Selective Cross-Subject Transfer Learning Based on Riemannian Tangent Space for Motor Imagery Brain-Computer Interface

Transferring a deep learning model from healthy subjects to stroke patients in a motor imagery brain-computer interface

Objective.Motor imagery (MI) brain-computer interfaces (BCIs) based on electroencephalogram (EEG) have been developed primarily for stroke rehabilitation, however, due to limited stroke data, current deep learning methods for cross-subject classification rely on healthy data. This study aims to assess the feasibility of applying MI-BCI models pre-trained using data from healthy individuals to detect MI in stroke patients.Approach.We introduce a new transfer learning approach where features from two-class MI data of healthy individuals are used to detect MI in stroke patients. We compare the results of the proposed method with those obtained from analyses within stroke data. Experiments were conducted using Deep ConvNet and state-of-the-art subject-specific machine learning MI classifiers, evaluated on OpenBMI two-class MI-EEG data from healthy subjects and two-class MI versus rest data from stroke patients.Main results.Results of our study indicate that through domain adaptation of a model pre-trained using healthy subjects' data, an average MI detection accuracy of 71.15% (±12.46%) can be achieved across 71 stroke patients. We demonstrate that the accuracy of the pre-trained model increased by 18.15% after transfer learning (p<0.001). Additionally, the proposed transfer learning method outperforms the subject-specific results achieved by Deep ConvNet and FBCSP, with significant enhancements of 7.64% (p<0.001) and 5.55% (p<0.001) in performance, respectively. Notably, the healthy-to-stroke transfer learning approach achieved similar performance to stroke-to-stroke transfer learning, with no significant difference (p>0.05). Explainable AI analyses using transfer models determined channel relevance patterns that indicate contributions from the bilateral motor, frontal, and parietal regions of the cortex towards MI detection in stroke patients.Significance.Transfer learning from healthy to stroke can enhance the clinical use of BCI algorithms by overcoming the challenge of insufficient clinical data for optimal training.

Classification Algorithm for Electroencephalogram-based Motor Imagery Using Hybrid Neural Network with Spatio-temporal Convolution and Multi-head Attention Mechanism

Motor imagery (MI) is a brain-computer interface (BCI) technique in which specific brain regions are activated when people imagine their limbs (or muscles) moving, even without actual movement. The technology converts electroencephalogram (EEG) signals generated by the brain into computer-readable commands by measuring neural activity. Classification of motor imagery is one of the tasks in BCI. Researchers have done a lot of work on motor imagery classification, and the existing literature has relatively mature decoding methods for two-class motor tasks. However, as the categories of EEG-based motor imagery tasks increase, further exploration is needed for decoding research on four-class motor imagery tasks. In this study, we designed a hybrid neural network that combines spatiotemporal convolution and attention mechanisms. Specifically, the data is first processed by spatiotemporal convolution to extract features and then processed by a Multi-branch Convolution block. Finally, the processed data is input into the encoder layer of the Transformer for a self-attention calculation to obtain the classification results. Our approach was tested on the well-known MI datasets BCI Competition IV 2a and 2b, and the results show that the 2a dataset has a global average classification accuracy of 83.3% and a kappa value of 0.78. Experimental results show that the proposed method outperforms most of the existing methods.

Segment alignment based cross-subject motor imagery classification under fading data

Motor imagery (MI) aims to use brain imagination without actual body activities to support motor learning, and machine learning algorithms such as common spatial patterns (CSP) are proven effective in the analysis of MI signals. In the conventional machine learning-based approaches, there are two main difficulties in feature extraction and recognition of MI signals: high personalization and data fading. The high personalization problem is due to the multi-subject nature when collecting MI signals, and the data fading problem as a recurring issue in MI signal quality is first raised by us but is not widely discussed at present. Aiming to solve the above two mentioned problems, a cross-subject fading data classification approach with segment alignment is proposed to classify the fading data of one single target with the model trained with the normal data of multiple sources in this paper. he effectiveness of proposed method is verified via two experiments: a dataset-based experiment with the dataset from BCI Competition and a lab-based experiment designed and conducted by us. The experimental results obtained from both experiments show that the proposed method can obtain optimal classification performance effectively under different fading levels with data from different subjects.

Supervised and Semisupervised Manifold Embedded Knowledge Transfer in Motor Imagery-Based BCI

