Automated Cardiac Disease Classification from Portable Ultrasound Images Using Deep Learning

This research focuses on specific issues in the provision of cardiac examination services. Patients have to visit specialized ultrasound rooms for heart examinations, which can be challenging for those with poor health conditions. The primary obstacle stems from the bulky nature of conventional ultrasound machines, which require stationary setups and are not easily portable. To overcome this, a compact and cloudenabled system has been designed to automatically classify heart conditions by analyzing the motion of the left ventricle. The system employs the Lucas-Kanade optical flow algorithm to monitor the movement of the heart's chambers. Through this tracking process, two essential features are derived: motion distance and direction. The motion distance is computed using the Euclidean formula, while the direction is determined through angular calculations utilizing the cosine rule in triangle geometry. Furthermore, after obtaining all the features, a classification process is performed using deep learning methods. The research results demonstrate the success of the tracking and feature extraction processes. During the classification process, the system achieves an accuracy rate of 94.29 % for two chambers and four chambers, and an accuracy rate of 84.21% for short axis. This indicates the potential utilization of deep learning methods in classifying heart conditions based on left ventricle motion.