Mask R-CNN for segmentation of left ventricle

Document Type

Conference Item

Publication Date

1-1-2021

Abstract

Globally, cardiovascular diseases (CVDs) remain the major cause of death among citizens. With echocardiography, doctors are able to diagnose and determine vital parameters for the evaluation of these diseases. Segmentation of left ventricular (LV) from echocardiography is a significant tool for cardiovascular medical analysis. Besides calculating important clinical indices (e.g. ejection fraction), segmentation also can be useful for the investigation of the basic structure of ventricle. Automatic segmentation of the LV has become a valuable means in echocardiography as we can achieve fast and accurate results and a large number of cases can be handled with limited availability of experts. The Convolutional Neural Networks (CNN) have shown outstanding outcomes for image classification, detection, and segmentation in numerous fields. Recently Mask Regions Convolutional Neural Network (Mask R-CNN) has emerged as a very good segmentation model. In this work, Mask R-CNN is proposed for the segmentation of LV. The Mask R-CNN model is first fine-tuned with Common Object in Context (COCO) weights and then the model is trained with our own data. The model first finds out the region of interest (ROI) in the image that contains the desired object i.e. LV. In the ROI, the model segment LV by generating the mask around it. The results demonstrated by the proposed method segments the LV accurately and efficiently with limited training data. © 2021, Springer Nature Switzerland AG.

Keywords

Deep learning, Left ventricle, Medical images, Segmentation

Divisions

biomedengine,sch_ecs

Funders

Ministry of Higher Education, Malaysia

Publication Title

IFMBE Proceedings

Volume

81

Event Title

3rd International Conference for Innovation in Biomedical Engineering and Life Sciences, ICIBEL 2020

Event Location

Kuala Lumpur

Event Dates

6 - 7 December 2019

Event Type

conference

This document is currently not available here.

Share

COinS