Toward calibration-free motor imagery brain-computer interfaces: a VGG-based convolutional neural network and WGAN approach.

OBJECTIVE: Motor Imagery (MI) represents one major paradigm of Brain-Computer Interfaces (BCIs) in which users rely on their Electroencephalogram (EEG) signals to control the movement of objects. However, due to the inter-subject variability, MI BCIs require recording subject-dependent data to train machine learning classifiers that are used to identify the intended motor action. This represents a challenge in developing MI BCIs as it complicates its calibration and hinders the wide adoption of such a technology. APPROACH: This study focuses on enhancing cross-subject MI EEG classification using EEG spectrum images. The proposed calibration-free approach employs deep learning techniques for MI classification and Wasserstein Generative Adversarial Networks (WGAN) for data augmentation. The proposed WGAN generates synthetic spectrum images from the recorded MI-EEG to expand the training dataset; aiming to enhance the classifier's performance. The proposed approach eliminates the need for any calibration data from the target subject, making it more suitable for real-world applications. MAIN RESULTS: To assess the robustness and efficacy of the proposed framework, we utilized the BCI competition IV-2B, IV-2A, and IV-1 benchmark datasets, employing leave one-subject out validation. Our results demonstrate that using the proposed modified VGG-CNN classifier in addition to WGAN-generated data for augmentation leads to an enhancement in cross-subject accuracy outperforming state-of-the-art methods. SIGNIFICANCE: This approach could represent one step forward towards developing calibration-free BCI systems and hence broaden their applications.

Generative adversarial networks in EEG analysis: an overview.

Electroencephalogram (EEG) signals have been utilized in a variety of medical as well as engineering applications. However, one of the challenges associated with recording EEG data is the difficulty of recording large amounts of data. Consequently, data augmentation is a potential solution to overcome this challenge in which the objective is to increase the amount of data. Inspired by the success of Generative Adversarial Networks (GANs) in image processing applications, generating artificial EEG data from the limited recorded data using GANs has seen recent success. This article provides an overview of various techniques and approaches of GANs for augmenting EEG signals. We focus on the utility of GANs in different applications including Brain-Computer Interface (BCI) paradigms such as motor imagery and P300-based systems, in addition to emotion recognition, epileptic seizures detection and prediction, and various other applications. We address in this article how GANs have been used in each study, the impact of using GANs on the model performance, the limitations of each algorithm, and future possibilities for developing new algorithms. We emphasize the utility of GANs in augmenting the limited EEG data typically available in the studied applications.

Dynamic time warping-based transfer learning for improving common spatial patterns in brain-computer interface.

OBJECTIVE: Common spatial patterns (CSP) is a prominent feature extraction algorithm in motor imagery (MI)-based brain-computer interfaces (BCIs). However, CSP is computed using sample-based covariance-matrix estimation. Hence, its performance deteriorates if the number of training trials is small. To address this problem, this paper proposes a novel regularized covariance matrix estimation framework for CSP (i.e. DTW-RCSP) based on dynamic time warping (DTW) and transfer learning. APPROACH: The proposed framework combines the subject-specific covariance matrix ([Formula: see text]) estimated using the available trials from the new subject, with a novel DTW-based transferred covariance matrix ([Formula: see text]) estimated using previous subjects' trials. In the proposed [Formula: see text], the available labelled trials from the previous subjects are temporally aligned to the average of the available trials of the new subject from the same class using DTW. This alignment aims to reduce temporal variations and non-stationarities between previous subjects' trials and the available trials from the new subjects. Moreover, to tackle the problem of regularization parameter selection when only a few trials are available for training, an online method is proposed, where the best regularization parameter is selected based on the confidence scores of the trained classifier on the upcoming first few labelled testing trials. MAIN RESULTS: The proposed framework is evaluated on two datasets against two baseline algorithms. The obtained results reveal that DTW-RCSP significantly outperformed the baseline algorithms at various testing scenarios, particularly, when only a few trials are available for training. SIGNIFICANCE: Impressively, our results show that successful BCI interactions could be achieved with a calibration session as small as only one trial per class.

Weighted Transfer Learning for Improving Motor Imagery-Based Brain-Computer Interface.

One of the major limitations of motor imagery (MI)-based brain-computer interface (BCI) is its long calibration time. Due to between sessions/subjects variations in the properties of brain signals, typically, a large amount of training data needs to be collected at the beginning of each session to calibrate the parameters of the BCI system for the target user. In this paper, we propose a novel transfer learning approach on the classification domain to reduce the calibration time without sacrificing the classification accuracy of MI-BCI. Thus, when only few subject-specific trials are available for training, the estimation of the classification parameters is improved by incorporating previously recorded data from other users. For this purpose, a regularization parameter is added to the objective function of the classifier to make the classification parameters as close as possible to the classification parameters of the previous users who have feature spaces similar to that of the target subject. In this paper, a new similarity measure based on the Kullback-Leibler divergence (KL) is used to measure the similarity between two feature spaces obtained using subject-specific common spatial patterns (CSP). The proposed transfer learning approach is applied on the logistic regression classifier and evaluated using three datasets. The results showed that compared with the subject-specific classifier, the proposed weighted transfer learning classifier improved the classification results, particularly when few subject-specific trials were available for training (p < 0.05). Importantly, this improvement was more pronounced for users with medium and poor accuracy. Moreover, the statistical results showed that the proposed weighted transfer learning classifier performed significantly better than the considered comparable baseline algorithms.

Estimation of joint angle based on surface electromyogram signals recorded at different load levels.

To control upper-limb exoskeletons and prostheses, surface electromyogram (sEMG) is widely used for estimation of joint angles. However, the variations in the load carried by the user can substantially change the recorded sEMG and consequently degrade the accuracy of joint angle estimation. In this paper, we aim to deal with this problem by training classification models using a pool of sEMG data recorded from all different loads. The classification models are trained as either subject-specific or subject-independent, and their results are compared with the performance of classification models that have information about the carried load. To evaluate the proposed system, the sEMG signals are recorded during elbow flexion and extension from three participants at four different loads (i.e. 1, 2, 4 and 6 Kg) and six different angles (i.e. 0, 30, 60, 90, 120, 150 degrees). The results show while the loads were assumed unknown and the applied training data was relatively small, the proposed joint angle estimation model performed significantly above the chance level in both the subject-specific and subject-independent models. However, transferring from known to unknown load in the subject-specific classifiers leads to 20% to 32% loss in the average accuracy.