). t indicates the frame number. The exported images with OME.TIFF
format in the described formatting order can be directly imported into analysis software and
script (Code Availability). The first and second step are only required when there is noticeable
spatial drifting of any channel. SHy-Cam in this work was built on top of a stabilized optical
table (T48H, Thorlabs, Inc.) inside of a temperature-controlled room at 20℃. We found the FoV
was able to stay within the defined rectangular regions for more than 1 month without
adjustment. The control points re-locating on the other hand should be done daily to make sure
the registration performance.

44

Image stitching
SHy-Cam images are first registered using a deformable warping-based image registration
method [64]. The registration process utilized one of the four channels as a fixed reference for
aligning the other three. For each system physical alignment change, an initial registering
Fig. 2.6 Image pre-processing pipeline. (a) Manual cropping of the four channels from a brighfield image is performed in
MATLAB (see Code Availability). Image shows the camera raw data with four channels distributed in the four quadrants of the
sensor. (b) Screenshot of ‘SIN’ channel here used as the reference channel. Green dots with numbers are the manually selected
control points from the reference channel. Their corresponding control points from a registering moving channel (ASIN) shown
in (c) are also manually selected. 8 control points pair on the hexagonal FOV edge and 1 control point pair in the center area of
the FOV are selected. (d) Overlay image of ‘SIN’ and ‘ASIN’ channels before registration with visible misalignment. (e)
Overlay image of ‘SIN’ and ‘ASIN’ after registration shows well aligned channels. (f) Screenshot of nine image tiles in Imaris
Sticher 9.6 (Bitplane, Switzerland) before stitching. (g) Screenshot of stitched image after rotation and crop correction.

45

process was conducted using the Big Warp plug-in function in Fiji [67] (Fig. 2.6a) to establish
the geometrical relations between the reference and registering channels. As long as there is no
physical change or misalignment, the registration is performed automatically through a
MATLAB script (Code Availability) which uses the previously located control points for
registering future data. The registered images are saved as a 4-channel OME-TIFF phasor cube,
with layers corresponding to SIN, COS, A-SIN and A-COS (Fig. 2.1b). In the case of tiled
acquisitions (Fig. 2.6f, g), image stitching was performed using Imaris Stitcher 9.6 (Bitplane,
Switzerland). Final results were stored in IMS Imaris file format. For analysis software and
script to load the stitched images, a format conversion from IMS to OME.TIFF was done using
Imaris 9.7 (Bitplane, Switzerland).
2.4.7 Image analysis
Spectral phasor analysis in HySP
We integrated hyperspectral phasor calculation function into an integrated hyperspectral phasor
analysis software previously published HySP (Code Availability) [12]. The software reads pre-
processed OME.TIF images from the disc and transform each pixel from the original data into
phasor coefficients following equation (2.3) – (2.7). The transformed phasor coefficients are
then organized into a 2-D histogram for visualization the distribution on the phasor plane (Fig.
2.7). Each bin on the phasor plane, the 2-D histogram, corresponds to the pixels from the image
with the same spectrum. The separation of multiple fluorescence emissions can be achieved by
geometrically applying region of interest masks on the corresponding bins on the phasor plane
(Fig. 2.2f, 2.7a). This 2-D geometric analysis is a fast and intuitive way for the multiplexing
purpose for spatially sparse fluorescence and for verifying the existence of certain fluorescence.
46

In addition, SEER [52], a hyperspectral phasor-based spectral visualization method was
integrated into also HySP. It overlays a specific colormap onto the phasor plane and
automatically visualizes the spectral information by mapping the pseudo color from the colormap
back to image plane (Fig. 2.2b, d).
Ratiometric spectral analysis using spectral linear unmixing
Spectral LU [7,8] can be applied directly to the pre-processed four-channel SHy-Cam image for
a quantitative and ratiometric spectral analysis, without hyperspectral phasors conversion. LU is
treated as a linear least-squares (LLS) problem which can be represented by equation (2.8),
min
𝒙 𝑛 1
2
‖𝑹 4×𝑛 ∙ 𝒄 𝑛 − 𝒔 4
‖
2
2
(2.8)
𝑹 4×𝑛 = [𝒓 1
, 𝒓 2
… 𝒓 𝑛 ] (2.9)
Fig. 2.7 Spectral analysis on spectral phasor plane (a) The phasor plot of a pre-processed SHy-Cam image of a three-color
4dpf transgenic zebrafish embryo Tg(krt4:lyn-EGFP;kdrl:mCherry;lyz:TagRFP) in HySP (see Code Availability). Three
polygonal selections are applied to unmix three fluorescent signatures (Green: EGFP labeled superficial epithelial cells; Yellow:
TagRFP labeled neutrophils; Magenta: mCherry labeled vasculatures). (b) Pseudo-color image in HySP representing, with
saturated colors, the selections on phasor plane, color-matching the ROIs in A. The image is a lateral view of the trunk area of a
zebrafish embryo, the slice shown is the 40th out of an 81-slice Z stack. (c) Maximum intensity projection image of the unmixed
volume.
47

where 𝑛 is the number of fluorescence signals. 𝑹 4×n
is a 4-by-n matrix of reference spectra.
Each column, 𝒓 𝑘 , of the matrix is a normalized four-channel reference spectra of a pure
fluorescence. A custom MATLAB script (Code Availability) loads a pre-processed image and
rectangular masks are applied manually and sequentially to the area where contains only a
specific pure fluorescence. The pixels inside of each mask are averaged to a four-channel
intensity vector 𝑰 𝑘 which is further normalized by its total intensity ‖𝑰 𝑘 ‖
1
and get the reference
spectra vector 𝒓 𝑘 . The reference spectra matrix 𝑹 4×n
is save as a .mat file to be loaded when
conducting spectral LU. 𝒙 𝑛 represents the optimal solution of the contribution vector of n
different fluorescence. 𝒔 4
denotes the normalized four-channel spectral vector from the pre-
processed analysis image pixel. 𝑰 n
is an n-Dimensional identity matrix. 𝑱 n
is an n-Dimensional
all-ones matrix. 𝟎 𝑛 and 𝟏 𝑛 are n-Dimensional all-zeros and all-ones vectors. Two constraints are
used to better define the problem. The first constraint ensures that the summation of all
contributions equals one. The second confines the range of contributions to be zero to one.
2.4.8 Photon-efficiency and -throughput estimation
Photon-efficiency and throughput of five commonly used fluorescence proteins was estimated by
summing the light coming out of the four channels at the last surfaces of the tube lenses. The
loss along the optical train of SHy-Cam was calculated by considering the photon loss at each
optical surface between relay lens 𝑅𝐿
1
and the four tube lenses 𝑇𝐿
1−4
into consideration (Fig.
2.1b). All lenses used in the SHy-Cam prototype were 𝑀𝑔𝐹 2
coated achromatic lenses
(Edmund Optics). The manufacturer-claimed light loss due to reflection is under 1.75%. We
used 1.75% as the number to estimate the light loss at the lens surfaces. Spectral specifications
of other off-the shelf optical components mentioned above can be found on the respective
products webpages. The transmittance and reflectance profiles of two custom sinusoidal
48

dichroic mirrors were provided by manufacturer. The throughput estimation of the filter-based
hyperspectral phasor approach [68] was estimated by the average signal of the sine and cosine
channel at the last surface of the tube lens.
Table 1 SHy-Cam photon efficiency of 5 commonly used fluorescence. The photon efficiency estimates the percentage of the
photons that are able to enter and exit SHy-Cam to the detector over the total emitted photons exciting the back aperture of the
detection objective lens. The estimation is based on the light with 0° angle of incidence (AOI) for simplicity reasons. The
photon loss along the optical train of SHy-Cam is calculated by taking the loss of each optical surface between the first surface of
the filter pair 𝐹 and the last surface of the tube lenses 𝑇𝐿
1−4
(see Figure S1). For achromatic doublet lenses, the first and last
surface photon loss is estimated by the average reflection value of the anti-reflection coatings within 400nm-700nm (Edmund
Optics: <1.75%, Thorlabs< 0.5%). The glass-to-glass photon loss by reflection is estimated by Fresnel equation under 0° AOI
condition between two glasses with different refractive indices. Photon loss at folding mirrors and beam splitter is estimated at
1% per the coating specification from Thorlabs. The transmittance and reflectance profiles of two custom sinusoidal dichroic
mirrors were provided by manufacturer

2.4.9 Sample preparation
Fluorescent dyes
Rhodamine 6G and Rhodamine B powders were respectively dissolved in distilled water to
create pure dye solutions with the concentration of 1mg/mL. Five mixtures of the two dye
solutions with different relative concentrations (Fig. 2.2b-d) were created by mixing the two pure
dye solutions.
Fluorescent beads
1-𝜇𝑚 , 1% solid fluorescent bead samples (L2778 fluorescent red, L1030 fluorescent yellow-
green, L9654 fluorescent orange, Sigma-Aldrich; Dragon Green, Flash Red, Suncoast Yellow,
Bangs Laboratories, Inc.) were diluted 1000-fold in distilled water, then further diluted another
100-fold in 1% low-melt agarose. The 10,000-fold diluted fluorescent bead sample in 1% low-
Fluorophore
Channel

CFP

eGFP

eYFP

mCherry

iRFP670
ASIN 11% 12% 19% 35% 30%
ACOS 27% 31% 33% 19% 5%
SIN 29% 27% 20% 7% 12%
COS 12% 7% 6% 23% 37%
Total 79% 77% 78% 84% 84%
49

melt agarose was pulled into a glass capillary (5-000-1025, Drummond Wiretrol) with a
stainless-steel plunger. After the agarose solidified at room temperature at 20℃ for 1-2 minutes,
the capillary was transferred to a distilled water-filled imaging chamber and a length of the
agarose containing the beads was extruded from the micropipettes allowing optical access.
Transgenic zebrafish embryo
For in-vivo imaging tests, zebrafish embryos were immersed in liquid solution of 1% low-melt
agarose (made with 30% Danieau solution) and pulled into a glass capillary, and agarose was
allowed to solidify at room temperature. The capillary was transferred to a Danieau solution-
filled imaging chamber and the agarose containing the embryo was extruded from the
micropipettes allowing optical access. 0.075% Tricaine was added to both the agarose solution
and Danieau solution-filled imaging chamber to prevent movement of embryos. During
imaging, the chamber temperature was set and kept at 28.5℃. For fixed transgenic zebrafish
embryos imaging (Fig. 2.2i-k), embryos at 4dpf were euthanized with 0.1% tricaine
methanesulfate (Sigma-Aldrich Corp.) and then fixed by 4% paraformaldehyde (Sigma-Aldrich
Corp.) overnight at 4℃, followed by washing with PBS three times. The embryo to be imaged
on Zeiss LSM 780 was immersed in liquid solution of 1% low-melt agarose inside a glass bottom
imaging dish. The agarose solidified at the room temperature before transferring the dish onto
the microscope stage. Upon finishing the experiment, the fixed embryo was carefully freed from
the agarose and got mounted into the glass capillary following the sample mounting procedure
for in-vivo live embryos, without adding Tricaine.
50

