A Smartphone-Enabled Deep Learning Approach for Myocardial Infarction Detection Using ECG Traces for IoT-Based Healthcare Applications

The development of artificial intelligence (AI)-based frameworks to detect myocardial infarction (MI) from electrocardiogram (ECG) data is important for Internet of Things (IoT)-based healthcare applications. This letter proposes a smartphone-enabled deep learning (DL)-based healthcare system for the automated detection of MI using ECG traces. An ensemble or majority voting-based DL model is formulated based on the decisions of five branches of network layers, with each branch consisting of a frozen transfer learner block followed by global average pooling, batch normalization, drop-out, and Softmax layers. The weight parameters of the suggested ensemble model are quantized using the floating point 16-bit and integer 8-bit formats. After quantization, the model is deployed on the Android mobile device to detect MI and abnormal heart diseases (AHD). The results demonstrate that the suggested ensemble DL model has obtained the accuracy, precision, and recall values of 99.34%, 99.35%, and 99.34%, respectively, for classifying normal, MI, and AHD. The proposed healthcare framework is simple and inexpensive for the automated prediction of MI and AHD using 12-lead ECG trace images.