A novel approach to heart disease classification using echocardiogram videos with transfer learning architecture and MVCNN integration

Abstract

The echocardiogram, also known as a cardiac ultrasound, captures real-time images of the heart’s chambers and valves. Ultrasonic waves are used in this method to penetrate the skin and generate the pattern of the heart’s movement, allowing healthcare professionals to assess its overall function. In this research study, we propose a novel approach for classifying heart diseases relying on echocardiogram videos using transfer learning and ensemble methods. The approach involves using pre-trained convolutional neural network models such as VGG19, Densenet201, and Inceptionv3 as feature extractors and then training a classifier on top of these extracted features. The pre-trained models have been trained on large datasets with millions of images, making them highly effective feature extractors for various computer vision tasks. The main objective is to leverage the learned representations from these models and apply them to echocardiogram videos for accurate classification of heart diseases. The novel integration of MVCNN (pre-trained convolutional neural network models VGG19, Densenet201, and Inceptionv3) with ensemble methods has led to a significant increase in accuracy, achieving an overall accuracy of 98.09% in classifying heart diseases using echocardiogram videos and achieved AUC-0.82% After implementing the novel integration.