2.4.10 Zebrafish lines
Lines were raised and maintained following standard literature practice and in accordance with
the Guide for the Care and Use of Laboratory Animals provided by the University of Southern
California. Fish samples were part of a protocol approved by the IACUC (permit number: 12007
USC). krt4:lyn-egfp and krtt1c19e:lyn-tdtomato transgenic lines [69] were kind gifts from
Thomas J. Carney (A*STAR, Singapore). kdrl:mCherry transgenic line [70] was a kind gift
from Ching-Ling Lien (Children’s Hospital Los Angeles). TgBAC(sox10:BirA-
mCherry)ox104a line was reported previously [71]. mpv17
a9/a9
;mitfa
w2/w2
(casper) line [72] was
purchased from Zebrafish International Resource Center (ZIRC) and csf1r
j4e1/j4e1
(panther)
line [73] was a kind gift from David Parichy (Univ. Virginia). We crossed casper with panther
to produce triple heterozygote mpv17
a9/+
;mitfa
w2/+
;csf1r
j4e1/+
F1 generation fish, which were
subsequently incrossed to produce F2 generation with 27 combinations of mutational state of
these genes. Since csf1r
j4e1
phenotype was not clear in F2 adult with casper phenotype, we
outcrossed these fish with panther fish to determine the zygosity of csf1r
j4e1
mutation based on
the frequency of larva with xanthophores (heterozygote and homozygote produced 50%- and
0%-fraction of xanthophore-positive larva, respectively) by fluorescent microscopy. The
casper;csf1r
j4e1/j4e1
line is viable and reproducible; we outcrossed either casper;csf1r
j4e1/j4e1
line or
casper;csf1r
j4e1/+
line with other fluorescent transgenic lines over several generations to obtain
fish harboring multiple transgenes on casper background either in the presence or absence of
xanthophores.
The coding sequences for human Histone 2b region (H2B) and fluorescent protein iRFP670 were
amplified from the vector for Tg(PGK1:H2B-chFP) [74] using primers #1 and #2, and
piRFP670-N1 (Addgene # 45457) using primers #3 and #4, respectively. The PCR products
51

were fused to generate H2B-iRFP670 fusion fragment and cloned into pDONR221 (Thermo
Fisher Scientific). Subsequent MultiSite Gateway reaction was performed using Tol2kit vectors
according to developer’s manuals [75]. pENTR5’ubi (Addgene #27320), pDONR221-H2B-
iRFP670, and pDONR P2R-P3-WPRE [52] were assembled into pDestTol2pA2 (Tol2kit
#394) [76,77]. The resultant pDestTol2-ubi:H2B-iRFP670 was co-injected with tol2 mRNA
into one-cell-stage casper zebrafish embryos. Injected F0s were raised and screened for founders.
Positive F1s grown to reproductive age were subjected to Splinklerette PCR analysis [78,79] to
determine genomic integration sites. Embryos that had specific expressions of transgenic
fluorescent proteins were collected and raised in a low-salt embryo medium per established
procedures until the appropriate time (4dpf) for imaging.
2.5 Code availability
The MATLAB functions can be downloaded from SHy-Cam GitHub repository at
https://github.com/PaulWZZtoLA/Singleshot-Hyperspectral-Phasor-Camera-SHyCam.
HySP software and instruction can be found at
http://bioimaging.usc.edu/software.html#HySP.
2.6 Publication information
The results presented in this chapter have been submitted for Cell Reports Methods. The
manuscript is now in revision. [80]

Author list
Pu Wang
1,2
, Masahiro Kitano
1,3
, Kevin Keomanee-Dizon
1,4
, Thai V. Truong
1,3
, Scott E.
Fraser
1,2,3
and Francesco Cutrale
1,2,3

1
Translational Imaging Center, University of Southern California, 1002 West Childs Way, Los
Angeles, CA 90089, USA.
52

2
Biomedical Engineering, University of Southern California, 1002 West Childs Way, Los
Angeles, CA 90089, USA.
3
Molecular and Computational Biology, University of Southern California, 1002 West Childs
Way, Los Angeles, CA 90089, USA.
4
Joseph Henry Laboratories of Physics, Princeton University, Princeton, NJ 08544, USA.

Author contributions
P.W. designed and built SHy-Cam system, conducted experiments and research study. M.K.
raised zebrafish lines and set up cross for embryos that were used as in-vivo imaging samples.
T.V.T built the SPIM. K.K.D. provided advice on adapting SHy-Cam prototype onto SPIM.
F.C. and S.E.F. conceived the original idea and provided systematic advice in this project. All
authors discussed the conclusions and commented on the manuscript.

53

Chapter 3
Hyperspectral phasor autofluorescence characterization of
unstained Barrett’s esophagus dysplasia and adenocarcinoma
unstained label-free biopsies
3.1 Introduction
Esophageal cancer (EC) has been an important health problem across the globe because of its
poor prognosis and high mortality rate. It ranked as the seventh most common cancer (604,000
new cases) and the sixth in terms of cancer mortality (544,000 deaths) in 2020 [81]. There are
mainly two distinct histologic subtypes of EC, squamous cell carcinoma and adenocarcinoma. In
recent years, adenocarcinoma has become the most prevalent EC subtype arising on Barrett’s
esophagus among new cases in developed countries [82–84]. Dysplasia and adenocarcinoma are
the neoplastic progression of Barrett’s esophagus which are traditionally diagnosed and staged
with the evaluation of cell and tissue morphological changes within Hematoxylin and Eosin
(H&E) stained biopsy samples [85,86]. Pathologists play an important role for evaluating
biopsies for features of such neoplastic progression and both the sensitivity and specificity can be
hampered by the dependance of subjective evaluation process, often requiring the need to seek
opinion from more than one pathologist specialized in gastrointestinal tissue. In addition, the
complete and proper tissue processing procedure is time consuming. Therefore, the biopsies
prepared and stained with H&E are unsuitable for time-demanding intra-surgical evaluation. A
54

more objective and faster-to-prepare contrast will potentially improve the detection accuracy and
help doctor make the right intra-surgical decision by being able to easily conduct fast imaging
test, which ultimately reduces the morbidity and mortality from this aggressive malignancy.
Tissue contains a variety of autofluorescent signals such as collagen, elastin, and reduced
nicotinamide adenine dinucleotide (NADH) [87]. Tissue metabolic and morphological changes
during cancerous progression can alter the autofluorescence spectral and intensity properties,
which makes autofluorescence a great label-free biomarker for cancer diagnosis. [87] [88] [86]
Researchers have conducted studies on finding new diagnostic methodologies for dysplasia in
BE and esophageal cancer that is within epithelium based on tissue autofluorescence with in vivo
fluorescence endoscopic imaging [13–17] as well as ex vivo imaging with confocal fluorescence
microscopes [89].
In this paper, we extend the effort to observe and characterize the autofluorescence properties
beyond the epithelium layers. We conducted imaging experiments with the unstained biopsy
slides from three esophageal adenocarcinoma patients using Singleshot Hyperspectral Phasor
Digital Slide Scanner (SHy-DSS), a high-sensitivity imaging device optimized for
autofluorescence imaging of unstained biopsy slides. As a proof of concept, we used
hyperspectral [9–12] that is readily available from SHy-DSS images to intuitively characterize
the subtle tissue autofluorescence spectral and intensity changes at different esophagus layers
between adenocarcinoma and the surrounding tissue.
55

3.2 Methods
3.2.1 Singleshot Hyperspectral Phasor Digital Slide Scanner (SHy-DSS)
Hyperspectral phasor (HySP) [9–12] is a computational analysis method originally designed for
an intuitive 2-D representation of the high-dimension hyperspectral fluorescence data. It
contains two values, G and S, called phasor coefficients, which are essentially the normalized
first harmonic Fourier coefficients of the original spectral vector 𝑰 (Eq. 3.1, 3.2) and numerically
located inside of a 2-D unit circle named phasor plane (Fig. 3.1e). Fluorescence spectral
difference and changes can be easily visualized on the phasor plane (Fig. 3.1d, e). The phase
represents the overall hue of the fluorescence spectra. Larger phase corresponds to a spectrum
with a longer wavelength (Fig. 3.1d, e). The distance from the phasor coefficients to the phasor
plane origin represents the bandwidth of the spectrum. A longer distance corresponds to a
narrower spectral bandwidth (Fig. 3.1d, e).
HySP is a great tool to visualize subtle tissue autofluorescence wavelength shift and
bandwidth change potentially caused by the cancerous progression of esophageal
adenocarcinoma [13–17]. As previously mentioned, HySP is derived from hyperspectral data
acquired using fluorescence microscopes with a hyperspectral detector. Such imaging
instrument is very expensive. In addition, hyperspectral detector has low sensitivity due to the
narrow bandwidth of each color channel. Tissue autofluorescence normally has much lower
quantum yield compared with fluorophores or fluorescent proteins which hyperspectral detectors
are more suitable to image. In order to acquire the autofluorescence images with sufficient
signal-to-noise ratio (SNR), the excitation light source power and exposure time need to be
56

increased, which will cause more photobleaching to the tissue sample and hamper the diagnostic
results.
𝐺 =
∑ 𝐼 (𝜆 𝑛 )∙cos(
2𝜋 𝑁 𝜆 )∗∆𝜆 𝑁 −1
𝑛 =0
∑ 𝐼 (𝜆 𝑛 )∗∆𝜆 𝑁 −1
𝑛 =0
(3.1)
Fig. 3.1 Single snapshot hyperspectral phasor camera (SHy-DSS) system overview. (a) Diagram of SHy-DSS optical design.
The autofluorescence is excited with a collimated 385nm UV LED and collected by an infinity-corrected objective lens OL with
20x magnification. A rectangular field stop FS is placed at the intermediate image plane created by a relay lens 𝑅𝐿
1
for confining
the effective field of view to avoid overlapping on the final sensor image plane. After being collimated by a second relay lens
𝑅𝐿
2
, the light is first equally divided into two beam paths via a 50/50 beam splitter BS and then being divided into four spectrally
encoded channels with two sinusoidal dichroic mirrors 𝐷𝑀
𝑆 and 𝐷𝑀
𝐶 . Eight mirrors mounted on kinematic mounts
(𝑣𝑒𝑟𝑡𝑖𝑐𝑎𝑙 𝑚𝑖𝑟𝑟𝑜𝑟 𝑉𝑀 𝑆 ,𝐶 ,𝐴 𝑆 ,𝐴𝐶
𝑎𝑛𝑑 ℎ𝑜𝑟𝑖𝑧𝑜𝑛𝑡𝑎𝑙 𝑚𝑖𝑟𝑟𝑜𝑟 𝐻𝑀
𝑆 ,𝐶 ,𝐴𝑆 ,𝐴𝐶
) with two axis adjustability along with two fixed folding
mirrors 𝐹𝑀
𝐶 and 𝐹𝑀
𝑆 steer the four channels into four tube lenses 𝑇𝐿
𝑆 ,𝐶 ,𝐴𝑆 ,𝐴𝐶
with the correct incidence angles to focus on four
quadrants of a 5 Mega-pixel CMOS sensor. Other abbreviations 𝐹 𝑒𝑥
: excitation band pass filter; 𝐹 𝑒𝑚
: emission long pass filter.
CL: collimating lens. (b) Theoretical (solid line) and real (dashed line) transmission (tr) and reflection (re) profiles of the two
custom sinusoidal dichroic mirrors (𝐷𝑀
𝑆 : Sine, 𝐷𝑀
𝐶 : Cosine dichroic mirrors) in the visible spectral range (400nm-700nm). (c)
The diagram showing the normalization operation for calculating phasor coefficients G and S from the registered and aligned four-
channel image (see Method 2.2 Image processing and analysis). (d) Two sample fluorescence emission spectral profiles, EM1and
EM2, and the corresponding spectral phasors on the 2-D phasor plane (e). EM1 has peak emission at a shorter wavelength and a
narrower spectral bandwidth compared with EM2, which reflects a smaller phase angle and larger distance to the spectral phasor
origin.
57

𝑆 =
∑ 𝐼 (𝜆 𝑛 )∙cos (
2𝜋 𝑁 𝜆 )∗∆𝜆 𝑁 −1
𝑛 =0
∑ 𝐼 (𝜆 𝑛 )∗∆𝜆 𝑁 −1
𝑛 =0
(3.2)
Singleshot Hyperspectral Phasor Camera (SHy-Cam) (Fig. 2.1) in the previous chapter is a
high-sensitivity and cost-effective add-on imaging device of a light sheet microscope originally
designed for multiplexed fluorescence imaging applications [80,90]. It utilizes two custom
dichroic mirrors with sinusoidal transmission profiles (Fig. 2.1a) to optically convolve with the
fluorescence spectrum and acquire the hyperspectral phasors directly with the acquisition and a
simple normalization operation. The much wider transmission profiles of the sinusoidal dichroic
mirrors achieve over 10-fold sensitivity increase compared with a traditional 32-channel
hyperspectral detector. The image splitter (Fig. 2.1b) in SHy-Cam allows a fast snapshot
acquisition with a single sCMOS camera.
Based on the concept of SHy-Cam, we designed a standalone widefield version of SHy-Cam
called SHy-DSS (Fig. 3.1a, b, Fig. 3.2). SHy-DSS is optimized for autofluorescence imaging of
unstained biopsy slides. The system is equipped with a 385nm UV LED as the autofluorescence
excitation light source, a 20x infinity-corrected air objective lens with 0.5 numerical aperture
(NA) and a 3-axis motorized sample stage (Fig. 3.2). SHy-DSS is a bench-top imaging device
which takes less than the area of 18 inches by 24 inches. [91–95] [9–12] [80,90] [80,90]SHy-
DSS was built mostly with off-the-shelf optical, opto-mechanical components and custom opto-
mechanical parts that are easy to fabricate with bench-top 3D printers and 3-axis computer
58

numerical control (CNC) milling machines. The build cost of the core components excluding the
sample stage and sCMOS camera was under $7,000.
Fig. 3.2 The 3D rendering and picture of SHy-DSS. (a) The 3D rendering of SHy-DSS computer-aided design (CAD)
model showing the physical size of the system. SHy-DSS uses inverted setup for convenient sample mounting/unmounting
from the 3-axis motorized sample stage. The objective lens 𝑂𝐿 can also be easily changed to another one with a different
magnifying power. The picture inside of the dashed rectangular box shows four tube lenses 𝑇𝐿
𝑆 ,𝐶 ,𝐴𝑆 ,𝐴𝐶
mounted on 3D
printed lens mounts which have independent translational adjustability for compensating the focal shift of each channel.
(b) The picture of SHy-DSS with the 3D printed enclosure sitting on a 18’’x24’’ aluminum optical breadboard. SHy-DSS
is designed to be operated on the bench. There are no moving components inside of the system, which increases the
stability.
59

