An Energy-Efficient Cardiac Arrhythmia Classification Processor using Heartbeat Difference based Classification and Event-Driven Neural Network Computation with Adaptive Wake-Up

Wearable intelligent ECG sensors integrating cardiac arrhythmia classification processor have been used to detect and classify arrhythmia to alert users for potential cardiovascular diseases [1] [2]. The state-of-the-art arrhythmia classification processors using neural network (NN) can achieve high accuracy, but the high complexity of NN computation brings significant energy consumption. Another challenge is that the accuracy of the NN is affected by the patient-to-patient variation, leading to accuracy degradation when applying a trained NN to the patients whose ECG features differ from that in the training database. To address the above issues, in this work, we proposed an arrhythmia classification processor using heartbeat difference encoding and event-driven NN to achieve high energy efficiency and high accuracy against patient-to-patient variation. The main features of the proposed processor include a) heartbeat difference based classification to improve the accuracy under the patient-to-patient variation and reduce the energy consumption. b) event-driven NN computation with shared feature extraction to reduce the energy consumption. c) adaptive NN wake-up technique to reduce the energy consumption while maintaining accuracy.