Core Concepts
BCIモデルをキャリブレーションから制御に効果的に移行する新しいパラダイムを示す。
Abstract
ABSTRACT:
Motor Imagery-based BCI datasets are used to develop good classifiers.
Uncertainty in EEG patterns during control tasks with BCIs may lead to errors.
New paradigm with calibration and EMG-based BCI control session demonstrated.
INTRODUCTION:
Transparent progress in Machine Learning models for EEG processing using public BCI datasets.
Discrepancy between benchmark datasets and real BCI usability requirements.
Introduction of a new dataset addressing limitations of existing datasets.
CONTRIBUTIONS:
New paradigm for transferring BCI-based ML models from calibration to control tasks.
Visual cued calibration session followed by driving a simulated car based on EMG signals.
METHODS:
Calibration session following Graz-BCI Motor Imagery paradigm, EMG used for mock BCI development.
Offline analysis of EEG data recorded during calibration and driving sessions.
RESULTS:
SMR and MRCP analysis showed similarities and differences between calibration and driving sessions.
CSP classifier performance best in calibration, followed by driving, then transfer scenario.
DISCUSSION:
Challenges in continuous decoding for online BCI control tasks highlighted.
Low classification performance for some subjects may impact successful control with EEG-based BCIs.
CONCLUSION:
Feasibility of SMR-based classifier for transferring from calibration to driving demonstrated.
Challenges remain for implementing online EEG-based BCIs for control tasks.
Stats
この論文では、20人の健康な被験者が使用されました。
EMG分類器は、平均精度94%で訓練されました。