Fully automated segmentation of long-axis MRI strain-encoded (SENC) images using active shape model (ASM)

Authors:
Ahmed A. Harouni;David A. Bluemke;Nael F. Osman
Affiliations:
Departments of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD;National Institutes of Health, Bethesda MD;Departments of Electrical and Computer Engineering, Baltimore and Departments of Radiology, Johns Hopkins University, Baltimore, MD
Venue:
ISBI'09 Proceedings of the Sixth IEEE international conference on Symposium on Biomedical Imaging: From Nano to Macro
Year:
2009

Citing 2
Cited 0

Active shape models—their training and application

Computer Vision and Image Understanding
Snakes, shapes, and gradient vector flow

IEEE Transactions on Image Processing

Quantified Score

Hi-index	0.00

Visualization

Abstract

Myocardial strain is an important measure used for assessing regional function, which could help in detecting myocardial infarction as well as following up with patients with heart diseases. MRI Strain-Encoding technique (SENC) produces strain values throughout the cardiac cycle. SENC has proved to be one of the few techniques that can quantify right ventricle (RV) regional function. However, SENC images suffer from low signal-to-noise ratio (SNR). In this paper we present a fully automatic method to detect, segment, and track the myocardium throughout the cardiac cycle using prior knowledge of the shape of the 4-champers long-axis (LA) view. Our detection algorithm has a success rate of 91% (33/36 cases). The dice similarity coefficient was 0.81 ± 0.07 and 0.71 ± 0.15 for the left ventricle-septum (LV-SEP) and RV respectively, yielding a high correlation R ≥ 0.91 between strain values measured from automatic and manual segmentation.