An Investigation of Deep Supervised Approaches for Ventricle Region Segmentation using Cardiac MRI

Abstract

The precise delineation of ventricle regions through Cardiac MRI segmentation is a vital aspect of quantitative analysis of cardiac function. This facilitates the early detection and follow-up of many heart conditions. The goal of this investigation is to evaluate the most advanced deep learning methods presently in use, such as UNet, Segnet, FCN(Fully Convolutional Network), UNet++, and UNet-CBAM. The techniques are capable of extracting significant elements from the input images without receiving any manually constructed characteristics. The integration of deep learning techniques has brought notable improvements to the field of cardiac MRI segmentation. According to the evaluation metrics, all five models were highly meticulous at segmenting ventricle sections, whereas UNet-CBAM outperformed than other four models. Numerous studies have demonstrated that deep learning-based segmentation techniques perform more accurately and effectively than conventional segmentation techniques. Overall, the adaptation of deep learning models exposed a raise in the semantic segmentation framework of Cardiac MRI.