Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
A high frequency array- based photoacoustic microscopy imaging system
(USC Thesis Other)
A high frequency array- based photoacoustic microscopy imaging system
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
A HIGH FREQUENCY ARRAY- BASED PHOTOACOUSTIC MICROSCOPY
IMAGING SYSTEM
by
Rachel Rinat Bitton
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(BIOMEDICAL ENGINEERING)
December 2007
Copyright 2007 Rachel Rinat Bitton
ii
Dedication
I dedicate this work to my family. I dedicate this to my hilarious sisters Ayali and Galit,
with whom I share an unbreakable bond, and to my beautiful mother for her amazing
strength and all the years of hard work in raising three children alone. Her faith and drive
has made me the woman I am today. I would also like to dedicate this work to my old
next door neighbor who touched my life in more ways than I could have ever imagined.
Ben, you began as my neighbor and became an inspiration, a respected mentor, a patient
teacher, and an amazing father to me. You have taught me to believe in myself, and that
things never turn out the way you plan, they turn out better. Without your love, support,
a pinch of mathematical help, and encouragement I would have made this journey. I
thank my calculus teacher Judy, whose eccentricity and love of calculus was infectious in
all the right ways. I also dedicate this work to Giordano, from high school, to college,
and now to colleague, I truly cherish your friendship and support. Through many good
times and hard times in the last few years, I thank my friends for being a support
structure, an escape, and reminding me “don’t stop believin”. I thank my roommate
Sarah for her kindness, her graphic advice, and for enduring summer after summer
without television while I work. Graduate school has been my dream for many years; I
only hope that the years to come are just as formative, challenging, and fulfilling.
iii
Table of Contents
Dedication ii
List of Tables vii
List of Figures viii
Abstract xiv
Chapter 1: Introduction 1
Chapter 2: Ultrasound Theory 6
2.1 General Ultrasound Principles 6
2.1.1 Propagation of Acoustic Waves 6
2.1.2 Attenuation 7
2.2 Transducer Fundamentals 7
2.2.1 Beam Patterns 8
2.2.2 Transducer Arrays 9
2.2.3 Beamforming 13
2.2.4 Axial and Lateral Resolution 17
Chapter 3: Photoacoustic Theory 18
3.1 The Opto-Acoustic Effect 18
3.2 Qualitative Theory 18
3.2.1 Light Interaction 18
3.2.2 Photoacoustic Sound Generation 21
3.2.2.1 Thermal Stress and Confinement 21
3.2.2.2 Initial Pressure Distribution 23
3.2.2.2.1 Mechanisms 23
3.2.2.2.2 The Photoacoustic Wave Equation 26
Chapter 4: Photoacoustic Imaging 28
4.1 Current Technologies 28
4.1.1 High Frequency Ultrasound 28
4.1.2 Magnetic Resonance Imaging 29
4.1.3 Optical Imaging 30
4.2 Photoacoustic Imaging 32
4.2.1 Introduction 32
4.2.2 Depth 32
4.2.3 Speckle 34
4.2.4 Contrast 34
4.3 Applications of Photoacoustic Imaging 35
iv
4.3.1 Atherosclerosis 35
4.3.2 Skin Cancer 36
4.3.3 Angiogenesis 37
4.4 Previous Work in Photoacoustic Imaging 38
4.4.1 Introduction 38
4.4.2 Photoacoustic Tomography 38
4.5 Research Goals 42
Chapter 5: Design of a High Frequency Photoacoustic
Microscopy System: Part I 44
5.1 Purpose 44
5.2 Introduction – High Frequency Design 44
5.3 Materials and Methods 45
5.3.1 System Design 45
5.3.1.1 30MHz Transducer Array 48
5.3.1.2 Analog Receiver Design 51
5.3.1.3 PCB Layout Considerations 57
5.3.2 Software Development 58
5.3.2.1 System Control 58
5.3.2.2 Image Reconstruction 58
5.3.3 Phantom Description 60
5.3.4 Experimental Setup 60
5.4 Results 63
5.4.1 System Performance 63
5.4.2 Images 66
5.4.2.1 Carbon Fiber in Water 66
5.4.2.2 Carbon Fiber and Human Hair in Intralipid 71
5.4.2.3 Carbon Fiber Matrix in Water 72
5.4.2.4 Animal Microvasculature 73
5.5 Discussion 75
5.6 Summary 79
Chapter 6: Design of a High Frequency Photoacoustic
Microscopy System: Part I 80
6.1 Purpose 80
6.2 Introduction – High Speed Digital Design 80
6.2.1 Sampling Frequency and the Knee Frequency 80
6.2.2 Distributed Systems 81
6.2.3 Terminations 83
6.2.4 Clock Distribution 84
6.2.5 Technology 86
6.3 Materials and Methods 87
6.3.1 System Design 87
6.3.2 Digital Implementation in Hardware 90
v
6.3.2.1 Sampling, Timing, & Control 90
6.3.2.2 Interconnect 93
6.3.2.3 Data Bus 94
6.3.3 Software Development 95
6.3.3.1 System Control 95
6.3.3.2 Image Reconstruction in 3D 98
6.4 Results 98
6.4.1 Digital System Perform 98
6.4.2 Images 100
6.4.2.1 Carbon Fiber in Water 100
6.4.2.2 Carbon Fiber and Hair in Intralipid 104
6.4.2.3 Carbon Fiber Matrix in Water 105
6.4.2.4 In Vivo Images 106
6.5 Discussion 109
6.5.1 System Observations 109
6.5.2 Noise 111
6.6 Summary 114
Chapter 7: Future Work 115
7.1 Immediate Aims in Real Time Imaging 115
7.1.1 Purpose 115
7.1.2 Materials and Methods 116
7.1.2.1 System Design 116
7.1.3 Results 117
7.1.4 Discussion 120
7.1.5 Fast Acquisition Multi-Channel
Parallel Processing 120
7.1.5.1 Design Solution 120
7.1.5.2 Hardware Implementation 122
7.1.5.3 Image Reconstruction 123
7.1.6 Suggestions on Image Improvement 124
7.2 Long Term Aims 125
Chapter 8: General Summary 127
References 129
Appendix A: Field II Array Directivity Simulation Code 136
Appendix B: PAM Sector Scan Image Formation Code 139
Appendix C: PAM I User Control Code 145
Appendix D: Receiver Board Schematics and PCB Layout 141
vi
Appendix E: PAM II System Control Code 153
Appendix F: Logic Analyzer Screen Capture 156
Appendix G: Digital System Schematics and PCB Layout 157
Appendix H: VHDL Code and Simulation for Real Time PAM 162
Appendix I: Real Time PAM Schematics and PCB Layout 171
Appendix J: Real Time PAM Control Code 173
vii
List of Tables
Table 1.1 Depth/Lateral resolution comparison
of conventional US, high frequency US,
OCT, and PA imaging 3
Table 5.1 Summary of gain and loss for each
receiver signal processing stage. 64
viii
List of Figures
Figure 1.1 System flow for the array based
photoacoustic microscopy system (PAM). 4
Figure 2.1 Behavior of incident ultrasonic
energy arriving at an angle to the
acoustic impedance boundary. 6
Figure 2.2 Beam pattern for a single element transducer. 9
Figure 2.3 a) Linear switched transducer array
b) Linear phased transducer array. 11
Figure 2.4 Angular beam pattern of a typical
transducer array. 12
Figure 2.5 Beamforming focusing during echo reception. 14
Figure 2.6 Geometrical representation of the
time delay differences between elements
in an array from one image point. 15
Figure 2.7 Coherent and non-coherent delayed
scan lines. 16
Figure 3.1 Light absorption spectrum of
deoxy-hemoglobin (Hb), oxy-hemoglobin
(HbO), and melanin. 20
Figure 3.2 The thermoelastic effect of short
laser pulse heating on tissue generating
acoustic waves. 23
Figure 3.3 Propagation of photoacoustic waves
for spherical and line absorbers. 25
Figure 4.1 Speckle patterns present in ultrasound
and optical coherence tomography scans. 31
Figure 4.2 High Resolution Imaging Technologies. 33
ix
Figure 4.3 a) PAT image of a superficial lesion
(1mm x 4mm) on a rat brain acquired with
the skin and skull intact (b) Open-skull
photograph of the rat cerebral surface
acquired after the PAT experiment. 40
Figure 4.4 Functional photoacoustic dark field microscopy. 41
Figure 5.1 PAM I overall receive system structure
for noninvasive high frequency array-based imaging. 47
Figure 5.2 Two way pulse and FFT frequency
response for array element 19 in the
30MHz linear array. 49
Figure 5.3 Combined Acoustic/Electric crosstalk for the
30MHz linear array as a function of frequency. 50
Figure 5.4 Field II simulated directivity pattern for a
30MHz, 48 element array spaced at 2 λ. 50
Figure 5.5 PAM front end pre-amplifier circuit design. 52
Figure 5.6 Complex frequency response for a
Butterworth 4
th
order filter. 54
Figure 5.7 Simulated (PSPICE) frequency response
for Butterworth 4
th
order band pass filter. 55
Figure 5.8 Measured frequency response for
Butterworth 4
th
order band pass filter. 55
Figure 5.9 Imaging phantoms of carbon fiber and human hair. 61
Figure 5.10 Animal experimental setup for PAM imaging. 62
Figure 5.11 Photograph of transducer array and
optical fiber mounted on translation stages. 63
Figure 5.12 Setup for measuring minimum
detectable signal, noise floor, and
dynamic range of analog receiver. 64
x
Figure 5.13 Photograph of the complete analog
receiver PCB for the PAM system. 65
Figure 5.14 PAM I raw RF data from element #19
using a 6µm carbon fiber target. 68
Figure 5.15 PAM I envelope detected raw RF
data from all active channels form a single matrix. 68
Figure 5.16 Point spread function of PAM I
6µm carbon fiber a) without coherence
factor (CF) weighting b) with CF weighting. 69
Figure 5.17 PAM I 3D mesh representation of
6µm carbon fiber in water. 70
Figure 5.18 PAM I B-mode image of a 6µm
carbon fiber in water. 70
Figure 5.19 PAM I image of 80µm human hair
at 7.3mm depth in 1% Intralipid solution. 71
Figure 5.20 PAM I image of 6µm human hair
at 7.2mm depth in 1% Intralipid solution. 72
Figure 5.21 PAM I dynamic focusing: Composite
image of a 6µm carbon fiber matrix in water. 73
Figure 5.21 PAM I photoacoustic B-scan of
subcutaneous micro-vessels in a young rat
visualized 2-3mm below the skin’s surface. 74
Figure 5.23 A series of PAM scans with
increasing depth are projected onto a single
image to form a C-scan image. 74
Figure 5.24 a) Photograph of the underside of
excised rat skin showing micro-vessels and
b) PAM I C-scan in the identical region as part a). 75
Figure 6.1 Excessive overshoot of a digital
switching signal can cause false triggers
within the logic threshold. 82
xi
Figure 6.2 Active line termination through the
Schottky diode scheme. 84
Figure 6.3 H-Tree design with parallel termination
for high speed clock distribution. 86
Figure 6.4 PAM II overall system structure and
flow for noninvasive high frequency
array-based imaging. 89
Figure 6.5 Timing diagram of master clock,
transfer clock and control operations
for data transfer. 92
Figure 6.6 PAM digital system software flow chart. 97
Figure 6.7 Photograph of the digital receiver
PCB for the PAM II system. 99
Figure 6.8 PAM II raw RF data from element #20
using a 6µm carbon fiber target. 101
Figure 6.9 PAM II envelope detected raw RF data
from all active channels form a single matrix. 101
Figure 6.10 Point spread function of PAM II
6µm carbon fiber. 102
Figure 6.11 PAM II 3D mesh representation
of 6µm carbon fiber in water. 103
Figure 6.12 PAM II B mode image of a 6µm
carbon fiber in water. 103
Figure 6.13 PAM II B-mode image of 6µm
carbon fiber in 0.5% Intralipid solution. 104
Figure 6.14 PAM II B-mode image of 80µm
human hair in 1% Intralipid solution. 105
Figure 6.15 PAM II dynamic focusing:
Composite image of a 5x9, 6µm
carbon fiber matrix in water. 106
xii
Figure 6.16 PAM II B-mode image of a cross
section of blood vessels in the lower portion
of a human hand in vivo. 107
Figure 6.17. 3D PAM images of micro-vessels
below the surface of the skin in two
Sprague Dawley rats. 108
Figure 6.18 PAM II PSF for a 6µm carbon fiber
in water with a 90° scan angle a) with CF
weighting b) with out CF weighting. 113
Figure 7.1 Fast acquisition PAM B-scan of
subdermal rat micro-vessels. 119
Figure 7.2 A general CPLD structure diagram. 122
Figure 7.3 Photograph of PAM fast acquisition
CPLD motherboard. 123
Figure C-1. PAM I Labview Front Panel 145
Figure C-2. PAM I Labview Back Panel 146
Figure D-1. AnalogMain.schdoc 147
Figure D-2. Preamp.schdoc 148
Figure D-3. Mux.schdoc 149
Figure D-4. Muxi.schdoc 150
Figure D-5. Bpass_VGA.schdoc 150
Figure D-6. 2Fx_Amp.schdoc 151
Figure D-7. SMA.schdoc 151
Figure D-8. Analog Receiver PCB 152
Figure E-1. Front Panel 153
Figure E-2. Back Panel: Windows a b and c are
consecutive screen shots of the same Labview program. 155
xiii
Figure F-1. Logic Analyzer Data of 8 Channel Boards 156
Figure G-1. Digitalmain.schdoc 157
Figure G-2. Connectorland.schdoc 158
Figure G-3. Channelboard.schdoc 159
Figure G-4. Channel Board PCB Layout 160
Figure G-5. Digital Board PCB Layout 161
Figure H-1. VHDL Schematic for Real Time PAM 162
Figure H-2. VHDL Simulation Test Panel 170
Figure I-1. CPLDmain.Schdoc 171
Figure I-2. CPLD.schdoc 172
Figure J-1. Real Time Back Panel 173
xiv
Abstract
Photoacoustic microscopy is an imaging technique which draws from the specific
strengths of two imaging modalities by capturing the contrast of optical imaging, while
retaining the high resolution of ultrasonic imaging. It provides great promise for studying
the structure and dynamics of tissue micro-vasculature in development and pathogenesis.
Previous work in photoacoustic imaging has been mostly limited to single element
transducers. This thesis presents results of a novel photoacoustic microscopy system
using a 30MHz linear array and a custom receive electronics. There are two versions of
the system, PAM I and PAM II. Both systems are comprised of three main components,
a short pulsed laser, a high frequency transducer, and a custom multi-channel electronics
system. The attraction towards high frequency arrays over single element transducers is
natural; they offer the same resolution advantage of higher frequencies, while
diminishing the need for mechanical scanning through steering of the beam, delivering
aperture flexibility, tighter focusing capabilities through beamformation, and the
capability to image in real-time.
The PAM I system includes an Nd:YAG pumped tunable dye laser, delivering a 6.5ns
pulse duration, and a 10Hz pulse repetition rate to the sample via an optical fiber.
Furnishing an incident energy of approximately 6mJ/cm
2
at 584nm, the laser induced
acoustic waves via thermoelastic expansion. Using a 30MHz linear array and a custom
multi-channel receive system, both phantom and in situ photoacoustic images were
obtained. The receiving transducer array is a piezo-composite 48 element linear array,
xv
with an 8mm focal depth, and a -6dB fractional bandwidth of 50%. Multi-channel
receive electronics were developed to include multiplexing and signal processing stages.
Four-to-one multiplexers are used to select between elements. The signals are passed
through filtering stages, followed by variable and fixed gain stages. The system receiver
gain varies from 33dB-73dB, with a -3dB system response between 8MHz and 55MHz.
The channels are further multiplexed to acquire data from a 4 channel oscilloscope.
Using offline delay and sum beamforming, initial results provided phantom images from
an 80µm hair in water, and a 6µm carbon fiber in an optically scattering medium similar
to biological tissue. Photoacoustic images in situ clearly showed subcutaneous vessels
less than 100µm in diameter imaged at depths of 3mm below the skin surface in a
Sprague Dawley rat.
The PAM II system is a 16 channel fully automated parallel multi-channel system that
acquires data from all elements on the receive board at once. Controlled by the user
interface and the PC, the PAM II system uses the receiver front end analog board and is
complimented by custom digital electronics. The digital portion of the system is a
backplane motherboard/channel board scheme. Individual channel boards
simultaneously digitize the echoes from the receiver at a 100MHz sampling rate. Digital
data are then stored in temporary memory and transferred via the PCI bus to the PC with
an NI-6534 (National Instruments, Houston, TX) acquisition board. A Labview program
was developed to handle system triggering, and control signals to the digital board and
receiver multiplexers. PAM II uses an Isonnolab Edgewave laser pumping 6ns pulses at
xvi
598nm using an electro-optic Q-switch, delivering an incident energy of below
15mJ/cm
2
. Phantom images in water and Intralipid solution were formed to characterize
the system. Photoacoustic images of micro-vessels in a human hand and 3D images of
vasculature in two Sprague Dawley rats were obtained in vivo. The axial and lateral
spatial resolutions for both systems were found to be 45±5µm and 100±5µm,
respectively. Ongoing research is also presented for development of a real time PAM
system. Initial experiments provided in vivo rat images differentiating micro-vessels in
systole and diastole.
