System Challenges in Image Analysis for Radiation Therapy Stephen M. Pizer Kenan Professor Medical Image Display & Analysis Group University.

Slides:



Advertisements
Similar presentations
Surface contour scanning system
Advertisements

Patient information extraction in digitized X-ray imagery Hsien-Huang P. Wu Department of Electrical Engineering, National Yunlin University of Science.
Independent Component Analysis Personal Viewpoint: Directions that maximize independence Motivating Context: Signal Processing “Blind Source Separation”
In the past few years the usage of conformal and IMRT treatments has been increasing rapidly. These treatments employ the use of tighter margins around.
These improvements are in the context of automatic segmentations which are among the best found in the literature, exceeding agreement between experts.
PET/CT Working Group Update Jayashree Kalpathy-Cramer Sandy Napel.
Background:  IMRT has become the choice of treatment for disease sites that require critical structure sparing such as head and neck cancer.  It has.
Effects of Sampling in IMRT Optimization by Ronald L. Rardin Professor of Industrial Engineering Purdue University West Lafayette, Indiana, USA Caesarea.
1 Skeletal Models Correspondence: Improved Interface, Organization, and Speed Clients: Prof. Stephen Pizer Comp.
Multiscale Profile Analysis Joshua Stough MIDAG PI: Chaney, Pizer July, 2003 Joshua Stough MIDAG PI: Chaney, Pizer July, 2003.
Medial Object Shape Representations for Image Analysis & Object Synthesis Stephen M. Pizer Kenan Professor Medical Image Display & Analysis Group.
University of North Carolina Comparison of Human and M-rep Kidneys Segmented from CT Images James Chen, Gregg Tracton, Manjari Rao, Sarang Joshi, Steve.
Medical Image Synthesis via Monte Carlo Simulation James Z. Chen, Stephen M. Pizer, Edward L. Chaney, Sarang Joshi Medical Image Display & Analysis Group,
Session - 25 MULTIDATABASE CASE Electronic Health Matakuliah: M0184 / Pengolahan Data Distribusi Tahun: 2005 Versi:
Clustering on Image Boundary Regions for Deformable Model Segmentation Joshua Stough, Stephen M. Pizer, Edward L. Chaney, Manjari Rao Medical Image Display.
Yujun Guo Kent State University August PRESENTATION A Binarization Approach for CT-MR Registration Using Normalized Mutual Information.
The Uses of Object Shape from Images in Medicine Stephen M. Pizer Kenan Professor Medical Image Display & Analysis Group University of North.
12-Apr CSCE790T Medical Image Processing University of South Carolina Department of Computer Science 3D Active Shape Models Integrating Robust Edge.
Clustering on Local Appearance for Deformable Model Segmentation Joshua V. Stough, Robert E. Broadhurst, Stephen M. Pizer, Edward L. Chaney MIDAG, UNC-Chapel.
Interpolation Snakes Work by Silviu Minut. Ultrasound image has noisy and broken boundaries Left ventricle of dog heart Geodesic contour moves to smoothly.
McDaniels – Feb 29, Outline Patient 6 question Patient 11 ADC results Abstract for AAPM conference.
Caudate Shape Discrimination in Schizophrenia Using Template-free Non-parametric Tests Y. Sampath K. Vetsa 1, Martin Styner 1, Stephen M. Pizer 1, Jeffrey.
Medical Image Synthesis via Monte Carlo Simulation An Application of Statistics in Geometry & Building a Geometric Model with Correspondence.
ANALYSIS OF A LOCALLY VARYING INTENSITY TEMPLATE FOR SEGMENTATION OF KIDNEYS IN CT IMAGES MANJARI I RAO UNIVERSITY OF NORTH CAROLINA AT CHAPEL HILL.
Towards a Multiscale Figural Geometry Stephen Pizer Andrew Thall, Paul Yushkevich Medical Image Display & Analysis Group.
Statistics of Anatomic Geometry Stephen Pizer, Kenan Professor Medical Image Display & Analysis Group University of North Carolina This tutorial.
Method for Determining Apparent Diffusion Coefficient Values for Cerebral Lesions from Diffusion Weighted Magnetic Resonance Imaging Examinations T.H.
Image Guided Surgery in Prostate Brachytherapy Rohit Saboo.
Introduction Background In image-guided interventions, anatomical structures are typically derived from medical images through segmentation. In radiation.
به نام خداوند بخشایندۀ بخشایشگر
Radiotherapy - the art of the invisible Terry Kehoe Consultant Clinical Scientist Head of Oncology Physics Edinburgh Cancer Centre “How to crack a walnut”
