An Intelligent IoT-Compatible Arrhythmia Detection System Using a Hybrid Vision Transformer-LSTM Framework.

BACKGROUND: In electrocardiogram (ECG) signal classification, advanced IoT-compatible systems and medical signal processing solutions have become feasible due to the rapid growth of the Internet of Things (IoT). However, it is challenging to detect and classify arrhythmias because of the complex nature and large volumes of ECG data. OBJECTIVE: This article proposes a hybrid deep learning (DL) approach by combining long short-term memory (LSTM)and a Vision Transformer (ViT) for the automatic arrhythmia classifi cation from ECG signals. OBJECTIVE: This article proposes a hybrid deep learning (DL) approach by combining long short-term memory (LSTM)and a Vision Transformer (ViT) for the automatic arrhythmia classifi cation from ECG signals. METHODS: Initially, the proposed framework gathers the data from the MIT-BIH database for training. Then, preprocessing strategies such as min-max normalization and signal-to-spectrogram conversion using the Stockwell transform are applied. Then, the imbalance problem in the collected dataset is rectifi ed by applying the K-Mean scentered Adaptive Synthetic Sampling (KMADASYN) technique. Afterward, the proposed framework employs the Hybrid Optimized ViT with Long Short-Term Memory (HOVLSTM) for classification, where spatial and temporal features are extracted to achieve higher classification performance. RESULTS: The developed model achieved a classification accuracy of 99.17%, out performing several existing ECG classification systems. CONCLUSION: These outcomes demonstrate that the proposed technique is suitable for IoT-compatible healthcare systems. While the IoT-compatible system does not employ data acquisition, the model is evaluated in a simulated IoT environment, demonstrating its potential for future deployment in IoT-enabled clinical decision-support systems.