Scoring upper-extremity motor function from EEG with artificial neural networks: a preliminary study

Motor function of chronic stroke survivors is generally accessed using clinical motor assessments. These motor assessments are partially subjective and require prior training for the examiners. Additionally, those motor function assessments require the health professionals to be present in person. The method proposed in this paper has the potential to radically change the way motor function is assessed.

This work investigates the feasibility of automatically scoring upper-extremity motor function from EEG using artificial neural networks. Twelve healthy participants and fourteen participants with chronic stroke participated in this study. EEG data were recorded while the participants were clicking a button. Convolutional neural network models were trained based on the participants' Fugl Meyer motor assessment score.

The result showed that the proposed method achieved high prediction accuracy both within (n  =  14, r  =  0.9921, p   =  3.3907  ×  10^-12) and cross (n  =  14, r  =  0.9867, p   =  7.9342  ×  10^-11) participant testing.

This evidence suggests the proposed method is feasible to be used as a stable and objective measurement for motor function assessment.