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

▶
direct cardiac image analysis without segmentation
SMART_READER_LITE
LIVE PREVIEW

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

Sep 01, 2022 343 likes •714 views

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

slide-1
SLIDE 1

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

1/36

slide-2
SLIDE 2

Outline

1

Introduction Existing Approach

2

Direct Volume Estimation Direct LV Volume Estimation Direct Bi-Ventricle Volume Estimation Direct Four Chambers Volume Estimation

3

Direct Diagnosis Background and Old School Our Direct Method

4

Conclusion

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

2/36

slide-3
SLIDE 3

Section 1 Introduction

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

3/36

slide-4
SLIDE 4

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

4/36

slide-5
SLIDE 5

Functionality of Segmentation

Volume Estimation Diagnosis

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

5/36

slide-6
SLIDE 6

Existing Approach

Subsection 1 Existing Approach

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

6/36

slide-7
SLIDE 7

Existing Approach

Old School Example: Cardiac Image Segmentation

Current Status and Challenges

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

7/36

slide-8
SLIDE 8

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

8/36

slide-9
SLIDE 9

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

9/36

slide-10
SLIDE 10

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

10/36

slide-11
SLIDE 11

Section 2 Direct Volume Estimation

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

11/36

slide-12
SLIDE 12

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

12/36

slide-13
SLIDE 13

Direct LV Volume Estimation

Subsection 1 Direct LV Volume Estimation

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

13/36

slide-14
SLIDE 14

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

14/36

slide-15
SLIDE 15

Direct Bi-Ventricle Volume Estimation

Subsection 2 Direct Bi-Ventricle Volume Estimation

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

15/36

slide-16
SLIDE 16

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

16/36

slide-17
SLIDE 17

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

17/36

slide-18
SLIDE 18

Direct Bi-Ventricle Volume Estimation

EF: Automatic vs. Manual

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

18/36

slide-19
SLIDE 19

Direct Four Chambers Volume Estimation

Subsection 3 Direct Four Chambers Volume Estimation

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

19/36

slide-20
SLIDE 20

Direct Four Chambers Volume Estimation

Direct Four Chambers Volume Estimation: CT and MRI

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

20/36

slide-21
SLIDE 21

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

21/36

slide-22
SLIDE 22

Section 3 Direct Diagnosis

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

22/36

slide-23
SLIDE 23

A Comparasion: Old School vs. Direct Analysis

Background: Global and Regional Abnormality Old School Our Approach

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

23/36

slide-24
SLIDE 24

Background and Old School

Subsection 1 Background and Old School

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

24/36

slide-25
SLIDE 25

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

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

25/36

slide-26
SLIDE 26

Background and Old School

Global Abnormality

Technical challenges

Significant overlap between normal and abnormal distributions

Our solution

The Shannon’s Differential Entropy — global measure Analysis

Receiver Operating Characteristics

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 False positive Sensitivity Mean systolic velocity Mean radial displacement Fisher information Renyi entropy Shannon’s differential entropy

Distribution of Normal and Abnormal

−40 −20 20 40 0.02 0.04 0.06 0.08 0.1 Mean systolic velocity Abnormal Normal −20 −15 −10 −5 0.1 0.2 0.3 0.4 Mean radial displacement Abnormal Normal

B = 0.32 B = 0.53

−500 500 1000 0.005 0.01 0.015 Fisher information Abnormal Normal −1 1 2 0.5 1 1.5 2 2.5 Shannon’s differential entropy

B = 0.59 B = 0.62 Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

26/36

slide-27
SLIDE 27

Background and Old School

Heart Abnormality Detection - Regional Diagnosis

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

27/36

slide-28
SLIDE 28

Background and Old School

Regional Abnormality Analysis

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

28/36

slide-29
SLIDE 29

Background and Old School

Regional Abnormality Analysis

The proposed method

Anatomical cine MRI Nonrigid Image Registration Unscented Kalman Smoother The SDE The Bayesian Classifier 1. Kumaradevan Punithakumar, Ismail Ben Ayed, Ali Islam, Aashish Goela and Shuo Li, Regional Heart Motion Abnormality Detection via Multiview Fusion, MICCAI 2012 2. Kumaradevan Punithakumar, Ismail Ben Ayed, Ali Islam, Aashish Goela, Ian G. Ross, Jaron Chong, Shuo Li, Regional Heart Motion Abnormality Detection: An Information Theoretic Approach, Medical Image Analysis, 2013 Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

29/36

slide-30
SLIDE 30

Background and Old School

Quantitative Analysis

ROC

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Specificity Sensitivity SDE of radial distance SDE of radial velocity SDE of segment arc length SDE of segment area 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

(a) Our method (b) Ground truth Distribution of normal and abnormal myocardial segments

−2.5 −2 −1.5 −1 −2.5 −2 −1.5 −1 −0.5 SDE of segement area SDE of normalized radial distance Normal Abnormal −2.5 −2 −1.5 −1 −2.5 −2 −1.5 −1 −0.5 SDE of segement area SDE of normalized radial distance Normal Abnormal −2.5 −2 −1.5 −1 −2.5 −2 −1.5 −1 −0.5 SDE of segement area SDE of normalized radial distance Normal Abnormal

(a) apical (b) mid-cavity (b) basal Accuracy (%) Sensitivity (%) Specificity (%) Apex 92.5 90.0 93.3 Mid-cavity 93.3 93.1 94.1 Base 87.2 100.0 84.9

The classification accuracy: 90.8% Sensitivity — 94.5% Specificity — 90.0%

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

30/36

slide-31
SLIDE 31

Our Direct Method

Subsection 2 Our Direct Method

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

31/36

slide-32
SLIDE 32

Our Direct Method

Our Direct Method: Simple, Fast and Accurate

1. Mariam Afshin, Ismail Ben Ayed, Kumaradevan Punithakumar, Max W. K. Law, Ali Islam, Aashish Goela, Ian G. Ross, Terry M. Peters and Shuo Li, Myocardial Function via Statistical Features in MR Images, MICCAI 2011 2. Mariam Afshin, Ismail Ben Ayed, Kumaradevan Punithakumar, Max W. K. Law, Ali Islam, Aashish Goela, Terry M. Peters, and Shuo Li, Regional Assessment of Cardiac Left Ventricular Myocardial Function via MRI Statistical Features. IEEE TMI, 2014 Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

32/36

slide-33
SLIDE 33

Section 4 Conclusion

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

33/36

slide-34
SLIDE 34

Summary and Significance

Direct Diagnosis without Segmentation

Beyond Simple and Efficient Focus on Clinical Goal

Till today: Beyond Cardiac; Cross modality Very Customizable

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

34/36

slide-35
SLIDE 35

Summary and Significance

Recognition from RSNA

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

35/36

slide-36
SLIDE 36

Conclusion and Future Work

Introduction Direct Volume Estimation Direct Diagnosis Conclusion

  • Dr. Shuo Li – Direct Cardiac Image Analysis without Segmentation

36/36