An Improved Deep Learning Approach for Heart Attack Detection from Digital Images

The mortality rate due to different diseases is alarmingly rising day by day across the world. The major reason for this death rate includes heart-related problems occurring due to age factors, blood pressure, and diabetes. Normally, old people like living by on their own which creates problems in cases of an emergency, and it gets hard for the paramedical staff to provide them with prompt help. Several people die just because of not getting emergency medical attention during a heart attack. The patients usually cannot convey a request for help due to severe pain in the chest which stops them to do any activity. Hence, timely identification of a patient with an ongoing heart attack becomes a matter of life and death. In this research, we propose a new methodology for the identification of people with an ongoing heart attack in color images. For this, we implement various pre-trained deep learning Convolutional Neural Networks (CNNs) models including a modified version of ResNet-50 to identify a person with a heart attack by detecting special heart attack-related postures. A special set of images containing the people having a heart attack are input to these models for comprehensive training. As compared to the other implemented pre-trained models, our modified ResNet-50 model achieved an accuracy of 92% during the classification of infarcts.