Subject-independent EEG classification of motor imagery based on a dual-branch feature fusion

Abstract

Background: Brain-computer interface (BCI) system helps people with motor dysfunction interact with external environment. With the advancement of technology, the BCI system has been applied in practice, but its practicability and usability are still greatly challenged. A large amount of calibration time is often required before BCI systems are used, which can consume the patient’s energy and patience and can easily lead to anxiety. We propose a subject-independentzero calibration approach to solve this problem. Methods: A dual-branch multiscale autoencoder network(MSAENet) is proposed realize the subject-independent classification in motor imagery, aiming to realize the plug-and-play of BCI. Firstly, the network consists of a multiscale branch and an autoencoder (AE) for feature learning from different perspectives. Secondly, the covariance between the EEG signal and the common spatial pattern in the 8-30 Hz band was used as spatio-spectral features, and the feature pre-extracted information is used as the input of MSAENet. Finally, The network introduces a central loss function to improve the classification ability. Testing network generalizability on three publicly available datasets BCIIV2a, SMR-BCI, and OpenBMI. Results: The results show that our proposed network shows good results on all three datasets. In the case of subject-independence, MSAENet outperformed the other four comparison methods on the BCIIV2a and SMR-BCI datasets,while achieving F1 score values as high as 69. 34% on the OpenBMI dataset.Subject-dependent results with the best classification performance were significantly better than the other four advanced comparison methods. Our method can maintain a better classification accuracy while ensuring a small amount of parameters and a short prediction time. Conclusions: MSAENet verifies the following three points: (1)Spatial frequency domain features can extract effective information from raw EEG signals. (2)Two-branch multiscale feature fusion can extract features more comprehensively. (3)The integration of the central loss function compensates for the fact that Softmax classifiers only consider class spacing and ignore intra-class distance. Our proposed method achieves zero calibration, and the problem that requires a lot of calibration time in BCI applications is effectively solved.Keywords: BCI; Motor imagery; Subject-independent; Feature fusion; Zero Calibration

A novel ANN adaptive Riemannian-based kernel classification for motor imagery

“…MSAENet [35]: The authors suggest a dual-branch multiscale autoencoder network (MSAENet), which starts by building a network with a multiscale branch and an autoencoder (AE) for feature learning from various perspectives. Second, the covariance between the EEG signal and the common spatial pattern was employed as spatio-spectral features, and MSAENet's input is the feature-pre-extracted data.…”

Exploring the Potential of Attention Mechanism-Based Deep Learning for Robust Subject-Independent Motor-Imagery Based BCIs

“…Although these models can be quite accurate, they require ongoing calibration for each new user, leading to resource-intensive implementations [15], [16]. The second approach focuses on subject-independent models, aiming to identify features that are consistent across different individuals to avoid the need for calibration [17], [18]. While this approach is less resource-intensive, it often results in compromised performance [19].…”