Heart Defect Monitoring System Based on Hybrid Machine Learning Approach

Abstract

Early diagnosis of heart defects is crucial in order to reduce mortality from cardiovascular diseases and improve survival of patients. This paper proposes IoT-enabled heart defect monitoring system by the use of a novel machine learning-based technique applied to echocardiography (ECHO) images that can be used for continuous and remote cardiac health assessment. The proposed system fusion of the IoT-based physiological data acquisition with a novel hybrid feature learning approach, namely, the fusion of deep feature representations and optimized handcrafted features for the improved diagnostic accuracy. A lightweight ma-chine learning classifier is used to guarantee the real-time performance and suitability for IoT environment. The validity of the system is tested on a dataset of 1500 ECHO im-ages which include normal and pathological cardiac cases. Experimental results have shown that the classification accuracy of the proposed method is 97.1%, sensitivity is 96.4%, specificity is 97.8%, and F1 score is 96.9%, which is better than the conventional machine learning models by an average margin of 6.2% in classification accuracy. The IoT framework enables real-time monitoring with an average data transmission latency of 1.6 seconds and reduces the diagnostic time to 45% as compared to the manual ECHO interpretation. The obtained results validate the effectiveness, scalability and reliability of the proposed novel IoT and machine learning based heart defect monitoring system for smart healthcare applications.