3.2.2 SHy-DSS image acquisition and processing pipeline
SHy-DSS image acquisition is performed using the open-source microscope hardware control
software 𝜇 Manager [62,63]. The unstained biopsy slide is placed on a motorized sample stage
(ASI Instrumentation, Inc., Newport, Inc.) capable of 3-axis xy and z movement. Fig. 3.3a
shows what a single camera frame looks like. Each channel takes up an area that is slightly
smaller than one quadrant of the sensor with the field of view (FoV) being confined by a
physical rectangular field stop. With total magnification of 20x, the FoV on the sample plane is
300μm x 240μm. Each frame is captured using 100 millisecond (ms) exposure time with a 2x2
pixel binning. After inputting the bounding area of the sample region of interests (ROI),
𝜇 Manager software automatically controls the sample stage for a tiled scanning (Fig. 3.3b).
Upon finishing acquiring the sample image, three important correction images are acquired for
image processing including the dark image 𝑫 𝐼 of the sample image 𝑰 , the flat field image 𝑭 and
the dark image 𝑫 𝐹 for F (Fig. 3.3c). The dark images are used to correct both the background
signals from the sample fixation solution and the dark noise that is present on the camera sensor.
𝑫 𝐼 and 𝑫 𝐹 are respectively subtracted from the sample image 𝑰 and flat field image 𝑭 (Eq. 3.3,
3.4). Non-uniformity of the excitation and the autofluorescence emission are corrected by the
flat field image (Eq. 3.5) using a fluorescence dye slide prepared by following the procedures
previously reported [96,97].
Hyperspectral phasors are derived directly from the SHy-DSS image by pixel-wise
calculations (Fig. 3.1c), which requires precise alignment of the four channels. In order to align
the four channels and establish their spatial correlations, we need to locate the control points on
the four-channel images that represent the same physical locations on the sample plane. The
sample image is first converted to a black-white image (Fig. 3.3d) by image binarization with an
60

adaptive threshold in MATLAB. Next, an automatic control points detection algorithm called
Speed-Up Robust Feature (SURF) [101] is applied to the black-white image to find the control
points (Fig. 3.3d) between the reference (SIN) channel with other three registering channels
(COS, ACOS, ASIN). A geometric affine transformation is applied to the three registering
channels and the aligned four channels are save into an image stack (Fig, 3,3f) for phasor
analysis or image stitching. There are two things worth mentioning for the image registration.
First, SURF control points locating only needs to be conducted for the one of the many raster-
scanned image, normally the first image, and the control points are saved for aligning and
registering the rest of the sample images. Second, the dark frame correction and flat field
correction are applied during the automatic registration step (Fig. 3.3E) using the previously
acquired dark image and flat field image. The dark correction image 𝑫 𝐼 , 𝑫 𝐹 are subtracted from
the sample image I and flat field image F (Eq. 3.3, 3.4) before the dark corrected sample image
𝑰 𝐷𝐹𝐶
and flat field image 𝑭 𝐷𝐹𝐶
are sub-divided into four channels for registration. After the registration,
flat field correction is applied individually to each channel using its corresponding sub-divided flat field
image channel (Eq. 3.5).
𝑰 𝐷𝐹𝐶
= 𝑰 − 𝑫 𝐼 (3.3)
𝑭 𝐷𝐹𝐶
= 𝑭 − 𝑫 𝐹 (3.4)
𝑰 𝑐𝑜𝑟𝑟𝑒𝑐𝑡𝑒𝑑 =
𝑰 𝐷𝐹𝐶 𝑭 𝐷𝐹𝐶
∗ 𝑚𝑒𝑎𝑛 (𝑭 𝐷𝐹𝐶
) (3.5)
61

Image stitching is further applied to the output images from the registration step in a
microscope image stitching software called Imaris Stitcher (Oxford Instruments) if raster
scanning (tiling) is used during image acquisition.
3.2.3 Biopsy slide preparation and standard H&E imaging
The esophagus biopsy slides were prepared at the Translational Pathology Core of the University
of Southern California. Each sample consists of two vertically sequential 5um cuts, one left
Fig. 3.3 SHy-DSS image acquisition and processing pipeline. (a) The camera frame showing four channels on the four
quadrants of the sensor. Each channel takes up one quadrant of the sensor with a 300μm x 240μm FoV at the object plane. FoV is
physically confined by the field stop FS (Fig. 3.1, 3.2). (b) Raster scanning is conducted by capturing images of multiple adjacent
sample areas. The motorized sample stage is automatically controlled to move the slide in x-y plane. (c) Two dark images (𝑫 𝑰 and
𝑫 𝑭 ) and one flat field image F for processing are acquired after the sample image I acquisition. 𝑫 𝑰 is acquired using the same
camera and excitation light settings for I and with the sample stage being moved to a position where there is no tissue sample in the
camera FoV. 100 frames are acquired, and the average frame is used for 𝑫 𝑰 . The same procedure is applied to 𝑫 𝑭 . The flat field
image F is acquired using the fluorescent dye slide prepared by the protocol previously reported [96,97]. The camera and excitation
light setting can be different from the ones used for I. F is calculated by the average of 100 frames as well. (d) One sample image
is converted to a black-white image for detecting the control points (the color dots) from the four channels that represent the same
physical points on the sample plane. Affine image transformation (registration) is applied to three registering channels (COS,
ACOS and SIN) to be aligned with the reference channel (SIN). (e) The dark frame correction is applied to both the sample image
and flat field image before the registration. Since flat field correction requires pixel-wise calculation (Eq. 3.5). Registration is
applied to both sample images and the flat field image. Flat field correction is then applied to the registered samples images
channel by channel using the registered flat field image. (f) Processed sample images are saved as a 3-D data cube (x, y, 𝜆 ). 𝜆
represents the sinusoidal channels with the order of ASIN, ACOS, SIN and COS. Note the data cube can either be single sample
image or a stitched image from multiple raster-scanned images.
62

unstained for SHy-DSS autofluorescence imaging while the other stained with H&E. The
standard H&E imaging was performed utilizing a Hamamatsu Nanozoomer S60 digital slide
scanner in bright field modality at 20x magnification. Digital pathology images of H&E samples
were reviewed, and the dysplastic and cancerous tissue areas were digitally labelled for the
reference of autofluorescence imaging by registered pathologists.
3.2.4 Autofluorescence characterization with SHy-DSS image
Autofluorescence images were acquired with the unstained esophagus cross-section biopsy slides
from three Barrett’s esophagus patients (Fig. 3.5 caption), who were respectively diagnosed with
pT1a, pT2 and pT3N2 pathologic stage of esophageal adenocarcinoma. We compared the
autofluorescence intensity between the dysplastic and cancerous tissue areas with the
surrounding non-invaded tissue areas at different depths of invasion as well as the
autofluorescence spectral change with hyperspectral phasor and hyperspectral phasor color-
mapped images (Fig. 3.4, 3.5). The ASIN channel was used to represent and show the intensity
images. The averaged hyperspectral phasor of the dysplastic and cancerous tissue and the
surrounding normal tissue are shown on the phasor plane (Fig. 3.4 g, h, Fig. 3.5. p-t) to
statistically represent the spectral characteristics of the sample. Color-coded images (Fig. 3.4e, f,
Fig. 3.5k-o) based on the phase information of hyperspectral phasors are used for visualizing the
spectral difference.

63

3.3 Results
3.3.1 High-grade dysplasia (HGD) in Barrett’s esophagus (BE)
HGD in epithelium shows distinct curvy epithelial lining and enlarged nuclei (Fig. 3.6a)
compared with normal squamous epithelium (Fig. 3.6b) in the bright field H&E image. The
autofluorescence shows increased intensity in HGD area (Fig. 3.4c) compared with normal
squamous epithelium (Fig. 3.4d), in conformity with literature results from confocal microscopy
ex vivo autofluorescence studies of HGD in BE [89]. The bright autofluorescence signal (Fig.
3.4c), emitted from hemoglobin inside of red blood cells, is, sparsely present in the slide and is
not necessarily related to the dysplasia itself. In addition to the autofluorescence intensity
changes, HGD area shows subtle spectral shift toward longer wavelength which can be observed
with the slightly increased hyperspectral phasor phase angle and (Fig. 4g, h). The color-mapped
images are generated base-on the phase angle to intuitively visualize the subtle spectral
wavelength shift (Fig. 4g, h). The HGD autofluorescence spectrum also becomes flatter or less
saturated compared with normal squamous epithelium which is represented with a shorter
distance of the hyperspectral phasor to the phasor plane origin (Fig. 4g, h).
64

3.3.2 Esophageal adenocarcinoma at different depths of invasion
Besides the outermost epithelium layer, esophagus contains multiple tissue layers, each of which
is formed by connective tissue or muscle tissue (Fig. 3.5). Since different layer types have
different constituents, the autofluorescence can present different spectral and intensity changes
during the cancerous progression. We used SHy-DSS to capture the autofluorescence images of
adenocarcinoma-invaded areas at different depths of invasion thus different tissue layers and
characterized the autofluorescence property changes based on hyperspectral phasors and
Fig. 3.4 Autofluorescence characterization of high-grade dysplasia (HGD) in esophagus epithelium. (a) The standard
bright field digital slide scanner image of a H&E-stained slide containing HGD at the esophagus epithelium from the biopsy
of a patient diagnosed with pT1a stage esophageal adenocarcinoma. The curvy epithelium lining and enlarged nuclei can be
observed in this H&E image. (b) The standard bright field digital slide scanner image of a normal squamous epithelium area
from the same slide with (a). (c, d) SHy-DSS autofluorescence intensity (ASIN channel) image of the unstained biopsy slide
containing the same cross section area with (a) and (b). The HGD area (c) shows increased autofluorescence intensity
compared with the normal squamous epithelium area in (d). Note (c) and (d) are normalized into the same intensity scale for
the direct intensity comparison. (e, f) Color-mapped images of the same HGD area and squamous epithelium area based on
the phase information of the hyperspectral phasor. The color scale bar on the phasor plane (g, h) shows the color mapping
relation. The average hyperspectral phasors of the HGD area and squamous epithelium area are respectively represented
with a red dot and blue dot on the phasor plane. 𝜎 1
is the standard deviation of the phase in the squamous epithelium area
and 𝜎 2
is the one in the HGD area. The color scale bar ranges from one standard deviation 𝜎 2
below the blue dot and one
standard deviation above the red dot, which provides clear color mapping results and intuitive visualization of the spectral
shift. Note this range of color scale bar is an empirical value which can be adjusted to yield a clear visualization for different
images.
65

compared the results with the surrounding tissue areas like what we did with HGD in the
epithelium layer.
The bright field H&E image (Fig. 3.6a) shows a syncytial growth pattern with the back-to-
back glands inside of the lamina propria layer, which is a pathological feature of the
intramucosal adenocarcinoma [85]. The glandular tissue has a distinct papillary shape with a
smoother autofluorescence pattern and slightly decreased intensity with its surrounding single-
and cluster-cell connective tissue compared to the connective tissue that is not invaded by the
adenocarcinoma (Fig. 3.6f). The adenocarcinoma area also emits autofluorescence at slightly
longer wavelength with a flatter emission spectrum (Fig. 3.6k, p).
Fig. 3.5 Anatomy of esophagus cross section. Esophagus consists of six tissue layers. Epithelium is the outer most layer
which forms the lining of esophagus. Lamina propria, submucosa and adventitia are three connective tissue layers (blue texts)
which contain nerve cells, lymphatic tissue, and large vessels. Note that the actual adventitia layer is not shown in this anatomy
picture. In addition, there are two layers of muscle tissue (green texts) that are located in between the connective tissue layers
called muscularis mucosa and muscularis externa. The tissue autofluorescence properties can vary differently during cancerous
progression in those two different tissue layer types as the connective tissue and muscle tissue have different constituents.
66

