VLSI Design of LSTM-Based ECG Classification for Continuous Cardiac Monitoring on Wearable Devices

A portable and efficient electrocardiogram (ECG) classification system is essential for continuous cardiac monitoring in wearable healthcare devices. This paper presents a highly efficient very large scale integration architecture optimized for real-time ECG classification. The proposed system integrates a feature extraction module that utilizes a four-level daubechies discret wavelet transform and a classification module comprising multiple long-short-term memory recurrent neural networks, fully connected layers, and a multilayer perceptron. The design achieves a classification accuracy of 99%$99\%$. The hardware architecture demonstrates low resource utilization and operates at a power consumption of 41 mW with a clock frequency of 54 MHz, ensuring real-time classification. The presented design is verified on a Xilinx field-programmable gate array and tested using the publicly available ECG data set. Compared to state-of-the-art implementations, our approach achieves a superior balance between classification accuracy, power efficiency, and hardware resource optimization, making it suitable for wearable cardiac monitoring applications.