Application of Deep Learning to Enhance Finger Movement Classification Accuracy From UHD-EEG Signals

Abstract

This study investigates the classification of Ultra-High-Density Electroencephalography (UHD-EEG) signals corresponding to finger movements through the application of machine learning techniques, namely Support Vector Machines (SVM) and Multi-Layer Perceptrons (MLP). We analyzed UHD-EEG data from five subjects engaged in motor tasks involving finger extensions, applying binary classification to each pair of fingers. The MLP models achieved an average classification accuracy of 65.68%, demonstrating a considerable improvement over SVMs (60.4%). Further, we utilized saliency maps generated from the MLP models to identify the periods most critical for classification, uncovering the phases of finger flexion and relaxation as particularly informative. These saliency maps succesfully visualized the most important time periods and channels in the deep learning predictions. This work not only sheds light on the neural mechanisms of finger movement but also underscores the efficacy of advanced machine learning methodologies in decoding UHD-EEG signals, marking a substantial contribution to the field of neural engineering and rehabilitation technology.

Keywords

• Ultra-high-density electroencephalography
• support vector machine
• multi-layer perceptron