Edge-Optimized Hybrid CNN-BiLSTM Framework for Real-Time ECG Classification

In modern times, real-time Electrocardiogram (ECG) signal classification has received significant attention for continuous cardiac observation, specifically in Internet of Medical Things (IoMT) applications. The precise and timely detection of arrhythmias is essential to avoid life-threatening cardiac events. However, installing Deep Learning (DL) models on edge devices has several drawbacks, including a high computational complexity and fewer memory and energy constraints. Moreover, traditional approaches, such as Convolutional Neural Networks (CNN), frequently struggle to capture long-range temporal dependencies in ECG sequences, affecting the classification performance. To overcome these drawbacks, a lightweight and efficient ECG classification framework uses a hybrid approach which combines CNN with Bidirectional Long Short-Term Memory (BiLSTM). First, ECG data are collected from Massachusetts Institute of TechnologyBeth Israel Hospital (MIT-BIH) arrhythmia dataset. Then, using band-pass filtering, min-max normalization, segmentation, and Short-Time Fourier Transform (STFT), data is preprocessed. Furthermore, a CNN is employed for extraction of spatial features, and a BiLSTM to capture temporal dependencies. The results clearly show that proposed CNNBiLSTM achieves better results in terms of accuracy (99.32 %) for real-time health-monitoring applications than existing CNN approach.