Abstract

Aims

1. To propose a baseline model with RNNs to classify ECG beats effectively and efficiently.
2. To propose a lightweight model with fused RNN for speeling up the prediction time on CPUs. 

Methods

Used 48 ECGs from the MIT-BIH (Massachusetts Institute of Technology-Beth Israel Hospital) Arrhythmia Database, and 76 ECGs were collected with S-Patch devices developed by Samsung SDS. Developed both baseline and lightweight models on the MXNet framework. Trained both models on graphics processing units and measured both models’ inference times on CPUs. 

Results

Our models achieved overall beat classification accuracies of 99.72% for the baseline model with RNN and 99.80% for the lightweight model with fused RNN. Moreover, our lightweight model reduced the inference time on CPUs without any loss of accuracy. The inference time for the lightweight model for 24-hour ECGs was 3 minutes, which is 5 times faster than the baseline model. 

Conclusion 

Both our baseline and lightweight models achieved cardiologist-level accuracies(RNN: 99.72%, fused RNN: 99.8%).
Lightweight model reduced the inference time on CPUs without any loss of accuracy(Interference time: 3 minures, which is 5 times faster than the baseline model)