Enhanced Classification of Individual Finger Movements with ECoG

• Published in: Conference record - Asilomar Conference on Signals, Systems, & Computers pp. 2063 - 2066
• Main Authors: Yao, Lin, Shoaran, Mahsa
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.11.2019
• Subjects: Accuracy; Brain-computer interfaces; Brain-machine interface (BMI); Conditional random fields; Decision trees; Decoding; ECoG; finger movement classification; Fingers; machine learning; Machine learning algorithms; Minimally invasive surgery; Motors; Prosthetics; temporal dynamics
• ISSN: 2576-2303
• DOI: 10.1109/IEEECONF44664.2019.9048649

Motor decoding at the level of individual finger movements is critical for high-performance brain-machine interface (BMI) applications. In this work, we propose to exploit the temporal dynamics of the multi-channel electrocorticography (ECoG) signal from human subjects and modern machine learning algorithms to improve the finger-level movement classification accuracy. Using a decision tree ensemble as the classifier and the temporally-concatenated features of ECoG as input, we achieved an average classification accuracy of 71.3%±7.1% on 3 subjects, 6.3% better than the state-of-the-art approach based on conditional random fields (CRF) on the same dataset. Our proposed method could enable a high-performance and minimally invasive cortical BMI for paralyzed patients.