EEG Artifact Removal At the Edge Using AI Hardware

Wearable electroencephalography (EEG) devices enable non-invasive brain monitoring for conditions like epilepsy but are often affected by artifacts. While many AI models for EEG artifact removal exist, real-time deployment on edge hardware has not been achieved. This paper presents the first implementation of a deep autoencoder for EEG artifact removal on edge hardware using Arduino Nano 33 BLE, Coral Dev Board Micro, and Coral Dev Board Mini. We compare these systems in terms of power consumption and inference time for 4-second EEG segments. The Coral Dev Board Mini demonstrated the fastest inference time (8.9 ms) but high power consumption (1.7 W), while the Coral Dev Board Micro balanced inference time (273 ms) with power consumption (0.6 W). The Arduino Nano 33 BLE had the lowest power draw (0.1 W) but longer inference time (1.3 s). These results show that edge AI for EEG artifact removal is feasible, with power consumption being the primary limitation for long-term battery-powered operation. This first-of-its-kind edge deployment of EEG processing represents a significant step toward artifact-free, real-time, portable EEG monitoring.