Left ventricle segmentation and quantification using deep learning

Document Type

Book Chapter

Source of Publication

Cardiovascular and Coronary Artery Imaging

Publication Date

1-1-2023

Abstract

Cardiac MRI is a widely used noninvasive tool that can provide us with an evaluation of cardiac anatomy and function. It can also be used for heart diagnosis. Heart diagnosis through the estimation of physiological heart parameters requires careful segmentation of the left ventricle (LV) from the images of cardiac MRI. Therefore we aim at building a new deep learning method for the automated delineation and quantification of the LV from cine cardiac MRI. Our goal is to reach lower errors for the calculated heart parameters than the previous works by introducing a new deep learning cardiac segmentation method. Our pipeline starts with an accurate LV localization by finding LV cavity center point using a fully convolutional neural network (FCN) model called FCN1. Then, from all heart sections, we extract a region of interest (ROI) that encompasses the LV. A segmentation for the LV cavity and myocardium is performed from the extracted ROIs using FCN called FCN2. The FCN2 model is associated with multiple bottleneck layers and uses less memory footprint than traditional models such as U-net. Furthermore, we introduced a novel loss function called radial loss that works on minimizing the distance between the ground truth LV contours and the predicted contours. After myocardial segmentation, we estimate the functional and mass parameters of the LV. We used the Automated Cardiac Diagnosis Challenge (ACDC-2017) dataset to validate our pipeline, which provided better segmentation, accurate calculation of heart parameters, and produced fewer errors compared to other approaches applied on the same dataset. Additionally, our segmentation approach showed that it can generalize well across different datasets by validating its performance on a locally collected cardiac dataset. To sum up, we propose a novel deep learning framework that we can translate it into a clinical tool for cardiac diagnosis.

ISBN

9780128219836

Publisher

Elsevier

First Page

113

Last Page

147

Disciplines

Computer Sciences | Medicine and Health Sciences

Keywords

Left ventricle, segmentation, cardiac MRI, deep learning, diagnosis

Indexed in Scopus

no

Open Access

no

Link to Full Text

Share

COinS