Transcript Image Guided Surgery in Robotic Biopsy

Image Guided Surgery in
Prostate Brachytherapy
Rohit Saboo
Prostate Cancer


A growing problem in US and world
over with increasing longevity
Methods of treatment



Surgery
Irradiation
Many problems of existing methods
Where is it?
Brachytherapy procedure


Localized and prolonged dose
Brachytherapy overview
Brachytherapy - old way

Pre-planning CT



Outline prostate
Develop plan for needle movement
Guide needle at time of surgery with help
from Ultra sound images
“Dynamic Brachytherapy of the prostate under active image guidance”, Gang
Cheng et. al, MICCAI 2001
Brachytherapy – old way

Problems





Prostate movement
Prostate size/shape variance
Due to anesthesia effects
Time and hormonal therapy
Drawbacks:


Lots of error due to prostate movement
More than necessary needle insertions
Brachytherapy – new way




General outline of prostate from preplanning CT
Outline prostate in real-time during
surgery
Provide guidance for the needle with
real time prostate outlining.
Track needle errors in real time
Steps in automation



Acquire volumetric ultra-sound images
on the fly
Automatically recognize the
prostate/rectum and other structures
(segmentation)
Analyze dose distribution
Segmentation


The process of outlining the prostate
(or any organ) is called segmentation
Two chief ways to deal with it


Model-based
Image based
Segmentation problem
Ultrasound problems


Noise!
Speed of sound is not uniform

Image distance incorrect in one axis
Approaches to segmentation

Model based

Shape model




Probable shapes
Probable intensity/texture variations
Examples: ASM, AAM, M-reps
Image based



Outline drawn by expert on one image (atlas)
Image intensity/feature based registration
Outline carried over
Feature Model – Ruo et. al




Set of boundary points
i - Sample object
Xi – Tuple representing ith object
Each object has m points on the
boundary
Feature Model
Xi =
(Li, ri1, ri2, … rim )T
Shape variation

Mean shape

Covariance matrix
Feature Model

Eigenvalue decomposition of covariance
matrix



pi Principal components (eigenvectors)
Eigen-values
Sort the eigenvalues, choose the largest
t
Eigenvalues
New plausible models
Optimization - GA
Image match


Fitness function
outi and ini average of
intensities along a
profile (15 pixels long)
Image Match

Simplify and speed it up

vi unit normal
GA parameters
GA parameters





90% crossover rate
1% mutation rate
population size 200
2000 generations
repeated 15 times
Experiment





40 images
training from 27, 3 poor quality
test on remaining 10
Expert segmentation by two different
experts
human-human disagreement vs
automated-human disagreement
Results
Results
Results
Results
Summary




Very good analysis
Point based boundary models are poor
Parameter tuning
No reasoning for fitness function
Methods


Model based methods
Image based methods



Fully automatic, registration
Deformable registration – Wei, 2004
Snakes
Image based
Overview
 MRI/MRSI and US data
 prostate carefully outlined on MRI data
 US image is acquired during operation time
 The two images are brought into alignment
during operation time.

They do for biopsy, but same techniques
work for brachytherapy
Global Alignment




US data is poor
Model is pre-segmented in MRI
Surface to volume alignment methods
are used
Gradient of image is computed in US
and model information is used to
roughly find the correspoding boundary
in US
Global alignment


GA based optimization
fitness function
Registration

Deformation is elastic


Two orthogonal directions
Therefore 2-parameter model for
deformation
Registration



Obtain Curvature images
Obtain mapping g for a few points
Use these points to drive a TPS
deformation.
Validation

Phantom



Gelatin made prostate phantom
15 fiducial markers implanted inside the
prostate
Soft container filled with water placed on
top to simulate pubic bone.
Validation
Results on phantom
Patient study
Patient Study
Summary



Two step registration technique
Phantom Studies
Only tested over one patient
Methods


Model based methods
Image based methods



Fully automatic, registration
Deformable registration – Wei, 2004
Snakes


semi-automatic
In between both
Snakes


Give an initial approximate contour
Two forces act on a snake

Internal force


External force


Based on curvature
Based on image gradients
Let the model evolve using ordinary
force equations till equilibrium
Snake based methods

Approach used by Zhouping Wei, 2005
Snake based methods

1.19 +/- 0.14 mm
on average
Questions?
References





“Dynamic Brachytherapy of the prostate under active image guidance”,
Gang Cheng et. al, MICCAI, 2001
“Automatic Prostate Boundary Recognition in Sonographic Images Using
Feature Model and Genetic Algorithm”, Ruo Yun et. al, Journal of
Ultrasound in Medicine, Vol 19, Issue 11, 2000
“Deformable Registration Between MRI/MRSI and Ultrasound Images for
Targeted Robotic Prostate Biopsy”, Wei Shao et. al, Proceedings of the
2004 IEEE Conference on Cybernetics and Intelligent Systems
“A Discrete Dynamic Contour Model”, Steven Lobregt et. al, IEEE
Transactions on Medical Imaging, vol 14, no 1, March 1995
“Dynamic Intraoperative Prostate Brachytherapy Using 3D TRUS Guidance
with Robot Assistance”, Zhouping Wei et. al, Proceedings of the 2004 IEEE
Engineering in Medicine and Biology, 2005