A Deep Learning MI - EEG Classification Model for BCIs

Abstract

Reliable signal classification is essential for using an electroencephalogram (EEG) based Brain-Computer Interface (BCI) in motor imagery (MI) training. While deep learning (DL) is used in many areas with great success, only a limited number of works investigate its potential in this domain. This study presents a DL approach, which could improve or replace current stateof-the-art methods. Here, an end-to-end convolutional neural network (CNN) model is presented, which can be applied to raw EEG signals. It consists of a temporal and spatial convolution layer for feature extraction and a fully connected (FC) layer for classification. The global models were trained on 3s segments of EEG data. Training a subject-independent global classifier reaches 80.10%, 69.72%, and 59.71% mean accuracy for a dataset with two, three, and four classes, respectively, validated in 5-fold crossvalidation. Retraining the global classifier with data from single individuals improves the overall mean accuracy to 86.13%, 79.05%, and 68.93%, respectively. The results are superior to the results reported in the literature on the same data. Generally, the reported accuracy values are comparable with related studies, which shows that the model delivers competitive results. As raw signals are used as input, no pre-processing is needed, which qualifies DL methods as a promising alternative to established EEG classification methods.

Major Depressive Disorder Classification Based on Different Convolutional Neural Network Models: Deep Learning Approach

“…EEG, which is one of these devices, is more preferred than other tools because it is noninvasive, economical, practical, and easy to operate. The accuracy, robustness, and reliability of the EEGrelated methodology combined with deep learning (DL) has proven with many research in brain-computer interface (BCI), especially in motor imagery task classification, [5][6][7][8][9][10][11][12][13] epileptic seizure prediction and detection, [14][15][16][17][18][19][20][21][22][23] drivers fatigue prediction, 24,25 emotion and affective state classification, [26][27][28][29][30][31][32] sleep stage detection, [33][34][35][36] prognosis in rapid eye movement behavior disorder, 37 EEG-based diagnosis of various neurodegenerative diseases, including attention deficit/hyperactivity disorder, 38 schizophrenia, 39,40 Creutzfeldt-Jacob disease, 41 Parkinson's disease, 42 Alzheimer's disease, 43 mild cognitive impairment, 44 predicting transcranial direct current stimulation treatment outcomes of patients with MDD has been studied in the recent literature. 45 The performance of a machine learning methodology based on the pretreatment EEG for MDD pro...…”

Multiclass Support Matrix Machines by Maximizing the Inter-Class Margin for Single Trial EEG Classification

“…Vector based methods such as linear discriminant analysis (LDA) [8], [9], [10], support vector machines (SVM) [11], [12], [13], K nearest neighbor (KNN) [14], [15] and neural network [16], [17], [18], [16], [19] have been successively applied for EEG classification. In most cases, the data has to be reshaped into vectors for further classification which could in-turn destroy the structural information embedded in it.…”

Compressing Subject-specific Brain-Computer Interface Models into One Model by Superposition in Hyperdimensional Space

“…MI-BCIs strive to decode the cognitive process of thinking of a motion, e.g., the left or the right hand movement, without actually performing it. Recently, deep learning algorithms strive to decode EEG signals to deliver compact yet accurate MI-BCI models [1], [2], [6], [7].…”

“…To further improve classification accuracy and deal with the high variability of EEG signals between subjects, most approaches [2], [5], [6], [8], [9] train a personalized model per subject. This subject-specific model training requires a calibration session for each subject, which can be exhausting for clinical users with impaired cognitive abilities, or tedious for healthy users [10].…”

“…However, the global model often comes with a significant degradation in classification accuracy. For instance, an accuracy loss of 9.22% is observed when using a global model instead of subject-specific models [6].…”