A Novel Transfer Learning Architecture Integrating 3D CNN and MVCNN for Heart Disease Classification from Echocardiogram Videos Focusing on The Heart Valves Region

The echocardiogram, also known as a cardiac ultrasound, captures real-time images of the heart's chambers and valves using ultrasonic waves, allowing healthcare professionals to assess cardiac function, including the critical movement and operation of heart valves. In this research study, we propose a novel approach for heart disease classification using echocardiogram videos, with a specific focus on the heart valves region, leveraging transfer learning and ensemble methods. Our approach integrates 3D Convolutional Neural Networks (3D CNN) and Multi-View Convolutional Neural Networks (MVCNN) within a transfer learning framework. Pretrained models such as VGG19, Densenet201, and Inceptionv3 are employed as feature extractors, leveraging their learned representations from large-scale datasets for effective feature extraction. The innovation in our method lies in the integration of 3D CNN with MVCNN and ensemble learning, which collectively enhances the ability to capture both spatial and temporal patterns in the heart valves region of echocardiogram videos. This novel architecture significantly improves classification accuracy, achieving an overall accuracy of 98.09% in classifying heart diseases. Additionally, our approach yields an Area Under the Curve (AUC) of 0.82, precision of 98.15%, recall of 97.95%, and an F1 score of 98.05%. These results demonstrate a significant advancement in heart disease classification using echocardiogram videos, underscoring the effectiveness of our integrated framework.