MRI Guided Radiation Therapy: Brachytherapy
Parameter selection in prostate IMRT Renzhi Lu, Richard J. Radke 1, Andrew Jackson 2 Rensselaer Polytechnic Institute 1,Memorial Sloan-Kettering Cancer.
Multimodal Interaction Dr. Mike Spann
Analyzing Configurations of Objects in Images via
1 4D: Adaptive Radiotherapy & Tomotherapy Bhudatt Paliwal, PhD Professor Departments of Human Oncology & Medical Physics University of Wisconsin Madison.
Marching Cubes: A High Resolution 3D Surface Construction Algorithm William E. Lorenson Harvey E. Cline General Electric Company Corporate Research and.
AUTOMATIZATION OF COMPUTED TOMOGRAPHY PATHOLOGY DETECTION Semyon Medvedik Elena Kozakevich.
Enhanced Correspondence and Statistics for Structural Shape Analysis: Current Research Martin Styner Department of Computer Science and Psychiatry.
A Challenging Example Male Pelvis –Bladder – Prostate – Rectum.
Introduction Magnetic resonance (MR) imaging is recognised as offering potential benefits in the delineation of target volumes for radiotherapy (RT). For.
Clinical decisions in the optimization process I. Emphasis on tumor control issues Avi Eisbruch University of Michigan.
Object Orie’d Data Analysis, Last Time SiZer Analysis –Zooming version, -- Dependent version –Mass flux data, -- Cell cycle data Image Analysis –1 st Generation.
Common Property of Shape Data Objects: Natural Feature Space is Curved I.e. a Manifold (from Differential Geometry) Shapes As Data Objects.
Visualization of Tumors in 4D Medical CT Datasets Visualization of Tumors in 4D Medical CT Datasets Burak Erem 1, David Kaeli 1, Dana Brooks 1, George.
1 CLUSTER VALIDITY  Clustering tendency Facts  Most clustering algorithms impose a clustering structure to the data set X at hand.  However, X may not.
Somvilai Mayurasakorn, MD. Division of Therapeutic Radiology and Oncology, Faculty of Medicine, Chiang Mai University Somvilai Mayurasakorn, MD. Division.
The PET/CT Working Group: CT Segmentation Challenge Informatics Issues Multi-site algorithm comparison Task: CT-based lung nodule segmentation.
NA-MIC National Alliance for Medical Image Computing NAMIC UNC Site Update Site PI: Martin Styner UNC Site NAMIC folks: C Vachet, G Roger,
Optimization of Volumetric Modulated Arc Therapy (VMAT) Planning Strategy Using Ring-shaped ROI for Localized Prostate cancer Kentaro Ishii, Masako Hosono,
Statistical Shape Analysis of Multi-object Complexes June 2007, CVPR 2007 Funding provided by NIH NIBIB grant P01EB and NIH Conte Center MH
1 UNC, Stat & OR Hailuoto Workshop Object Oriented Data Analysis, I J. S. Marron Dept. of Statistics and Operations Research, University of North Carolina.
Flair development for the MC TPS Wioletta Kozłowska CERN / Medical University of Vienna.
Methods Conclusions References ResultsIntroduction After all tests were performed, the optimal tolerance value was This tolerance value had an overall.
Anatomic Geometry & Deformations and Their Population Statistics (Or Making Big Problems Small) Stephen M. Pizer, Kenan Professor Medical Image.
Evaluation of an Automatic Algorithm Based on Kernel Principal Component Analysis for Segmentation of the Bladder and Prostate in CT Scans Siqi Chen and.
A comparison between soft tissue and bone registration techniques for prostate radiotherapy Richard Small, Paul Bartley, Audrey Ogilvie, Nick West and.
Clustering on Image Boundary Regions for Deformable Model Segmentation Joshua Stough, Stephen M. Pizer, Edward L. Chaney, Manjari Rao, Gregg.
Landmark Based Shapes As Data Objects
Dr. Malhar Patel DNB (Radiation Oncology)
Segmentation of Single-Figure Objects by Deformable M-reps
Dynamic management of segmented structures in 3D Slicer
You Zhang, Jeffrey Meyer, Joubin Nasehi Tehrani, Jing Wang
In Search of the Optimal Set of Indicators when Classifying Histopathological Images Catalin Stoean University of Craiova, Romania
Radiological Sciences Department Ph.D., Paris-Sud 11 University
CNRS applications in medical imaging
Interpolation Snakes Work by Silviu Minut.
Chapter 17 Intensity-Modulated Radiation Therapy
Clinical and radiobiological evaluation of a method for planning target volume generation dependent on organ-at-risk exclusions in magnetic resonance.
Computed Tomography (C.T)
Presentation transcript:

