L1-norm based time-varying brain neural network and its application to dynamic analysis for motor imagery

Multi-source deep domain adaptation ensemble framework for cross-dataset motor imagery EEG transfer learning

Objective. Electroencephalography (EEG) is an important kind of bioelectric signal for measuring physiological activities of the brain, and motor imagery (MI) EEG has significant clinical application prospects. Convolutional neural network has become a mainstream algorithm for MI EEG classification, however lack of subject-specific data considerably restricts its decoding accuracy and generalization performance. To address this challenge, a novel transfer learning (TL) framework using auxiliary dataset to improve the MI EEG classification performance of target subject is proposed in this paper.Approach. We developed a multi-source deep domain adaptation ensemble framework (MSDDAEF) for cross-dataset MI EEG decoding. The proposed MSDDAEF comprises three main components: model pre-training, deep domain adaptation, and multi-source ensemble. Moreover, for each component, different designs were examined to verify the robustness of MSDDAEF.Main results. Bidirectional validation experiments were performed on two large public MI EEG datasets (openBMI and GIST). The highest average classification accuracy of MSDDAEF reaches 74.28% when openBMI serves as target dataset and GIST serves as source dataset. While the highest average classification accuracy of MSDDAEF is 69.85% when GIST serves as target dataset and openBMI serves as source dataset. In addition, the classification performance of MSDDAEF surpasses several well-established studies and state-of-the-art algorithms.Significance. The results of this study show that cross-dataset TL is feasible for left/right-hand MI EEG decoding, and further indicate that MSDDAEF is a promising solution for addressing MI EEG cross-dataset variability.

EEG based emotion recognition by hierarchical bayesian spectral regression framework

Spectral regression (SR), a graph-based learning regression model, can be used to extract features from graphs to realize efficient dimensionality reduction. However, due to the SR method remains a regularized least squares problem and being defined in L2-norm space, the effect of artifacts in EEG signals cannot be efficiently resisted. In this work, to further improve the robustness of the graph-based regression models, we propose to utilize the prior distribution estimation in the Bayesian framework and develop a robust hierarchical Bayesian spectral regression framework (named HB-SR), which is designed with the hierarchical Bayesian ensemble strategies. In the proposed HB-SR, the impact of noises can be effectively reduced by the adaptive adjustment approach in model parameters with the data-driven manner. Specifically, in the current work, three different distributions have been elaborately designed to enhance the universality of the proposed HB-SR, i.e., Gaussian distribution, Laplace distribution, and Student-t distribution. To objectively evaluate the performance of the HB-SR framework, we conducted both simulation studies and emotion recognition experiments based on emotional EEG signals. Experimental results have consistently indicated that compared with other existing spectral regression methods, the proposed HB-SR can effectively suppress the influence of noises and achieve robust EEG emotion recognition.

An Adaptive Joint CCA-ICA Method for Ocular Artifact Removal and its Application to Emotion Classification

BACKGROUND

The quality of Electroencephalogram (EEG) signals is critical for revealing the neural mechanism of emotions. However, ocular artifacts decreased the signal to noise ratio (SNR) and covered the inherent cognitive component of EEGs, which pose a great challenge in neuroscience research.

NEW METHOD

We proposed a novel unsupervised learning algorithm to adaptively remove the ocular artifacts by combining canonical correlation analysis (CCA), independent component analysis (ICA), higher-order statistics, empirical mode decomposition (EMD), and wavelet denoising techniques. Specifically, the combination of CCA and ICA aimed to improve the quality of source separation, while the higher-order statistics further located the source of ocular artifacts. Subsequently, these noised sources were further corrected by EMD and wavelet denoising to improve SNR of EEG signals.

RESULTS

We evaluated the performance of our proposed method with simulation studies and real EEG applications. The results of simulation study showed our proposed method could significantly improve the quality of signals under almost all noise conditions compared to four state-of-art methods. Consistently, the experiments of real EEG applications showed that the proposed methods could efficiently restrict the components of ocular artifacts and preserve the inherent information of cognition processing to improve the reliability of related analysis such as power spectral density (PSD) and emotion recognition.

COMPARISON WITH EXISTING METHODS

Our proposed model outperforms the comparative methods in EEG recovery, which further improve the application performance such as PSD analysis and emotion recognition.

CONCLUSIONS

The superior performance of our proposed method suggests that it is promising for removing ocular artifacts from EEG signals, which offers an efficient EEG preprocessing technology for the development of brain computer interface such as emotion recognition.