1
Chapter 1: Introduction
The development of novel approaches to biomedical imaging is stimulated by the
manifest need for high speed, high resolution, non-invasive techniques. An imaging
technology capable of visualizing micro-vessels non-invasively would lend the ability to
identify the angiogenic process, one of the hallmarks of tumor growth (Liu, 2006). Laser
induced photoacoustic microscopy is an imaging modality based on the intrinsic optical
properties of biological tissue and ultrasonic detection at high frequencies (>20MHz).
Photoacoustic microscopy can be used as a diagnostic tool which combines the strengths
of both optical and acoustic imaging techniques.
Photoacoustic imaging has its physical basis in a phenomenon called the photoacoustic
effect. In biological tissue, incident laser light will experience both scattering and
absorption. The photoacoustic effect occurs when the pulsed light energy is absorbed
locally in the tissue, causing a small rapid temperature rise in the medium, and inducing
thermoelastic expansion. This expansion produces pressure transients, which propagate
as acoustic waves throughout the tissue. A photoacoustic (PA) wave is unique in the
sense that although it is an ultrasonic wave, it carries optical absorption information.
Photoacoustic waves are generated from within the tissue and propagate outwards
towards the medium surface. At the surface, the photoacoustic echoes are detected
through ultrasonic piezoelectric transducers. They are then processed to create an image
which maps the optical absorption distribution of the medium. Since photoacoustic
2
imaging is a hybrid modality, a review of both the ultrasonic and optical theory is
included in this thesis (chapters 2 and 3).
Current available technologies available suffer from either poor penetration depth, as in
optical imaging techniques or poor contrast resolution, as in ultrasonic imaging. Optical
imaging such as optical coherence tomography (OCT) can provide excellent spatial
resolution (Singh, 1998). However, its penetration depth is severely limited as a result of
the strong diffusivity of light biological tissue (Fercher, 1997). Also, while optical
imaging techniques are sensitive to the light backscattering that is related to tissue
morphology, they are insensitive to the optical absorption that is related to important
biochemical information. High frequency ultrasonic imaging (US) can achieve good
resolution throughout a larger range of image depth, but has a contrast resolution
determined by weak acoustic backscatter, falling short of its optical counterparts. The
varied optical absorption of states of hemoglobin allows photoacoustic techniques to also
extract some functional imaging information (Zhang, 2006). Photoacoustic imaging can
essentially bridge the gap between technologies by providing optically based contrast
with an ultrasonic resolution at greater depths. Table 1.1 provides a comparison of
current imaging modalities. Further discussion is found in section 4.1.
Photoacoustic imaging offers the following qualities:
1. Non-invasive imaging
2. Non-ionizing radiation
3. Spatial resolution and imaging depth dependant upon the transducer frequency
4. No speckle artifacts
3
5. High contrast resolution
6. The capability for functional imaging
7. The capability for real time data acquisition
Table 1.1 Depth/Lateral resolution comparison of conventional US, high frequency US,
OCT, and PA imaging.
Photoacoustic imaging has been generally confined to mechanical scans with single
element transducers until now. The draw to higher frequencies is palpable, since lateral
spatial resolution improves with increased frequency. The following research
investigates the design and use of array technology for high frequency photoacoustic
imaging in order to increase acquisition speed, and facilitate electronic steering and
focusing of the receive beam.
This thesis presents the design of a novel photoacoustic microscopic imaging system,
PAM. The PAM system utilizes a high frequency piezoelectric transducer array and 16
channel custom receive electronics to visualize micro-vascular structures in Sprague
Dawley rats. The photoacoustic microscopy system is comprised of three main
Modality Image Depth (mm) Lateral Resolution (µm)
Conventional US
2-12MHz Excellent (2500-100) Poor (800-200) Fair Contrast
High Frequency US
20-50MHz Good (15-3) Good (200-80) Fair Contrast
Optical Coherence
Tomography Poor (1.5) Excellent (~15) Excellent Contrast
Photoacoustic
Imaging Good (10-3)
No Speckle
Good (200-80) Good Contrast
4
components: An Nd:YAG or Nd:YLF laser source used to irradiate the tissue and induce
photoacoustic waves, a 48 element piezo-composite transducer array which receives the
photoacoustic echoes, and a 48 to 16 channel custom receive electronics system which
processes and digitizes the data (Figure 1.1).
Figure 1.1 System flow for the array based photoacoustic microscopy system (PAM).
[1:16]
[1:16]
[1:48]
[1:16]
[1:16]
RF Signal
Processing
Multiplexing
Front End
Dye
Laser
Lens
Optical Fiber
Synch/
Trigger
Beam
Shaper
Nd:
YLF/YAG
Laser
Digitization
Memory
Storage
PC
Control
Display
FIFO Control
Transducer
Array
High Frequency Array-Based
PAM System
5
The process initiates when a TTL signal from the motherboard triggers the laser pulse to
irradiate the sample through the optical fiber positioned obliquely to the sample and
array. After each laser pulse each sub-aperture is processed through the analog and
digital boards, and then stored to the PC. Chapters 5 and 6 discuss the development of
the first and second generations of the PAM system. PAM images displaying resolution,
dynamic focusing of carbon fiber phantoms and in vivo micro-vessels in both a human
hand and rats are reported (Bitton, 2006).
6
Chapter 2: Ultrasound Physics
2.1 General Ultrasound Principles
Ultrasonic imaging employs sound waves that have frequencies above the human audible
range of 20KHz. In the following sections, ultrasonic imaging fundamentals will be
discussed.
2.1.1 Propagation of Acoustic Waves
As an acoustic wave travels through a medium and approaches a boundary of two
different acoustic impedances, some of the energy is reflected at an angle while the
remainder is transmitted with a change in direction (Shung, 1992). In conventional
ultrasound, many useful echoes come not from simple specular reflection at a medium
interface, but from a diffuse scattering of small, wavelength or smaller sized particles.
These diffuse reflections, termed Raleigh scattering, or simply scatter, are derived from
low amplitude signals that emit omni-directionally, allowing the angle of incidence to be
a negligible parameter (Papadakis, 1999).
Figure 2.1 Behavior of incident ultrasonic energy arriving at an angle to the acoustic
impedance boundary.
Incident
Ultrasound
Specular
Reflection
Diffuse reflection,
scattering
Ultrasound
Refraction
Impedance Boundary
7
2.1.2 Attenuation
Propagating acoustic waves through inhomogeneous media, such as biological tissues,
experience loss as a function of distance. The pressure of a plane wave p(z) traveling in
the Z direction displays the following exponential behavior (Shung, 1992).
z
a
e p z p
α −
= ) 0 ( ) (
(2.1)
where a
a
is the acoustic attenuation coefficient of the medium in Nepers/cm.
There are a few primary causes for this attenuation: reflection and scattering at
interfaces, and absorption. While attenuation from reflection and scattering divert
energy, attenuation caused by absorption converts the energy into heat within the
medium. In this instance the sound energy is lost, not to be recovered. Absorption of the
acoustic wave is frequency dependent. It is assumed to be attributed to viscous properties
and a relaxation phenomenon.
2.2 Transducer Fundamentals
Transducers are the devices which make ultrasonic imaging possible. The physical and
electromechanical characteristics of an ultrasound transducer greatly affect the properties
of the image. Piezoelectric based ultrasonic transducers convert electrical energy to
mechanical energy (acoustic waves), and vice versa via a phenomenon called
piezoelectric effect. When an electrical signal is applied across the material it forces the
electric dipoles to realign and thus, produces a physical change in thickness. Likewise,
8
when a mechanical force such as an ultrasonic pressure wave is applied, there is a
resulting electrical potential generated (Shung, 1996).
2.2.1 Beam Patterns
The beam pattern from a transducer consists of a near field, or Fresnel zone and a
divergent far field. The beam can be calculated following Huygens’ principle and adding
up wavelets from point sources distributed all across the transducer surface. The near
field beam profile is complex as it includes waves that are out of phase, and waves that
exhibit constructive and destructive interference (Shung, 1996). It is also characterized
by fluctuations in the amplitude and intensity from one point in the beam to another. The
beam in the far field is smoother, and more predictable (Figure 2.2). The divergence
angleγ , of the far field is given by the relationship
d
λ
γ
2 . 1
sin =
(2.2)
where λ is the wavelength and d is the diameter of the transducer. Although most of the
energy is contained in the main beam, side lobes contain some energy radiated in other
directions due to constructive interference.
9
Figure 2.2 Beam pattern for a single element transducer. The near field is characterized
by fluctuations in the amplitude intensity. The beam is narrowest at the near-far field
transition point and diverges in the far field at angle γ.
2.2.2 Transducer arrays
Transducer arrays are the current and future trend for ultrasonic imaging. Arrays employ
many piezoelectric elements, where each element has its own electrical connection.
Elements can be excited individually or in groups, effectively forming a sub-aperture.
They provide two enormous advantages over single element transducers: the ability to
electronically steer and electronically focus a beam. Unlike single element transducers,
linear arrays have the advantage of obtaining multiple scan lines without mechanical
repositioning to create an image. Transducer arrays create a system where different
delays can be assigned to the received echoes from each element by beamforming, and
allowing for an adjustable electronic focus without mechanical scanning. Although there
are a variety of transducer arrays, such as annular and 2D arrays, the scope of this
research applies to the use of linear and phased arrays. Linear arrays form scan lines by
using a sub-aperture of elements that slides across the face of the array. Each subaperture
forms a single scan line, and is then translated by one array element to the next position,
γ
Near field Far field
10
producing a rectangular image field. Phased arrays generate angular scan lines by using
all elements at once to steer the main beam at different angles (Figure 2.3).
Side lobes and grating lobes are unwanted energies directed outside the main beam of the
array. Grating lobes are caused by the periodic nature of regularly spaced elements in the
linear array. The main lobe, side lobes, and grating lobes in an array are often visualized
by plotting the angular directivity pattern. It shows how the amplitude of the wave varies
as a function of angle in the far field of the array. Figure 2.4 illustrates the angular beam
pattern of a typical linear array.
11
Figure 2.3 a) Linear switched transducer array. Linear scan lines are formed as the
effective aperture slides across the array. b) Linear phased transducer array. Scan lines
are formed in an angular manner producing a sector image.
Lateral/Azimuth
Elevation
Image/Depth
a)
b)
12
Figure 2.4 Angular beam pattern of a typical transducer array.
Grating lobes are dictated by the pitch of the array. They are caused by constructive
interference (or a 1 λ shift in path length) between the fields emitted by adjacent elements.
This interference is observed at an angle
g
φ from the normal to the array face. The
angular location is defined and can be predicted by
=
−
g
n
g
λ
φ
1
sin
(2.3)
Where n=±1, ±2, … Grating lobes also vary in amplitude as the main lobe is steered in
either scan direction (Pompei, 2002). Even when an array exceeds the ideal pitch value
to avoid grating lobes, acceptable imaging can still be achieved. Decreasing the pulse
length, randomizing the pitch, or a reduction in element width can help reduce the affect
of grating lobes (Cannata, 2004). A short pulse length can lessen the amplitude of the
grating lobe, especially when a large number of elements are used for beamforming.
Since elements constructively interfere at the grating lobe angle due to a 1 λ difference in
13
path length from element to element, when the pulse is only a few λ long, only a limited
number of elements contribute to the grating lobe.
2.2.3 Beamforming
Appropriately delaying pulses from array elements can achieve both beam steering and
beam focusing. Time delays are applied to individual signals to compensate for
differences in echo arrival time to elements further away from the center reference. This
arranges all the echoes in phase. The emerging beam is the sum of the delayed echoes
from all elements (Figure 2.5).
The required delay time needed for beamforming can be calculated by considering a
point target at particular depth (Figure 2.6). Assuming an acoustic velocity c
a
, and a
distance R to the target point at an angle θ from the element in question relative to the
reference center element, the time delay ∆τ
n
for element number n can be calculated
using geometry.
a
n
n
c
R R R x − + +
= ∆
θ θ
τ
2 2 2
cos ) sin (
(2.4)
Using the Taylor series expansion we obtain the following formula:
a
n
a
n
n
Rc
x
c
x
2
) ( cos ) sin(
2
θ θ
τ + ≈ ∆ (2.5)
Where the first term describes the steering and the second describes focusing.
Accounting for the total trip distance by applying c
a
t=2R we have the final form:
14
t c
x
c
x
a
n
a
n
n 2
2 2
) ( cos sin θ θ
τ + ≈ ∆ (2.6)
This method creates a sector scan where all elements are involved in producing the beam
for each acoustic line. After applying the delays, the signals are added together creating a
scan line where the directivity angle of the beam can vary based on the delays. This is
repeated for each increment in angle of the image.
Figure 2.5 Beamforming focusing during echo reception. Because of slightly different
distances from the focal point echoes S(i) arrive at different times. Time delays align the
echoes and are summed coherently to form a single scan line.
S(i)
t5
τ
1
τ
2
τ
3
τ
4
τ
5
τ
6
τ
7
ΣS(i)
Focal Point
S(i)
t1
S(i)
t3
S(i)
t4
S(i)
t2
S(i)
t6
S(i)
t7
15
Figure 2.6 Geometrical representation of the time delay differences between elements in
an array from one image point.
In addition to beam steering and beam focusing other post processing and filtering
techniques have been shown useful for image improvement (Busse, 1995). An effective
method developed by Pai Chi Li and Meng Lin Li is called adaptive coherence factor
(CF) weighting (Li, 2003). The coherence factor is applied to the beamformed data to
reduce focusing errors resulting from sound velocity inhomogeneities and phase
aberrations, as well as some correction for steering errors. The coherence factor
weighting equation is
∑
∑
=
−
=
−
=
1
0
2
2
1
0
) (
) (
N
i
D
N
i
D
i S N
i S
CF
(2.7)
Where N is the number of elements and S
D
is the channel data after the time delays for
steering and focusing have been applied. The numerator represents the energy of the
R
D
x
n
θ
Image
point
16
coherent sum and the denominator denotes is the total incoherent energy that is N times
the coherence sum. It is used as an adaptive weighting factor that is maximal (essentially
a one to one ratio) when the delayed signals are identical across the synthesized aperture,
or perfectly coherent. Figure 2.7 shows two cases of received echoes after applied
delays. The image on the left represents the case when applied delays produce an error
free coherent result from all elements. The image on the right represents echoes after
delays when the source is not in the direction of the formed beam, representing a steering
error.
Figure 2.7 Coherent and non-coherent delayed scan lines. The z axis is in the depth
direction and the i axis is the azimuth direction (Li, 2003).
The CF of an on axis point source without focusing errors equals 1. The case in which
the point source is not on the expected beam axis may correspond to a steering error. In
this case if the CF is low, it indicates a destructive summation in the numerator. The
final result can then be expressed by the CF weighted signal as
) ( ) ( ) ( t CF t S t S
D Dweighted
⋅ =
(2.8)
17
For image reconstruction, the factor is applied to each image point. This technique also
offers some other perks, including an increased signal to noise ratio (SNR) since most
noise is incoherent, and reduced side lobe and grating lobe levels.
2.2.4 Axial and Lateral Resolution
Image resolution is an issue which is important in any imaging modality. Axial
resolution refers to the minimum reflector spacing along the axis of a beam that results in
separate, distinguishable echoes. It is the spatial extent of the echo in the axial direction
and is mostly determined by pulse duration. Lateral resolution refers to the ability to
distinguish two closely spaced reflectors positioned perpendicular to the beam axis
(Lockwood, 1996). It is more concisely defined as the spatial extent of the beam profile
in the lateral direction within -3dB or -6 dB of the peak intensity. Basic imaging theory
has established that lateral resolution is a function of the wavelength of sound, the depth
of the scan, and the size of the aperture used to form an image. A common measurement
of the aperture size is f number (f#), which is defined as the ratio of focal depth to
aperture size (McKeighan, 1998). The best estimate for lateral resolution is given by
λ # f R
Lateral
=
(2.9)
Naturally, as frequency increases, this linear relationship reveals that lateral spatial
resolution improves.
18
Chapter 3: Photoacoustic Theory
3.1 The Opto-Acoustic effect
The optoacoustic effect was first described in 1880 by Alexander Graham Bell, who
observed sound emanating from a sheet of rubber that was being illuminated by an
intermittent beam of sunlight (Kruger, 1999). This phenomenon extends to other
materials such as gaseous and liquid media, as well as biological tissue, wherever pulsed
electromagnetic energy can be absorbed. The optoacoustic effect refers specifically to
the excitation of sound in a medium that absorbs a variable electromagnetic flux. The
fields of optoacoustics and photoacoustics both describe the same interaction of
electromagnetic and acoustic waves. Optoacoustics deals with the excitation of sound
from the action of electromagnetic radiation on the medium (Gusev, 1993). When the
electromagnetic energy is absorbed by a biological tissue, a small temperature rise in the
medium causes a thermoelastic expansion, producing pressure waves, which propagate as
acoustic waves throughout the tissue omni-directionally. More specifically,
photoacoustics describes the use of short pulsed laser light as the electromagnetic source,
and is based on the opto-acoustic effect.
3.2 Qualitative Theory
3.2.1 Light Interaction
In many ways acoustics and optics can be seen as sister modalities. Behaving as another
electromagnetic wave, we observe the light intensity in a medium decrease exponentially
as it is affected by attenuation. Just as in acoustics, at interface boundaries, some of the
19
energy is specularly reflected, some is transmitted through the medium, while the
remaining portion is scattered. Almost analogous to ultrasonic imaging, optical
coherence tomography, OCT, uses backscattered light to determine optical absorption.
