Direct Cardiac Image Analysis without Segmentation Dr. Shuo Li the - PowerPoint PPT Presentation

Direct Cardiac Image Analysis without Segmentation Dr. Shuo Li the Digital Imaging Group (DIG) of London Lawson Health Research Center the University of Western Ontario Introduction Direct Volume Estimation Direct Diagnosis Conclusion 1/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Outline Introduction 1 Existing Approach Direct Volume Estimation 2 Direct LV Volume Estimation Direct Bi-Ventricle Volume Estimation Direct Four Chambers Volume Estimation Direct Diagnosis 3 Background and Old School Our Direct Method Conclusion 4 Introduction Direct Volume Estimation Direct Diagnosis Conclusion 2/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Section 1 Introduction Introduction Direct Volume Estimation Direct Diagnosis Conclusion 3/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Segementation or Not Segmentation? ”Plays a fundamental role in understanding medical images” ”Too many segmentation methods” ”Is segmentation really necessary?” Introduction Direct Volume Estimation Direct Diagnosis Conclusion 4/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Functionality of Segmentation Volume Estimation Diagnosis Introduction Direct Volume Estimation Direct Diagnosis Conclusion 5/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Existing Approach Subsection 1 Existing Approach Introduction Direct Volume Estimation Direct Diagnosis Conclusion 6/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Existing Approach Old School Example: Cardiac Image Segmentation Current Status and Challenges Introduction Direct Volume Estimation Direct Diagnosis Conclusion 7/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Existing Approach Model the Challenge Learn the similarity between the regions then recover it 1. Ismail Ben Ayed, Yingli Lu, Shuo Li, and Ian Ross, Left Ventricle Tracking Using Overlap Priors . MICCAI08 2. Ismail Ben Ayed, Shuo Li, and Ian Ross, Embedding overlap priors in varia- tional left ventricle tracking , IEEE TMI, 2009 3. Ismail Ben Ayed, Kumaradevan Punithakumar, Shuo Li, Distribution Matching with the Bhattacharyya Similarity: a Bound Optimization Framework , IEEE TPAMI, 2015 Introduction Direct Volume Estimation Direct Diagnosis Conclusion 8/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Existing Approach Generalization Learn the similarity between the regions then recover it 1. Ismail Ben Ayed, Shuo Li, and Ian Ross, Tracking distributions with an overlap prior , CVPR 2008 2. Ismail Ben Ayed...Shuo Li, A Statistical Overlap Prior..., IJCV 2009 Introduction Direct Volume Estimation Direct Diagnosis Conclusion 9/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Existing Approach Optimization 1. Kumaradevan Punithakumar, Jing Yuan, Ismail Ben Ayed, Shuo Li and Yuri Boykov, A Convex Max-Flow Approach to Distribution Based Figure-Ground Separation , SIAM Journal on Imaging Sciences, 2012 2. Mohammad Saleh Nambakhsh, Jing Yuan, Ismail Ben Ayed, Kumaradevan Punithakumar, Aashish Goela, Ali Islam, Terry Peters, and Shuo Li, A Convex Max-Flow Segmentation of LV using Subject-Specific Distributions on Cardiac MRI , IPMI 2011 3. Ismail Ben Ayed, Kumaradevan Punithakumar, Shuo Li, Ian Ross, Jaron Chong, Graph Cut Kernel Tracking of the Left Ventricle . MICCAI, 2008 4. Ismail Ben Ayed, Hua-mei Chen, Kumaradevan Punithakumar, Ian Ross, and Shuo Li, Max-flow segmentation of the left ventricle by recovering subject-specific distributions via a bound of the Bhattacharyya measure , MedIA 2011 Introduction Direct Volume Estimation Direct Diagnosis Conclusion 10/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Section 2 Direct Volume Estimation Introduction Direct Volume Estimation Direct Diagnosis Conclusion 11/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct Volume Estimation without Segmentation Direct LV Volume Estimation Direct LV and RV Volume Estimation Direct Four Chambers Volume Estimation Introduction Direct Volume Estimation Direct Diagnosis Conclusion 12/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct LV Volume