Data-driven Energy-efficient Adaptive Sampling Using Deep Reinforcement Learning

This article presents a resource-efficient adaptive sampling methodology for classifying electrocardiogram (ECG) signals into different heart rhythms. We present our in two folds: ( i ) the design of novel real-time neural network architecture capable ECG with rates and ii runtime implementation rate control using deep reinforcement learning (DRL). By essential morphological details contained heartbeat waveform, DRL agent can effectively reduce energy consumption at runtime. To evaluate classifier, we use MIT-BIH database recommendation AAMI to train classifiers. The classifier is designed recognize three major types arrhythmias, which are supraventricular ectopic beats (SVEB), ventricular (VEB), normal (N). performance arrhythmia classification reaches an accuracy 97.2% SVEB 97.6% VEB beats. Moreover, system 7.3× more energy-efficient compared baseline architecture, where not utilized. proposed provide reliable accurate signal analysis performances comparable state-of-the-art methods. Given its time-efficient, low-complexity, low-memory-usage characteristics, also suitable practical applications, case classification, resource-constrained devices, especially wearable healthcare devices implanted medical devices.