Longitudinal Study of Mitral Valve Stenosis Prognosis using Deep Learning Techniques

This research presents a deep learning-driven framework for Mitral Valve Stenosis (MVS) detection from echocardiographic images, integrating advanced image processing and classification techniques. Segmentation isolates the Left Ventricle (LV) and Left Atrium (LA), followed by bounding box and contour detection to delineate the Mitral Valve. InceptionV3 and ResNet-50 models classify MVS, achieving test accuracies of 87.71% and 92.88%, respectively, with ResNet-50 demonstrating superior performance through deeper architecture and skip connections. This non-invasive approach enhances diagnostic precision, reduces redundant scans, and improves patient outcomes. Future efforts focus on real-time dataset expansion and clinical application development for widespread adoption.