Estimation Subsection 1 Direct LV Volume Estimation Introduction Direct Volume Estimation Direct Diagnosis Conclusion 13/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct LV Volume Estimation Left Ventricle Volume Estimation Mariam Afshin, Ismail Ben Ayed, Ali Islam, Aashish Goela, Ian G. Ross, Terry M. Peters and Shuo Li, Global Assessment of Cardiac Function using Image Statis- tics in MRI , Medical Image Computing and Computer-Assisted Intervention (MIC- CAI), 2012. Introduction Direct Volume Estimation Direct Diagnosis Conclusion 14/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct Bi-Ventricle Volume Estimation Subsection 2 Direct Bi-Ventricle Volume Estimation Introduction Direct Volume Estimation Direct Diagnosis Conclusion 15/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct Bi-Ventricle Volume Estimation Direct Bi-Volume Estimation - Baysian Model 1. Zhijie Wang, Mohamed Ben Salah, Bin Gu, Ali Islam, Aashish Goela, Shuo Li, Direct Estimation of Cardiac Biventricular Volumes With an Adapted Bayesian Formulation , IEEE TBME 2014 2. Zhijie Wang, Mohamed Ben Salah, Ismail Ben Ayed, Ali Islam, Aashish Goela, Shuo Li, Bi-ventricular Volume Estimation for Cardiac Functional Assessment , RSNA 2013 Introduction Direct Volume Estimation Direct Diagnosis Conclusion 16/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct Bi-Ventricle Volume Estimation Direct Bi-Volume Estimation - Deep Learning 1. Xiantong Zhen, Zhijie Wang, Ali Islam, M. Bhaduri, Ian Chan and Shuo Li, Direct Estimation of Cardiac Bi-ventricular Volumes with Regression Forests , MICCAI 2014 2. Xiantong Zhen, Zhijie Wang, Ali Islam, M. Bhaduri, Ian Chan and Shuo Li, A Comparative Study of Methods for Cardiac Ventricular Volume Estimation , RSNA, 2014 3. Xiantong Zhen, Zhijie Wang, Ali Islam, M. Bhaduri, Ian Chan and Shuo Li, Multi-Scale Deep Networks and Regression Forests for Direct Bi-ventricular Volume Estimation , MedIA, 2015 Introduction Direct Volume Estimation Direct Diagnosis Conclusion 17/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct Bi-Ventricle Volume Estimation EF: Automatic vs. Manual Introduction Direct Volume Estimation Direct Diagnosis Conclusion 18/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct Four Chambers Volume Estimation Subsection 3 Direct Four Chambers Volume Estimation Introduction Direct Volume Estimation Direct Diagnosis Conclusion 19/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct Four Chambers Volume Estimation Direct Four Chambers Volume Estimation: CT and MRI Introduction Direct Volume Estimation Direct Diagnosis Conclusion 20/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Direct Four Chambers Volume Estimation Four Chambers: Automatic vs. Manual 1. Xiantong Zhen, Zhijie Wang, Ali Islam, M. Bhaduri, Ian Chan and Shuo Li, Four Chamber Volume Estimation... , MICCAI 2015 2. Xiantong Zhen, ... and Shuo Li, Supervised Descriptor Learning for Multi-Output Regression , CVPR, 2015 Introduction Direct Volume Estimation Direct Diagnosis Conclusion 21/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Section 3 Direct Diagnosis Introduction Direct Volume Estimation Direct Diagnosis Conclusion 22/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

A Comparasion: Old School vs. Direct Analysis Background: Global and Regional Abnormality Old School Our Approach Introduction Direct Volume Estimation Direct Diagnosis Conclusion 23/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Background and Old School Subsection 1 Background and Old School Introduction Direct Volume Estimation Direct Diagnosis Conclusion 24/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

Background and Old School Global Abnormality: Technical challenges Significant overlap between normal and abnormal distributions (a) Typical normal (b) Typical abnormal (c) Borderline normal (d) Borderline abnormal Proposed solution: The Shannon’s Differential Entropy — a global measure Recursive Bayesian filtering — temporal smoothing Introduction Direct Volume Estimation Direct Diagnosis Conclusion 25/36 Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation