Online semi-supervised learning for motor imagery EEG classification

• Published in: Computers in biology and medicine Vol. 165; p. 107405
• Main Authors: Zhang, Li, Li, Changsheng, Zhang, Run, Sun, Qiang
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 01.10.2023; Elsevier Limited
• Subjects: Algorithms; Artificial neural networks; Brain-computer interface; Calibration; Classification; Classifiers; Data augmentation; Deep learning; Distance learning; Edited nearest neighbor rule; Electroencephalography; Extreme learning machine; Fatigue; Human-computer interface; Image classification; Internal Medicine; Internet; Machine learning; Mental task performance; Methods; Motor skill learning; Other; Semi-supervised learning; Synthetic minority oversampling technique; Brain-computer interface; Extreme learning machine; Synthetic minority oversampling technique; Semi-supervised learning; Edited nearest neighbor rule
• DOI: 10.1016/j.compbiomed.2023.107405
• ISSN: 0010-4825; 1879-0534

Time-consuming data labeling in brain-computer interfaces (BCIs) raises many problems such as mental fatigue and is one key factor that hinders the real-world adoption of motor imagery (MI)-based BCIs. An alternative approach is to integrate readily available, as well as informative, unlabeled data online, whereas this approach is less investigated. We proposed an online semi-supervised learning scheme to improve the classification performance of MI-based BCI. This scheme uses regularized weighted online sequential extreme learning machine (RWOS-ELM) as the base classifier and updates its model parameters with incoming balanced data chunk-by-chunk. In the initial stage, we designed a technique that combines the synthetic minority oversampling with the edited nearest neighbor rule for data augmentation to construct more discriminative initial classifiers. When used online, the incoming chunk of data is first pseudo-labeled by RWOS-ELM as well as an auxiliary classifier, and then balanced again by the above-mentioned technique. Initial classifiers are further updated based on these class-balanced data. Offline experimental results on two publicly available MI datasets demonstrate the superiority of the proposed scheme over its counterparts. Further online experiments on six subjects show that their BCI performance gradually improved by learning from incoming unlabeled data. Our proposed online semi-supervised learning scheme has higher computation and memory usage efficiency, which is promising for online MI-based BCIs, especially in the case of insufficient labeled training data. •An online semi-supervised learning method is proposed to deal with the problem of limited EEG data during calibration.•The proposed method does not rely on stored data to retrain the model online, which decreases memory usage burden.•An auxiliary classifier is introduced to help select pseudo-labeled data to avoid the performance degradation.•The proposed method can solve the problem of class imbalance during online learning.