Since the speed of light is much faster than the speed of sound, time delay measurements
as in ultrasound are too difficult to obtain by similar means (Sainter, 2004). Instead,
OCT uses interferometry, the study of the interference pattern of two or more waves, to
compare the incident light with the received light for image reconstruction. Contrast in
optical imaging is greater than that in ultrasound (Wang, 2005). This is because of the
varied optical properties of biological tissue are related to their molecular composition
and not to acoustic impedance properties as in ultrasound.
As light interacts with tissue, it is either absorbed or scattered in various proportions
depending on the optical properties of the tissue. Biological tissues are turbid media, and
the light attenuation due to absorption is augmented by light scattering (Kienly, 2004).
Tissue also absorbs light differently at specific wavelengths. The characteristic
wavelength at which tissue absorbs light depends on its structure and composition, and
will vary for different tissues. The two largest components of tissue that effect
absorption of laser light are pigmentation and water concentration (Schmitt, 1990).
Chromophores are light-absorbing substances within tissue. Most organic molecules
display strong absorption in the ultraviolet region, so penetration in the UV spectrum is
very weak (a few microns). In the visible spectrum (blue, green and yellow), absorption
20
is principally due to the chormophores hemoglobin and melanin. In the visible regime,
the light can penetrate beyond a few millimeters (Xu, H. 2006). Therefore, the visible
range is preferred for use in photoacoustic imaging.
Photoacoustic signal generation is based on the optical absorption of the medium, and is
discussed further in section 3.2.2. The light absorption spectrum varies for some
components of biological tissue (Fig. 3.1). Absorption of hemoglobin and the different
oxygenation states of hemoglobin, as well as the absorption of melanin, validate the use
of light for imaging the near-infrared and visible spectrum.
Figure 3.1 Light absorption spectrum of deoxy-hemoglobin (Hb), oxy-hemoglobin
(HbO), and melanin. Strong absorption of melanin and blood is present in the visible
spectrum (Horecker, 1942).
21
The dominating attenuation effect in biological tissue in the visible region is scattering.
Before diffusion theory there was no good model to describe how scattered light behaved,
as a result, all the work was based on empirical measurements. A commonly used
method to characterize tissue absorption and scattering requires the removal of a thin
slice of tissue and is invasive (Pickering, 1993). The problem of determining light
propagation in tissue is often modeled based on approximations derived from diffusion
theory (Arnfield, 1992). It has been modeled by Gamert and Star as an approximate
solution to the transport equation (Gamert, 1987), and more recently by Monte Carlo
simulation (Wang, 1995). Both of these methods represent the optical properties of the
tissue with the effective attenuation coefficient, µ
eff
, which is a function of both the
absorption coefficient µ
a
(cm
-1
) and scattering coefficient, µ
s
(cm
-1
)(Wang, 1997).
3.2.2 Photoacoustic Sound Generation
The following sections describe the conditions and method by which a photoacoustic
transient is generated.
3.2.2.1 Thermal Stress and Stress Confinement
A medium exposed to electromagnetic radiation, can provide sound excitation through a
few mechanisms such as electrostriction and thermal expansion (Tam, 1986). Sound
excitation via the thermal effect is the dominant mechanism for generation of
photoacoustic signals within biological tissue.
22
To generate photoacoustic waves by coupling thermal energy to mechanical energy two
conditions must be met; thermal confinement and stress confinement (Gusev, 1993).
Thermal confinement occurs when the energy deposition happens faster than the time
needed for heat relaxation. In the case of photoacoustic tomography, the laser pulse
duration, τ
L
, is on the order of nanoseconds. The laser pulse duration is much shorter
than the heat diffusion time, τ
th
in seconds, given by
κ
τ
4
2
p
th
d
= (3.1)
where d
p
is the characteristic length, or light penetration depth, and κ(cm
2
/s) is the
thermal diffusivity or thermal conductivity constant of the medium. We see that if the
laser pulse duration τ
L
is short enough, thermal diffusion can be neglected and we assume
an instantaneous heating of the medium (Frenz, 1996).
The second condition is stress confinement. This condition relates to the time for the
stress generated from thermoelastic expansion to move beyond the heated region. The
stress time, τ
s
in seconds is given by a steadfast relation in ultrasound,
a
p
s
c
d
= τ (3.2)
where c
a
is the speed of sound. During stress confinement, the high thermoelastic
pressure in the sample can build up quickly (Gusev, 1993). A shorter laser pulse will
establish a more stringent stress confinement condition. Essentially, a short laser
duration will ensure that the pulse will cease before the acoustic wave travels, and that
the surrounding tissue will not damage tissue by diffusing the heat. Using a short laser
23
pulse duration, the two conditions (thermal stress and confinement) are met. Thus, we
examine the method of acoustic generation through thermoelastic expansion alone.
3.2.2.2 Initial Pressure Distribution
Initial pressure distribution can be determined by assuming that conditions of thermal
stress and confinement have been met. In this case, the circumstance to satisfy equation
3.1 is met and instantaneous heating is applied. The slight temperature rise is usually in
the millikelvin range, and is related to the energy deposition inside the medium.
Thermoelastic expansion induces the initial change in pressure. The following section
examines the process by which a photoacoustic pressure wave is generated.
3.2.2.2.1 Mechanisms
To illustrate the phenomenon which generates an acoustic wave by thermoelastic
expansion, consider the sphere shown in Figure 3.2. The sphere has a radius R and
uniform optical absorption properties in a highly scattering medium. When the light
pulse illuminates the medium, the high scattering of light yields the assumption that the
radiation will be isotropic by the time it penetrates the absorbing sphere. In other words,
the high scattering of light essentially illuminates the absorber uniformly, causing a
uniform energy deposition.
24
Figure 3.2 The thermoelastic effect of short laser pulse heating on tissue generating
volumetric change which induces acoustic waves. R is the radius of the sphere, ∆T and
∆P denote change in temperature and pressure.
The absorbed light is converted into heat. The general heating function has a both a
spatial and temporal component, depending on the position r, and time t. The heating
function, H, assumes instantaneous delta function heating, and can be written as
) ( ) ( ) , ( t A t H δ r r = (3.3)
A(r) is the spatial portion of the heating function and is the heat deposited per unit
volume (Wang, 2006). The spatial portion of the heating function can further be
expressed in terms of the locally absorbed laser energy by
) , ( ) ( ) (
eff a
F A µ µ r r r = (3.4)
where F is the local fluence in (J/m
2
), a function of the effective attenuation µ
eff
, due to
both scattering and absorption parameters µ
s
(cm
-1
)
and µ
a
(cm
-1
).
R
∆T
∆P
R+ ∆R
25
In response to the temperature rise of the sphere, there is a change in volume. The related
volumetric expansion is completely reversible (Karabutov, 2000). This localized change
in volume will produce a rise in pressure, p
i
(r) in (kg/ms
2
) given by
) ( ) ( ) (
2
r r r A A
c
c
p
p
a
i
Γ = =
β
(3.5)
where β is the coefficient of volume thermal expansion, c
p
is the specific heat, c
a
is the
acoustic velocity, and Γ is the unitless Grüneisen parameter, which represents the
efficiency of the conversion of heat to pressure (Oreavsky, 2003). Thus, the heating
prompts a thermoelastic expansion and induces acoustic pressure waves (Kostli, 2001).
Photoacoustic waves generated from spherical absorbers will propagate spherically, and
waves generated from line targets, such as wire phantoms and vessels, produce
cylindrical propagation through the medium (Figure 3.3)
Figure 3.3 Propagation of photoacoustic waves for spherical and line absorbers.
26
3.2.2.2.2 The Photoacoustic Wave Equation
To understand the full picture of this effect, we consider that beyond the initial change in
pressure, there is a pressure wave produced by the pulse of light that propagates not in
one direction, but omni directionally from the surface of the sphere. In photoacoustic
imaging techniques such as tomography, the behavior of the radiating sphere as a whole
is imperative for reconstruction. When examining the rigorous theory of the produced
wave we find that the photoacoustic wave that is generated looks much like a pulse. The
laser pulse will generate the time dependant pressure wave p(r,t) related to the time
dependant electromagnetic absorption, or heating function H(r,t) at position r and time t
given by (Kruger, 1995).
t
t H
c
t p c
t
t p
p
a
∂
∂
= ∇ −
∂
∂ ) , (
) , (
) , (
2 2
2
2
r
r
r β
(3.6)
The basic form of the resulting pressure wave using the general time retarded solution to
the wave equation and the Green’s function (Scott, 1998) can be written as
∫∫∫
′ −
′
′ ∂
′ ∂
=
r r
r r
r
d
t
t H
c
c
t p
p
a
) , (
4
) , (
π
β
(3.7)
Where integrating in three dimensions, we consider a 3D surface where the wave front
travels spherically. The Green’s function also introduces the use of source coordinates,
denoted with primes, and field coordinates, unprimed, unlike solutions using Maxwell’s
equations (Scott, 1998). During backprojection reconstruction, the characteristics of the
transducer can be taken into account by considering the photoacoustic signal as the result
the signal generated within the tissue convolved with the impulse response of the
transducer. However, while this formula is largely used in modeling the photoacoustic
27
waves generated for tomography reconstruction algorithms, it does not take into account
the acoustic attenuation through the medium. The acoustic attenuation is perhaps
assumed a negligible parameter, although the reasons for this exclusion are not explicitly
stated in the literature. Additionally, this can be compensated for in photoacoustic
microscopy by applying a time gain compensation function to the received photoacoustic
echoes.
From equation 3.3 and 3.4, we can clearly see that stronger absorbers will produce
greater amplitude photoacoustic waves and provide better contrast for imaging. What is
also important to note, is that the photoacoustic waves produced from different diameter
targets will produce waves of different frequencies for the same incident light pulse. This
is because the photoacoustic wave is modulated by the heating function which is
temporally and spatially dependent. A larger diameter absorber will produce a different
spatial profile for the same temporal profile of a smaller diameter absorber. For this
reason, as the diameter of the absorbing target decreases, the generated photoacoustic
wave frequency increases. Therefore, a sample of biological tissue can produce a wide
range of photoacoustic frequencies. Even though a spectrum of waves can be generated,
the bandwidth of the transducer determines the waves used to construct the image. For
this reason, wideband transducers are essential for the detection of thermoelastically
generated acoustic waves.
28
Chapter 4: Photoacoustic Imaging
4.1 Current Technologies
Although there are many available medical imaging methods, the following discussion
looks at competing technologies that are comparable to photoacoustic microscopy.
Comparable modalities include those that are non-ionizing, non-invasive, and capable of
sub-millimeter resolutions for in vivo imaging.
4.1.1 High Frequency Ultrasound
Ultrasonic imaging is one of the more mature non-ionizing imaging techniques in use
currently, having been commercially available and adopted by the medical community
since the 1970’s (Enderle, 1999). Developments in high frequency transducer technology
have increased the spatial resolution of ultrasound. Using single element transducers,
ultrasonic images of skin dermal layers have been produced with lateral spatial
resolutions from 30µm to 94µm at depths of 4mm and 8mm, respectively (Turnbull,
1993). Vascular images have been formed offline using mechanical scanning of micro-
vessels (>100µm) with a lateral spatial resolution of 80µm at 3mm depths (Thind, 2006).
More recently, single element real time ultrasonic mouse heart images have been formed
in vivo using mechanical scanning with a lateral spatial resolution of 72µm (Figure
4.1)(Sun, 2007).
Compared to single element ultrasonic systems, transducer arrays offer steering and
dynamic focusing capabilities throughout the image depth. However, at high frequencies,
29
arrays pose challenges in transducer fabrication and multi-channel system design.
Transducer arrays >30MHz have a characteristic wavelength of less than 50µm. The
pitch of the array (spacing between adjacent elements) is designed close to this
wavelength to avoid unwanted constructive interference patterns such as grating lobes
(Cannata, 2002). This poses a difficult fabrication demand of producing very small
elements with a narrow pitch, necessary for high frequency transducer arrays (Cannata,
2004). Additionally, high frequency multi-channel electronic systems are not
commercially available due to high speed and bandwidth requirements of electronic
components, cost, and system complexity (Ritter, 2002). Using a high frequency 48
element linear array and digital beamformer, in vivo real time images of the eye have
been formed providing a 370µm lateral spatial resolution at depths of up to 6.5mm (Hu,
2006). A major pitfall of ultrasonic imaging is that amongst competing technologies,
ultrasonic imaging displays a reduced contrast. The echoes detected in ultrasound are
based on the acoustic properties of the tissue inhomogeneities, which are weakly
backscattered versions of the incident signal, causing poor image contrast.
4.1.2 Magnetic Resonance Imaging
Another current non-ionizing imaging modality is magnetic resonance imaging (MRI).
Conventional MRI images provide fair resolution (~1mm) per image depth. In vivo brain
scans of high resolution MRI systems have provided between 280µm - 500µm in plane
resolutions (Klien, 2005)(Walters, 2003). MRI provides excellent imaging depth well
beyond the ultrasonic and optical limits with the ability to images centimeters in tissue
30
surrounded by bone (Walters, 2003). It also retains a strong image contrast without the
speckle pattern known to optical and ultrasonic imaging techniques. However, MRI
techniques have been unable to deliver the real time imaging capabilities, affordable cost,
and portability of its ultrasonic counterparts.
4.1.3 Optical Imaging
Optical technologies for in vivo real time high resolution imaging include confocal
microscopy and optical coherence tomography (OCT). Confocal microscopy and OCT
are limited to resolutions between 1-2µm at a 0.5mm image depth, and around 10µm at a
1mm image depth, respectively (Fercher, 1997) (Sing, 1998). Light scatters heavily in
turbid media like biological tissue (Cheong, 1990). This high scattering phenomenon,
unlike ultrasound, whose dominant attenuation contribution in biological tissue is from
absorption, results in an imaging depth limitation for optical techniques. This is due to
the fact that the scattering coefficient of light in biological tissue is orders of magnitude
greater than that of ultrasound and causes the light to diffuse quickly. Thus, confocal
microscopy and OCT suffer from severely degraded spatial resolution with increased
depth. Alternately, an advantage of the greater scattering coefficient of light in tissue is
that the optically based backscatter used to create the image is also greater, allowing OCT
to provide a higher contrast than ultrasound. A governing trade off exists between the
two technologies, strong image contrast in optical imaging for superior resolution versus
increased image depths in ultrasound (Duck, 1990).
31
The method behind OCT is largely analogous to ultrasonic imaging. Using
interferometry, it relies on the singly backscattered light to form images (Huang, 1991).
OCT and US are two modalities dependent on a backscattered version of the incident
signal, and both suffer from the effects of speckle. Speckle patterns can degrade the
image and are key issues in OCT and ultrasound. The origin of the speckle pattern is
from multiple, random scattering of particles less than a wavelength in size. The negative
impact of the presence of speckle in ultrasound has experimentally shown a reduction of
the ability to detect a lesion by approximately a factor of eight (Bamber, 1986).
Figure 4.1 Speckle patterns present in ultrasound and optical coherence tomography
scans. a) high frequency ultrasound image of an adult mouse heart RA: right atrium, LV:
left ventricle (Sun, 2007). b) OCT image of a section of human skin. Arrows point to
dermal layers (Weissman, 2004).
100µ
m
a) b)
32
4.2 Photoacoustic Imaging
4.2.1 Introduction
Photoacoustic imaging uses a short pulsed laser to illuminate a medium and generates
photoacoustic waves by means of thermoelastic expansion, described in section 3.2.2.
The photoacoustic waves propagate towards the surface, and are detected by a
piezoelectric transducer. The time-of-flight measurements are then processed to map the
optical absorption distribution within the medium.
Photoacoustic imaging seeks to overcome some of the limitations of current technologies
such as optical imaging and ultrasonic imaging. When comparing the competing
technologies, the role of photoacoustic imaging is to bring some of the advantages of
optical imaging techniques into the high frequency ultrasonic imaging depth range
(Figure 4.2).
4.2.2 Depth
Pure optical imaging, like optical coherence tomography, while providing exceptional
images for very superficial structures, is crippled by its meager penetration depth of no
more than 2mm (Welzel, 1997). Unlike optical coherence tomography, photoacoustic
imaging is diffraction limited, meaning limited by the frequency of the detected
photoacoustic signals and not by optical diffusion. This technology does not depend on
single backscattered light as OCT does. In contrast to OCT, the scattering of light works
as an advantage in photoacoustic techniques because any light, including both singly and
33
multiply scattered photons, contributes to the tissue absorption and improves tissue
illumination homogeneity (Niederhauser, 2005). As a result, the imaging depth
capability far exceeds that of OCT.
Figure 4.2 High Resolution Imaging Technologies. Photoacoustic imaging brings the
contrast of optical imaging into the high frequency ultrasonic depth range without speckle
noise.
34
4.2.3 Speckle
Since photoacoustic waves travel one way to reach the ultrasonic transducer
photoacoustic imaging is essentially speckle free. Just as in ultrasound tomography,
temporally detected transmission echoes are stronger than reflected or backscattered
echoes. The photoacoustic wave is representative of the optical absorption information
of the medium, but is also comprised of an acoustic one way traveling echo. The acoustic
portion of the traveling wave is still subject to the acoustic effects, such as random
scattering of wavelength and smaller sized particles. However, the acoustic wave is
generated from within the tissue. This means it is a one way traveling wave that carries
predominately optical contrast information, not ultrasonic backscatter. Therefore,
speckle resulting from the acoustic properties of the medium does not offer a significant
contribution to degrade the image, and the dominant contrast mechanism is optical
contrast.