Adenocarcinoma in muscularis mucosa layer has a similar pathological feature found in
lamina propria, which is the glandular crowding (Fig. 3.6b). The autofluorescence intensities
(Fig. 3.6g) decreases slightly in the adenocarcinoma area compared with the surrounding muscle
tissue. In addition, we can observe that the autofluorescence emission shifts slightly toward a
shorter wavelength and the emission spectrum becomes flatter (Fig. 3.6i, q).
Ulceration can be seen in submucosa, muscularis externa and adventitia where
adenocarcinoma invades (Fig. 3.6c-e). Similar with lamina propria layer, the adenocarcinoma
tissue in lamina propria shows slightly decreased autofluorescence intensities (Fig. 3.6h) and a
longer emission wavelength with a flatter spectrum (Fig. 3.6m, r). In muscularis externa, we
observe decreased autofluorescence intensity, especially in the red-bounded ulcerated tissue (Fig.
3.6i), with a shorter wavelength and a flatter spectrum (Fig. 3.6n, s), which is also visible in
muscularis mucosa layer. In the innermost connective layer called adventitia, autofluorescence
intensities subtly decrease inside of the ulcerated tissue (Fig. 3.5j). The slightly longer
67

wavelength and flatter spectrum are experimentally similar to what we previously observed in
another two connective tissue layers (Fig. 3.5k, p).
Fig. 3.6 Autofluorescence characterization of esophageal adenocarcinoma. (a-e) Bright field images of H&E-stained
biopsy slides containing esophageal adenocarcinoma at different depth of invasion. The red-bounded regions represent the
adenocarcinoma invaded areas. (a-b) stage: pT2, grade: poorly differentiate; (c-e) stage: pT3N2, grade: moderately differentiate.
(f-j) Autofluorescence intensity (ASIN channel) images of unstained biopsy slides from the near cross sections of the H&E-
stained slides showing the same sample areas with (a-e). The five intensity images are normalized into the same intensity range
represented by the intensity scale bar. (k-o) Color-mapped autofluorescence images based on the phase information of the
hyperspectral phasor shown in (p-t). The color scale bar has the same defined range with Fig. 3.4.
68

3.4 Discussion
Autofluorescence is an emerging diagnostic imaging contrast for its simplified sample
preparation compared with traditional staining-based imaging contrast like H&E.
Autofluorescence can also provide more objective diagnostic information which can be extracted
from the images. In this work, we introduced SHy-DSS, an imaging device optimized for the
label-free autofluorescence imaging of unstained biopsy slides. SHy-DSS utilizes the same
snapshot hyperspectral phasor acquisition strategy with SHy-Cam presented in Chapter 2, which
was designed for fast and high-sensitivity multiplexed fluorescence imaging applications. SHy-
DSS inherits the high sensitivity and the fast snapshot imaging capabilities from its predecessor,
essential features for the relatively low quantum yield imaging applications like tissue
autofluorescence imaging in a fast-pace clinical environment. SHy-DSS also has the capability
of visualizing small spectral changes in an intuitive way by taking the advantage of the 2-D
analysis with hyperspectral phasors, intrinsically acquired with the data. Phasors’ sensitivity and
simplified interpretation can provide an intuitive diagnostic methodology to interpret the subtle
autofluorescence spectral changes related to disease.
As a proof of concept, we used SHy-DSS to study the potential correlation of the
autofluorescence spectral and intensity changes with one of the most common cancer types,
esophageal adenocarcinoma. We conducted the autofluorescence imaging of three unstained
esophageal adenocarcinoma biopsy slides and characterized the fluorescence intensity and
spectral changes in different esophagus layer types. As a result, we found subtle but consistent
variation of autofluorescence spectral and intensity between the adenocarcinoma areas and the
surrounding normal tissue areas in the same type of tissue layers. Owing to their anatomical
position, these differences are potentially triggered by localized changes induced by the
69

cancerous progression. We also presented an intuitive pipeline of visualizing such subtle change
in the autofluorescence characteristics using hyperspectral phasors, whose components are
obtained during acquisition of the SHy-DSS data. SHy-DSS coupled with hyperspectral phasor
data visualization pipelines, like SEER [52], can potentially overcome the limitations in
sensitivity and quantification of the diagnostic golden standard, pathologic evaluation with H&E
staining. Future studies with larger sample sizes can further validate the robustness of our
proposed method.
The image acquisition speed of SHy-DSS depends on the instrument magnification,
illumination, stage, and camera, as well as on the sample area to be acquired and, consequently,
the number of image tiles that compose the mosaic acquisition. With our current setup each
image tile covers a field of view of ~300μm x 240μm on the sample and requires ~500ms
acquisition time, of which 100ms for camera exposure time and the remaining for stage
movement. For example, the dataset presented in Fig. 3.5g presents a sample area consisting of
5x4 tiles that required approximately ~16 seconds to acquire. Optimizations in stage movement
combined with the use of more sensitive cameras and more powerful illumination could
considerably reduce this acquisition time.
Image pre-processing, registration, alignment, and spectral phasor analysis are currently
conducted post acquisition and, for the datasets presented in this manuscript, required 2-5
minutes computational time. In future implementations, a real-time processing and analysis
could be implemented during acquisition, we estimate this would reduce the time to results by
over 10 times. Embedded System-on-Chip (SoC) processors with integrated GPUs like the
Nvidia Jetson platform can be used for data processing and analysis to reduce the system cost
and minimize the footprint of SHy-DSS.
70

In conclusion, SHy-DSS was applied to label-free biopsy slide imaging exploiting
autofluorescence to potentially assist clinical decision making. The versatility of this imaging
platform also allows for In-vivo imaging implementations that could facilitate intra-surgical or
endoscopic patient assessment. These implementations could utilize increasingly available
gradient-index (GRIN) lens-based endoscopes as a vehicle for delivering and collecting
autofluorescence data and take full advantages of a high efficiency real-time acquisition and
analysis.
3.5 Data Availability
The datasets generated during and/or analyzed during the current study are not publicly available
but are available from the corresponding author on reasonable request.
3.6 Publication information
This results presented in this chapter have been prepared for the submission for Biomedical
Optics Express, OSA.
Author list
Pu Wang
1,2
, Bing Zhang
3
, Scott E. Fraser
1,2,4
, and Francesco Cutrale
1,2

1
Translational Imaging Center, University of Southern California, 1002 West Childs Way, Los
Angeles, CA 90089, USA.
2
Biomedical Engineering, University of Southern California, 1002 West Childs Way, Los
Angeles, CA 90089, USA.
3
Keck School of Medicine, University of Southern California, 1520 San Pablo St., Los Angeles,
CA 90033, USA.
4
Molecular and Computational Biology, University of Southern California, 1002 West Childs
Way, Los Angeles, CA 90089, USA.
Author contributions
71

P.W. designed and built SHy-DSS system, conducted experiments. B.Z. provided advice on
correlating the findings from the autofluorescence image acquired and processed from SHy-DSS
to traditional brightfield pathology. F.C. and S.E.F. conceived the original idea and provided
systematic advice in this project. All authors discussed the conclusions and commented on the
manuscript.

72

Chapter 4
Conclusions
4.1 Summary of findings
In this thesis, we presented a fast and efficient spectral encoding and acquisition strategy for
multiplexed fluorescence and autofluorescence imaging applications, which was inspired by the
high photon throughput sinusoidal filter-based hyperspectral phasor acquisition method [30,31]
previously introduced by Gratton lab. Compared with the previous method, our new strategy
achieves a more than 10-fold imaging speed increase by enabling the single snapshot acquisition
instead of using sequential acquisition with filter switching. The single snapshot acquisition is
achieved by the novel image splitting and refocusing mechanism. In addition, two highly
efficient sinusoidal dichroic mirrors inside of the image splitter improve the photon efficiency by
collecting the originally rejected photons via two additional reflection channels. The high
photon efficiency helps gather more spectral information during the acquisition and enables
quantitative analysis with the spectral linear unmixing.
In Chapter 2, we realized the strategy by instrumenting SHy-Cam and validated its
performance with varieties of in-vivo fluorescence imaging tests on the multi-color transgenic
zebrafish embryos. We demonstrated SHy-Cam’s fast imaging speed with a clean separation of
five spectrally and spatially overlapping fluorescent proteins using spectral linear unmixing (Fig.
2.3). The whole volume was imaged in 4 minutes, more than 2-fold faster than a hyperspectral
73

fluorescence golden standard imaging instrument, the confocal laser scanning microscopes. We
also explored SHy-Cam’s long-duration multiplexed volumetric timelapse (Fig. 2.4) capability
with a 2-hour long continuous imaging test, in which we successfully tracked the neutrophil
migrating toward a cutting wound inside of a zebrafish embryo with a minimal photodamage to
the live sample. The 2-hour continuous timelapse only caused 11% photobleaching thanks to
SHy-Cam’s high sensitivity, which allowed lower laser excitation power to be applied. The high
sensitivity also enables SHy-Cam to use a much faster frame rate with a shorter exposure time to
capture images with sufficient SNRs. The heart rate of zebrafish embryos is around 180 beats
per minute or 3 beats per second which can be very challenging for other multiplexed
fluorescence imaging methods as the frame rate is either limited by the sensitivity or the
sequential acquisition, which will inevitably induce motion blur. We were able to capture a 20-
fps video of a multi-color zebrafish embryo beating heart cross section with minimum motion
blur using SHy-Cam while still having a sufficient SNR to cleanly unmix the three fluorescent
colors using linear unmixing (Fig. 2.4e-h). Those validation imaging tests demonstrated SHy-
Cam as a high-speed and high-throughput multiplexed fluorescence imaging method, especially
with the improved detection at low signal conditions.
Chapter 3 expanded SHy-Cam’s application beyond the fluorescence biological imaging to the
emerging field of tissue autofluorescence-based diagnostic imaging. We designed a standalone
widefield imaging system called Singleshot Hyperspectral Phasor Digital Slide Scanner (SHy-
DSS) based on SHy-Cam and optimized it for the autofluorescence imaging of unstained biopsy
slides. We also introduced a SHy-DSS image processing pipeline which was specifically
designed for an easy and intuitive visualization of the subtle tissue autofluorescence spectral and
intensity changes triggered by the cancerous progressions. As a proof of concept, we conducted
74

the validation imaging tests using SHy-DSS on three unstained esophageal adenocarcinoma
biopsy slides at different tissue layers and characterized the autofluorescence property changes
with our proposed image processing pipeline. Our results showed the consistent
autofluorescence spectral and intensity changes in the adenocarcinoma tissue areas that belonged
to the same tissue layer types. In the future work, a larger sample size can be used to further
validate the robustness of SHy-DSS as a potential candidate for cancer diagnostic imaging
applications.
This thesis not only demonstrates the scientific novelties of SHy-Cam and SHy-DSS but also
serves as an example of how modern prototyping and in-house manufacturing technologies can
facilitate the biomedical imaging instrumentation related research. With an ever-so-easy access
to the affordable additive manufacturing instruments such as the fused deposition modeling
(FDM) and stereolithography (SLA) 3D printers and R&D-grade milling machines, researchers,
especially from developing countries, now can realize complex prototypes with considerably
reduced costs.
4.2 Future applications
4.2.1 Multiplexed stem cell imaging
Cancer Stem Cells (CSCs) play an important role in tumor development [102–104]. In-vivo and
in-vitro imaging with fluorescence microscopy is one of the key tools for researchers to study
CSCs. Multiple fluorophores are normally used at the same time to study complex features and
processes happening inside of CSCs. Traditional multi-color fluorescence imaging with filter
switching induces more phototoxicity to the live cells and causes photobleaching, mostly due to
the added laser exposure during the sequential image acquisition. In addition, the traditional
75

