Imperial College London
Alternate

DrNavtejChahal

Faculty of Medicine, National Heart & Lung Institute

Honorary Clinical Senior Lecturer
 
 
 
//

Contact

 

navtej.chahal07

 
 
//

Location

 

Royal BromptonRoyal Brompton Campus

//

Summary

 

Publications

Citation

BibTex format

@article{Li:2018:10.1109/TMI.2017.2747081,
author = {Li, Y and Ho, CP and Toulemonde, M and Chahal, N and Senior, R and Tang, MX},
doi = {10.1109/TMI.2017.2747081},
journal = {IEEE Transactions on Medical Imaging},
pages = {1081--1091},
title = {Fully automatic myocardial segmentation of contrast echocardiography sequence using random forests guided by shape model},
url = {http://dx.doi.org/10.1109/TMI.2017.2747081},
volume = {37},
year = {2018}
}
Download

RIS format (EndNote, RefMan)

TY  - JOUR
AB  - Myocardial contrast echocardiography (MCE) is animaging technique that assesses left ventricle function and myocardialperfusion for the detection of coronary artery diseases.Automatic MCE perfusion quantification is challenging and requiresaccurate segmentation of the myocardium from noisy andtime-varying images. Random forests (RF) have been successfullyapplied to many medical image segmentation tasks. However, thepixel-wise RF classifier ignores contextual relationships betweenlabel outputs of individual pixels. RF which only utilizes localappearance features is also susceptible to data suffering fromlarge intensity variations. In this paper, we demonstrate howto overcome the above limitations of classic RF by presentinga fully automatic segmentation pipeline for myocardial segmentationin full-cycle 2D MCE data. Specifically, a statisticalshape model is used to provide shape prior information thatguide the RF segmentation in two ways. First, a novel shapemodel (SM) feature is incorporated into the RF frameworkto generate a more accurate RF probability map. Second, theshape model is fitted to the RF probability map to refineand constrain the final segmentation to plausible myocardialshapes. We further improve the performance by introducinga bounding box detection algorithm as a preprocessing stepin the segmentation pipeline. Our approach on 2D image isfurther extended to 2D+t sequences which ensures temporalconsistency in the final sequence segmentations. When evaluatedon clinical MCE datasets, our proposed method achieves notableimprovement in segmentation accuracy and outperforms otherstate-of-the-art methods including the classic RF and its variants,active shape model and image registration.
AU  - Li,Y
AU  - Ho,CP
AU  - Toulemonde,M
AU  - Chahal,N
AU  - Senior,R
AU  - Tang,MX
DO  - 10.1109/TMI.2017.2747081
EP  - 1091
PY  - 2018///
SN  - 0278-0062
SP  - 1081
TI  - Fully automatic myocardial segmentation of contrast echocardiography sequence using random forests guided by shape model
T2  - IEEE Transactions on Medical Imaging
UR  - http://dx.doi.org/10.1109/TMI.2017.2747081
UR  - http://hdl.handle.net/10044/1/50535
VL  - 37
ER  -
Download