Echocardiographic Assessment of Heart Disease Using CNN and Image Processing

Heart diseases are leading causes of premature death in adults over the age of 30, particularly for individuals with high cholesterol and diabetes. The diagnosis of these diseases often requires an echocardiogram, which provides a detailed view of the heart. However, the proper functioning of the echocardiogram machine and the expertise of a competent radiologist are essential for accurate analysis, which is not always feasible and possible. Manual annotation of echocardiograms to determine the presence of heart diseases is a laborious task that is prone to human errors like misdiagnosis and can be automated by using image processing algorithms like CNN. The probability of misdiagnosis increases in remote areas, where the training of radiologists may be inadequate compared to those in metropolitan cities. Additionally, faulty or defective equipment can contribute to inaccurate diagnoses in these remote areas. To address these challenges, we propose a system that automates the prediction of heart diseases based on echocardiograms. By reducing human error, our goal is to ensure a consistent standard of healthcare across both urban and rural medical institutions. For the initial analysis of this project, we implemented Convolutional Neural Network (CNN), ResNet50, and VGG 16 algorithms. After testing their accuracies we choose CNN as the optimal algorithm due to accuracy concerns. Specifically, the system focuses on detecting Angina Pectoris, Cardiovascular disease, Coronary Artery Disease, and Left Ventricular Hypertrophy(LVH). During preliminary testing, our CNN model achieved an accuracy score of 95.27%.