bandpass filters have narrow bandwidths and allow limited number of photons to pass though in
order to minimize the cross talks and increase the signal specificity. SHy-Cam has been
introduced in Chapter 2 as a fast, high sensitivity and gentler imaging device designed for
multiplexed fluorescence imaging applications. CSC imaging can benefit from the added
imaging sensitivity and the fast snapshot acquisition that SHy-Cam provides. In addition, SHy-
Cas was designed as an add-on imaging device and can be easily adapted to most widefield
microscopes. The versatile adaptability of SHy-Cam makes it a great choice for the those who
want to add the multiplexed imaging capabilities to their existing wide field microscopes. Aside
from the advantages on the hardware side, the hyperspectral phasors components that are
acquired in the SHy-Cam images provide an intuitive analysis and visualization toolset for the
multiplexed CSC imaging.
4.2.2 Autofluorescence-based endoscopic esophageal adenocarcinoma diagnosis
Chapter 3 demonstrated that SHy-DSS and its image processing pipeline can potentially provide
an easy-to-conduct and intuitive-to-analyze esophageal adenocarcinoma diagnostic imaging
Fig. 4.1 SHy-Cam for cancer stem cell imaging. (a) The CAD model of SHy-Cam without enclosure. This is a newly adapted
SHy-Cam based on the design presented in Chapter 2 and is designed to be an add-on imaging device for a Nikon Ti-E inverted
widefield microscope specifically for in-vivo multiplexed CSC imaging application. The adaptation is a very streamlined process
and this SHy-Cam inherits the overall design and construction found in SHy-DSS presented in Chapter 3. (b) The Nikon Ti-E
inverted microscope that SHy-Cam is going to be adapted. The microscope has a temperature and 𝐶𝑂
2
-controlled sample
chamber for conducting in vivo cancer stem cell imaging.
76

method. In future work, a less-invasive diagnostic method can be developed by coupling and
adapting an endoscope to SHy-DSS (Fig. 4.2). The endoscopic version of SHy-DSS will inherit
the benefits of the objective diagnostic metric that autofluorescence provides and the intuitive
analysis of the hyperspectral phasors. As an endoscopic in-vivo imaging method, it is less-
invasive and does not require physical biopsies taken from the patients as it could perform
optical observations in situ. The time that was originally used for preparing the biopsies into
slides can be saved, which is critical for a fast-paced surgical environment. This system can
potentially be the alternative to the traditional reflectance-based endoscope screening for
esophageal adenocarcinoma and we name it Singleshot Hyperspectral Phasor Endoscopic
Scanner (SHy-ES).
SHy-ES consists of three main components. The first component is the image splitter and a
CMOS camera similar to the one in SHy-DSS and SHy-Cam, with the addition of an objective
lens and a long-pass filter (Fig. 4.2). The objective lens is to collect the autofluorescence image
that is relayed to the distal end of the imaging fiber bundle. A small rectangular pinhole made of
tungsten foil is placed at the imaging fiber bundle distal end as a field stop to confine the FoV
and prevent the spatial overlapping on the sensor. A long-pass filter is placed right behind the
back aperture of the objective lens to reject the excitation light from the autofluorescence light.
The second main component is the excitation light source. The one equipped in SHy-DSS is a
long-wave UV LED centering at 385nm. However, UV exposure may cause tissue
irritation [105] during the in-vivo imaging. Instead, a deep blue LED with 405nm center
wavelength can be equipped onto SHy-ES to minimize the tissue irritation while still being able
to excite most of the autofluorescence molecules in the tissue. However, the autofluorescence
77

excitation efficiency could be hampered by switching from the 385nm to the 405nm excitation
and further tests should be conducted to find the optimal excitation power.
The third main component is the endoscope, which includes an imaging fiber bundle with a
(gradient-index) GRIN lens for collecting the autofluorescence, a multi-mode imaging fiber for
delivering the deep blue excitation light, and a prism mirror for folding the excitation and
autofluorescence light 90 degrees to be orthogonal to the esophagus epithelial lining. The
imaging probe requires two separate light paths for the excitation and autofluorescence to avoid
the fluorescence contamination from the imaging fiber itself. Imaging fibers are normally made
of fused silica and fluoresce under the UV or deep blue excitation. The fluorescence from the
imaging fiber is strong enough to contaminate the tissue autofluorescence and render the images
unusable. For this practical reason, the UV excitation light cannot be delivered via the same
imaging fiber bundle that is used for delivering the autofluorescence light and a separate multi-
mode imaging fiber is needed for delivering excitation light (Fig. 4.2).
Fig. 4.2 Diagram of SHy-ES. SHy-ES consists of three main components, an image splitter coupled with a CMOS camera for
autofluorescence spectral encoding and detection, a deep blue LED excitation light source, an endoscope to guide the excitation
light to the sample and relay the autofluorescence light to the image splitter. A prism mirror is placed in the front of the
endoscope tip to turn both the excitation and autofluorescence light 90 degrees to be orthogonal to the esophagus epithelial
lining, which increases the excitation and signal collection efficiency.
78

4.2.3 Behavior study based on zebrafish embryo whole-brain functional imaging
Calcium imaging with fluorescent calcium indicators has been a popular fluorescence
microscopy technique for behavior studies and brain functional imaging since changes in
intracellular calcium ion (Ca
2+
) and calcium flux are closely related to many cellular functions in
the brain [107]. Zebrafish embryo is a great animal model for conducting in-vivo whole-brain
calcium imaging as the size is suitable for volumetric imaging when using proper fluorescence
imaging techniques such as light sheet microscopes. Multi-color transgenic zebrafish embryos
can provide fluorescent labels that selectively bond other entities which can be used to study the
correlation of certain neural activity with other biological processes.
SHy-Cam can be a great imaging tool for calcium imaging especially for the multiplexed
imaging applications when multiple fluorescent labels are present in the brain area. The high
photon throughput of SHy-Cam with gentler illumination of light sheet microscopes reduces the
chances of photobleaching effect and minimizes the phototoxicity to the brain tissue especially in
the long-duration timelapse imaging applications such as sleep study [108]. The snapshot
acquisition allows fast imaging speed for time-resolved imaging of neural activities. Multi-
camera setup can replace the single-camera architecture introduced previously in SHy-Cam and
SHy-DSS to increase the FoV, which is important for imaging larger area such as the whole
brain section without compromising the imaging speed.
4.2.4 Beyond biomedical: hyperspectral phasor handheld fruit quality inspection
camera
Cameras have become one of the most important sensing units that drive many decisions in
almost every field. Hyperspectral cameras have been a popular choice for the applications in
which the color information is critical such as fruit sorting and remote sensing. As we already
79

mentioned previously, one of the many advantages of hyperspectral phasor acquisition is the
intuitive color change visualization with the 2-D hyperspectral phasors that are readily available
from the images, which makes hyperspectral phasor acquisition a great candidate for imaging
applications that can benefit from such spectral change information.
The United States has more than 5,000 apple producers, who grow over 240 million bushels of
apples each year [109]. Fruit inspections need to be conducted regularly in order to estimate the
production yield and quality. Sampled visual inspection is the most commonly used inspection
method which does not require specialty equipment. However, it is based on subjective visual
inspection which is prone to human error. A cost-effective handheld device that can provide the
inspectors with the objective inspection information will ultimately simplify their workflow and
help fruit producers identify problems in time.
The reflection color information in the near infrared (NIR) wavelength range of the apple skin
is a great fruit quality indicator. For example, the apple peel reflectance directly correlates with
the fruit sugar content and the internal firmness especially between 800 to 900nm [110,111].
Here we propose a handheld imaging device that can intuitively visualize such reflectance
changes based on the hyperspectral phasor acquisition and we name this device Snapshot
Hyperspectral Phasor Pocket Camera (SHy-PC).
SHy-PC shares the same concept with SHy-Cam and SHy-DSS, but it is fundamentally a
different imaging device. It has the form factor of a photographic device instead of a microscope
due to the difference in the imaging object sizes and imaging contrast, which is reflection instead
of fluorescence. SHy-PC uses sinusoidal filters instead of dichroic mirrors for color encoding
and achieves snapshot with 3 independent camera modules, a COS module, a SIN module, and
an intensity module (Fig. 4.3a, b). SHy-PC is designed to work within 700nm to 1000nm
80

spectral range to accommodate the most significant reflectance changes between 800nm-900nm
as previously mentioned. This system also has a streamlined workflow (Fig. 4.3c). The
acquisition includes the actual fruit sample images and a white balance (WB) image of the white
color checker. After the acquisition, the sample images from three camera modules are aligned
by the image registration. A dark correction is applied to the registered image with the dark
image previously stored in the camera sensor non-volatile memory (NVM) during a one-time
calibration process of evaluating the dark level using different exposure time values. The last
processing step is white balancing with the WB image captured during the acquisition. This is to
remove the spectral shift caused by different lighting conditions, e.g., the sunlight at different
time of the day. The processed image can then be converted into the hyperspectral phasor
coefficients, G and S as we previously introduced in Chapter 2 and 3. We can either conduct
analysis on the phasor plane or generating color mapped images like what we did with SHy-DSS
images. A more advanced color mapping method called Spectrally Encoded Enhanced
81

Representations (SEER) can also be used to further enhance the spectral difference on the apple
peel to facilitate interpretation of inspection results.
Fig. 4.3 SHy-PC concept and workflow. (a) SHy-PC is a handheld imaging device based on three separate camera
modules. The left module serves as the COS channel which has a built-in COS sinusoidal filter. The middle camera module
is the intensity channel and has a bandpass filter. The right camera module is the SIN channel with a SIN sinusoidal filter.
SHy-PC is designed to work within 700nm-1000nm NIR range and both the sinusoidal filters have the transmission profile
defined in this spectral range. The cross section (A-B) image to the right shows the main components inside of the SIN
camera module. The lens is installed on a voice coil motor (VCM) that controls the focus of the camera. The sinusoidal
filter is installed behind the lens and in front of the NIR CMOS sensor. In the intensity camera module, the 700-1000nm
bandpass filter is installed instead. (b) The transmission profiles of the SIN and COS filters. In addition, the filters reject the
light coming below or above this range. (c) The workflow of SHy-PC includes acquisition, processing, and analysis which is
similar to SHy-Cam and DSS.
82

References
1. E. Hecht, "Optics (Fifth Edition)," 704 (2015).
2. L. R. F.R.S., "XXXI. Investigations in optics, with special reference to the spectroscope,"
https://doi.org/10.1080/14786447908639684 8, 261–274 (2009).
3. M. Ikawa, K. Kominami, Y. Yoshimura, K. Tanaka, Y. Nishimune, and M. Okabe, "Green
fluorescent protein as a marker in transgenic mice," Dev Growth Differ 37, 455–459
(1995).
4. J. Liao, T. Yan, C. L. Hew, T. Jin Lam, and Z. Gong, "Faithful Expression of Green
Fluorescent Protein (GFP) in Transgenic Zebrafish Embryos Under Control of Zebrafish
Gene Promoters," Dev. Genet 25, 158–167 (1999).
5. B. Valeur and M. N. Berberan-Santos, Molecular Fluorescence: Principles and
Applications, Second Edition (Wiley-VCH, 2012).
6. M. Monici, "Cell and tissue autofluorescence research and diagnostic applications,"
Biotechnol Annu Rev 11, 227–256 (2005).
7. M. E. Dickinson, G. Bearman, S. Tille, R. Lansford, and S. E. Fraser, "Multi-spectral
imaging and linear unmixing add a whole new dimension to laser scanning fluorescence
microscopy," Biotechniques (2001).
8. T. Zimmermann, "Spectral imaging and linear unmixing in light microscopy," Adv
Biochem Eng Biotechnol 95, 245–265 (2005).
9. F. Fereidouni, A. N. Bader, and H. C. Gerritsen, "Spectral phasor analysis allows rapid
and reliable unmixing of fluorescence microscopy spectral images," Opt Express (2012).
10. L. M. Andrews, M. R. Jones, M. A. Digman, and E. Gratton, "Spectral phasor analysis of
Pyronin Y labeled RNA microenvironments in living cells," Biomed Opt Express (2013).
11. F. Cutrale, A. Salih, and E. Gratton, "Spectral phasor approach for fingerprinting of photo-
activatable fluorescent proteins Dronpa, Kaede and KikGR," Methods Appl Fluoresc
(2013).
12. F. Cutrale, V. Trivedi, L. A. Trinh, C.-L. Chiu, J. M. Choi, M. S. Artiga, and S. E. Fraser,
"Hyperspectral phasor analysis enables multiplexed 5D in vivo imaging," Nat Methods
14, 149–152 (2017).
13. J. Haringsma, G. N. J. Tytgat, H. Yano, H. Iishi, M. Tatsuta, T. Ogihara, H. Watanabe, N.
Sato, N. Marcon, B. C. Wilson, and R. W. Cline, "Autofluorescence endoscopy: feasibility
of detection of GI neoplasms unapparent to white light endoscopy with an evolving
technology," Gastrointest Endosc 53, 642–650 (2001).
14. K. Niepsuj, G. Niepsuj, W. Cebula, W. Zieleźnik, M. Adamek, A. Sielanóczyk, J.
Adamczyk, J. Kurek, and A. Sieroń, "Autofluorescence endoscopy for detection of high-
83