System Challenges in Image Analysis for Radiation Therapy Stephen M. Pizer Kenan Professor Medical Image Display & Analysis Group University of North Carolina Co-authors: Edward L. Chaney, Julian G. Rosenman Credits to many others in UNC MIDAG Stephen M. Pizer Kenan Professor Medical Image Display & Analysis Group University of North Carolina Co-authors: Edward L. Chaney, Julian G. Rosenman Credits to many others in UNC MIDAG

Objective: Segmentation in Radiation Treatment Planning & Delivery ä ä Target problems, in everyday RT ä ä Planning radiotherapy ä ä Segmentation of normal organs to be spared ä ä New patients ä ä Kidney, liver, head and neck ä ä Segmentation of regions implied by segmented organs: lymph levels ä ä Adaptive radiotherapy, incl. IGRT ä ä Segmentation of organs to be spared and target organ ä ä Day to day changes within a patient ä ä Male pelvic organs: bladder, prostate, rectum ä ä Target problems, in everyday RT ä ä Planning radiotherapy ä ä Segmentation of normal organs to be spared ä ä New patients ä ä Kidney, liver, head and neck ä ä Segmentation of regions implied by segmented organs: lymph levels ä ä Adaptive radiotherapy, incl. IGRT ä ä Segmentation of organs to be spared and target organ ä ä Day to day changes within a patient ä ä Male pelvic organs: bladder, prostate, rectum bladder, protsate, rectum

Goal: Segmentation for Radiation Treatment Planning & Delivery ä ä Our objective: A whole new level of segmentation ability ä ä On tough images ä ä As good as humans in most cases ä ä The principle ä ä Make use of probability distribution of geometric variation: p(m) ä ä Make use of probability distribution of geometry-relative intensity patterns p(I | m) ä ä Posterior optimization ä ä arg max m p(m | I) = arg max m [log p(I | m) + log p(m) ] ä ä Our objective: A whole new level of segmentation ability ä ä On tough images ä ä As good as humans in most cases ä ä The principle ä ä Make use of probability distribution of geometric variation: p(m) ä ä Make use of probability distribution of geometry-relative intensity patterns p(I | m) ä ä Posterior optimization ä ä arg max m p(m | I) = arg max m [log p(I | m) + log p(m) ] Bladder CTs Prostate

To achieve segmentation via Obtaining Training Data To achieve segmentation via p(m) and p(I | m): Obtaining Training Data ä ä Fitting m to training binary images ä ä arg min m f(m | binary I) = arg min m [image match penalty + geometric penalty] ä ä Tight fit of geometric model to binary is critical ä ä Extracting object-relative intensity patterns from corresponding CT images ä ä The tight fit of geometric model to binary makes regional intensity patterns more informative ä ä Fitting m to training binary images ä ä arg min m f(m | binary I) = arg min m [image match penalty + geometric penalty] ä ä Tight fit of geometric model to binary is critical ä ä Extracting object-relative intensity patterns from corresponding CT images ä ä The tight fit of geometric model to binary makes regional intensity patterns more informative

Representing Representing m and I | m ä ä Principle: representation should support PCA ä ä Representing object geometry m ä ä M-rep: sheet of medial atoms ä ä Captures local twisting, bending, magnification of interior ä ä Unfamiliar to physicians ä ä Representing image pattern relative to geometry ä ä I | m = = I relative to m = RIQF(interior), RIQF(exterior) ä ä RIQF: regional intensity quantile function ä ä Unfamiliar to physicians ä ä Principle: representation should support PCA ä ä Representing object geometry m ä ä M-rep: sheet of medial atoms ä ä Captures local twisting, bending, magnification of interior ä ä Unfamiliar to physicians ä ä Representing image pattern relative to geometry ä ä I | m = = I relative to m = RIQF(interior), RIQF(exterior) ä ä RIQF: regional intensity quantile function ä ä Unfamiliar to physicians Prostate

Segmentation Program ä ä Initialize pose of mean according to image landmarks ä Conjugate gradient optimization of over coefficients of 9 principal geodesics of ä Conjugate gradient optimization of log p(I | m) + log p(m) over coefficients of 9 principal geodesics of p(m) ä ä Objects are thereby restricted to credible shapes ä ä Initialize pose of mean according to image landmarks ä Conjugate gradient optimization of over coefficients of 9 principal geodesics of ä Conjugate gradient optimization of log p(I | m) + log p(m) over coefficients of 9 principal geodesics of p(m) ä ä Objects are thereby restricted to credible shapes