4.2.4 Contrast
High frequency ultrasonic imaging bestows good resolution for the image depth
achieved, but is limited by a weak image contrast. Photoacoustic imaging combines the
contrast advantage of optical imaging with the resolution advantage of ultrasonic
imaging. The contrast of the reconstructed photoacoustic image is related to the optical
properties of the tissue, while the resolution is not limited by optical diffusion or photon
scattering like pure optical imaging. Natural sources for strong optical contrast in
biological tissue include hemoglobin and other chromophores present in the tissue
35
(Beard, 2001). In addition to structural visualization, the varying absorption of different
states of hemoglobin allow for some functional imaging capabilities of photoacoustic
imaging (Figure 3.1).
4.3 Applications of Photoacoustic Imaging
Photoacoustic imaging possesses good contrast resolution based on optical tissue
properties, and good lateral resolution per imaging depth. High frequency photoacoustic
imaging can be used as a novel diagnostic tool for characterizing atherosclerotic plaques,
early detection of skin cancer, and visualization of micro-vasculature involved in
angiogenesis in vivo. This type of imaging can provide the lateral resolution capable of
microscopic imaging of skin lesions and subdermal micro-vessels that can be accessed
non-invasively.
4.3.1 Atherosclerosis
Atherosclerosis accounts for more deaths in this nation than any other disorder. It is
caused by the formation plaque within the arteries. Atherosclerotic plaques have three
principal components:
1. Cells, including smooth muscle cells, macrophages, and other leukocytes
2. Connective tissue, including collagen, elastic fibers, and proteoglycans
3. Intracellular and extracellular lipid deposits.
These three components occur in varying proportions in different plaques, producing a
variety of lesions (Robins, 1994). Angioscopy, which is the direct visualization of the
36
blood vessel with a fiber optic, unfortunately cannot image through the vessel wall.
High-frequency ultrasound is superior to angiography, since it can image beyond the
vessel wall, but it has a limited ability to visualize the components of fat plaques with
high contrast. MRIs can differentiate the microstructure of plaque however; unlike
photoacoustic microscopy, they can not do it in real time, at an affordable price, or in a
portable manner which would be useful for monitoring the restenosis process that often
occurs after angioplasty. The powerful image contrast of photoacoustic imaging could
contribute to the differentiation between lipid and water based tissue within and
surrounding the plaque. Lipid and water based tissues have distinct optical absorption
properties that would allow photoacoustic imaging to identify the lipid-full
microstructure in atherosclerotic arteries.
4.3.2 Skin Cancer
Skin cancer including basal and squamous cell carcinoma and melanoma, is the most
common type of human cancer, affecting almost 1 million people each year in the United
States alone (Glass, 1989). Although a high optical contrast enables OCT detection of
superficial lesions associated with neoplasia, many of these lesions develop from deeper
structures >1.5mm, and neither confocal microscopy nor OCT can reach those depths
(Welzel, 1997). An imaging modality such as photoacoustic imaging is capable of
preserving a high contrast while possessing the range of depth that can image further into
the dermis. In addition to early stage detection of neoplasia in the dermal layer,
photoacoustic imaging could have applications in melanoma diagnosis and treatment by
37
identifying and monitoring the irregular borders common with dysplastic nevus, atypical
moles.
4.3.3 Angiogenesis
Angiogenesis, the growth of new blood vessels is an important natural process, occurring
both in health and in disease. Studies in oncology have shown that angiogenesis by
means of formation of new blood vessels within a tumor, or the growth of new blood
vessels between a tumor and surrounding tissues, plays a critical role in tumor growth
and metastasis of cancer (Folkman, 1996) (Yancopoulos , 2000) (Papetti , 2002). Tumors
need to be supplied by blood vessels, delivering oxygen and nutrients while removing
metabolic waste in order to propagate. Additionally, it has been researched that low
metastatic activity of some in situ tumors may be related to the inability of these tumors
to form new vessels (Carmeliet, 2000). While the formation of new micro-vessels can be
part of normal development and wound healing, it is also a key initial step in tumor
progression, since tumor cells induce angiogenesis (Liu, 1999).
Blood vessels have an intrinsic source of strong optical contrast, hemoglobin at its
various oxygenation states (Wang, 2003). This relationship between optical absorption
and hemoglobin makes photoacoustic imaging well suited to visualize microvasculature
dysfunction and monitor growth stages related to the pathogenic process.
38
4.4 Previous Work in Photoacoustic Imaging
4.4.1 Introduction
Currently, there are two main approaches to performing photoacoustic imaging; 1.
photoacoustic tomography (PAT) and 2. photoacoustic microscopy (PAM). The
experimental method of acoustic generation is the same for both techniques. A Q-
switched pulsed laser is used as the electromagnetic source to irradiate tissue. This setup
begins with a pumping laser (usually a Nd:YAG type), which provides a continuous wave
beam at a specific wavelength. The Q-switch is essentially an aperture which allows the
light to accumulate, then releases it, sending out a burst. The light can be coupled to a
fiber or directed through a lens at the target medium. The photoacoustic waves thereby
generated are detected by the ultrasonic transducer or transducer array, and traverse
receiving, signal processing, and digitization steps before the electromagnetic absorption
distribution is reconstructed to create the image.
4.4.2 Photoacoustic Tomography
Photoacoustic tomography uses very wide band transducers to acquire the echoes, by
following a 360° circular path over a surface that encloses the medium (Xu, Wang,
2005). The circular scanning technique used in PAT is termed forward mode detection.
The tomographic method applies a reconstruction based on the internal photoacoustic
source distribution of the tissue and the spherical wave equation described in 3.2.2.
Reconstruction algorithms using temporal backward projection (Kostli, 2000) and the
half time reflectivity tomography paradigm have been derived in the literature (Anastasio,
39
2005). Phantom images of a 4cm medium have been shown with this method (Su, 2005).
Additionally, in situ images of the lesions on the superficial layer of a rat brain have been
produced via photoacoustic tomography (Wang 2003). These images were obtained
using the time domain reconstruction algorithm for spherical geometries developed for
thermoacoustic tomography (Xu, Wang, 2002).
Another PAT study used a 3.5MHz single element wideband transducer (88% at -6dB)
with a Panametrics pulser/receiver to image a rat brain. The transducer was driven by a
step motor scanning around the rat head obtaining data in the image plane at each
position. The lateral spatial resolution of the system was measured at 200µm with a
phantom. Figure 4.3 shows visualization of the brain structure and the lesion (Wang,
2003). Although the photoacoustic tomography scans produce good resolution at
increased depths, they can only be accomplished with elevated objects such as the breast
or brain, which allow full 360° circular scanning. In addition, obtaining these images
requires precise motor control, and many scans. Scans of this nature can take hours to
complete and offline reconstruction demands high computational times, further
strengthening the allure of photoacoustic imaging with transducer arrays.
40
Figure 4.3 a) PAT image of a superficial lesion (1mm x 4mm, in the right cortex cerebri
area) on a rat brain acquired with the skin and skull intact. RH, right cerebral hemisphere;
LH, left cerebral hemisphere; L, lesion. b) Open-skull photograph of the rat cerebral
surface acquired after the PAT experiment (Wang, 2003).
The second method, photoacoustic microscopy, uses simple time-of-flight measurements
to obtain two dimensional scans, similar to ultrasound. All of the mechanisms for
inducing photoacoustic waves are the same as photoacoustic tomography, while the
method of scanning and reconstruction differ. This technique is faster and more intuitive
as, it does not require complicated iterative backprojection algorithms for reconstruction.
Photoacoustic microscopy is referred to as backward or reflection mode photoacoustic
imaging because the detection takes place at the site of laser irradiation. Consequently, it
becomes important to use focused ultrasonic transducers to localize the photoacoustic
sources in linear or sector scans. There is a significant advantage in scan time of
photoacoustic microscopy compared with photoacoustic tomography. Tomographic
images from circular mechanical scans can take hours to produce. Experiments in vivo or
41
in situ are effected by the by the passage of time. Difficulties arise in maintaining the
animal position for lengthy scan times, since animals must be restrained and/or
anesthetized during the process to reduce motion artifacts. Also, the condition of the live
animal will change over time and in some cases, the animal may expire before the scans
are complete. This is certainly a problem for functional imaging as oxygenation and
blood profusion vary after expiration.
Recently, a single element photoacoustic microscopy system has been able to image
vascular structures and demonstrate functional abilities by providing hemoglobin
saturation levels of animals in vivo (Zhang, 2006). This method uses “dark field”
illumination technique where the light is confocally focused to create an annulus which
overlaps the focus of the transducer. Functional photoacoustic images of micro-
vasculature were constructed by scanning the same area with multiple wavelengths of
light, then combining the absorption information from each wavelength’s scan into one
image (Figure 4.4). These images were constructed offline using a single element
transducer and a lens to create a high numerical aperture. Since this type of functional
image requires multiple scans, the only true potential for real-time imaging would require
an increase in the laser pulse repetition frequency as well as the advancement to high
frequency array technology and multi-channel receiving systems.
42
a) b)
Figure 4.4 Functional photoacoustic dark field microscopy. a) Image sub-dermal micro-
vessels in a Sprague Dawley rat. Images from four wavelengths (578nm, 584nm, 590nm
and 596nm) were combined using least squares fitting to create vessel sO
2
mapping. b)
Microsphere-perfusion image showing arterioles (red) and venules (blue).
4.5 Research Goals
Increased transducer frequencies improve lateral resolution in ultrasound and
photoacoustic imaging alike. Until now, all of the effort in high frequency photoacoustic
imaging has been executed with single element transducers. The advantage of high
frequency (>20MHz) photoacoustic imaging with arrays is evident. The same benefits
ultrasound technology gained from evolving to transducer array systems can be applied to
photoacoustic imaging. The aperture flexibility, removal of the need for mechanical
scanning, beamforming, and beam steering advantages attest to the validity of the
progression towards array technology in photoacoustic imaging. The barrier in
progression towards high frequency photoacoustic imaging with transducer arrays is
primarily due to two main challenges: 1. Fabrication of high frequency arrays and 2.
The need for custom high frequency systems that push the upper limit of electronic
component speed and bandwidth in a parallel multi-channel system environment. This
research will attempt to address the inadequacies of current photoacoustic imaging
43
technologies by offering the design of a 16 channel photoacoustic imaging system which
can provide optical absorption information at high frequency ultrasonic imaging depths.
The broadest aim of this research is to enhance the state of high resolution non-invasive
imaging by crafting the next step in the development of photoacoustic imaging.
Specifically, the goal of this thesis is to present the design and viability of a novel high
frequency array-based photoacoustic microscopy system (PAM), capable of multi-
channel parallel processing, and visualization of micro-vascular structures in biological
tissue. The PAM system holds promise for applications in oncology and in tracking
angiogenesis for real time in vivo imaging.
44
Chapter 5: Design of a High Frequency Photoacoustic Microscopy
System: Part I
5.1 Purpose
The goal of this project is to create a 16 channel high frequency photoacoustic imaging
system for a 48 element 30MHz array that provides analog receive signal processing for
proof of concept and system characterization. The first version of the system uses
oscilloscope data acquisition by down multiplexing. Photoacoustic images are produced
by offline reconstruction methods based on radio frequency (RF) data collected from all
transducer elements.
5.2 Introduction – High Frequency Design
The design rules held steadfast for conventional ultrasonic systems cannot be easily
applied to higher frequencies. The detection of ultrasonic frequencies above 20MHz
using transducer arrays imposes a few stringent demands on system design.
1. Wide bandwidth electronic components operating in high frequencies while
providing ultra low noise ratings.
2. Signal isolation of multi-channel systems.
3. Low noise tolerance over large system gains.
Advancement in the availability of high speed components has made amplifiers with
gain-bandwidth products well into the gigahertz range. However, many are not well
suited for detection in high frequency ultrasonic systems because of their noise ratings.
Although wideband, ultra low noise, stand alone, commercial amplifiers are accessible
45
for single element high frequency ultrasound and photoacoustic imaging, multi-channel
design requires a systems approach. In this sense, the behavior of a single component, or
single channel may not be suited to characterize the system as a whole. Each component
of each channel applies noise that behaves in an additive manner. Multiple signals of
interest in one high frequency system require isolation from other signals, and from
ground/power plane noise. The front end receive circuitry can set the tone for the entire
system. High frequency front end receiver design is governed by three components that
exist in a trade off, gain, linearity, and noise (Spiridon, 2003). Once noise or signal
distortion is established at the first stage, the signal corruption is virtually impossible to
remove in subsequent stages. Low level signals from exceptionally small transducer
elements undergo attenuation due to an impedance mismatch and the effect of lossy
cables. These signals demand large gains to ensure accurate detection for digital
conversion. Amplifiers with remarkably high gains can quickly become unstable and
oscillate, suggesting a multi-stage amplification approach. In addition, system noise,
sometimes within the signal bandwidth, is introduced by each amplification stage. A
careful balance must be struck between sensitivity of the front end, and available system
gain.
5.3 Materials and Methods
5.3.1 System Design
A short pulsed Nd: YAG laser irradiates the sample inducing acoustic waves. This laser
is a Q-switched (Quantel, Brilliant B, Harpenden, UK) operating at a wavelength of
46
532nm and delivers up to 500mJ per pulse. The YAG laser is then used to pump a dye
laser (Continuum ND6000, Santa Clara, CA). The dye laser is essentially used to tune
the light to a different wavelength. The broad emission from the dye then passes through
a cavity that amplifies a specific wavelength. It produces 6.5ns pulses of 584nm light, at
10Hz pulse repetition frequency (PRF). The dye laser emits around 80mJ per pulse,
which is still too high for biological tissue. To reduce pulse energy, the light passes
through a diffuser stage. Flexibility in target illumination includes coupling the laser
pulses into an optical fiber. After the ground glass diffuser stage, a lens is used to direct
the light into the core of the 600µm fiber. Approximately 3mJ of light is delivered
through the fiber to a 6mm x 3mm area. Positioned obliquely to the array, the fiber
causes an elliptical illumination pattern to form delivering 6mJ/cm
2
incident light.
Once the tissue is illuminated, the generated photoacoustic waves are received through a
48 element linear array, centered at 30MHz. Echoes detected by the transducers are
applied to the receiver, where they undergo several signal processing stages including
noise filtering and multi-stage amplification. Lastly, the data is digitized and stored for
post processing. Figure 5.1 shows the overall system structure. The process initiates
when a TTL signal from the parallel port of the PC initiates the first trigger which
controls the laser firing. The second trigger sent from the port is the data acquisition
trigger for the oscilloscope, the multiplexer trigger that selects the first group of 16
elements, and for the second set of multiplexers that select 4 of the 16 channels. These
triggers repeat for a 4 channel subaperture to accommodate the oscilloscope 4 channel
47
limit. The data acquisition’s trigger is delayed due to the latency between initial laser
triggering, the Q-switch for laser pumping, and actual time of the light delivery to the
sample.
Figure 5.1 PAM I overall receive system structure for noninvasive high frequency array-
based imaging.
Pre-Amp M
U
X
3:1
Dual Variable Gain
Amplifier
BP Filter
Fixed Amplifier
PC Interface
Data Storage
& Control
Gain
Control
1. Front End 2. Signal Processing
3. Interface/Control
Transducer Array
4:1
Multiplexer
1-48
1-16
1-16
GPIB
1
2
1
2
Analog Board
1-4
Signal Isolation
3
4
Pre-Amp
Pre-Amp
M
U
X
3:1
Pre-Amp
Pre-Amp
Pre-Amp
Oscilloscope
Laser Trig
Pulse Gen
48
5.3.1.1 30MHz Transducer Array
Difficult challenges in fabrication demand novel design and limit the availability of high
frequency transducer arrays (Ritter, 2002). The 48 element transducer array used in the
PAM system is constructed of a piezo-composite material. The fabrication process for
the transducer array is reported by Cannata, et al. (2005). The pulse echo response below
shows a -6dB fractional bandwidth of 50% (Figure 5.2).
The array bears a 100µm pitch (2 λ), a 2mm elevation aperture and an 8.2mm focal point.
The average measured center frequency of the array was 27.6MHz with the highest and
lowest resonances at 26.9MHz and 28.3MHz. Elements 1 and 5 are open, and the
element crosstalk was shown to be less than -27dB over the passband (Figure 5.3).
Measurements for the array are taken with a single representative element. In this case,
element #19 was used and the crosstalk was mapped for its adjacent elements.
Individual RF data from every array element is used to form a single scan line. In this
way, the array is used to steer and focus the beam as a phased array. However, the
original construction of this array was intended for use in linear scanning with a sliding
subaperture of fewer elements. Since elements are spaced at 2 λ instead of .5 λ, ideal for
phased arrays, it is important to consider and estimate the location of grating lobes. A
simulation of the directivity pattern from all elements in the array was generated using
Field II® (Jensen, 1992) (Directivity.m, Appendix A). Using equation 2.3 we confirm
the presence of grating lobes at ±30˚ in the plotted directivity pattern in Figure 5.4.
49
Typical Two Way Pulse (30 MHz Array Element 19)
Time (µs)
10.8 10.9 11.0 11.1 11.2
Amplitude (Volts)
-0.6
-0.4
-0.2
0.0
0.2
0.4
0.6
Typical Two Way Pulse FFT (30 MHz Array Element 19)
Frequency (MHz)
10 20 30 40 50
Amplitude (dB)
-50
-40
-30
-20
-10
0
Figure 5.2 Two way pulse and FFT frequency response for array element #19 in the
30MHz linear array.