grade dysplasia in short-segment Barrett’s esophagus," Gastrointest Endosc 58, 715–719
(2003).
15. B. Mayinger, P. Horner, M. Jordan, C. Gerlach, T. Horbach, W. Hohenberger, and E. G.
Hahn, "Light-induced autofluorescence spectroscopy for the endoscopic detection of
esophageal cancer," Gastrointest Endosc (2001).
16. R. S. DaCosta, B. C. Wilson, and N. E. Marcon, "Spectroscopy and fluorescence in
esophageal diseases," Best Pract Res Clin Gastroenterol (2006).
17. M. Panjehpour, B. F. Overholt, J. L. Schmidhammer, C. Farris, P. F. Buckley, and T. Vo-
Dinh, "Spectroscopic diagnosis of esophageal cancer: New classification model, improved
measurement system," Gastrointest Endosc (1995).
18. L. Song, E. J. Hennink, I. T. Young, and H. J. Tanke, "Photobleaching kinetics of
fluorescein in quantitative fluorescence microscopy," Biophys J 68, 2588–2600 (1995).
19. A. Diaspro, G. Chirico, C. Usai, P. Ramoino, and J. Dobrucki, "Photobleaching,"
Handbook of Biological Confocal Microscopy: Third Edition 690–702 (2006).
20. F. H. C. Wong, J. S. Ng-Kamstra, N. L. H. Chen, and C. Fradin, "Localized photodamage
of the human erythrocyte membrane causes an invagination as a precursor of
photohaemolysis," J Microsc 226, 6–17 (2007).
21. D. Satsoura, B. Leber, D. W. Andrews, and C. Fradin, "Circumvention of Fluorophore
Photobleaching in Fluorescence Fluctuation Experiments: A Beam Scanning Approach,"
ChemPhysChem 8, 834–848 (2007).
22. G. Weber and F. W. J. Teale, "Determination of the absolute quantum yield of fluorescent
solutions," Transactions of the Faraday Society 53, 646–655 (1957).
23. W. Denk, J. H. Strickler, and W. W. Webb, "Two-Photon Laser Scanning Fluorescence
Microscopy," Science (1979) 248, 73–76 (1990).
24. G. H. Patterson and D. W. Piston, "Photobleaching in Two-Photon Excitation
Microscopy," Biophys J 78, 2159–2162 (2000).
25. T. K. L. Meyvis, S. C. de Smedt, P. van Oostveldt, and J. Demeester, "Fluorescence
Recovery After Photobleaching: A Versatile Tool for Mobility and Interaction
Measurements in Pharmaceutical Research," Pharmaceutical Research 1999 16:8 16,
1153–1162 (1999).
26. D. Wüstner, L. M. Solanko, F. W. Lund, D. Sage, H. J. Schroll, and M. A. Lomholt,
"Quantitative fluorescence loss in photobleaching for analysis of protein transport and
aggregation," BMC Bioinformatics 13, 1–21 (2012).
27. W. T. Welford, "Advanced Light Microscopy. Vol. 1. Principles and Basic Properties,"
Advanced Light Microscopy 1, 1546–1547 (1989).
28. M. A. Browne, "Advanced Light Microscopy, Vol. 2 (Specialised Methods). By
Maksymilian Pluta," J Microsc 158, 416–416 (1990).
84

29. Savile. Bradbury, Brian. Bracegirdle, and Royal Microscopical Society (Great Britain),
"Introduction to light microscopy," 123 (1998).
30. A. Dvornikov and E. Gratton, "Hyperspectral imaging in highly scattering media by the
spectral phasor approach using two filters," Biomed Opt Express (2018).
31. P. N. Hedde, R. Cinco, L. Malacrida, A. Kamaid, and E. Gratton, "Phasor-based
hyperspectral snapshot microscopy allows fast imaging of live, three-dimensional tissues
for biomedical applications," Commun Biol (2021).
32. "Microscopy Resource Center | Olympus LS," https://www.olympus-
lifescience.com/en/microscope-resource/.
33. "Nikon’s MicroscopyU - The Source for Microscopy Education,"
https://www.microscopyu.com/.
34. "Microscopy Basics | Science Lab | Leica Microsystems," https://www.leica-
microsystems.com/science-lab/topics/basics-in-microscopy/.
35. "ZEISS Microscopy Online Campus | Introduction to Spectral Imaging," https://zeiss-
campus.magnet.fsu.edu/articles/spectralimaging/introduction.html.
36. W. Jahr, B. Schmid, C. Schmied, F. O. Fahrbach, and J. Huisken, "Hyperspectral light
sheet microscopy," Nat Commun 6, 1–7 (2015).
37. P. Mahou, M. Zimmerley, K. Loulier, K. S. Matho, G. Labroille, X. Morin, W. Supatto, J.
Livet, D. Débarre, and E. Beaurepaire, "Multicolor two-photon tissue imaging by
wavelength mixing," Nat Methods 9, 815–818 (2012).
38. A. M. Valm, S. Cohen, W. R. Legant, J. Melunis, U. Hershberg, E. Wait, A. R. Cohen, M.
W. Davidson, E. Betzig, and J. Lippincott-Schwartz, "Applying systems-level spectral
imaging and analysis to reveal the organelle interactome," Nature 546, 162–167 (2017).
39. R. M. Levenson and J. R. Mansfield, "Multispectral imaging in biology and medicine:
Slices of life," Cytometry Part A 69, 748–758 (2006).
40. Z. Lavagnino, J. Dwight, A. Ustione, T. U. Nguyen, T. S. Tkaczyk, and D. W. Piston,
"Snapshot Hyperspectral Light-Sheet Imaging of Signal Transduction in Live Pancreatic
Islets," Biophys J 111, 409–417 (2016).
41. J. Huisken, J. Swoger, F. del Bene, J. Wittbrodt, and E. H. K. Stelzer, "Optical sectioning
deep inside live embryos by selective plane illumination microscopy," Science (1979)
(2004).
42. J. Huisken and D. Y. R. Stainier, "Selective plane illumination microscopy techniques in
developmental biology," Development (2009).
43. R. M. Power and J. Huisken, "A guide to light-sheet fluorescence microscopy for
multiscale imaging," Nat Methods (2017).
44. K. Keomanee-Dizon, S. E. Fraser, and T. V. Truong, "A versatile, multi-laser twin-
microscope system for light-sheet imaging," Rev Sci Instrum (2020).
85

45. R. K. Chhetri, F. Amat, Y. Wan, B. Höckendorf, W. C. Lemon, and P. J. Keller, "Whole-
animal functional and developmental imaging with isotropic spatial resolution," Nat
Methods (2015).
46. M. Mickoleit, B. Schmid, M. Weber, F. O. Fahrbach, S. Hombach, S. Reischauer, and J.
Huisken, "High-resolution reconstruction of the beating zebrafish heart," Nat Methods
(2014).
47. P. Mahou, J. Vermot, E. Beaurepaire, and W. Supatto, "Multicolor two-photon light-sheet
microscopy," Nat Methods (2014).
48. N. Bedard, "Image mapping spectrometry: calibration and characterization," Optical
Engineering (2012).
49. Y. Zhang, L. K. Schroeder, M. D. Lessard, P. Kidd, J. Chung, Y. Song, L. Benedetti, Y.
Li, J. Ries, J. B. Grimm, L. D. Lavis, P. de Camilli, J. E. Rothman, D. Baddeley, and J.
Bewersdorf, "Nanoscale subcellular architecture revealed by multicolor three-dimensional
salvaged fluorescence imaging," Nat Methods 17, (2020).
50. T. V. C. Truong, W. Supatto, D. S. Koos, J. M. Choi, and S. E. Fraser, "Deep and fast live
imaging with two-photon scanned light-sheet microscopy," Nat Methods 8, 757–762
(2011).
51. V. Trivedi, T. V. Truong, L. A. Trinh, D. B. Holland, M. Liebling, and S. E. Fraser,
"Dynamic structure and protein expression of the live embryonic heart captured by 2-
photon light sheet microscopy and retrospective registration," Biomed Opt Express
(2015).
52. W. Shi, D. E. S. Koo, M. Kitano, H. J. Chiang, L. A. Trinh, G. Turcatel, B. Steventon, C.
Arnesano, D. Warburton, S. E. Fraser, and F. Cutrale, "Pre-processing visualization of
hyperspectral fluorescent data with Spectrally Encoded Enhanced Representations," Nat
Commun (2020).
53. D. Sancho, O. P. Joffre, A. M. Keller, N. C. Rogers, D. Martínez, P. Hernanz-Falcón, I.
Rosewell, and C. R. E. Sousa, "Identification of a dendritic cell receptor that couples
sensing of necrosis to immunity," Nature (2009).
54. H. T. K. Tse, D. R. Gossett, Y. S. Moon, M. Masaeli, M. Sohsman, Y. Ying, K. Mislick,
R. P. Adams, J. Rao, and D. Di Carlo, "Quantitative diagnosis of malignant pleural
effusions by single-cell mechanophenotyping," Sci Transl Med (2013).
55. K. S. Ralston, M. D. Solga, N. M. MacKey-Lawrence, Somlata, A. Bhattacharya, and W.
A. Petri, "Trogocytosis by Entamoeba histolytica contributes to cell killing and tissue
invasion," Nature (2014).
56. M. Maryanovich, Y. Zaltsman, A. Ruggiero, A. Goldman, L. Shachnai, S. L. Zaidman, Z.
Porat, K. Golan, T. Lapidot, and A. Gross, "An MTCH2 pathway repressing mitochondria
metabolism regulates haematopoietic stem cell fate," Nat Commun (2015).
57. D. Basiji and M. R. G. O’Gorman, "Imaging flow cytometry," J Immunol Methods
(2015).
86

