3D Near-infrared Imaging Based on a

Single-photon Avalanche Diode Sensor

Juan Mata Pavia

Cristiano Niclass

Martin Wolf

Edoardo Charbon

2

Experimental Techniques

Continuous wave (CW)

Time domain (TD)

Phase

I0

I

Frequency domain (FD)

time

I0

I

I0

time

tissueinitial signal

detected signal

time

3

Near Infrared Imaging (NIRI)

Low spatial resolution (~1cm resolution)

3D images require long acquisition times

CW systems can be miniaturized

TD and FD systems are bulky

Low number of sources/detectors

[Wells K et al Proc. SPIE 1997][Muehlemann T et al Opt.

Express 2008]

4

Single-photon avalanche diode (SPAD)

5

New 3D NIRI system:

• High resolution images: <1cm

• Almost real time operation

• Bedside applicable

SPAD imagers offer:

• High resolution images

• Time resolved measurements

Objective

6

SPAD Image Sensor: LASP Chip

128x128

SPAD array

32 TDCs

100ns range

97ps res.

I/O interface

[Niclass et al JSSC 2008]

7

Experimental Setup with Cylindrical Waves I

8

Experimental Setup with Cylindrical Waves II

Telecentric

objective

Collimator

& diffuser

Collimator

& diffuserIntralipid phantom

SPAD sensor &

acquisition electronics

4 cm

9

Phantom Measurements I

10

Phantom Measurements II

11

The Diffusion Equation

•Homogeneous medium

•Fourier in the time domain

•Born approximation

Measurement

Object to be reconstructed

Matrix coefficients

tts

2 ω,rUrO=ω,rUk+

0

2

'rdω,'rU'rOω,'r,rG=ω,rUttts

0

txtxyzsx

2

yzω,rUrOF=ω,ω,rUwk+

0

22

yztyzxyztxyzyztxyzs'rdω,'rUω,'rOω,ω,'r,rG=ω,ω,rU

0

Fourier in the X dimension

X independent

The triple integral is reduced to a double integral

Measurement

Object to be reconstructed

Photon density in the

homogeneous medium

12

Experimental Results: Setup 1

13

Experimental Results: Reconstruction 1

14

Experimental Results: Setup 2

15

Experimental Results: Reconstruction 2

16

Experimental Results: Setup 3

17

Experimental Results: Reconstruction 3

18

Summary & Outlook

Design a NIRI system based on a SPAD image

sensor

Develop the image reconstruction algorithm

Build the system

Performance evaluation

• Design a new SPAD image sensor

• Pre-clinical trials

• SPADs imagers make possible time-resolved high spatial resolution measurements for 3D NIRI

19

Thanks to

Oliver SieberThomas MühlemannMartin BiallasAndreas MetzDamien de CourtenFelix ScholkmannSonja SpichtigIvo TrajkovicDaniel OstojicChristoph KuhnHans-Ulrich BucherDominik MartiMartin Frenz

Questions ?

22

Outline

Introduction

Imaging system

Image reconstruction

Preliminary results

23

Absorption

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

600 700 800 900 1000

Wavelength [nm]

O2H

b,

HH

b [

1/(

mM

*mm

)]

0

0.01

0.02

0.03

0.04

0.05

0.06

H2O

, L

ipid

[1

/mm

]

HHb

O2Hb

Lipid

H2O

24

Experimental Techniques

I0

IContinuous wave (CW)

I0

time

Time domain (TD)

tissueinitial signal

detected signal

time

Phase

I0

I

Frequency domain (FD)

time

25

Scattering

Random walk theory

Photon random walk step in biological tissue ~1mm

Photon

Non-scattering medium Biological tissue

1mm

26

Time Correlated Single Photon Counting (TCSPC)

LaserSubject

Single-photon

detector

Lens

Time-to-digital

converter

Histogram

builder

StartStop

27

Applications

Perfusion state of tissue

Tissue oxygenation monitoring

Internal bleeding

Detection of infarcts

Tumor detection and analysis

Viability of tissue

Tissue function

28

SPAD Image Sensor: LASP Chip

CMOS 0.35µm technology

3.2mm x 3.2mm active area

Fill factor 6%

Microlenses improve the fill factor up to 50%

128x128

SPAD

array

32 TDCS

100ns range

97ps res.

