Real-Time ECG Feature Detection Algorithm using Moving Statistic Adaptive Thresholding Method for Signal Extremum Sampling in Microcontroller Systems: A Simulink Approach.

This work has implemented a real-time ECG feature detection system, a precursor to automated detection, using a moving statistic adaptive thresholding algorithm for signal extremum sampling. The lack of literature on implementing a holistic real-time feature detection system in affordable microcontroller devices motivates the pursuit of this topic. This system uses MATLAB Simulink to design the necessary components that detect the Heart Rate; Q, R, S, and T peaks; QRS complex, QTend, STpeak, STend, and TpkTend interval features. Peak feature detection systems use MATLAB programming to assess the extrema sampled and the adaptive threshold output from the moving statistic algorithm. On the other hand, the interval feature detection systems use peak detection system outputs, unit pulse interval sampler, and time-integrator algorithms that increment a particular constant per sample until it reaches the boundary of the periodic interval. Comparing the yielded ECG from the implemented system to the KenzECG108 device and Apple Watch Series 4 ECG application using Pearson correlation and two-sample independent t-test for evaluation determines a strong linear correlation that suggests a similarity between the displayed ECG Feature Detection System ECG and the other two ECG devices (KenzECG108 vs. ECG Feature Detection System: r = 0.9032 (p= 7.542e-279); Apple Watch ECG vs. ECG Feature Detection System: r= 0.8919 (p = 0)). No significant difference was also observed between the mean of the KenzECG108 Heart Rate and the ECG Feature Detection System (h = 0; p = 0.7597; ci = [-2.731; 1.9975]); and also the mean of the Apple Watch ECG (h = 0; p = 0.6038; ci = [-3.350; 1.949]). The system's detection performance was also evaluated wherein the average performance of the ECG Feature Detection System has: Accuracy = 0.98012; Error Rate = 0.0198; True Positive Rate = 0.9914; Precision = 0.9807.