Extraction of ECG features with spiking neurons for decreased power consumption in embedded devices

Abstract

In recent years, the computational efficiency of spike-based biomimetic information processing has received increasing interest. Here we show by simulation how two spiking neurons with different thresholds can be used to extract ECG features. One high-threshold neuron detects the location of the heartbeat, and the other low-threshold neuron detects other small-magnitude features. These detected features alone can then be transmitted to a nearby computer to classify the heartbeat instead of the entire ECG signal. Reducing transferred data by about 50 times, minimizing energy consumption and thus potentially extending the continuous use time for health monitoring applications. We show that a KNN algorithm classifies the heartbeat based on the obtained features with an overall accuracy of 96%, proving our method’s feasibility.