50
Frequency (MHz)
20 25 30 35 40 45 50
Crosstalk (dB)
-60
-55
-50
-45
-40
-35
-30
-25
-20
Nearest
Next Nearest
Third Nearest
Figure 5.3 Combined Acoustic/Electric crosstalk for a representative array element of
the 30MHz linear array as a function of frequency.
Figure 5.4 Field II simulated directivity pattern for a 30MHz, 48 element array spaced at
2 λ.
51
5.3.1.2 Analog Receiver Design
High speed design emphasizes the behavior of passive circuit elements including circuit
boards, integrated circuit packages, and signal traces which directly affect electrical
performance. The printed circuit board layout must be designed such that the effect of
the PCB is transparent to the circuit.
A useful parameter to characterize a system is dynamic range. Dynamic range however,
can have different meanings depending upon the application. With respect to analog
receiver electronics, the dynamic range can be defined as the ratio of the largest
undistorted signal to the smallest, defined by the noise floor and expressed in dB.
min
max
) (
) (
log 20
i S
i S
DR =
(5.2)
The receiver of the system consists of an analog front end shown in blocks 1 and 2 of
Figure 5.1. The array connects with the electronics through 48 RG-174 co-axial cables
with gold SMA board connectors (Ampenol, Wallingford, CT). The receive circuit
(Block 1-1, Figure 5.1) includes a broadband ultra low noise, 0.75nV/ √Hz, MAX4107
(Maxim/Dallas Semiconductor, Sunnyvale, CA) voltage feedback pre-amplifier dedicated
to each of the 48 elements in the array. The noise rating at this stage is imperative for
two reasons: 1. The low signal level from the transducer is worsened by attenuation
caused by “lossy” cables 2. Higher noise levels on the first amplification stage would be
amplified on subsequent stages therefore, the first stage sets the performance limit for the
entire system. The input of the pre-amplifier has a large, 1K Ω resistor in parallel to
create a voltage divider, driving most of the signal into the input. Since this is a receive-
52
only front end, cable and element impedance matching circuits which transmit maximum
power at the cost of half the signal strength are wasteful. Voltage is maximized since it is
the parameter of interest. Selection of the feedback resistor is key in the circuit. Large
values will increase voltage noise and interact with the amplifier’s input and board
capacitance to generate undesirable poles and zeros. Poles within the anticipated signal
bandwidth will jeopardize stability and cause oscillations. On the flip side, reducing the
feedback resistor will extend the pole outside of the amplifier bandwidth, but could limit
the output swing by adding more of a load in parallel with the amplifier load. R
feedback
was chosen at 30 Ω to provide a stable +20dB gain.
2
3
4
6
7
4106
U7
50
R3
30
R1
300
R2
50
R4
0.1u
C3
0.1u
C1
0.1u
C5
Bd
EX22
Bd
EX20
10u
C2
10u
C4
+5V
-5V
PRE_AMP_OUT
GND
GND
GND
TRANS_ELE
GND
Figure 5.5 PAM front end pre-amplifier circuit design.
The 16 active channels for the system are formed by utilizing three inputs of a 4:1
multiplexer (AD8184, Analog Devices, Norwood, MA). There are 16 multiplexers
53
which require two control bits to select between elements (Block 1-2, Figure 5.1). Both
these initial stages operate on -5V and +5V power supplies. Following channel
formation, the passive band pass filter (Block 2-1, Figure 5.1) is used to remove spurious
signals with frequencies out of the desired transducer response. A fourth order
Butterworth filter best fit the requirements. Butterworth filters are characterized by a flat
pass band response. Since the transducer response is wideband, a uniform response
around the resonant frequency is desirable. This is due to the even spacing of poles.
Visualizing the behavior of low pass filter is an excellent place to begin design. Figure
5.6 below shows the complex response of a 4
th
order low pass Butterworth (Maxim,
2001). The approximated normalized transfer function for this filter takes the form:
+ ⋅ +
+ ⋅ +
≅
1 26
500
181
1 26
20
3
1
) (
2 2
s s s s
s H
(5.3)
where s is the complex frequency (Boulter, 2000). As soon as the effect of one pole
begins to wear off, the next pole takes over, yielding a flat pass band.
54
Figure 5.6 Complex frequency response for a Butterworth 4
th
order filter(Maxim, 2001).
The trade off is that a Butterworth filter does not have the sharp cutoff of some of its
counterparts like the Chebychev filter which suffer from significant ripple in the pass
band region. Although active filters minimize loss, they require very high performance
op-amps for MHz frequencies. Passive lumped element filters are often used in radio
frequency, RF and microwave applications (Drozd, 1997). The pi structure filter was
first simulated in PSPICE (Figure 5.7). This simulation uses ideal Q parameters for the
frequency dependant components and does not take into account any stray capacitances
that may exist on the PCB. For comparison, the frequency response of the filter
implemented on the receiver board was measured with a spectrum analyzer (Figure 5.8)
(Agilent Inc., Santa Clara, CA). Both simulation and experimental spectrums show a
wide response, larger than the transducer frequency response.
55
Figure 5.7 Simulated (PSPICE) frequency response for Butterworth 4
th
order band pass
filter.
Figure 5.8 Measured frequency response for Butterworth 4
th
order band pass filter.
56
The multiplexing and filtering stages cause a loss in signal amplitude. Amplification is
achieved through the AD8332 (Analog Devices, Norwood, MA), a dual variable gain
amplifier (VGA) with a low noise amplifier (LNA) stage built in (Block 2-2, Figure 5.1).
The variable gain is user selected by a simple voltage divider circuit and manual turn
potentiometer ranging from 0 - 0.9V, 0.9V corresponding to the maximum possible gain.
This stage offers a gain range of +(0 - 40)dB. This VGA requires a good deal of external
circuitry. The output is immediately filtered by a simple capacitor inductor pair. The
dual package accepts a single ended input and converts the output to a differential signal.
Although a bulky solution, 1:1 transformers (T1 6T, Minicircuits, Brooklyn, NY) present
an appealing solution which not only converts back to a single ended signal, but also
provides additional signal isolation (Block 2-3, Figure 5.1). The analog signal processing
stage is concluded with a voltage feedback, fixed gain amplifier (OPA847, Texas
Instruments, Dallas, TX) which provides another +20dB fixed gain (Block 2-4, Figure
5.1). The full hierarchy and schematic design for front end, and analog signal processing
portions of the receiver are included in Appendix D.
After completion of the analog system, initial experiments showing system feasibility
were carried out by further multiplexing 16 channels to 4 channels for oscilloscope data
acquisition (Block 3, Figure 5.1). Using 4 MAX4141(Maxim/Dallas Semiconductor,
Sunnyvale, CA) the channel data were sampled with a Tektronix 5054 Digital Phosphor
Oscilloscope capable of sampling rates up to 1.5Gs/s. In oscilloscope fast acquisition
mode, the data sample record length taken is adjustable. To reduce acquisition time, the
57
record length was reduced to 2500 samples for each experiment. The oscilloscope
acquisition trigger was provided by the external pulse generator that also triggered the
laser. The acquisition was delayed from the trigger by 300-500µs to account for the
delay of light delivery due to laser pumping and Q-switching.
5.3.1.3 PCB Layout Considerations
Printed circuit board (PCB) layout and routing was implemented manually using
Altium© Protel DXP software. High frequencies in signal processing dictate the level of
complexity of board layout. A four layer board was used to facilitate component stacking
on top and bottom layers while reserving an internal layer for ground and another for a
split power plane (-5V and +5V). Large tantalum coupling capacitors are used at the
entrance junction of the power plane. These capacitors have a low ESR and can filter out
low frequency noise present in the power supply. Low level echoes subjected to a series
of high gains amplifiers must be protected from additional noise brought on by the range
of signals interacting with dielectric properties of the PCB.
The linear amplifiers are particularly sensitive to power rail fluctuations. To compensate
for this issue, local coupling capacitors are present at every via connected to the power
plane. However, every via placed on a multilayer PCB also introduces parasitic series
inductance, whose primary effect is to degrade the effectiveness of power supply bypass
capacitors (Johnson, 1993). Vias connected to power planes are kept closer to the
58
coupling capacitors than to power pins on the device. The 4 layer PCB layout for the
analog portion of the receiver can be found in Appendix D.
5.3.2 Software Development
The following sections discuss the software developed as a user interface to control
hardware operations, and the algorithm for beamformation and image reconstruction.
5.3.2.1 System Control
A Labview (National Instruments, Austin, TX) program was created as a user interface.
The program provides control signals, and initiates the transfer from the oscilloscope to
the PC memory. Two TTL multiplexer control bits were provided from the PC parallel
port to the 16 multiplexers on the analog main board. A General Purpose Interface Bus
(GPIB) was used to transfer data from the scope to the PC hard drive. The Labview
interface for PAMI can be found in Appendix C.
5.3.2.2 Image Reconstruction
To create a truly robust image reconstruction, raw channel data is stored in the PC
memory and all beamforming and post processing is achieved through software. This
lends a unique flexibility in reconstruction since all of the raw RF data from 48 elements
is accessible. Currently, the other high frequency 16 channel linear array ultrasonic
systems use analog and digital beamforming via hardware, leaving individual raw
channel data completely inaccessible (Xu, 2006) (Hu, 2006). Image reconstruction and
59
post processing was executed in MATLAB (MathWorks, Natick, MA). The program
Arraysectorscan.m (Appendix B) was written to calculate time delays and apply beam
focusing and steering. It should be noted that the photoacoustic equation to calculate
time delays for beamforming is slightly different from equation 2.6 used for conventional
ultrasound.
t c
x
c
x
a
n
a
n
n 2
2 2
2
cos sin ϕ ϕ
τ + ≈ ∆ (5.1)
The steering term is the same as equation 2.6, while the focusing term has an extra factor
of two in the denominator. Instead of the two way relation, c
a
t =2R, which accounts for a
roundtrip path, the distance the echo travels in photoacoustics is one way, thus c
a
t = R,
altering the denominator of the steering term. The image is calculated using 128 scan
lines and a 45° scan angle. All 48 elements are used in the reconstruction. This provides
a larger effective aperture yielding a tighter focus and better lateral resolution. Using
equation 2.9, we estimate the best possible theoretical lateral resolution to be around
85µm. Steering and focusing delays are applied to every element for every scan line with
respect to the center element as the reference. The delayed photoacoustic echoes are then
summed together, creating a single beamformed scan line. To further improve the quality
of the image, coherence factor weighting is used. This technique will help correct for
steering errors, increase the SNR, and reduce the effect of grating lobes. The coherence
weighting factor is applied to every point in the scan line. This is the new beamformed
and weighted data. For envelope detection, the Hilbert transform is applied, and
normalized amplitude measurements are converted to the log scale.
60
The reconstruction algorithm allows an image to be viewed in two ways: as a surface in
a 3D mesh, or as a 2D B-mode type grayscale scan. Since the information is based on
128 scan lines, an interpolation method is used as a scan converter to map the sector scan
image data onto an x-y raster line format. A 3D mesh is constructed by plotting the
photoacoustic data as the amplitude of lateral(x) and depth(z) dimensions. The dynamic
range of the 3D plot is mapped to an intensity color bar. This plot can be used to
calculate an appropriate dynamic range for displaying an image. It is also flattened to 2D
for system characterization calculations such as determining axial resolution and plotting
the point spread function. To create the photoacoustic image, the beamformed and
weighted data is treated as surface plotted by the pcolor Matlab function. This function
generates a 3D surface with a bird’s eye viewing angle parallel to the xy image plane.
This viewing angle provides a 2D image, displayed in grayscale.
5.3.3 Phantom Description
The phantoms used for imaging were a dark human hair with a diameter of 80µm and a
6µm carbon fiber. These phantoms were chosen because of their high optical absorption
properties. The fibers were laid taut across the length of the container and fixed with
epoxy to mounting blocks on each end illustrated in Figure 5.9. The blocks were then
fixed to the bottom and the bath filled with water. Intralipid™ solution often used in
optical imaging experiments was used as a tissue mimicking phantom. A fatty
suspension similar to milk, Intralipid™ consists of phospholipids micelles and water. It
is turbid and has no strong absorption for light in the visible region. It is a highly
61
scattering medium, like biological tissue, and has well defined scattering properties over
a broad spectrum of wavelengths (Jacques, 1992).
Figure 5.9 Imaging phantoms of carbon fiber and human hair.
5.3.4 Experimental Setup
The transducer array and optical fiber were mounted on a 3-axis translation stage. For
the phantom images, the array is lowered into the water bath and the laser is then
positioned by hand. Positioning of the coupling fiber and construction of a good
illumination pattern is a sensitive and time consuming process. Once the optimum signal
is received from the center element of the array the laser fiber is fixed (Figure 5.10). For
phantom imaging the tank shown in Figure 5.9 was used. For animal imaging the
transducer is lowered into a water tank with about a 5cm
2
aperture cut out of the bottom.
The Sprague Dawley rat is raised to fill the aperture. A layer of cellophane creates the
water seal between the rat’s skin and the water in the tank. Cellophane was chosen as a
62
mostly acoustically and optically transparent medium (Figure 5.10). The animal is
shaven and euthanized directly prior to imaging. It is then fixed on the stage beneath the
tank. Since the laser PRF is only 10Hz, this is done to reduce image artifacts resulting
from breathing, muscle twitch, and cardiac induced motion. Imaging experiments were
performed in the optical imaging laboratory at Texas A&M University.
Figure 5.10 Animal experimental setup for PAM imaging.
Transducer Array
63
Figure 5.11 Photograph of transducer array and optical fiber mounted on translation
stages.
5.4 Results
5.4.1 System Performance
The total system frequency response showed a -3dB bandwidth of 35MHz. The passband
frequencies ranged from 8MHz to 44MHz, which has a sharper cutoff than the simulated
model. Essentially, the filter may also double as an anti-aliasing filter for the benefit of
the analog to digital conversion. The passband also exhibited a ripple of <1dB. This
may be due to the contributions of all component frequency responses summed over the
pass band. Certain amplifiers may pull down response of the gain at a few select
frequencies while contributing more in other frequencies. Measured gain and loss stages
are summarized in Table 5.1. There is a significant loss due to the passive filter,
however, all of it is recovered and increased by the subsequent stages.
64
To characterize the sensitivity of the front end the minimum detectable signal was
measured using the setup in Figure 5.12. The minimum detectable signal is used for
sensitivity characterization instead of SNR since the variable gain causes SNR to vary
over the gain range. Here, the function generator provides a 10mVp-p, 20MHz sine
wave. The signal passes through an attenuator and the receiver output is connected to the
oscilloscope. The signal from the function generator is attenuated in steps until the sine
wave can no longer be distinguished and is lost below the noise floor. At 30dB
attenuation the minimum detected signal was 316µV.
Figure 5.12 Setup for measuring minimum detectable signal, noise floor, and dynamic
range of analog receiver.
Table 5.1 Summary of gain and loss for each receiver signal processing stage.
Experimental Measurements
Stage 1: Preamp +19.5dB
Stage 2: Mux -2.7dB
Stage 3: BP Filter -6.8dB
Stage 4: VGA +(0 – 37)dB
Stage 5: Amplifier +25.2dB
Total Gain ~ (33 – 73)dB
65
Figure 5.13 Photograph of the complete analog receiver PCB for the PAM system.
Dynamic range measurements for the receive board were inspected at minimum and
maximum VGA gains. Both measurements produced >75dB dynamic range. The
maximum dynamic range dictated by bit width of the A/D converter on the scope, which
uses 8 bits, thus providing 48dB. It is also a logical expectation for the noise to increase
drastically as it is amplified through such high gain stages. Despite noise filtering, any
unwanted signals within the pass band will tag along for the ride and also receive the
benefits of amplification. Figure 5.13 shows the completed populated receiver analog
board.
66
5.4.2 Images
The following section presents a progression of images from phantom targets in water to
animal micro-vasculature PAM imaging. Images from this version of the photoacoustic
microscopy system are referred to as PAM I.
5.4.2.1 Carbon Fiber in Water
To demonstrate the spatial resolution of the system, PAM I image data was collected for
a 6µm carbon fiber in water, placed at the transducer focal depth of 8mm. The
oscilloscope sampling rate was 250MHz and the delay between laser trigger and
acquisition was 2.4µs. The variable portion of the applied system gain was measured as
58dB. Figure 5.14 shows the time domain representation of a detected photoacoustic
wave from element 19, chosen arbitrarily. The envelope detected raw RF data from all
active channels form a single matrix, showing open elements and delays from the furthest
array elements (Figure 5.15). Photoacoustic echo arrival time increases as a function of
distance from the center array element.
Lateral spatial resolution was measured by projecting image data onto the X-axis (lateral
dimension) and creating a point spread function (Figure 5.16b). As a means of
comparison and validation of the coherence factor weighting method, a point spread
function of the same data was constructed without CF weighting (Figure 5.16a). Using
the same method, axial resolution can be measured through projection onto the Y-axis
(depth dimension). After beamforming and coherence factor weighting is applied, the
67
noise levels were reduced to less than -35dB with no averaging. The -6dB axial and
lateral spatial resolution for the carbon fiber target near the focus were measured as 50µm
and 97µm, respectively. An axial resolution of 25µm has been reported with this system
(Zemp, JBO, 2007). This axial resolution was measured by superimposing the signal
from a single carbon fiber with the signal from the same fiber translated vertically, and
examining the envelope. The separation distance between two distinguishable peaks was
used as the axial resolution figure of merit, 25µm. However, for the purposes of this
research the -6dB figure is used.