I/O

in

terf

ac

e

[Niclass et al JSSC 2008]

29

SPAD Image Sensor Architecture

[Niclass et al JSSC 2008]

30

Laser: Becker & Hickl BHL-700

780nm wavelength

Repetition rate 80MHz (fixed)

Pulse width ~100ps at 1mW

0.2mW to 10mW adjustable average CW power

300mW typical peak power

[www.becker-hickl.de]

31

Experimental Setup with Cylindrical Waves I

32

Trigger Adaptation Circuit

33

Experimental Setup with Plane Waves

34

Results with Plane Waves

Time gating-window (GW) does not improve the IRF

SPAD’s Impulse Response

Function (IRF) is too slow

[Torricelli et al. 2005]

[Niclass C et al JSSC 2008]

[Schwartz D et al JSSC 2008]

35

Time-to-digital Converter Linearity

0 128 256 384 512 640 768 896 1024-1.5

-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

TDC Differential Non-linearity

No dithering

Dithering

Improved dithering

TDC Code

LS

B

0 128 256 384 512 640 768 896 1024-3.0

-2.0

-1.0

0.0

1.0

2.0

TDC Integral Non-linearity

No dithering

Dithering

Improved dithering

TDC Code

LS

B

36

Detector: Photon Detection Probability (PDP)

PDP increases with the excess bias voltage, but so does the DCR.

The shallow p+ / n-well junction explains why the SPAD is more effective at blue/UV wavelengths than at red/IR.

[Niclass C et al JSSC 2008]

37

Dark Count Rate (DCR)

Pulses generated in the SPAD in the absence of light.

Causes:

• Thermal generation of carriers

• Electron-hole generation due to tunneling effects.

[Niclass C et al JSSC 2008]

[Niclass C et al JSSC 2008]

38

BHP-700 time response

39

The Diffusion Equation

•Homogeneous medium

•Fourier in the time domain

Using the first Born approximation

Measurement

Object to be reconstructed

Matrix coefficients

40

The Diffusion Equation for Plane Waves

Fourier in XY dimensions

Homogeneous field is XY independent

Infinite medium

41

The Diffusion Equation for Cylindrical Waves

Fourier in the X dimension

Homogeneous field is X independent

Infinite medium

42

Reconstruction Algorithm

Semi-infinite medium: the method of images can be applied

Multiple sources: superposition principle

TD measurements: high number of equations for only one acquisition

43

The Inverse Problem: Regularization

The problem is ill-posed by its nature.

Regularization defines restrictions in the solution’s complexity: smoothness, norm …

Tikhonov regularization: equivalent to minimize:

Linear operator

Problem's solution (object

to be reconstructed)Matrix to

be inverted

Regularization parameter

Measurement

Sub-space preconditioned LSQR (SP-LSQR):

• Iterative method

• Tikhonov regularization

• Predefined sub-space of possible solutions

44

Image Reconstruction Testbench

45

Simulation Results: ART

46

Simulation Results: LSQR

47

Simulation Results: SP-LSQR Norm

48

Simulation Results with SP-LSQR

49

FPGA Architecture for fast data acquisition

50

High definition imaging with NIR

CW systems with CCD cameras.

Long acquisition times: Full scan of the object.

Low depth resolution.

Only applied to small spaces. E.g. small animal imaging.

[A. Martin, Mol. Img. (2008) ]

[A. Martin, Mol. Img. (2009) ]

51

Future Modifications of the SPAD Image Sensor

Solve the Linearity problem in the TDC

Reduce the total time range of the TDC

Increase the number of pixels

Modify the TDC/pixel clustering to reduce the number of acquisitions per frame

52

Outlook

Design a NIRI system based on a SPAD image sensor

Develop an image reconstruction algorithm based on the system

Build a system to perform measurements on phantoms

Evaluate the performance of the new setup

Design a customized SPAD image sensor

Pre-clinical trials

53

Conclusions

SPADs enable the acquisition of time-resolved measurements with high spatial resolution for NIRI

They make possible the development of more efficient algorithms:

• Higher resolution images

• Less computation power