58. O. Otto, P. Rosendahl, A. Mietke, S. Golfier, C. Herold, D. Klaue, S. Girardo, S. Pagliara,
A. Ekpenyong, A. Jacobi, M. Wobus, N. Töpfner, U. F. Keyser, J. Mansfeld, E. Fischer-
Friedrich, and J. Guck, "Real-time deformability cytometry: On-the-fly cell mechanical
phenotyping," Nat Methods (2015).
59. D. B. Sykes, Y. S. Kfoury, F. E. Mercier, M. J. Wawer, J. M. Law, M. K. Haynes, T. A.
Lewis, A. Schajnovitz, E. Jain, D. Lee, H. Meyer, K. A. Pierce, N. J. Tolliday, A. Waller,
S. J. Ferrara, A. L. Eheim, D. Stoeckigt, K. L. Maxcy, J. M. Cobert, J. Bachand, B. A.
Szekely, S. Mukherjee, L. A. Sklar, J. D. Kotz, C. B. Clish, R. I. Sadreyev, P. A. Clemons,
A. Janzer, S. L. Schreiber, and D. T. Scadden, "Inhibition of Dihydroorotate
Dehydrogenase Overcomes Differentiation Blockade in Acute Myeloid Leukemia," Cell
(2016).
60. I. A. Vorobjev and N. S. Barteneva, Imaging Flow Cytometry Methods and Protocols
(2016).
61. H. Mikami, M. Kawaguchi, C. J. Huang, H. Matsumura, T. Sugimura, K. Huang, C. Lei,
S. Ueno, T. Miura, T. Ito, K. Nagasawa, T. Maeno, H. Watarai, M. Yamagishi, S.
Uemura, S. Ohnuki, Y. Ohya, H. Kurokawa, S. Matsusaka, C. W. Sun, Y. Ozeki, and K.
Goda, "Virtual-freezing fluorescence imaging flow cytometry," Nat Commun (2020).
62. A. D. Edelstein, M. A. Tsuchida, N. Amodaj, H. Pinkard, R. D. Vale, and N. Stuurman,
"Advanced methods of microscope control using μManager software," J Biol Methods
(2014).
63. A. Edelstein, N. Amodaj, K. Hoover, R. Vale, and N. Stuurman, "Computer control of
microscopes using manager," Curr Protoc Mol Biol (2010).
64. J. A. Bogovic, P. Hanslovsky, A. Wong, and S. Saalfeld, "Robust registration of calcium
images by learned contrast synthesis," in Proceedings - International Symposium on
Biomedical Imaging (2016).
65. A. Goshtasby, "Image registration by local approximation methods," Image Vis Comput
(1988).
66. I. G. Goldberg, C. Allan, J. M. Burel, D. Creager, A. Falconi, H. Hochheiser, J. Johnston,
J. Mellen, P. K. Sorger, and J. R. Swedlow, "The Open Microscopy Environment (OME)
Data Model and XML file: open tools for informatics and quantitative analysis in
biological imaging.," Genome Biol (2005).
67. J. Schindelin, I. Arganda-Carreras, E. Frise, V. Kaynig, M. Longair, T. Pietzsch, S.
Preibisch, C. Rueden, S. Saalfeld, B. Schmid, J. Y. Tinevez, D. J. White, V. Hartenstein,
K. Eliceiri, P. Tomancak, and A. Cardona, "Fiji: An open-source platform for biological-
image analysis," Nat Methods (2012).
68. N. Hedde, R. Cinco, L. Malacrida, A. Kamaid, and E. Gratton, "Ultrafast phasor-based
hyperspectral snapshot microscopy for biomedical imaging," bioRxiv (2020).
69. R. T. H. Lee, P. V. Asharani, and T. J. Carney, "Basal keratinocytes contribute to all strata
of the adult zebrafish epidermis," PLoS One (2014).
87

70. K. Proulx, A. Lu, and S. Sumanas, "Cranial vasculature in zebrafish forms by angioblast
cluster-derived angiogenesis," Dev Biol (2010).
71. L. A. Trinh, V. Chong-Morrison, D. Gavriouchkina, T. Hochgreb-Hägele, U. Senanayake,
S. E. Fraser, and T. Sauka-Spengler, "Biotagging of Specific Cell Populations in Zebrafish
Reveals Gene Regulatory Logic Encoded in the Nuclear Transcriptome," Cell Rep (2017).
72. R. M. White, A. Sessa, C. Burke, T. Bowman, J. LeBlanc, C. Ceol, C. Bourque, M.
Dovey, W. Goessling, C. E. Burns, and L. I. Zon, "Transparent Adult Zebrafish as a Tool
for In Vivo Transplantation Analysis," Cell Stem Cell (2008).
73. D. M. Parichy, D. G. Ransom, B. Paw, L. I. Zon, and S. L. Johnson, "An orthologue of the
kit-related gene fms is required for development of neural crest-derived xanthophores and
a subpopulation of adult melanocytes in the zebrafish, Danio rerio," Development (2000).
74. D. Huss, B. Benazeraf, A. Wallingford, M. Filla, J. Yang, S. E. Fraser, and R. Lansford,
"A transgenic quail model that enables dynamic imaging of amniote embryogenesis,"
Development (Cambridge) (2015).
75. K. M. Kwan, E. Fujimoto, C. Grabher, B. D. Mangum, M. E. Hardy, D. S. Campbell, J.
M. Parant, H. J. Yost, J. P. Kanki, and C. Bin Chien, "The Tol2kit: A multisite gateway-
based construction Kit for Tol2 transposon transgenesis constructs," Developmental
Dynamics (2007).
76. K. Kawakami, H. Takeda, N. Kawakami, M. Kobayashi, N. Matsuda, and M. Mishina, "A
transposon-mediated gene trap approach identifies developmentally regulated genes in
zebrafish," Dev Cell (2004).
77. A. Urasaki, G. Morvan, and K. Kawakami, "Functional dissection of the Tol2
transposable element identified the minimal cis-sequence and a highly repetitive sequence
in the subterminal region essential for transposition," Genetics (2006).
78. A. G. Uren, H. Mikkers, J. Kool, L. Van Der Weyden, A. H. Lund, C. H. Wilson, R.
Rance, J. Jonkers, M. Van Lohuizen, A. Berns, and D. J. Adams, "A high-throughput
splinkerette-pcr method for the isolation and sequencing of retroviral insertion sites," Nat
Protoc (2009).
79. L. A. Trinh, T. Hochgreb, M. Graham, D. Wu, F. Ruf-Zamojski, C. S. Jayasena, A.
Saxena, R. Hawk, A. Gonzalez-Serricchio, A. Dixson, E. Chow, C. Gonzales, H. Y.
Leung, I. Solomon, M. Bronner-Fraser, S. G. Megason, and S. E. Fraser, "A versatile gene
trap to visualize and interrogate the function of the vertebrate proteome," Genes Dev
(2011).
80. P. Wang, M. Kitano, K. Keomanee-Dizon, T. v. Truong, S. E. Fraser, and F. Cutrale,
"SHy-Cam: A Single-Shot Hyperspectral Phasor Camera for Fast, Multi-Color
Fluorescence Microscopy," SSRN Electronic Journal (2022).
81. H. Sung, J. Ferlay, R. L. Siegel, M. Laversanne, I. Soerjomataram, A. Jemal, and F. Bray,
"Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality
Worldwide for 36 Cancers in 185 Countries," CA Cancer J Clin (2021).
88

82. C. Lepage, B. Rachet, V. Jooste, J. Faivre, and M. P. Coleman, "Continuing rapid increase
in esophageal adenocarcinoma in England and Wales," American Journal of
Gastroenterology (2008).
83. H. Pohl and H. G. Welch, "The role of overdiagnosis and reclassification in the marked
increase of esophageal adenocarcinoma incidence," J Natl Cancer Inst (2005).
84. M. Barret and F. Prat, "Diagnosis and treatment of superficial esophageal cancer," Ann
Gastroenterol (2018).
85. S. Jain and S. Dhingra, "Pathology of esophageal cancer and Barrett’s esophagus," Ann
Cardiothorac Surg 6, 99 (2017).
86. R. D. Odze, "Diagnosis and grading of dysplasia in Barrett’s oesophagus," J Clin Pathol
59, 1029 (2006).
87. M. Monici, "Cell and tissue autofluorescence research and diagnostic applications,"
Biotechnol Annu Rev 11, 227–256 (2005).
88. J. R. Albani, Principles and Applications of Fluorescence Spectroscopy (John Wiley &
Sons, 2008).
89. M. A. Kara, R. S. Dacosta, C. J. Streutker, N. E. Marcon, J. J. G. H. M. Bergman, and B.
C. Wilson, "Characterization of tissue autofluorescence in Barrett’s esophagus by
confocal fluorescence microscopy," Diseases of the Esophagus 20, 141–150 (2007).
90. P. Wang, M. Kitano, K. K. Dizon, T. v. Truong, S. E. Fraser, and F. Cutrale, "Fourier
transform based high speed snapshot multi-color fluorescence acquisition,"
https://doi.org/10.1117/12.2609295 PC11966, PC119660D (2022).
91. R. M. Levenson and J. R. Mansfield, "Multispectral imaging in biology and medicine:
Slices of life," Cytometry Part A 69, 748–758 (2006).
92. Z. Lavagnino, J. Dwight, A. Ustione, T. U. Nguyen, T. S. Tkaczyk, and D. W. Piston,
"Snapshot Hyperspectral Light-Sheet Imaging of Signal Transduction in Live Pancreatic
Islets," Biophys J 111, 409–417 (2016).
93. W. Jahr, B. Schmid, C. Schmied, F. O. Fahrbach, and J. Huisken, "Hyperspectral light
sheet microscopy," Nature Communications 2015 6:1 6, 1–7 (2015).
94. P. Mahou, M. Zimmerley, K. Loulier, K. S. Matho, G. Labroille, X. Morin, W. Supatto, J.
Livet, D. Débarre, and E. Beaurepaire, "Multicolor two-photon tissue imaging by
wavelength mixing," Nature Methods 2012 9:8 9, 815–818 (2012).
95. A. M. Valm, S. Cohen, W. R. Legant, J. Melunis, U. Hershberg, E. Wait, A. R. Cohen, M.
W. Davidson, E. Betzig, and J. Lippincott-Schwartz, "Applying systems-level spectral
imaging and analysis to reveal the organelle interactome," Nature 2017 546:7656 546,
162–167 (2017).
96. "How to acquire flat-field correction images | Center for Advanced Light Microscopy,"
https://calm.ucsf.edu/how-acquire-flat-field-correction-images.
89

97. M. Model, "Intensity calibration and flat-field correction for fluorescence microscopes,"
Curr Protoc Cytom (2014).
98. H. Tian, B. Fowler, and A. el Gamal, "Analysis of temporal noise in CMOS photodiode
active pixel sensor," IEEE J Solid-State Circuits 36, 92–101 (2001).
99. M. Linkert, C. T. Rueden, C. Allan, J. M. Burel, W. Moore, A. Patterson, B. Loranger, J.
Moore, C. Neves, D. MacDonald, A. Tarkowska, C. Sticco, E. Hill, M. Rossner, K. W.
Eliceiri, and J. R. Swedlow, "Metadata matters: access to image data in the real world," J
Cell Biol 189, 777–782 (2010).
100. I. G. Goldberg, C. Allan, J.-M. Burel, D. Creager, A. Falconi, H. Hochheiser, J. Johnston,
J. Mellen, P. K. Sorger, and J. R. Swedlow, "The Open Microscopy Environment (OME)
Data Model and XML file: open tools for informatics and quantitative analysis in
biological imaging," Genome Biol 6, R47 (2005).
101. H. Bay, A. Ess, T. Tuytelaars, and L. van Gool, "Speeded-Up Robust Features (SURF),"
Computer Vision and Image Understanding 110, 346–359 (2008).
102. L. S. Hart and W. S. El-Deiry, "Invincible, but Not Invisible: Imaging Approaches
Toward In Vivo Detection of Cancer Stem Cells," J Clin Oncol 26, 2901–2910 (2008).
103. E. Vlashi, K. Kim, C. Lagadec, L. della Donna, J. T. McDonald, M. Eghbali, J. W. Sayre,
E. Stefani, W. McBride, and F. Pajonk, "In Vivo Imaging, Tracking, and Targeting of
Cancer Stem Cells," JNCI: Journal of the National Cancer Institute 101, 350–359 (2009).
104. H. Ghaffari, M. R. Atashzar, and H. Abdollahi, "Molecular imaging in tracking cancer
stem cells: A review," Med J Islam Repub Iran 34, 90 (2020).
105. S. E. Freeman, H. Hacham, R. W. Gange, D. J. Maytum, J. C. Sutherland, and B. M.
Sutherland, "Wavelength dependence of pyrimidine dimer formation in DNA of human
skin irradiated in situ with ultraviolet light.," Proc Natl Acad Sci U S A 86, 5605 (1989).
106. J. A. Udovich, N. D. Kirkpatrick, A. Kano, A. Tanbakuchi, U. Utzinger, and A. F. Gmitro,
"Spectral background and transmission characteristics of fiber optic imaging bundles,"
Applied Optics, Vol. 47, Issue 25, pp. 4560-4568 47, 4560–4568 (2008).
107. N. L. Rochefort, H. Jia, and A. Konnerth, "Calcium imaging in the living brain: prospects
for molecular medicine," Trends Mol Med 14, 389–399 (2008).
108. L. C. Leung, G. X. Wang, R. Madelaine, G. Skariah, K. Kawakami, K. Deisseroth, A. E.
Urban, and P. Mourrain, "Neural signatures of sleep in zebrafish," Nature 571, 198
(2019).
109. "Apples | Agricultural Marketing Resource Center," https://www.agmrc.org/commodities-
products/fruits/apples.
110. C. Tang, H. He, E. Li, and H. Li, "Multispectral imaging for predicting sugar content of
‘Fuji’ apples," Opt Laser Technol 106, 280–285 (2018).
111. X.-Y. Chen, W.-T. Chen, J.-S. Lv, X. Long, and T. Pang, "Apple Internal Quality
Inspection Using Hyperspectral Image Technology," 768–774 (2016).
90

112. "Angular Magnification," (n.d.).
113. "ZEISS Microscopy Online Campus | Microscopy Basics | Numerical Aperture and
Resolution," https://zeiss-campus.magnet.fsu.edu/articles/basics/resolution.html.
114. "File:Jablonski Diagram of Fluorescence Only.png - Wikimedia Commons,"
https://commons.wikimedia.org/wiki/File:Jablonski_Diagram_of_Fluorescence_Only.png.
115. "Jablonski diagram - Chemistry LibreTexts,"
https://chem.libretexts.org/Bookshelves/Physical_and_Theoretical_Chemistry_Textbook_
Maps/Supplemental_Modules_(Physical_and_Theoretical_Chemistry)/Spectroscopy/Elect
ronic_Spectroscopy/Jablonski_diagram.