A long calibration procedure limits the use in practice for a motor imagery (MI)-based brain-computer interface (BCI) system. To tackle this problem, we consider supervised and semisupervised transfer learning. However, it is a challenge for them to cope with high intersession/subject variability in the MI electroencephalographic (EEG) signals. Based on the framework of unsupervised manifold embedded knowledge transfer (MEKT), we propose a supervised MEKT algorithm (sMEKT) and a semisupervised MEKT algorithm (ssMEKT), respectively. sMEKT only has limited labelled samples from a target subject and abundant labelled samples from multiple source subjects. Compared to sMEKT, ssMEKT adds comparably more unlabelled samples from the target subject. After performing Riemannian alignment (RA) and tangent space mapping (TSM), both sMEKT and ssMEKT execute domain adaptation to shorten the differences among subjects. During domain adaptation, to make use of the available samples, two algorithms preserve the source domain discriminability, and ssMEKT preserves the geometric structure embedded in the labelled and unlabelled target domains. Moreover, to obtain a subject-specific classifier, sMEKT minimizes the joint probability distribution shift between the labelled target and source domains, whereas ssMEKT performs the joint probability distribution shift minimization between the unlabelled target domain and all labelled domains. Experimental results on two publicly available MI datasets demonstrate that our algorithms outperform the six competing algorithms, where the sizes of labelled and unlabelled target domains are variable. Especially for the target subjects with 10 labelled samples and 270/190 unlabelled samples, ssMEKT shows 5.27% and 2.69% increase in average accuracy on the two abovementioned datasets compared to the previous best semisupervised transfer learning algorithm (RA-regularized common spatial patterns-weighted adaptation regularization, RA-RCSP-wAR), respectively. Therefore, our algorithms can effectively reduce the need of labelled samples for the target subject, which is of importance for the MI-based BCI application.

Multi-Source Decentralized Transfer for Privacy-Preserving BCIs

Transfer learning, which utilizes labeled source domains to facilitate the learning in a target model, is effective in alleviating high intra- and inter-subject variations in electroencephalogram (EEG) based brain-computer interfaces (BCIs). Existing transfer learning approaches usually use the source subjects' EEG data directly, leading to privacy concerns. This paper considers a decentralized privacy-preserving transfer learning scenario: there are multiple source subjects, whose data and computations are kept local, and only the parameters or predictions of their pre-trained models can be accessed for privacy-protection; then, how to perform effective cross-subject transfer for a new subject with unlabeled EEG trials? We propose an offline unsupervised multi-source decentralized transfer (MSDT) approach, which first generates a pre-trained model from each source subject, and then performs decentralized transfer using the source model parameters (in gray-box settings) or predictions (in black-box settings). Experiments on two datasets from two BCI paradigms, motor imagery and affective BCI, demonstrated that MSDT outperformed several existing approaches, which do not consider privacy-protection at all. In other words, MSDT achieved both high privacy-protection and better classification performance.

Riemannian distance based channel selection and feature extraction combining discriminative time-frequency bands and Riemannian tangent space for MI-BCIs

OBJECTIVE

Motor imagery-based brain computer interfaces (MI-BCIs) have been widely researched because they do not demand external stimuli and have a high degree of maneuverability. In most scenarios, superabundant selected channels, fixed time windows, and frequency bands would certainly affect the performance of MI-BCIs due to the neurophysiological diversities between different individuals. In this study, we tended to effectively use the Riemannian geometry of spatial covariance matrix to extract more robust features and thus enhance the decoding efficiency.

APPROACH

First, we propose a Riemannian distance-based EEG channel selection method (RDCS), which preliminarily reduces the information redundancy in the first stage. Second, we extract discriminative Riemannian Tangent Space features of EEG signals of selected channels from the most discriminant time-frequency bands (DTFRTS) to further enhance decoding accuracy for MI-BCIs. Finally, we trained a support vector machine (SVM) model with a linear kernel to classify our extracted discriminative Riemannian features and evaluated our proposed method using publicly available BCI Competition Ⅳ dataset Ⅰ (DS1) and Competition Ⅲ dataset Ⅲa (DS2).

MAIN RESULTS

The experimental results showed that the average classification accuracy with the selected 10-channel EEG signals of our method is 88.1% and 91.6% in DS1 and DS2 respectively. The average improvements are 24.3% & 27.1% on DS1 and 4.4% & 14.2% on DS2 for 10 & 20 selected channels, respectively.

SIGNIFICANCE

These results showed that our proposed method is a promising candidate for performance improvement of MI-BCIs.

Control of a Robotic Arm With an Optimized Common Template-Based CCA Method for SSVEP-Based BCI

Recently, the robotic arm control system based on a brain-computer interface (BCI) has been employed to help the disabilities to improve their interaction abilities without body movement. However, it's the main challenge to implement the desired task by a robotic arm in a three-dimensional (3D) space because of the instability of electroencephalogram (EEG) signals and the interference by the spontaneous EEG activities. Moreover, the free motion control of a manipulator in 3D space is a complicated operation that requires more output commands and higher accuracy for brain activity recognition. Based on the above, a steady-state visual evoked potential (SSVEP)-based synchronous BCI system with six stimulus targets was designed to realize the motion control function of the seven degrees of freedom (7-DOF) robotic arm. Meanwhile, a novel template-based method, which builds the optimized common templates (OCTs) from various subjects and learns spatial filters from the common templates and the multichannel EEG signal, was applied to enhance the SSVEP recognition accuracy, called OCT-based canonical correlation analysis (OCT-CCA). The comparison results of offline experimental based on a public benchmark dataset indicated that the proposed OCT-CCA method achieved significant improvement of detection accuracy in contrast to CCA and individual template-based CCA (IT-CCA), especially using a short data length. In the end, online experiments with five healthy subjects were implemented for achieving the manipulator real-time control system. The results showed that all five subjects can accomplish the tasks of controlling the manipulator to reach the designated position in the 3D space independently.