Next-Gen Cardiac Segmentation with Swish-U-Net and Optimizer Innovation

cardiovascular disease (CVD) is the leading global cause of death, responsible for 17.9 million deaths annually and accounting for 32% of global mortality in 2024. As the incidence of CVD rises, the need for advanced diagnostic tools increases. Cardiac image segmentation plays a key role in accurate diagnosis by visualizing cardiac structures. This study optimizes the U-Net architecture for segmentation by incorporating the Swish activation function to improve performance. We use the "Automatic Cardiac Diagnosis Challenge" (ACDC) dataset of 150 CMRI recordings to focus on the left ventricular cavity, myocardium, and right ventricle for segmentation. We compare five optimizers—Adam, SGD, Nadam, RMSprop, and Adagrad—using metrics such as accuracy, IoU, Jaccard distance, and Dice coefficient. The results show that using both Swish activation and the optimal optimizer together improves the accuracy of segmentation. This makes diagnostics more accurate and helps find cardiovascular disease early.