System Challenges ä ä Challenges of doing the research within a clinical setting ä ä Challenges of getting the research results evaluated in a clinical context ä ä Challenges of clinical adoption of the research results ä ä Challenges of doing the research within a clinical setting ä ä Challenges of getting the research results evaluated in a clinical context ä ä Challenges of clinical adoption of the research results

System Challenges in the Research ä Acquisition of image data of adequate quality ä Meet HIPAA regulations: anonymization ä Homogeneous, high resolution data sets ä Full volume of interest ä As artifact free as possible, or with typical artifacts ä Training cases from target population (with cancer) and from patients with normal anatomy ä Conversion of images and segmentations images and segmentations in RT planning and delivery system into research system ä Acquisition of image data of adequate quality ä Meet HIPAA regulations: anonymization ä Homogeneous, high resolution data sets ä Full volume of interest ä As artifact free as possible, or with typical artifacts ä Training cases from target population (with cancer) and from patients with normal anatomy ä Conversion of images and segmentations images and segmentations in RT planning and delivery system into research system

ä ä High quality manual segmentations in RT planning and delivery system ä ä Consistency across training cases ä ä In adaptive RT: consistency with MD’s planning day ä ä Planning of RT: Include multi-expert variation ä ä Improved manual segmentation tools were developed ä ä High quality manual segmentations in RT planning and delivery system ä ä Consistency across training cases ä ä In adaptive RT: consistency with MD’s planning day ä ä Planning of RT: Include multi-expert variation ä ä Improved manual segmentation tools were developed System Challenges in the Research, cont.

ä ä Need research-tolerant and interested physicians on the team ä ä Need physician input all along, without too heavily disappointing them with early results ä ä New approach was expected to, and did, take a decade to develop ä ä Obtaining a large number of careful, manual segmentations ä ä Need research-tolerant and interested physicians on the team ä ä Need physician input all along, without too heavily disappointing them with early results ä ä New approach was expected to, and did, take a decade to develop ä ä Obtaining a large number of careful, manual segmentations

Evaluation experiments ä ä Retrospective ä ä Our data ä ä Other hospital’s data ä ä Prospective ä ä Within clinical practice, but not interfering with it ä ä “Jeopardy” that they might use the good results clinically during the test ä ä Complete their own segmentations first ä ä Retrospective ä ä Our data ä ä Other hospital’s data ä ä Prospective ä ä Within clinical practice, but not interfering with it ä ä “Jeopardy” that they might use the good results clinically during the test ä ä Complete their own segmentations first

System Challenges in the Evaluation ä ä Access to software that is used clinically ä ä Adding objects with geometry, not just image slice contours (in one orientation), to RT system and its philosophy ä ä Providing segmentations with clinically useful measures of tolerance ä ä Hiding the image analysis details from the clinical user, while allowing access to them by the image analysis researcher ä ä Software’s robustness, reproducibility, user independence, speed (also for clinical use) ä ä Access to software that is used clinically ä ä Adding objects with geometry, not just image slice contours (in one orientation), to RT system and its philosophy ä ä Providing segmentations with clinically useful measures of tolerance ä ä Hiding the image analysis details from the clinical user, while allowing access to them by the image analysis researcher ä ä Software’s robustness, reproducibility, user independence, speed (also for clinical use)

System Challenges in the Evaluation, cont. ä ä Need consensus performance standards ä ä No gold standard with real clinical material ä ä Comparisons of computer vs.human differences against inter-human or intra-human differences ä ä Means of generating synthetic but realistic cases with known truth ä ä Community-wide test case collections ä ä Need consensus performance standards ä ä No gold standard with real clinical material ä ä Comparisons of computer vs.human differences against inter-human or intra-human differences ä ä Means of generating synthetic but realistic cases with known truth ä ä Community-wide test case collections Case index distance Cf. human-to-human prostate agreement: 1.9mm average surface distance

System Challenges of the Clinical Context ä ä Adding indications of non-credibility to segmentations; they may believe too readily ä ä Adding editing capability to computer generated segmentations ä ä Software continuing to change, during clinical tests and clinical use ä ä Adding indications of non-credibility to segmentations; they may believe too readily ä ä Adding editing capability to computer generated segmentations ä ä Software continuing to change, during clinical tests and clinical use

ConclusionConclusion ä ä Research on IGT methods has not only technical and clinical challenges but also significant software system challenges