Selective Subject Pooling Strategy to Improve Model Generalization for a Motor Imagery BCI.

Kyungho Won,Sung Chan Jun,Moonyoung Kwon,Minkyu Ahn

doi:10.3390/s21165436

Abstract

Brain–computer interfaces (BCIs) facilitate communication for people who cannot move their own body. A BCI system requires a lengthy calibration phase to produce a reasonable classifier. To reduce the duration of the calibration phase, it is natural to attempt to create a subject-independent classifier with all subject datasets that are available; however, electroencephalogram (EEG) data have notable inter-subject variability. Thus, it is very challenging to achieve subject-independent BCI performance comparable to subject-specific BCI performance. In this study, we investigate the potential for achieving better subject-independent motor imagery BCI performance by conducting comparative performance tests with several selective subject pooling strategies (i.e., choosing subjects who yield reasonable performance selectively and using them for training) rather than using all subjects available. We observed that the selective subject pooling strategy worked reasonably well with public MI BCI datasets. Finally, based upon the findings, criteria to select subjects for subject-independent BCIs are proposed here.

Highlights

Accepted: 10 August 2021Brain–computer interfaces (BCIs) have shown great usefulness in facilitating communication for people with disabilities by extracting brain activities and decoding user intentions [1]
It is common to train subject-independent spatial filters and classifiers using all subjects available; in this work, we propose a selective subject pooling strategy that selects subjects who are likely to generate discriminative EEG patterns during motor imagery, and subject-independent classifiers are trained with the selective subject pool only
Note that subject-independent BCI (SI BCI)-All refers to SI BCI performance using all subjects available, and SI BCI-α refers to SI BCI

Summary

Introduction

Accepted: 10 August 2021Brain–computer interfaces (BCIs) have shown great usefulness in facilitating communication for people with disabilities by extracting brain activities and decoding user intentions [1]. MI BCIs are more intuitive than other BCI systems that use reactive brain activity because they does not require external stimulation to generate the brain activity and can be applied in various areas to provide new communication and control channels for people who cannot move their own bodies [1]. Among other applications, these systems may be used to rehabilitate motor functions for patients [4] and develop game content [5,6]; some issues remain that must be resolved to increase the ability to use MI BCI. Users can operate the BCI application via their brain activity using the classifier that was trained or updated in the calibration phase; because the Published: 12 August 2021

Objectives

Methods

Results

Conclusion