ECG Beat-By-Beat Classification Using Hybrid Transformer Neural Network Model in Smart Health

Wearable cardiac monitors can be used to detect potential heart attack by syncing with smartphone apps for instant data analysis and alerts. Our goal is to build an efficient smart health application to help patients prevent and early diagnose the risk of heart disease. Our novel hybrid transformer neural network model can effectively predict the occurrence of heart disease and prevent it early. We used Artificial Neural Network and Convolutional Neural Network and combined them with the transformer model to form a hybrid transformer neural network model, Transformer Artificial Neural Network and Transformer Convolutional Neural Network, respectively. Used three different feature modules, (1) top 10 ranked time series features, (2) noise-removed MIT BIH ECG data, (3) combination module (top 10 ranked time series features and noise-removed MIT BIH ECG data) as input data and sent into the hybrid transformer neural network model to analysis their accuracy and power consumption, respectively. Converted the best-performing hybrid transformer neural network model into a pre-trained model and applied it to our Smart-Health application to verify the correctness and functionality of the algorithm. TCNN model achieved the highest accuracy of 98% and an F1 score of 99%. The hybrid transformer neural network model can be used in our application and detects cardiac diseases more accurately than the neural network model.