Cardiac Left Ventricle Segmentation Using U-Net Network

This study aims to improve the segmentation of the left ventricle in cardiac magnetic resonance images, which is a crucial task for monitoring and diagnosing heart disease. We suggest an improved method based on a U-Net deep learning model that includes Grad-CAM for interpretability, a generalized dice loss to address class imbalance, and augmentation strategies specifically designed for cardiac MRI. Our approach achieves strong results on the Sunnybrook Cardiac Dataset using a pretrained model to speed up convergence and enhance segmentation performance. The results demonstrate improved model transparency and segmentation accuracy, providing a reliable and comprehensible clinical solution. This work closes a significant research gap and attempts to support clinical decision-making by focusing on the explainability and increase in cardiac magnetic resonance segmentation data.