Impact of Approximate QRS Identification on RR Interval-Based Arrhythmia Detection and Classification

Wearable devices for arrhythmia diagnosis are battery-powered and require improved power efficiency. In our previous study, we applied approximate computing, which is effective for power reduction, to QRS identification algorithm and achieved significant power reduction. However, the impact of this approximation on arrhythmia detection and classification was not fully investigated. In this study, we investigate how the approximation --- specifically, the use of two approximate multipliers and eight approximate squarers applied to the Pan-Tompkins algorithm --- affects RR interval-based arrhythmia detection and classification. We use 48 samples from the MIT-BIH Arrhythmia Database to evaluate the accuracy of arrhythmia detection and classification. It was found that the approximate multiplier mul16s_GQU is the best choice, achieving nearly 100^% accuracy and less than 1 % false negatives in abnormal/normal classification. Additionally, the most power-efficient approximate squarer, E10C1, achieves relatively high accuracy with less than 5 % false negatives.