A 3D mesh image of the carbon fiber shows the magnitude and dynamic range of the
photoacoustic data mapped onto a color bar (Figure 5.17). By changing the viewing
angle to be parallel with the XY plane, a grayscale photoacoustic B-mode image of the
carbon fiber in water was also created (Figure 5.18). The image is displayed with a -
35dB dynamic range.
68
Figure 5.14 PAM I raw RF data from element #19 using a 6µm carbon fiber target.
Figure 5.15 PAM I envelope detected raw RF data from all active channels form a single
matrix. Open elements 1, 5 and element 34 have no data.
69
Figure 5.16 Point spread function of PAM I 6µm carbon fiber a) without coherence
factor (CF) weighting b) with CF weighting, measured -6dB lateral spatial resolution of
100±5µm. Clutter levels are reduced to below -35dB.
a)
b)
70
Figure 5.17 PAM I 3D mesh representation of 6µm carbon fiber in water.
Figure 5.18 PAM I B-mode image of a 6µm carbon fiber in water.
71
5.4.2.2 Carbon Fiber and Human Hair in Intralipid
After measuring the spatial resolution of the system, sensitivity was tested by imaging
phantoms in tissue mimicking solution. Highly optically scattering 1% Intralipid solution
was used in place of a water bath. The images in Intralipid were sampled at a rate of
1.25GHz. An image of an 80µm dark human hair in solution, at 58dB gain with no
averaging is shown. The B mode image was also generated at the same gain condition
for the carbon fiber in the Intralipid solution shown. As a method of truncation, a
selection on the scope was set to acquire a smaller data record length, thus shrinking the
buffer size. This setting applies only to images in Figures 5.19 and 5.20. The displayed
image depth is only a section of the image with the target inside the focal range. Both
images are shown with no signal averaging.
Figure 5.19 PAM I image of 80µm human hair at 7.3mm depth in 1% Intralipid solution.
72
Figure 5.20 PAM I image of 6µm human hair at 7.2mm depth in 1% Intralipid solution.
5.4.2.3 Carbon Fiber Matrix in Water
To express the dynamic focusing capability a composite image was constructed by
imaging the 6µm carbon fiber in evenly spaced positions. A single carbon fiber was
imaged, then repositioned and imaged while the array was in a fixed position. The
images from each carbon fiber represented a matrix of target locations and were stitched
together to create the composite image. Increasing the data record buffer back to the
original conditions, the sampling frequency was again set to 250MHz. The carbon fiber
matrix shows the extent of the focusing capabilities of PAM throughout the depth of view
(displayed with 35dB dynamic range) (Figure 5.20) (Zemp, JBO, 2007).
73
Figure 5.21 PAM I dynamic focusing: Composite image of a 6µm carbon fiber matrix in
water.
5.4.2.4 Animal Microvasculature
The ability to create phantom images in tissue mimicking solution justified the
progression towards animal imaging. A young Sprague Dawley rat (approximately 100g
in weight) was imaged, using an arbitrary averaging index of 40, to demonstrate the
ability to visualize micro-vessels in biological tissue. Figure 5.22 shows a photoacoustic
B scan image at one imaging location with several small vessels visible at 2-3mm depths
below the skin’s surface, which is approximately 6mm below the transducer array.
Bright echoes indicate cross sections of small vessels in the rat’s subdermal layers.
74
Figure 5.22 PAM I photoacoustic B-scan of subcutaneous micro-vessels in a young rat
visualized 2-3mm below the skin’s surface displayed with 35dB dynamic range.
A photoacoustic C-scan was formed using a sequence of 51 B-scans parallel to the
transducer face, acquired at 254mm elevational intervals (Figure 5.23). The maximum
intensity along the axial direction was projected onto the C scan imaging plane (Figure
5.24). It is essentially a flattened image where the scans at different depths are overlaid
onto the same plane. The image can be compared to the photograph of the underside of
the excised skin from that region, removed after photoacoustic imaging. These vessels
were not visible to the human eye from the outside of the skin’s surface.
Figure 5.23 A series of PAM scans with increasing depth are projected onto a single
image to form a C-scan image.
Scans parallel to the
transducer surface
75
Figure 5.24 a) Photograph of the underside of excised rat skin showing micro-vessels
and b) PAM I C-scan in the identical region as part a) (Zemp, JBO, 2007).
5.5 Discussion
A photoacoustic imaging system using a 30MHz linear array and custom receive
electronics has been presented. Receive electronics were designed from schematics, to
board layout, to assembly using a 4 layer PCB and commercial integrated circuits.
The frequency response from the filtering circuit is in good agreement with the
simulation. The frequency response of the analog receiver as a whole provides enough
76
bandwidth to not only accommodate the bandwidth of the 30MHz array, but also has
enough spectral characteristics to provide for another array with a greater bandwidth in
the future. A wider bandwidth transducer response could provide great image
improvement. Because the photoacoustic effect will produce waves at varying
frequencies, a larger bandwidth transducer will allow photoacoustic signals to lie within
the full width half maximum (FWHM) of the frequency response. For example, the
spectrum of the photoacoustic echo produced for a larger diameter target, such as the
hair, generated an echo in the 18-20MHz frequency range; while the carbon fiber
produced a photoacoustic pulse closer to the transducer resonant frequency, around
30MHz (Figure 5.14). By inspecting the RF data in Figure 5.14, we can also see a
ringing occur just after the detected echo. This effect can be seen in some of the
photoacoustic phantom images (Figures 5.18 – 5.20). This may be due to the absence of
impedance matching between the element output and pre-amp input. Matching resistors
to ground were removed in order to detect the lower level voltages close to the noise
floor, maximizing system sensitivity.
The front end electronics have a sensitive response that can detect echoes well into the
microvolt range. However, measurements of minimum detectable signal, although taken
from a populated PCB, were taken when a signal was applied to a single channel. This is
the best possible figure under normal operation because signals applied to all 48 channels
may contribute additively to power plane noise. This measure of sensitivity will also
77
naturally vary for higher gains applied to the system, where the noise floor may prevent
microvolt signals from being detected.
Images of carbon fibers demonstrated a -6dB axial resolution of 50µm and a -6dB lateral
spatial resolution of 97µm, with the target in the focus (Zemp, JBO, 2007). Coherence
factor weighting shows a great improvement in noise reduction by suppressing clutter
levels from -18dB to below -35dB (Figure 5.16). Some of the system noise may be
attributed to the connection of the noisy digital ground of the computer’s parallel port to
sensitive analog ground reference on the receiver board. This connection was necessary
to deliver the control bits to the multiplexers. The resolution of the carbon fiber images
is pleasantly superior to analogous ultrasonic array systems at the same frequency, which
report lateral spatial resolution over twice that amount (Hu, 2006). This is a result of the
ultrasonic system’s smaller effective aperture, using 16 elements instead of 48 elements
to focus the beam. As discussed in 5.2.1.1, we see that grating lobes are expected at ±30°
(Figure 5.4). Their effect is not immediately seen in the phantom images due to the 45°
scan angle, but can also be reduced significantly as a result of a short echo pulse, and
particularly, the effects of CF weighting.
The 3D mesh image of the carbon fiber in water shows the noise floor of the image as
well as what might be a hint of a grating lobe artifact in the negative lateral direction. It
confirms the use of -35dB as an appropriate dynamic range for the photoacoustic image.
78
Images obtained from the hair and carbon fiber in tissue mimicking solution were in the
first experimental setup. In that setup, the small data record lengths caused the image to
only show a portion of the depth where the target was just barely present, and the light
delivery setup was not maximized as in subsequent PAM experiments. However, these
images were to serve as a base sensitivity validation for biological tissue scans. The
carbon fiber image and human hair in Intralipid fluid introduced a good deal of noise into
the image (Figures 5.19, 5.20). Since the echo is weaker, noise from high gain was
introduced and image display dynamic range was limited to 30dB. Fluctuations in noise
levels for the composite image may be attributed to the inconsistent sample illumination
between photoacoustic scans.
Rat PAM I scans were initially intended to be in vivo, with the animal under sedation to
reduce motion artifacts. This was the case for the first 8 scans; however, the animal
unexpectedly expired due to a water leak in the animal jig. The remaining PAM scans
were acquired in-situ. The averaging index of 40 was chosen arbitrarily and was likely
more than necessary to obtain a similar quality image. Feasibility for the system for
animal imaging was shown with photoacoustic visualization of microvessels to depths of
3mm in a Sprague Dawley rat. The PAM I images were verified with a portion of the
rat’s excised skin showing corresponding vessels (Figure 5.22, 5.24).
79
5.6 Summary
A first stage receiver system for photoacoustic microscopy has been built and
characterized, and tested. Following target illumination by an Nd:YAG laser, 48 receive
front end circuits were multiplexed to 16 channels, sent through signal conditioning
stages, and multiplexed further to 4 channels. The PAM I system required 12 laser
firings to complete a single scan. Images were obtained using the full aperture (48
elements), beamforming, beam steering, and CF weighing. Images of phantoms in water
and Intralipid fluid around the 8mm focal depth were produced as well as in situ animal
images 3mm below the skin surface. B mode and C mode type photoacoustic images of
micro-vessels in rats were created using software reconstruction of the signals from all 48
elements. The PAM I system also used a 250MHz and 1.2GHz sampling rate while
reporting a variable 50-73dB measured gain over the bandwidth. The noise floor
measured by minimum detectable signal at max gain on the front end was 316µV. Axial
and lateral resolution was measured at approximately 50µm and 97µm, respectively.
80
Chapter 6: Design of High Frequency Photoacoustic Microscopy
System: Part II
6.1 Purpose
The aim of this project was to create a 16 channel photoacoustic imaging system for a 48
element transducer array. The second version of PAM needed to provide receive signal
processing, as well as a high speed digital data acquisition system, the compliment to the
analog system, capable of transferring and storing all 16 channels of raw data at once.
6.2 Introduction - High Speed Digital Design
At high speeds signal rise times exaggerate the influence of analog effects. It is
imperative that digital design consider the importance of signal integrity issues. The
following sections describe the most common design issues affecting signal integrity in
high speed digital hardware design.
6.2.1 Sampling Frequency and the Knee Frequency
When digitizing signals the sampling frequency must meet the Nyquist criterion, which
states that sampling frequency must be at least twice the highest frequency component of
the signal in order to prevent aliasing. The sampling frequency alone is a characteristic
that defines the system, however in digital design there are other frequencies that need to
be considered. As frequencies move up the spectrum electrical parameters change. The
spectral power density of a digital signal drops steadily at a rate of 20db/decade as
frequency increases until it reaches F
knee
, the knee frequency when the rolloff becomes
81
very steep. The knee frequency is a loose measurement used as a guide to determine if
frequency sensitive events in a digital system are insignificant or critical. Essentially, it
is the frequency above which harmonics present in the pulse edge may be ignored. It is
most useful when considering signal termination due to the effects of frequency and
distance on a trace. The knee frequency can be calculated by
r
knee
T
F
5 . 0
=
(6.1)
where T
r
is the signal rise time (Seams, 2005). The important message here is that the
knee frequency for any digital signal is related to the rise and fall time of its digital edges
and not simply the clock frequency.
6.2.2 Distributed Systems
Multi-channel systems require a good deal of PCB real estate. At higher frequencies,
long board traces become transmission lines. They begin to behave like capacitors,
inductors, and combinations of the two. This is the case of a distributed system.
Alternately, lumped systems are physically small enough so that traces can be ignored
and be considered (as they are schematically) as shorts. All points along the trace from
source to load react together with uniform potential. Naturally, in digital design one
strives for lumped element systems, however the distance required to lay out multiple
channels force the consideration of distributed elements’ effect on the signal. This is
especially the case with long traces associated with shared busses and motherboard
daughter card backplane design. Circuits with an electrical trace length greater than one
82
sixth the length of the signal rising edge are classified as distributed. The rising edge
length is given by
.) / (
) (
in ps Delay
ps Risetime
l
r
=
(6.2)
The propagation delay is fixed by the dielectric constant and the construction of the layer
stack on the PCB, and is generally between (140 - 180)(ps/in), (Johnson, 1993).
Distributed circuits cause the load viewed by the driving source to change. Just as in
transducer theory, the case of impedance mismatch will cause reflections and ringing on
the lines. Ringing of a digital line can be caused from undershoot or overshoot (Figure
6.1). Excessive overshoot of a digital signal will cause false triggers and unreliable
system behavior. This type of error in an ADC sample clock for received echoes would
be catastrophic because it would misreport the signal position by obtaining samples at
inconsistent frequencies.
Figure 6.1 Excessive overshoot of a digital switching signal can cause false triggers
within the logic threshold.
S
T
Overshoot
Logic High
Logic Low
83
Reflection of a signal on a line could bounce in between the source and load a few times
before stabilizing. High speed signals arriving on the output of the driving source can be
corrupted by the reflection noise of the previous signal. This is especially true for long
digital busses in which multiple drivers output on the same signal trace.
6.2.3 Terminations
High speed digital signals can avoid ringing and reflection problems with proper
termination. There are many methods of termination investigated in the literature, the
most common and easily implemented are: series, parallel, and active termination
(Canright, 1991).
In series termination, a resistor is added in series with the driver, and the transmission
line impedance must be matched in such a way that the sum of the driver’s output
impedance and the termination resistor are equal to the transmission line impedance.
This can be difficult for a few reasons. The driver output impedance may vary from one
state to another, and in the case of multiple drivers sharing one long bus, the driving
signal is reduced.
Parallel termination connects a resistor from the signal line to the ground plane at the end
of the line. The impedance of the trace must be calculated or estimated, and this method
introduces high power dissipation (Gopalan, 1996). A constant flow of DC current
through the termination resistor will also result in an extra load on the driver itself. The
84
advantage of this method is that resistors can easily be added post fabrication, although
the power dissipated can make this an impractical solution for many terminations on one
board. Parallel termination can also be achieved through the Chadkeren method, using a
pull up resistor tied to the voltage power plane to help drive the state high. Active
termination can be achieved with a Schottky diode scheme. This method completely
eliminates the task of impedance matching and does not consume excess DC power.
Schottky termination clamps the overshoot and undershoot by no more than the diode’s
forward voltage drop (Figure 6.2). Schottky termination for long signals which leave the
motherboard is the method used in this research.
Figure 6.2 Active line termination through the Schottky diode scheme. The devices
require ~0.6V forward voltage drop before clamping undershoot and overshoot.
6.2.4 Clock Distribution
In a synchronous digital system, the clock signal is used to define a time reference for the
movement of data within the system. The clock distribution network is critical to the
system. It distributes the clock signal from a common point to all the integrated circuits
that need it. It is typically the axis around which other aspects of design revolve. The
clock signal tends to be the most heavily loaded, required to travel the furthest distances,
85
and operate at the highest speeds of the system. Without a proper distribution method,
clock signals can suffer from skew and jitter. Several methods have been researched
including automated synthesis of clock distribution networks and process insensitive
design of clock distribution networks (Efstathiou, 2006)(Friedman, 2001).
Jitter is a temporal variation in clock arrival times, as in the case when two successive
clock edges are seen by the same device in a shorter period than the full clock cycle. It is
caused by statistical variations of the input reference clock. The source of jitter is usually
the oscillator and can be an intrinsic property of the component itself. Power supply
noise is a main source of jitter outside the jitter intrinsic to the part. Clock skew is a
spatial variation in clock signal arrival time between two sequentially adjacent registers.
Different signal paths can have different delays for a variety of reasons; the most
apparent is the differences in line lengths from the clock source to the different clock
registers (Wann, 1983). A solution is to implement a symmetric tree structure, such as
the H-tree (Figure 6.3), where the effective source to register trace lengths are equidistant
(Johnson, 1993). With this scheme, it is important that the signal be driven with enough
current for distribution.
86
Figure 6.3 H-Tree design with parallel termination for high speed clock distribution. The
main trunk is the master clock line.
6.2.5 Technology
The different types of logic devices are classified in "families", the only two applied in
this research are TTL (Transistor-Transistor Logic) and CMOS (Complimentary Metal
Oxide Silicon). TTL chips require a fairly narrow range of supply voltage, 5V+/- 0.5V.
For TTL circuits, logic 1 is usually defined at or above a signal of 2.4V, while logic 0 is
at or below 0.5V above ground. Any signal between those values is undefined. In
contrast to TTL logic, CMOS uses almost no power in the static state and is more
forgiving on the input threshold for a logic level of 1. Since they require less power, they
are often available as a 3.3V device. Interfacing the two technologies causes contentions
in the definitions of logic levels. However, some of the CMOS chips used in the PAM
system are fabricated to accept TTL levels, since logic families are mixed.
87
6.3 Materials and Methods
6.3.1 System Design
A photoacoustic scan initiates when a TTL signal from the motherboard triggers the laser
to irradiate the sample with a pulse. The laser for this system is an Innosolab Edgewave
laser (INNOSLAB Edgewave, Würselen, Germany). The Edgewave is an Nd: YLF laser,
a derivative of a YAG crystal, pumping a continuous wave at 35W. The wavelength of
the light at this stage is 524nm. The electro-optic Q-switch is an externally triggered
attenuator which modulates the Q factor of the optical resonator cavity, and accumulates
the light to send out a burst. This pumping laser and Q-switch produces pulses at 14mJ
when the switch is deactivated. The Q-switch for the Nd:YLF laser is faster than the
Nd:YAG and only requires a 300ns delay between trigger and pulse firing. The light is
then tuned using a Sirah ruby dye laser (Cobra, Sirah Laser, Germany). After the dye
laser, a 2-3mJ per pulse energy is available. The light pulses are coupled into a 600µm
optical fiber, using a beam shaper and a microscope objective to focus the light into the
fiber core. The beam shaper was necessary to convert the elliptically-shaped beam to a
circular shape before fiber coupling. After the fiber, the delivered energy is 0.8mJ per
pulse. The final output from the optical fiber is a 6.5ns pulse at 598nm. The beam
profile after fiber coupling is cylindrical. The energy delivered varied for different
photoacoustic scans, but always maintained a level below the ANSI limit of 20mJ/cm
2
.
