Automated Detection of Myocardial Infarction(MI) Complications Using Deep Learning Models on ECG Images

Abstract

Automated Myocardial Infarction (MI) Detection System is aimed at increasing the accuracy and speed of MI diagnosis by processing electrocardiogram (ECG) images. The system categorizes ECG images as normal ECG, abnormal heartbeats, history of MI, and active MI by using the sophisticated deep learning models. The system makes use of the following models: ResNet50, ResNet101, ResNet50V2, Xception, DenseNet201, DenseNet121, InceptionV3, MobileNetV2, EfficientNet B3, EfficientNet B5 and EfficientNet V2B2. Some of the models were tested by k-fold cross-validation, and ResNet50V2 was the highest-performing model with a score of 99.03, with DenseNet201 and MobileNetV2 coming in at 99% and 98.82ood, respectively. EfficientNet B3 (98.70%) and EfficientNet V2B2 (99.35) were also proved to be good models. Its system can be interoperable with Electronic Health Records (EHRs) and medical professionals can obtain the results of the diagnostic process effectively. This is because, not only does this integration shorten the diagnostic time but it also eradicates the chances of human error that help clinicians make more accurate and timely decisions by utilizing automated MI detection. Grad-CAM visualizations used in the system can generate interpretability, which is in line with the HIPAA and GDPR framework, and ensure the secure processing of data and compliance fulfillment. Finally, the Automated MI Detection System enhances the patient care experience by guaranteeing better and more timely MI diagnosis, improving workflow and providing healthcare experts with more sophisticated diagnostic resources.