Improving Generalization Based on l1-Norm Regularization for EEG-Based Motor Imagery Classification

An Explainable and Generalizable Recurrent Neural Network Approach for Differentiating Human Brain States on EEG Dataset

Electroencephalogram (EEG) is one of the most widely used brain computer interface (BCI) approaches. Despite the success of existing EEG approaches in brain state recognition studies, it is still challenging to differentiate brain states via explainable and generalizable deep learning approaches. In other words, how to explore meaningful and distinguishing features and how to overcome the huge variability and overfitting problem still need to be further studied. To alleviate these challenges, in this work, a multiple random fragment search-based multilayer recurrent neural network (MRFS-MRNN) is proposed to improve the differentiating performance and explore meaningful patterns. Specifically, an explainable MRNN module is proposed to capture the temporal dependences preserved in EEG time series. Besides, a MRFS module is designed to cut multiple random fragments from the entire EEG signal time course to improve the effectiveness of brain state differentiating ability. MRFS-MRNN is concatenatedto effectively overcome the huge variabilities and overfitting problems. Experiment results demonstrate that the proposed MRFS-MRNN model not only has excellent differentiating performance, but also has good explanation and generalization ability. The classification accuracies reach as high as 95.18% for binary classification and 89.19% for four-category classification on the individual level. Similarly, 95.53% and 85.84% classification accuracies are obtained for the binary and four-category classification on the group level. What's more, 94.28% and 85.43% classification accuracies of binary and four-category classifications are achieved for predicting brand new subjects. The experiment results showed that the proposed method outperformed other state-of-the-art (SOTA) models on the same underlying data and improved the explanation and generalization ability.

Sparse representation-based classification with two-dimensional dictionary optimization for motor imagery EEG pattern recognition

BACKGROUND

Sparse representation-based classification (SRC) has more advantages in motor imagery EEG pattern recognition, and the quality of dictionary construction directly determines the performance of SRC. In this paper, we proposed a two-dimensional dictionary optimization (TDDO) method to directly improve the performance of SRC.

NEW METHOD

Firstly, an initial dictionary was constructed with multi-band features extracted by filter band common spatial pattern (FBCSP). Then Lasso regression is used to select significant features in each atom synchronously in the horizontal direction, and the KNN-based method is used to clean up noise atoms in the vertical direction. Finally, an SRC method by training samples linearly representing test samples was implemented in classification.

RESULTS

The results show the necessity and rationality of TDDO-SRC method. The highest average classification accuracy of 86.5% and 92.4% is obtained on two public datasets.

COMPARISON WITH EXISTING METHOD(S)

The proposed method has more superior classification accuracy compared to traditional methods and existing winners' methods.

CONCLUSIONS

The quality of dictionary construction has a great impact on the robustness of SRC. And compared with the original SRC, the classification accuracy of the optimized TDDO-SRC is greatly improved.

Multi-objective optimization approach for channel selection and cross-subject generalization in RSVP-based BCIs

Objective.Achieving high precision rapid serial visual presentation (RSVP) task often requires many electrode channels to obtain more information. However, the more channels may contain more redundant information and also lead to its limited practical applications. Therefore, it is necessary to reduce the number of channels to enhance the classification performance and users experience. Furthermore, cross-subject generalization has always been one of major challenges in electroencephalography channel reduction, especially in the RSVP paradigm. Most search-based channel selection method presented in the literature are single-objective methods, the classification accuracy (ACC) is usually chosen as the only criterion.Approach.In this article, the idea of multi-objective optimization was introduced into the RSVP channel selection to minimize two objectives: classification error and the number of channels. By combining a multi-objective evolutionary algorithm for solving large-scale sparse problems and hierarchical discriminant component analysis (HDCA), a novel channel selection method for RSVP was proposed. After that, the cross-subject generalization validation through the proposed channel selection method.Main results.The proposed method achieved an average ACC of 95.41% in a public dataset, which is 3.49% higher than HDCA. The ACC was increased by 2.73% and 2.52%, respectively. Besides, the cross-subject generalization models in channel selection, namely special-16 and special-32, on untrained subjects show that the classification performance is better than the Hoffmann empirical channels.Significance.The proposed channel selection method could reduce the calibration time in the experimental preparation phase and obtain a better accuracy, which is promising application in the RSVP scenario that requires low-density electrodes.

Representation Learning for Motor Imagery Recognition with Deep Neural Network

This study describes a method for classifying electrocorticograms (ECoGs) based on motor imagery (MI) on the brain–computer interface (BCI) system. This method is different from the traditional feature extraction and classification method. In this paper, the proposed method employs the deep learning algorithm for extracting features and the traditional algorithm for classification. Specifically, we mainly use the convolution neural network (CNN) to extract the features from the training data and then classify those features by combing with the gradient boosting (GB) algorithm. The comprehensive study with CNN and GB algorithms will profoundly help us to obtain more feature information from brain activities, enabling us to obtain the classification results from human body actions. The performance of the proposed framework has been evaluated on the dataset I of BCI Competition III. Furthermore, the combination of deep learning and traditional algorithms provides some ideas for future research with the BCI systems.

Regularization of Deep Neural Networks for EEG Seizure Detection to Mitigate Overfitting

Seizure detection is a major goal for simplifying the workflow of clinicians working on EEG records. Current algorithms can only detect seizures effectively for patients already presented to the classifier. These algorithms are hard to generalize outside the initial training set without proper regularization and fail to capture seizures from the larger population. We proposed a data processing pipeline for seizure detection on an intra-patient dataset from the world's largest public EEG seizure corpus. We created spatially and session invariant features by forcing our networks to rely less on exact combinations of channels and signal amplitudes, but instead to learn dependencies towards seizure detection. For comparison, the baseline results without any additional regularization on a deep learning model achieved an F1 score of 0.544. By using random rearrangements of channels on each minibatch to force the network to generalize to other combinations of channels, we increased the F1 score to 0.629. By using random rescale of the data within a small range, we further increased the F1 score to 0.651 for our best model. Additionally, we applied adversarial multi-task learning and achieved similar results. We observed that session and patient specific dependencies were causing overfitting of deep neural networks, and the most overfitting models learnt features specific only to the EEG data presented. Thus, we created networks with regularization that the deep learning did not learn patient and session-specific features. We are the first to use random rearrangement, random rescale, and adversarial multitask learning to regularize intra-patient seizure detection and have increased sensitivity to 0.86 comparing to baseline study.

EEG Feature Extraction Based on a Bilevel Network: Minimum Spanning Tree and Regional Network

Feature extraction is essential for classifying different motor imagery (MI) tasks in a brain–computer interface (BCI). Although the methods of brain network analysis have been widely studied in the BCI field, these methods are limited by differences in network size, density, and standardization. To address this issue and improve classification accuracy, we propose a novel method, in which the hybrid features of the brain function based on the bilevel network are extracted. Minimum spanning tree (MST) based on electroencephalogram (EEG) signal nodes in different MIs is constructed as the first network layer to solve the global network connectivity problem. In addition, the regional network in different movement patterns is constructed as the second network layer to determine the network characteristics, which is consistent with the correspondence between limb movement patterns and cerebral cortex in neurophysiology. We attempt to apply MST to the classification of the MI EEG signals, and the bilevel network has better interpretability. Thereafter, a vector is formed by combining the MST fundamental features with the directional features of the regional network. Our method is validated using the BCI Competition IV Dataset I. Experimental results verify the feasibility of the bilevel network framework. Furthermore, the average classification performance of the proposed method reaches 89.50%, which is higher than that of other competing methods, thereby indicating that the bilevel network is effective for MI classification.