This study enhances the accuracy of semantic segmentation models in cardiology using a custom echocardiogram dataset. The goal is to adapt an existing deep learning model for better segmentation of heart structures in echocardiographic images, crucial for automated cardiac disease diagnosis. The performance improvement is evaluated using cardiology-specific metrics, showing enhanced segmentation accuracy of cardiac structures. This approach increases the model's clinical utility for cardiologists in diagnostics and treatment planning. The results highlight the potential of customized deep learning models in medical imaging and emphasize the importance of specialized datasets for precision in medical applications. This research contributes significantly to artificial intelligence in healthcare, offering advancements in automated echocardiographic analysis for clinical use.