1291 Towards an Autonomous Brain-Computer Interface for Chronic Stroke Neuromodulation: The Importance of Gamma Power in Movement Classification

INTRODUCTION: Chronic stroke affects over 7 million people in the US. Brain computer interfaces (BCIs) have the potential to improve quality of life for affected individuals given their ability to augment existing neuromodulatory therapies such as transcranial direct current stimulation (tDCS). A simple yet flexible BCI leveraging machine learning (ML) classifiers to classify movement state would allow neuromodulation to be delivered autonomously and continuously. METHODS: 10 participants with chronic stroke and 11 healthy controls were included in this study. Participants were fitted with an EEG cap and tDCS anodal electrode positioned at the ipsilesional motor cortex. Participants were then randomly assigned to either the tDCS stimulation or sham groups. After receiving stimulation or sham therapy, participants used a VR headset to perform reach tasks while EEG data was sampled at 1024 Hz. The recorded EEG data were then z-score normalized and binned. PSD and coherence were extracted and subsequently used to train 13 ML models to classify movement state. RESULTS: We observed that gamma PSD produced the highest classification accuracies. Classification was improved for stroke patients and those who received tDCS. We did not observe significant differences between our ML models with regards to accuracy. CONCLUSIONS: Simple ML models are able to classify movement state in stroke patients from minimally pre-processed EEG data with gamma PSD being most indicative of movement state.