Positioned obliquely to the array by a translation stage, or fixed to the array, illumination
is directed towards the target. Unlike the laser used in PAM I, the Nd:YLF laser does not
88
require a 10Hz trigger. Instead, the Nd:YLF can be triggered with any pulse repetition
frequency up to 1KHz.
The photoacoustic waves are detected and processed through the receiver in the same
manner as in Chapter 5. Instead of further multiplexing the sixteen channels to four, all
16 channels are digitized at a100MHz sampling rate, with 16 separate analog to digital
converters (ADC). They are then stored in temporary memory located on each channel
board. The data is transferred to the PC from each memory block one at a time. The
digital system is a backplane design with a motherboard – channel board scheme in
which the channel boards plug into the motherboard. This creates a long data bus where
channel cards plug into and load the bus. Each channel board contains two SMA inputs
to connect with the analog board, two ADCs, a temporary memory chip, and one channel
board enable circuit. The motherboard contains the clock distribution network, driving
buffers, and the data buses. The transfer to the computer is achieved through the PCI bus
using a digital data acquisition card (NI-6534, National Instruments, Houston, TX). The
entire PAM system is managed through a Labview G programmed interface. Figure 6.4
shows the system architecture for PAMII.
89
Figure 6.4 PAM II overall system structure and flow for noninvasive high frequency
array-based imaging.
Pre-Amp M
U
X
3:1
Dual Variable Gain
Amplifier
BP Filter
Fixed Amplifier
1. Front End 2. Signal Processing
1-48
1-16
1
2
:
1
2
Signal Isolation
3
4
Pre-Amp
Pre-Amp
M
U
X
3:1
Pre-Amp
Pre-Amp
Pre-Amp
Gain
Control
3. Digital/Control
A/D Converter
Memory
FIFO
Channel Board
A/D Converter
Transducer Array
1-16
8
8
1-16
1-16
PC
DAQ
Control &
Display
90
6.3.2 Digital Implementation in Hardware
The following sections describe the process of implementation of the digital system into
hardware, and present considerations and design parameters for PCB fabrication and
preservation of accurate system timing. Although most components were hand soldered,
fine pitch components were placed using a stencil. Solder paste was spread over a
custom made stencil and oven reflow was executed with a simple toaster oven.
Schematics and board layout for the digital portion of the PAM system can be found in
Appendix G.
6.3.2.1 Sampling, Timing, & Control
In the digital PAM system, the photoacoustic signals are sampled at a rate of 100MHz.
This is the Nyquist frequency appropriate for the upper limit of bandwidth of the
transducer, around 50MHz. Despite the transducer array’s proximate resonance at
30MHz, the variations in target size produce a wide range of frequencies for
photoacoustic echoes. Photoacoustic echoes are expected beyond 30MHz, dictating a
sampling frequency of at least 100MHz. The clock signal is generated by a crystal
oscillator (JITO-2, Fox Electronics, Ft. Myers, FL) and distributed by a low skew, high
fanout buffer/clock driver. The driver provides a 1:10 ratio of non-inverting synchronous
outputs (IDT74FCT3807, IDT, San Jose, CA), and can operate up to 133MHz. To
prevent ringing and overshoot errors, both series and parallel termination schemes were
used for 100MHz clock lines. This offers the most flexibility for post fabrication
adjustments because pads can be left open or populated with zero ohm resistors if unused.
91
There are eighteen, 8-bit ADCs (AD9054, Norwood, MA) constantly enabled and
operating at 100MHz on TTL logic. Sixteen ADCs are devoted to the channel data while
the remaining two are available for measuring the delivered laser energy with a diode.
The acquisition window is set by the capture into storage first in first out (FIFO) memory
blocks (SN74V225, Texas Instruments, Dallas, TX). These devices are 3.3V, CMOS
technology chips with TTL compatibility. This means they have 5V tolerant input pins
which allow them to interface with the ADCs. The FIFOs are 1Kx18 bit arrays enabled
by 4 main control signals, and can run asynchronous or synchronously. FIFOs were
selected to accommodate the maximum perceivable imaging depth. The image depth can
be calculated by multiplying the acoustic velocity by the time it takes to fill the FIFO at
100MHz.
mm bits ns IDepth
s
m
1 . 15 ) ( 1480 ) 1024 10 ( = ⋅ =
This estimate is of course limited both by the transducer’s performance in the far field
and mostly, by high optical scattering.
Since the memory has a wide data bus, every two 8-bit ADC’s, share one FIFO, each
contributing half of the bus width. The write operation (placing data into memory) is
trigger synchronized with the main system clock, and allows data to be clocked in at the
same rate as the conversion (100MHz). The read operation (pulling data out of memory)
is trigger synchronized with the main system clock and is read out at 12.5MHz. The
timing diagram for this operation is shown in Figure 6.5.
92
Figure 6.5 Timing diagram of master clock, transfer clock and control operations for data
transfer.
The pulled from the FIFO is read out slower than it is written in because it is limited by
the transfer rate of the data acquisition card to the PC. For deterministic timing
applications, it is important to that multiple clocks be synchronized with each other.
There is no way to predict the timing relation between separately generated clocks.
When considering the PAM digital system, there are three groups of 16 channel transfers
that make up one image. If the clocks are not synchronized and running independently,
there will be an irregular temporal offset in rising edges of the write and read clocks
between the first and second group of sixteen. This would cause a misrepresentation of
the actual position of photoacoustic data. Furthermore, this could cause the image to
appear discontinuous between sixteen element groups, and would corrupt averaging of
data points. The clock reading the data must be synchronized with the 100MHz master
11µs
IK
FIFO
Fill
Data
Transfer
10ns
Trigger/Aq
Delay >= 80ns
Step Size = 80ns
100MHz
12.5MHz
Laser
Trig
FIFO
Write
FIFO
Full
Division Latency
93
clock, keeping in mind that it is not absolute alignment of both the write and read clock
we want to preserve; it is their relationship to each other that must remain consistent. In
order to accomplish this synchronicity, the 100MHz clock is divided down in steps of 2
n
by a frequency counter (TC74VHC4040, Toshiba, Irvine, CA) to produce 12.5MHz.
This frequency, although delayed, preserves the latency between itself and the master
clock.
A 16-bit wide bus is shared by all memory blocks for all channels. Signals on the data
bus are sent one FIFO at a time. Even though all devices share control buses and data
buses, channel boards are selected to interface with the bus individually with a line driver
(SN74HCT244, Texas Instruments, Dallas, TX). The line driver buffer has inputs tied to
the control buses and a chip enable, used as a channel select. The read and write control
signals are essentially not seen by a memory device unless the line driver is enabled.
6.3.2.2 Interconnect
The channel boards plug into the motherboard with a right angle high speed connector
(QSS-050-01-F-D-A-GP, Samtec, New Albany, IN). Almost half of the signal pins are
dedicated to delivering both the 3.3V and 5V power needs. Although there is an integral
ground plane along the center of the connector, the 0.635mm pitch allows higher
frequency 5V voltage signals like the clock, to impose electrical crosstalk on adjacent
pins. Ground pins are placed on alongside the clock signal to minimize its effect on
neighboring pins. The motherboard connects to the NI-6534 card through a shielded 68
94
pin cable. A 12.5MHz external clock is provided from the motherboard to the NI-6534
card to ensure synchronization between the two.
6.3.2.3 Data Bus
Nine devices on the same data buses attempting to communicate over long distances can
cause some of the errors discussed in section 6.2. To determine if line termination is
needed, rise time parameters were calculated. The knee frequency related to the rise time
for the FIFO is considered instead of the 12.5MHz data rate of the transfer. According to
equation 6.1, the knee frequency is 166.7MHz. This F
knee
does not present an imposing
demand for termination. However, when evaluating equation 6.2,
in
in ps
ps l
78 . 2
6 ) / ( 180
) ( 3000
6
=
⋅
=
(6.3)
it is apparent that channel boards placed any distance greater than 2.78in constitute a
distributed system and call for line termination. Parallel termination was attempted but
resulted in a large power draw which loaded the 3.3V power plane, and eventually caused
other CMOS devices operating on that plane to fail. The high density and fine pitch of
the board to board Samtec connectors do not provide enough space for active termination
of every data line. Only non-local control signals traveling off the motherboard are
actively terminated using the Schottky diode method.
95
6.3.3 Software Development
The following section describes addition software that was developed to accommodate
the 16 channel digital system and fully orchestrate the scanning process. To create 3D
images, an additional algorithm was created to compile and render 2D B-scan images.
6.3.3.1 System Control
The system control, acquisition, and display were programmed using Labview 7.1
(National Instruments, Houston, TX). A program flow chart shows the PAM sequential
operations (Figure 6.6). Control loops and handshaking schemes are provided for each
sub-acquisition, and then looped three times to complete a photoacoustic scan. The front
panel (Appendix E) allows the user to specify the number of continuous acquisitions for
averaging, and the delay in microseconds between the laser trigger and the memory
acquisition. The user can also specify where to store the data and in what form. The
individual channel data is accessed in tabs, which display graphs of the photoacoustic
echoes from all 48 elements. The laser energy data is displayed on a graph and the final
tab is a pre-beamformed scan of the data from all elements. The G language back panel
(Appendix E) consists of arrays which build the timing waveform for all 9 channel
boards.
Each acquisition initiates a new task which outputs the control waveform, then initiates
an acquisition for a fixed number of samples. When the buffer is full, the data is stored in
an array, the task is cleared, and the process loops for all channel boards. In other words,
96
Labview closes and initiates a new task or program loop for every channel board
acquisition, repeating 9 times. The entire write and acquire process is repeated three
times for each sixteen element group, triggering the laser roughly at a 2-3Hz PRF. The
large data array is indexed and points are averaged along acquisitions before being stored
by element number as either binary or text files.
97
Figure 6.6 PAM digital system software flow chart.
For
i=1:3
For
i=1:NO
A
N
Build
Control
Waveform
Start
Start
Aq?
Number
of
Averages
DIO Config
Timing
Config
Y
N
DIO Write
Laser
Trig
Aq
DIO Clear
For
i=1:9
DIO Config
DIO Config
Timing
Config
DIO Write
Timing
Config
FIFO
OE
Trigger
Config
Read
Trig?
DIO Clear
1K
FIFO
Data
Error
DIO Clear
DIO Read
Laser-
Aq
Y
Index Array
by
Group/Chann
Save
Data
?
Y Number of
Data
Sets to
Skip
Average
Channel
Data SubVI
Text
File
?
Write Binary
Sub VI
Write Text
Sub VI
Y
N
End
N
Display Chnl
Data
Y
Image
Display
98
6.3.3.2 Image Reconstruction in 3D
A Matlab algorithm was developed to create 3D photoacoustic images. 3Dvis.m
(Appendix G) calls the 2D reconstruction algorithm ArraySectorScan.m (Appendix A).
To form a 3D image a given number of B-scans are loaded into a 3D array (Lateral x
Depth x No. of Elevation Scans). Once a 3D matrix or mesh grid is formed, the data is
plotted to vertices on the mesh by interpolation. The data points are then patched and
surfaced rendered forming 3D structures.
6.4 Results
A complimentary 16 channel digital system was designed, fabricated and assembled to
interface with the PAM analog receive electronics. Using the same experimental setup
explained in section 5.3.3, the digital electronics replaced the external multiplexers and
oscilloscope, allowing for an automated 16 channel system. To verify the performance of
the PAM system, measurements of sensitivity, spatial resolution, and focus throughout
the depth of view were taken. The following sections characterize the PAM system and
present photoacoustic images of biological tissue.
6.4.1 Digital System Performance
The digital system viability was tested using a logic analyzer (TLA5202, Tektronix,
Richardson, TX). 30MHz sinusoidal waveforms were fed into the RF cables just before
analog to digital conversion. A screen capture from the logic analyzer validated digital
waveforms from a channel card using the 16 bit data bus (Appendix F). Using the
99
minimum detectable signal test setup from Figure 5.10, the digital board replaced the
oscilloscope block and the photoacoustic data was run through both the analog and digital
system, and then stored to the PC. The minimum detectable signal was ~450µV at 65dB
gain. An image of the completed digital system is shown in Figure 6.7.
a)
b)
Figure 6.7 Photograph of the digital receiver PCB for the PAM II system. a) Each
channel board accommodates two array elements each. b) Digital system including
motherboard and 9 plug in channel boards.
100
6.4.2 Images
The following section presents a progression of images providing the basis of feasibility
and performance for imaging of the full 16 channel PAM system. Photoacoustic images
were formed of phantom targets and in vivo micro-vasculature in a human hand and a
Sprague Dawley rat.
6.4.2.1 Carbon Fiber in Water
To demonstrate the spatial resolution of the 16 channel photoacoustic microscopy system,
image data was collected for a 6µm carbon fiber in water, at the transducer focus of 8mm.
The photoacoustic RF echo data from a single representative channel (element #20) was
plotted in the time domain (Figure 6.8). Envelope detection of the raw data without
applying beamforming delays forms a grayscale matrix of each element response in time
(Figure 6.9). The gain applied system gain was roughly 66dB.
Images were created with the same reconstruction algorithm, using beam focusing, beam
steering, and coherence factor weighting (Appendix A). The carbon fiber data was used
to demonstrate -6dB spatial resolutions in the axial and lateral dimensions. Lateral
spatial resolution was measured by projecting image data onto the X-axis and creating the
point spread function (Figure 6.10). Clutter levels were between -50dB and -60dB.
Resolutions measurements were preserved in PAM II. With the target in the 8mm
transducer focus, spatial resolution of the system was approximately 46µm and 102µm
respectively (Bitton, 2006).
101
5.5 5.6 5.7 5.8 5.9 6 6.1 6.2 6.3 6.4
x 10
-6
-0.78
-0.39
0
0.39
0.78
Time(s)
Voltage (V)
Figure 6.8 PAM II raw RF data from element #20 using a 6µm carbon fiber target.
Figure 6.9 PAM II envelope detected raw RF data from all active channels form a single
matrix. Open elements 1, 5 and element 34 have no data.
102
Figure 6.10 Point spread function of PAM II 6µm carbon fiber showing a -6dB lateral
spatial resolution of ~103µm, -(50-60)dB clutter level.
103
Figure 6.11 PAM II 3D mesh representation of 6µm carbon fiber in water.
Figure 6.12 PAM II B mode image of a 6µm carbon fiber in water.
104
6.4.2.2 Carbon Fiber and Hair in Intralipid
To asses viability for biological tissue imaging, the PAM II system imaged two phantoms
in tissue mimicking solution. Samples in Itralipid had a 1.15µs delay between triggering
and echo arrival and used 39A of pumping laser current. For the carbon 6µm fiber, 0.5%
Intralipid solution was used in place of a water bath. The image is displayed with -35dB
dynamic range (Figure 6.12). The 80µm dark human hair was imaged in 1% Intralipid
solution, which introduces more noise since it is more optically scattering. The hair
phantom could still be identified within the medium (Figure 6.13). Both the carbon fiber
and hair photoacoustic B-mode images were generated applying 58dB gain with no
averaging.
Figure 6.13 PAM II B-mode image of 6µm carbon fiber in 0.5% Intralipid solution.
105
Figure 6.14 PAM II B-mode image of 80µm human hair in 1% Intralipid solution.
6.4.2.3 Carbon Fiber Matrix in Water
The carbon fiber matrix was made up of 5 rows x 9 columns. The rows were divided by
a 1mm separation in depth and the columns were spaced by lateral separations of
0.04inch (1.1016mm). This matrix image employed an alternate method of illumination
from that of PAM I. In this data set, the optical fiber was fixed relative to a single carbon
fiber, while the transducer position was moved. This ensures a more uniform
illumination of the target regardless of the matrix position. Each fiber image was
constructed separately and then added together to create the composite matrix. The
targets are well focused throughout the region of interest and beyond the focus of the
transducer (Figure 6.13) (Bitton, 2006).
106
Figure 6.15 PAM II dynamic focusing: Composite image of a 5x9, 6µm carbon fiber
matrix in water.
6.4.2.4 In Vivo Images
In vivo B-scan photoacoustic images were obtained from the lower portion of a human
hand. The laser fluence was 7mJ/cm
2
using a wavelength of 568nm and a pumping
current of 36A. The delay from laser triggering to light delivery was 1.6µs. The variable
gain was set to deliver 62dB total gain to the system. The subject’s hand, while still
attached to the body, was submerged in the water tank with the palm facing up, towards
the direction of the transducer and optical fiber. The hand vessel images were averaged
over two scans. A number of bright signals from micro-vessels less than 100µm in
diameter can be seen in the center of the image (Figure 6.14) (Bitton, 2007).
