Left Ventricular Wall Segmentation Using U-Net and Link-Net

Myocardial Infarction (MI) is categorized as the world's deadliest illnesses. It can be identified by looking for anomalies in the left ventricular wall through the heart ultrasound imaging method. Unfortunately, these types of images are often of poor quality, making it difficult for paramedics to identify them. Segmentation of the left ventricular wall in cardiac ultrasound images will be carried out to aid paramedics in seeing the left ventricular wall more clearly. U-Net architecture and its extension, Link-Net architecture, are employed in this study to do this segmentation. Both architectures share the use of skip-connections to enhance spatial information. The publicly available data from Tampere University, Qatar University, and Hamad Medical Corporation (HMC-QU) is chosen to demonstrate the work of these two architectures in segmenting the left ventricular wall. The results of the training indicate that Link-Net has a 13% faster training time when compared to U-Net. The highest training and validation accuracies achieved by Link-Net are 0.9973 and 0.9941. Link-Net is also superior by obtaining the highest F1 Score of 0.9636 throughout the testing phase. Both architectures have proven effective segmenting the left ventricular wall. The findings of this study have been shown to facilitate and improve the accuracy of medical practitioners in diagnosing MI.