A Study of Inferencing ECG Heartbeat Classification based on CNN and LSTM on Edge Devices

The classification of electrocardiogram (ECG) heartbeats plays a vital role in the early diagnosis and treatment of various cardiac conditions. This study evaluates the performance of deep learning techniques, particularly convolutional neural networks (CNN) and long short-term memory (LSTM) networks, for classifying ECG heartbeats on edge devices. CNN and LSTM are the two most widely used deep learning architectures for ECG classification. Using publicly available ECG datasets, we trained and compared multiple models, aiming to identify the best classifier that balances accuracy and efficiency while also addressing the limitations of edge devices in terms of computational constraints, memory limitations, and hardware-specific variability. Our study highlights the superior performance of the LSTM-FCN model, particularly when utilizing the R-R interval (RRI) feature configuration, compared to CNN-based models. The RRI sequence represents heart rhythm changes, and it is less affected by baseline drift, muscle noise, and motion artifacts, improving classification robustness. This finding underscores the advantages of LSTM-FCN, which demonstrates higher computational efficiency and lower memory requirements, making it well-suited for real-time edge-device applications. To validate its practicality, we deployed the LSTM-FCN-based heartbeat classification model on edge devices and tested it with real-time ECG signals from wearable sensors. Our experimental results confirm that LSTM-FCN effectively balances high classification accuracy, low computational overhead, and minimal memory usage, making it an optimal choice for real-time ECG monitoring and classification in resource-constrained environments.