An Ultra-Lightweight Time Period CNN Based Model with AI Accelerator Design for Arrhythmia Classification

This work proposes an arrhythmia classification system. The algorithm includes naive electrocardiography (ECG) data preprocessing procedures that apply to various ECG databases. Additionally, the paper presents an ultra-lightweight model designed for arrhythmia classification, which combines a Convolutional Neural Network (CNN) with long-term heart rate information to enhance the performance of the model. The proposed model was trained and tested using the MIT-BIH and NCKU-CBIC database, following the classification standards of the Association for the Advancement of Medical Instrumentation (AAMI), achieving an accuracy of 98.5% and 97.1%. Furthermore, this work proposes a customized artificial intelligence (AI) accelerator for hardware implementation, which leverages a parallelized processing element (PE) array architecture and hybrid stationary techniques to achieve high-performance computing. The chip implementation achieves a power consumption of 122 μW, a classification latency of 6.8 ms, and an energy efficiency of 0.83 μJ/classification.