91

Appendices

A Part list and CAD diagram of SHy-Cam-SPIM

92

Fig.A 1 The CAD model of SHy-CAM adapted to hSPIM in Chapter 2. Part numbers shown in this figure correspond to the
part list (Table A1). a#: optical components; b#: opto-mechanical components available from vendors; c#: 3D printed opto-
mechanical components; d#: fasteners; e#: components used for table mounting.

93

Table A 1 A SHy-Cam part list
Item Category Assembly
part
number
Acronym
(Fig. A1)
Quantity Manufacture
part number
Manufacture
Ø2"
Achromatic
doublet,
f=150mm,
400-700nm
Optics a1 𝑅𝐿
1
1 AC508-150-A Thorlabs
Ø2"
Achromatic
doublet,
f=75mm,
400-700nm
Optics a2 𝑅𝐿
2
1
AC508-075-A

Thorlabs
Broadband
Dielectric
Mirror, 400-
750nm
Optics a3 𝐺𝑀 1−4

4 BB1-E02 -
Ø1"
Thorlabs
Ø1"
Achromatic
doublet,
f=175mm,
400-700nm
Optics a4 𝑇𝐿
1−4

4 32-884 Edmund
Optics
30mm Cage
Cube-
Mounted
Dielectric
Tunning
Mirror,400-
750nm
Optics a5 𝑅𝑀
1−3

3 CCM1-E02 Thorlabs
50:50
Unpolarized
Beam
splitter, 400-
700nm
Optics a6 𝐵𝑆

1 BSW10R Thorlabs
Custom Sine
Dichroic
Mirror, 400-
700nm
Optics a7 𝐷𝑀
𝑆

1 Chroma
Custom
Cosine
Dichroic
Mirror, 400-
700nm
Optics a8 𝐷𝑀
𝐶

1 Chroma
Ø25mm
unmounted
380nm Long-
Pass Filter
Optics a9 𝐹 𝐿𝑃 380

1 FF01-
380/LP25
Semrock
Ø25mm
unmounted
715nm
Optics
a10

𝐹 𝑆𝑃 715

1
FF01-715/SP-
25

Semrock
94

Short-Pass
Filter
Ø25mm
unmounted
dual-notch
filter,
488/647nm
Optics a11 𝐹 𝑁 641

1 NF01-
488/647-
25x5.0
Semrock
Ø25mm
unmounted
single-notch
filter, 561nm
Optics a12 𝐹 𝑁 641

1
NF03-561E-25

Semrock
Ø1'' SM1 Lens
Tube, 1,00''
Thread
Depth

Opto-
mechanics

b1 𝐿𝑇
𝐹

1
SM1L10

Thorlabs
30mm to
60mm Cage
Plate
Adapter,
0.5'' thick
Opto-
mechanics

b2 𝐶𝑃𝐴

1
LCP02

Thorlabs
Cage
Assembly
Rod, 6''
Long, Ø6mm
Opto-
mechanics

b3 𝐸𝑅 6

4
ER6

Thorlabs
Cage
Assembly
Rod, 4''
Long, Ø6mm
Opto-
mechanics

b4 𝐸𝑅 4

14
ER4

Thorlabs
Cage
Assembly
Rod, 1.5''
Long, Ø6mm
Opto-
mechanics

b5 𝐸𝑅 1_5

2 ER1.5 Thorlabs
Cage
Assembly
Rod, 1''
Long, Ø6mm
Opto-
mechanics

b6 𝐸𝑅 1

10 ER1 Thorlabs
Cage
Assembly
Turret
(Gimbal)
Mount
Opto-
mechanics

b7 𝐺𝑀𝑀

4
KC45D

Thorlabs
30mm Cage
Cube with
Dichroic
Filter Mount
Opto-
mechanics

b8 𝐷𝐶𝑀

3 CM1-DCH Thorlabs
Ø6mm Cage
Assembly
Rod Right
Angle
Bracket
Opto-
mechanics

b9 𝑅𝑅𝐵

4
ER90B

Thorlabs
60mm Cage
Plate, SM2
Threads,
0.5'' Thick
Opto-
mechanics

b10 𝐿𝐶𝑃 01

1 LCP01 Thorlabs
95

SM2
(external) to
SM1
(internal)
Thread
Adapter
Opto-
mechanics

b11 𝑆𝑀 2𝐴 6

1
SM2A6

Thorlabs
SM2 Lens
Tube. 0.5''
Thread
Depth
Opto-
mechanics

b12 𝑆𝑀 2𝐿 05

1
SM2L05

Thorlabs
SM1
Graduated
Ring-
Actuated Iris
Diaphragm
Opto-
mechanics

b13 𝐹𝑆

1 SM1D12C Thorlabs
2'' Lens Tube
for Relay
Lens for RL2
Opto-
mechanics
(3D
Printed)
c1 𝐿𝑇
𝑅𝐿 2
1 3D Printed
2'' Lens Tube
and Cube
Mating Plate
Opto-
mechanics
(3D
Printed)
c2 𝑀𝑃 1 3D Printed
Dichroic
Cube Spacer,
1.2mm thick
Opto-
mechanics
(3D
Printed)
c3 𝐶𝑆𝑃 1.2

1 3D Printed
Dichroic
Cube Spacer,
2.7mm thick
Opto-
mechanics
(3D
Printed)
c4 𝐶𝑆𝑃 2.7

1 3D Printed
Cube
Mounting
Plate 1
Opto-
mechanics
(3D
Printed)
c5 𝐶𝑀𝑃 1
1 3D Printed
Cube
Mounting
Plate 2
Opto-
mechanics
(3D
Printed)
c6 𝐶𝑀𝑃 2
1 3D Printed
Cube
Mounting
Plate 3
Opto-
mechanics
(3D
Printed)
c7 𝐶𝑀𝑃 3
1 3D Printed
Cube
Mounting
Plate 4
Opto-
mechanics
(3D
Printed)
c8 𝐶𝑀𝑃 4
1 3D Printed
Tube Lens
Mount
Opto-
mechanics
(3D
Printed)
c9 𝑇𝐿𝑀 1 3D Printed
96

Tube Lens
Mount and
Camera
Mount
Mating Plate
Opto-
mechanics
(3D
Printed)
c10 𝑇𝐶𝑀𝑃

1 3D Printed

Camera
Mount (PCO
Edge 5.5)
Opto-
mechanics
(3D
Printed)
c11 𝐶𝑀 1 3D Printed
Cage rod
supporting
plate
Opto-
mechanics
(3D
Printed)
c12 𝐶𝑅𝑀𝑃

1 3D Printed
Modified
gimbal
mirror rod
mount
Opto-
mechanics
(3D
Printed)
c13 𝑀𝐺𝑀𝑅𝑀

1 3D Printed
Modified
gimbal
mirror
mount
Opto-
mechanics
(3D
Printed)
c14 𝑀𝐺𝑀𝑀

1 3D Printed
4-40 Cap
Screw, 1/4''
Long
Fastener

d1 𝑆𝐻 4𝑆 025

46 SH4S025 Thorlabs
8-32 Cap
Screw, 3/8''
Long
Fastener

d2 𝑆𝐻 8𝑆 038

4 SH8S038 Thorlabs
4-40 Nylon-
Tipped
Setscrew,
1/8'' Long
Fastener

d3 𝑆𝑆 4𝑁 013

2 SS4N013 Thorlabs
Ø1/2"
Optical Post,
3'' Long
Table
mounting
e1 𝑇𝑅 3 4 TR3 Thorlabs
Ø1/2"
Pedestal
Post Holder,
3.19'' Long
Table
mounting
e2 𝑃𝐻 3𝐸

4 PH3E Thorlabs
30 mm Cage
Mounting
Bracket
Table
mounting
e3 𝐶𝑃𝑀𝐴 2

2 CPMA2 Thorlabs
Ø1/2"
Optical Post
Holder
Clamping
Fork
Table
mounting
e4 𝑃𝐶𝐹

4 CF125 Thorlabs

Abstract (if available)

Abstract This thesis summarizes my efforts in defining a new imaging method for faster, more efficient multi-color fluorescence imaging with a lower instrumentation cost. Chapter 1 provides the readers with a brief introduction of light microscopes, fluorescence and existing fluorescence imaging methods and their limitations, specifically for multi-color imaging applications.
Chapter 2 presents the efforts to build an optical imaging module for a selective plane illumination microscope (SPIM) to achieve fast fluorescence multiplexing. The system was named Single-Shot Hyperspectral Phasor Camera (SHy-Cam) and validated with in-vivo imaging experiments with transgenic multi-color zebrafish embryos. Chapter 3 extends the concept of SHy-Cam from biological imaging applications to medical applications of diagnostic imaging with label-free biopsy slides based-on tissue autofluorescence. Chapter 4 summarizes the research findings and further discusses the potential future applications of SHy-Cam.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Extending multiplexing capabilities with lifetime and hyperspectral fluorescence imaging

PDF

Multiplexing live 5d imaging with multispectral fluorescence: Advanced unmixing through simulation and machine learning

PDF

Machine learning and image processing of fluorescence lifetime imaging microscopy enables tracking and analysis of subcellular metabolism

PDF

Analytical tools for complex multidimensional biological imaging data

PDF

Democratizing optical biopsy with diffuse optical imaging

PDF

Metabolic profiling of single hematopoietic stem cells for developing novel ex vivo culture strategies

PDF

Development of a quantitative proteomic molecular imaging platform using multiplexed nanoscale Raman scattering contrast agents

PDF

Hyperspectral optical imaging for detection, diagnosis and staging of cancer

PDF

Hyperspectral and multispectral optical bioluminescence and fluorescence tomography in small animal imaging

PDF

Neural network integration of multiscale and multimodal cell imaging using semantic parts

PDF

Machine learning based techniques for biomedical image/video analysis

PDF

A fluorescence microscopy study of quantum dots as fluorescent probes for brain tumor diagnosis

PDF

Exploring the role of vegfr transcripts during vascular development in the zebrafish embryo

PDF

Wastewater-based epidemiology for emerging biological contaminants

PDF

Process implications of executable domain models for microservices development

PDF

A polynomial chaos formalism for uncertainty budget assessment

PDF

Imaging cortical noradrenergic dynamics with genetically-engineered fluorescent indicators

PDF

Learning from limited and imperfect data for brain image analysis and other biomedical applications

PDF

Improving language understanding and summarization by leveraging auxiliary information through self-supervised or unsupervised learning

PDF

Improving arterial spin labeling in clinical application with deep learning

Asset Metadata

Creator Wang, Pu (author)

Core Title Hyperspectral phasor for multiplexed fluorescence microscopy and autofluorescence-based pathologic diagnosis

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Biomedical Engineering

Degree Conferral Date 2022-12

Publication Date 09/17/2022

Defense Date 07/07/2022

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag autofluorescence,diagnostic imaging,fluorescence microscopy,hyperspectral fluorescence microscopy,hyperspectral imaging,light microscopy,OAI-PMH Harvest

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Fraser, Scott E. (committee chair), Cutrale, Francesco (committee member), Lansford, Rusty (committee member), Zavaleta, Cristina (committee member)

Creator Email puw@usc.edu,puwusc@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111996067

Unique identifier UC111996067

Legacy Identifier etd-WangPu-11218

Document Type Dissertation

Format application/pdf (imt)

Rights Wang, Pu

Type texts

Source 20220917-usctheses-batch-981 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu