A common assumption in traditional supervised learning is the similar probability distribution of data between the training phase and the testing/operating phase. When transitioning from the training to testing phase, a shift in the probability distribution of input data is known as a covariate shift. Covariate shifts commonly arise in a wide range of real-world systems such as electroencephalogram-based brain–computer interfaces (BCIs). In such systems, there is a necessity for continuous monitoring of the process behavior, and tracking the state of the covariate shifts to decide about initiating adaptation in a timely manner. This paper presents a covariate shift-detection and -adaptation methodology, and its application to motor imagery-based BCIs. A covariate shift-detection test based on an exponential weighted moving average model is used to detect the covariate shift in the features extracted from motor imagery-based brain responses. Following the covariate shift-detection test, the methodology initiates an adaptation by updating the classifier during the testing/operating phase. The usefulness of the proposed method is evaluated using real-world BCI datasets (i.e. BCI competition IV dataset 2A and 2B). The results show a statistically significant improvement in the classification accuracy of the BCI system over traditional learning and semi-supervised learning methods.

The non-stationary nature of electroencephalography (EEG) signals makes an EEG-based brain-computer interface (BCI) a dynamic system, thus improving its performance is a challenging task. In addition, it is well-known that due to non-stationarity based covariate shifts, the input data distributions of EEG-based BCI systems change during inter- and intra-session transitions, which poses great difficulty for developments of online adaptive data-driven systems. Ensemble learning approaches have been used previously to tackle this challenge. However, passive scheme based implementation leads to poor efficiency while increasing high computational cost. This paper presents a novel integration of covariate shift estimation and unsupervised adaptive ensemble learning (CSE-UAEL) to tackle non-stationarity in motor-imagery (MI) related EEG classification. The proposed method first employs an exponentially weighted moving average model to detect the covariate shifts in the common spatial pattern features extracted from MI related brain responses. Then, a classifier ensemble was created and updated over time to account for changes in streaming input data distribution wherein new classifiers are added to the ensemble in accordance with estimated shifts. Furthermore, using two publicly available BCI-related EEG datasets, the proposed method was extensively compared with the state-of-the-art single-classifier based passive scheme, single-classifier based active scheme and ensemble based passive schemes. The experimental results show that the proposed active scheme based ensemble learning algorithm significantly enhances the BCI performance in MI classifications.

Appropriately combining mental practice (MP) and physical practice (PP) in a poststroke rehabilitation is critical for ensuring a substantially positive rehabilitation outcome. Here, we present a rehabilitation protocol incorporating a separate active PP stage followed by MP stage, using a hand exoskeleton and brain-computer interface (BCI). The PP stage was mediated by a force sensor feedback-based assist-as-needed control strategy, whereas the MP stage provided BCI-based multimodal neurofeedback combining anthropomorphic visual feedback and proprioceptive feedback of the impaired hand extension attempt. A six week long clinical trial was conducted on four hemiparetic stroke patients (screened out of 16) with a left-hand disability. The primary outcome, motor functional recovery, was measured in terms of changes in grip-strength (GS) and action research arm test (ARAT) scores; whereas the secondary outcome, usability of the system was measured in terms of changes in mood, fatigue, and motivation on a visual-analog-scale. A positive rehabilitative outcome was found as the group mean changes from the baseline in the GS and ARAT were +6.38 kg and +5.66 accordingly. The VAS scale measurements also showed betterment in mood ( $-$ 1.38), increased motivation (+2.10) and reduced fatigue ( $-$ 0.98) as compared to the baseline. Thus, the proposed neurorehabilitation protocol is found to be promising both in terms of clinical effectiveness and usability.

Corticomuscular coupling has been investigated for long, to find out the underlying mechanisms behind cortical drives to produce different motor tasks. Although important in rehabilitation perspective, the use of corticomuscular coupling for driving brain-computer interface (BCI)-based neurorehabilitation is much ignored. This is primarily due to the fact that the EEG-EMG coherence popularly used to compute corticomuscular coupling, fails to produce sufficient accuracy in single-trial based prediction of motor tasks in a BCI system. In this study, we have introduced a new corticomuscular feature extraction method based on the correlation between band-limited power time-courses (CBPT) associated with EEG and EMG. 16 healthy individuals and 8 hemiplegic patients participated in a BCI-based hand orthosis triggering task, to test the performance of the CBPT method. The healthy population was equally divided into two groups; one experimental group for CBPT-based BCI experiment and another control group for EEG-EMG coherence based BCI experiment. The classification accuracy of the CBPT-based BCI system was found to be 92.81 ± 2.09% for the healthy experimental group and 84.53 ± 4.58% for the patients' group. The CBPT method significantly (p-value < 0.05) outperformed the conventional EEG-EMG coherence method in terms of classification accuracy. The experimental results clearly indicate that the EEG-EMG CBPT is a better alternative as a corticomuscular feature to drive a BCI system. Additionally, it is also feasible to use the proposed method to design BCI-based robotic neurorehabilitation paradigms.

A major issue in electroencephalogram (EEG) based brain-computer interfaces (BCIs) is the intrinsic non-stationarities in the brain waves, which may degrade the performance of the classifier, while transitioning from calibration to feedback generation phase. The non-stationary nature of the EEG data may cause its input probability distribution to vary over time, which often appear as a covariate shift. To adapt to the covariate shift, we had proposed an adaptive learning method in our previous work and tested it on offline standard datasets. This paper presents an online BCI system using previously developed covariate shift detection (CSD)-based adaptive classifier to discriminate between mental tasks and generate neurofeedback in the form of visual and exoskeleton motion. The CSD test helps prevent unnecessary retraining of the classifier. The feasibility of the developed online-BCI system was first tested on 10 healthy individuals, and then on 10 stroke patients having hand disability. A comparison of the proposed online CSD-based adaptive classifier with conventional non-adaptive classifier has shown a significantly (p≤0.01) higher classification accuracy in both the cases of healthy and patient groups. The results demonstrate that the online CSD-based adaptive BCI system is superior to the non-adaptive BCI system and it is feasible to be used for actuating hand exoskeleton for the stroke-rehabilitation applications.