An Efficient Method for Detection and Localization of Myocardial Infarction Using Ramanujan Sums Wavelet Transform and Machine Learning Framework

A myocardial infarction (MI) can be fatal and cause irreversible damage to the functioning of the human heart. Therefore, it is crucial to save the lives of such patients by providing early diagnosis. The primary step involves MI detection through automated or manual inspection by experts with the help of an electrocardiogram (ECG). Manual detection is time-consuming and less accurate. In contrast, an automated algorithm provides higher accuracy as well as saves valuable treatment time. This article proposes a novel, efficient technique for MI detection and localization based on Ramanujan sums wavelet transform (RSWT). The process involves statistical and morphological feature extraction with Ramanujan sums (RSs) and from detailed subbands decomposed by RSWT. These characteristics are used with the deep neural network (DNN) classifier and compared with 12 lead accuracy. The best classification accuracy of 99.93% is obtained with DNN with 40 features from lead 10 (V4). In addition, we have added localization of five types of MI and healthy with 99.50% accuracy from Lead 7 (V1) using an intrapatient scheme. Furthermore, we tested the model for the interpatient scheme with a detection accuracy of 89.94% from lead 11 (V5). All the above results are obtained using single lead ECG for localization and detection. This helps in the reduction in device cost and improvement of patient comfort. These are the best results obtained through chest lead (V1, V4, and V5) and are better than the existing state of the art.