This progression led to 3D photoacoustic rat micro-vessel images in vivo. A Sprague
Dawley rat was prepared by depilating a section on the back to reduce excessive signal
107
loss from the fur and then fixing the animal position below the water tank. Spaced at
0.005 inch (0.127mm) intervals in the elevation direction (perpendicular to the B-scan
plane), 100 B-scans were acquired along the image plane using the 3 axis translation
stage for the transducer array. With 10mJ/cm
2
incident fluence at 598nm and an
averaging index of 16, subcutaneous vessels were imaged at depths of 3 mm below the
skin’s surface (in the 8mm focal zone) in two Sprague Dawley rats (Figure 8). To obtain
the most image data, rat micro-vessel images were reconstructed without CF weighting.
Because of the dense vasculature within the tissue, the images shown are reconstructed
from a truncated portion of the data in the depth direction, in order visualize the vessel
structures at a given range. Micro-vessels of varying diameters and vessel bifurcation
can be identified at varying depths within the truncated range. These vessels were not
visible from the skin’s surface. The animals were then sacrificed and portions of skin
excised to verify vessel structures (Bitton, 2007).
Figure 6.16 PAM II B-mode image of a cross section of blood vessels in the lower
portion of a human hand in vivo.
108
Figure 6.17. 3D PAM images of micro-vessels below the surface of the skin in two
Sprague Dawley rats graphed on axis X, Y, Z, representing lateral, scan direction, and
depth dimensions, respectively. a) Rat1 PAM image showing mircro-vessels with
corresponding photo of excised skin b) alternate viewing angle for Rat1 image, and c)
Rat2 PAM image with corresponding photo of excised skin. Markers r, p, and q denote
micro-vessels within the tissue (Bitton, 2007).
q
p
r
p
b)
1mm
r
a)
q
109
Figure 6.17 Continued
6.5 Discussion
6.5.1 System Observations
This version of the high frequency photoacoustic microscopy system uses a different
laser, wavelength of light, and a different crystal to deliver the energy. Compared with
the PAM I system, PAM II has a slightly lower sensitivity, with a difference in minimum
detectable signal of approximately +200µV. However, more efficient illumination
techniques provide PAM II with a better dynamic range. For matrix and in vivo images
the optical fiber was fixed to the transducer housing at a slight oblique angle. We found
this setup to cause less interference of the laser energy with the transducer array. Images
c)
1mm
110
of phantoms preserved the spatial resolution measurements of the previous system and
furthermore, dropped the clutter level below -50dB. Some vessels in the 3D rat images
are disconnected by a missing slice, which may have been caused by missing data sets,
motion artifacts, or could possibly be improved by decreasing the slice intervals at which
the scans are taken.
The speed of this system was limited primarily by the operation time of Labview in the
windows platform, and in a secondary fashion by the need to use the NI-6534 card for
both outgoing control signals and incoming data acquisition. A delay between these two
functions was due to the need to re-initialize each task phase of Labview. Although the
data acquisition from the analog receiver is deterministic and hardware based, the restart
of the entire process for the next set of 16 channels it is not immediately re-triggerable,
and must be reset via software by an internal Labview process. This poses a notable cost
to computational time and total transfer time. The pulse repetition frequency, is also
controlled by the Labview program, where timing may be within the ms error margin.
Although the diode-pumped Nd:YLF laser permits triggering at arbitrary rates unlike
flashlamp-pumped lasers, a faster deterministic PRF could provide more uniform energy
deposition between laser pulses.
Beyond the inherent noise level of the system, during the reconstruction static
thresholding techniques were used to form the boundaries of the vessel structures as 3D
surfaces. A more robust approach might include a dynamic thresholding technique which
111
could optimize the surface boundary threshold for areas outside of the focal zone of the
transducer.
6.5.2 Noise
There are two main sources of electronic noise in the PAM system, digital noise
introduced on the analog ground plane, and quantization bit errors on the data bus.
Digital switching noise that is introduced to the analog system through SMA connectors
joining both ground planes, could easily degrade the performance of the analog receiver
front end. Although the timing network for the system was successful in providing
synchronicity between control, triggering, and clocking signals, the inherent performance
of the ADC device causes it to occasionally, and randomly miss data samples. In
addition, since the board to board connectors were densely spaced, long data bus lines
could not be terminated. Ringing and overshoot may have caused some of the digital
switching error observed on some of the least significant bits. Consequently, the
effective number of bits (ENOB) can be reduced at the cost of dynamic range. However,
switching noise was observed on channel cards unconnected to the data bus with clean
signals provided by a function generator. Insinuating the ADC performance itself may
also be a contributor. The stencil fabrication process used for the channel boards and fine
pitch components often caused multiple shorts on reflow. As a result, some channel
boards were baked 3 or more times, a process that causes dielectric changes of the
board’s internal layers due to heat. This may explain some performance variation
between channel boards (Appendix F). Despite several switching errors, overall, the
112
digital system performed well compared with the commercial oscilloscope sampling
fewer channels at over twice the rate. On the receiver front end, variable gain was held
below the maximum to prevent signal distortion while the digital system faithfully
captured the RF data.
Laser energy fluctuations between 16 channel group acquisitions were observed. This
type of performance could largely be a result of the laser triggering. Using Labview, the
laser pulse repetition frequency is based on the loop acquisition time, which is a non
deterministic software property. This form of non-deterministic timing can affect the
uniformity of laser energy delivered to the sample. An external fully deterministic
control source such as a complex programmable logic device (CPLD), could handle FIFO
read operations, triggering, and synchronization, allowing Labview to handle PC transfer
only.
There is good deal of noise present in the in vivo images. While some of the 3D image
noise may be attributable to hardware, the good dynamic range of the phantom images
reveals that it is clearly not the sole source of image noise. CF weighting was removed
for the final set of images. Removing the weighting factor allowed for more visible data
and facilitated a simpler search for vessel structures. However, in the future, a vessel
filtering algorithm could be applied and coherence factor re-instated. Grating lobes due
to the 2 λ pitch of the transducer array may be rather impactful on the image quality. To
verify this theory, phantom images were reconstructed extending the scan angle from 45°
113
to 90°. The larger viewing angle encompassed the total response of the transducer array
including the affect of grating lobes. The point spread function of the 90° scan angle for
both CF weighted and non weighted data are plotted for comparison (Figure 6.17).
a)
b)
Figure 6.18 PAM II PSF for a 6µm carbon fiber in water with a 90° scan angle a) with
CF weighting b) with out CF weighting.
114
6.6 Summary
A complete array based photoacoustic microscopy system was built capable of acquiring
16 parallel channels of data per single laser irradiation. A high speed digital backplane
motherboard and daughter card configuration was designed and fabricated to compliment
the analog signal processing for the transducer array. The system was characterized, and
then used to create phantom images of 6µm carbon fibers in water. The front end
minimum detectable signal was measured at 500µV. The sensitivity of the system was
validated with image phantoms of a carbon fiber and dark human hair in Intralipid tissue
mimicking solution. The -6dB axial and lateral spatial resolution of the system was
measured as 46µm and 102µm, respectively. The dynamic focusing capability was
demonstrated through a 12.5 mm depth using a composite image of a carbon fiber matrix.
2-D in vivo images were formed of micro-vessel structures in the human hand. 3-D in
vivo images were also formed of micro-vessels below the surface of the skin in two
Sprague Dawley rats.
115
Chapter 7: Future Work:
7.1 Immediate Aims in Real Time Imaging
The aim of ongoing and future work in the short term is to increase the frame rate and
speed of the PAM system, thereby enabling real time high frequency photoacoustic
microscopy.
7.1.1 Purpose
Real time ultrasonic systems using hardware analog and digital beamformers for high
frequency arrays at sixteen channels have been developed at the NIH Transducer
Resource Center (Hu, 2006)(Xu, 2005). These systems are capable of displaying roughly
30 frames per second. They use either classic delay lines with analog beamformation, or
digital delays implemented with field programmable gate arrays (FPGA) for
beamforming. The RF transducer element data is lost as a result of hardware
beamformation and cannot be recovered. Speed and efficacy for these systems has been
shown, nevertheless, a more robust imaging system would include the access to the
individual RF channel data. For the first version of the PAM system, the Nd: YAG laser
had a limited operational frequency of 10Hz, thus making the fast frame rates needed for
real time imaging impossible. The Nd: YLF used in PAM II is a fast triggered laser
capable of accepting variable input pulse repetition rates up to 1KHz. This laser
overcomes the fundamental speed limitation of previous photoacoustic imaging systems.
There are two objectives that must be met in order to create a real time PAM system:
116
1. Increase data transfer time from digital system memory blocks to the PC. 2. Provide
ultra fast processing of data in software for beamforming and display.
7.1.2 Materials and Methods
The following section describes the development of an initial stage real time acquisition
PAM imaging system.
7.1.2.1 System Design
This system architecture was similar to the PAM I design. It utilized the fast Nd: YLF
laser, the receiver board with down multiplexing, and the oscilloscope to capture data
(Zemp, 2007). Using the same laser setup as PAM II, the laser to produced 14mJ pulses
of 6.5ns duration, this time at 1 KHz PRF. Higher PRFs were possible at the cost of
pulse energy. The dye laser produced ~2mJ of pulse energy which was coupled into a
600µm optical fiber. Approximately 0.7mJ of fiber coupled 566nm light was delivered to
the skin surface. In an area of approximately 1.5 by 4mm, the estimated fluence was
12mJ/cm
2
. A fast photodiode was used to record the energy of each laser pulse to
compensate for pulse-to-pulse amplitude fluctuations. The photodiode signal was
recorded on a second oscilloscope communicating with the host oscilloscope, using GPIB
communication protocols. Signals from the array were passed processed through the
analog receiver and a 73dB gain was applied to the system. They were then down
multiplexed from 16 to 4 channels to accommodate the scope input (see section 5.3.1.3).
Instead of a Labview program, the multiplexers were controlled by using 2 asynchronous
117
4-bit counters (CD74HCT163E, Texas Instruments, Dallas, TX) which looped from
binary 0 to 11, sequentially selecting each of the 12 multiplexer states. The control
signals for enable were provided via the parallel port on the PC portion of the
oscilloscope. MATLAB running on the oscilloscope controlled the data acquisition. The
fast-frame mode of the oscilloscope allowed for roughly 200 records of 500 samples each
at a 250MHz sampling rate. This setting for channel memory size allowed for 20 image
frames (with 12 pulses per frame) to be acquired. Each image frame was acquired by
pulsing the laser at 1KHz for 12 pulses, thus a frame could be acquired in 120ms, and the
acquisition frame rate could be as high as 83 frames per second. This high frame rate is
appropriate for mice or rats that have heart rates of 200-400 beats per minute or 4-6 beats
per second. Methods for spatial resolution measurements were identical to those
described in chapters 5 and 6.
7.1.3 Results
Using a 6µm carbon fiber, this system reported a -6dB lateral resolution of just under 100
µm, and axial resolution of 45µm (25µm using the method described in section 5.4.2.1),
identical to the previous designs. Vessels in a Sprague Dawley rat were visualized to
depths of 3mm (Figure 7.1). Images were obtained at 8 frames per second to capture 5
image frames every cardiac cycle. A pulse oximeter (8600V, NONIN, Plymouth, MN)
provided a TTL trigger signal every cardiac cycle. After arming the oscilloscope, a
handheld switch enabled the pulse oximeter to trigger the function generator to fire a
burst sequence for triggering image acquisition. The burst sequence consisted of 20
118
bursts of 12KHz-PRF TTL pulses at 0.125s intervals, leading to 20 frames of 5 images
per cardiac cycle for 4 cardiac cycles.
119
a)
b)
Figure 7.1 Fast acquisition PAM B-scan of subdermal rat micro-vessels in (a) diastole
and (b) systole as correlated using a pulse-oximeter (35dB dynamic range) (Zemp, 2007).
120
7.1.4 Discussion
The system produced a feasibility of fast acquisition in vivo. The averaging index was set
fairly high due to system noise. Again, this may have been caused by the switching
power noise associated with combining the analog receiver ground with that of the PC
parallel port. There was also an average of 11% amplitude fluctuation between recorded
pulses. This could be due to the fluctuations in delivered laser energy, as the laser was
triggered by the pulse oximeter instead of a precise and deterministic digital pulse
generator. Improved system noise or higher frame rate averaging may help decrease this
problem in future studies. This work paves the road for future developments to focus on
development of a higher channel count data parallel data acquisition system, which can
handle streaming of data at a high frame rate.
7.1.5 Fast Acquisition Multi-Channel Parallel Processing
The future work of the PAM system is presented in the following section by exploring
real time 16 channel parallel data acquisition, investigating a new real time approach in
software image reconstruction, and suggestions for further image improvement.
7.1.5.1 Design Solution
The sampling and data transfer speed of the current PAM system is fast enough for real
time acquisition. However, the bottleneck exists in the control timing provided by the
Labview software. This limitation mostly exists between the data transfer rate to the PC,
and in a secondary fashion by the NI-6534 card. The NI-6534 bears a maximum transfer
121
rate of 20Ms/s. For the current system, the card driven at 12.5MHz could still provide
fast enough transfer rates for real time imaging.
The hardware solution to providing a faster frame rate PAM system is to institute
separate sources for control signals and data acquisition. The NI-6534 card can only
execute one operation at a time. The card is capable of either sending control signals out,
or receiving data coming in, but not both simultaneously. The lag time between initiating
ports for outgoing control signals and assigning ports for incoming data signals causes a
large source of speed loss. Since the digital waveform for control is a well defined
sequence, signals from a microprocessor or complex programmable logic device (CPLD)
operating on the master clock or some division thereof, could provide synchronized
signals that select between channel boards and manage data transfer. CPLDs have a large
number of I/O (input output) pins available and could easily accommodate outgoing
control signals and their timing requirements. ISE 8.2i (Xilinx Inc., San Jose, CA) is a
robust software provided for implementation and full post fit
Abstract (if available)
Abstract
Photoacoustic microscopy is an imaging technique which draws from the specific strengths of two imaging modalities by capturing the contrast of optical imaging, while retaining the high resolution of ultrasonic imaging. It provides great promise for studying the structure and dynamics of tissue micro-vasculature in development and pathogenesis. Previous work in photoacoustic imaging has been mostly limited to single element transducers. This thesis presents results of a novel photoacoustic microscopy system using a 30MHz linear array and a custom receive electronics. There are two versions of the system, PAM I and PAM II. Both systems are comprised of three main components, a short pulsed laser, a high frequency transducer, and a custom multi-channel electronics system. The attraction towards high frequency arrays over single element transducers is natural
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Array transducers for high frequency ultrasound imaging
PDF
High-frequency ultrasonic transducers for photoacoustic applications
PDF
High frequency ultrasonic phased array system and its applications
PDF
Development of high-frequency (~100mhz) PZT thick-film ultrasound transducers and arrays
PDF
High-frequency ultrasound array-based imaging system for biomedical applications
PDF
Development of back-end processing system for high frequency ultrasound b-mode imaging
PDF
High-frequency ultrasound imaging system with Doppler features for biomedical applications using 30~35 mHz linear arrays and an analog beamformer
PDF
Development of novel 1-3 piezocomposites for low-crosstalk high frequency ultrasound array transducers
PDF
Intravascular imaging on high-frequency ultrasound combined with optical modalities
PDF
High frequency ultrasonic imaging for the development of breast biopsy needle with miniature ultrasonic transducer array
PDF
Development of front-end circuits for high frequency ultrasound system
PDF
High frequency ultrasound array for ultrasound-guided breast biopsy
PDF
Development of high frequency focused transducers for single beam acoustic tweezers
PDF
Highly integrated 2D ultrasonic arrays and electronics for modular large apertures
PDF
Transducers and signal processing techniques for simultaneous ultrasonic imaging and therapy
PDF
Single-cell analysis with high frequency ultrasound
PDF
Development of high frequency annular array ultrasound transducers
PDF
Multi-modality intravascular imaging by combined use of ultrasonic and opticial techniques
PDF
Novel beamforming with dual-layer array transducers for 3-D ultrasound imaging
PDF
Piezoelectric ultrasonic and acoustic microelectromechanical systems (MEMS) for biomedical, manipulation, and actuation applications
Asset Metadata
Creator
Bitton, Rachel Rinat
(author)
Core Title
A high frequency array- based photoacoustic microscopy imaging system
School
Viterbi School of Engineering
Degree
Doctor of Philosophy
Degree Program
Biomedical Engineering
Publication Date
12/20/2007
Defense Date
08/15/2007
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
high frequency ultrasound,high resolution medical imaging,microscopy,OAI-PMH Harvest,optical imaging,optoacoustic imaging,photoacoustic imaging,photoacoustic receiver,transducer array,vascular imaging
Language
English
Advisor
Shung, K. Kirk (
committee chair
), Cannata, Jonathan Matthew. (
committee member
), Kim, Eun Sok (
committee member
), Meng, Ellis F. (
committee member
), Yen, Jesse T. (
committee member
)
Creator Email
bitton@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-m988
Unique identifier
UC1471300
Identifier
etd-Bitton-20071220 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-595729 (legacy record id),usctheses-m988 (legacy record id)
Legacy Identifier
etd-Bitton-20071220.pdf
Dmrecord
595729
Document Type
Dissertation
Rights
Bitton, Rachel Rinat
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Repository Name
Libraries, University of Southern California
Repository Location
Los Angeles, California
Repository Email
cisadmin@lib.usc.edu
Tags
high frequency ultrasound
high resolution medical imaging
optical imaging
optoacoustic imaging
photoacoustic imaging
photoacoustic receiver
transducer array
vascular imaging