Exploration of Pattern Recognition Methods for Motor Imagery EEG Signal with Convolutional Neural Network Approach

Abstract

As an application of EEG, Motor Imagery based Brain-Computer Interface (MI BCI) plays a significant role in assisting patients with disability to communicate with their environment. MI BCI could now be realized through various methods such as machine learning. Many attempts using different machine learning approaches as MI BCI applications have been done with every one of them yielding various results. While some attempts managed to achieve agreeable results, some still failed. This failure may be caused by the separation of the feature extraction and classification steps, as this may lead to the loss of information which in turn causes lower classification accuracy. This problem can be solved by integrating feature extraction and classification by harnessing a classification algorithm that processed the input data as a whole until it produces the prediction, hence the use of convolutional neural network (CNN) approach which is known for its versatility in processing and classifying data all in one go. In this study, the CNN exploration involved a task to classify 5 different classes of fingers’ imaginary movement (thumb, index, middle, ring, and pinky) based on the processed raw signal provided. The CNN performance was observed for both non-augmented and augmented data with the data augmentation techniques used include sliding window, noise addition, and the combination of those two methods. From these experiments, the results show that the CNN model managed to achieve an averaged accuracy of 47%, meanwhile with the help of augmentation techniques of sliding window, noise addition, and the combined methods, the model achieved even higher averaged accuracy of 57,1%, 47,2%, and 57,5% respectively.

EEG-based finger movement classification with intrinsic time-scale decomposition

“…Despite the fact that the hidden layers in deep learning structures create a significant amount of workload and necessitate a significant amount of time for training, the reported classification performance in all subjects was not as high as expected (over 90.00%). In another study (Zahra et al, 2022 ), another deep learning-based classification with very high training time was adopted and considering the same drawbacks of the previous study (Anam et al, 2020 ), although a significant improvement in performance was achieved since the sample size of this study (i.e., only four subjects) and number of EEG channels (i.e., only four channels) were limited when compared with the sample size and number of EEG channels in our study. Thus, a comparison between the results of this study (Anam et al, 2020 ) and ours would not be meaningful.…”

“…Table 10 presents a comparison of the suggested study to relevant prior studies. Clearly, both subject-dependent (Kaya et al, 2018 ; Anam et al, 2019 , 2020 ; Kato et al, 2020 ; Mwata-Velu et al, 2021 , 2022 ; Azizah et al, 2022 ) and subject-independent (Kaya et al, 2018 ; Zahra et al, 2022 ) studies were conducted for FM classification in literature. In general, the highest performance values were achieved in subject-dependent classification as in our study.…”

“…Therefore, we preferred the k-fold CV method as in our similar studies (Isler, 2009;Isler and Kuntalp, 2009;Degirmenci et al, 2022a,b) and the recent literature (Anam et al, 2019(Anam et al, , 2020Kato et al, 2020;Mwata-Velu et al, 2021Azizah et al, 2022;Zahra et al, 2022). Using k-fold cross-validation (CV), the training data set was divided into k equal-sized subsets.…”

“…Different time-domain, frequency-domain, and spatial-domain EEG features have been calculated to predict FM in the past decade. The raw EEG time series (Kaya et al, 2018 ; Mwata-Velu et al, 2022 ; Zahra et al, 2022 ), different amplitude-based, and statistical-based EEG signal features (Degirmenci et al, 2024 ) were utilized to examine the effectiveness of the time domain. As for the spectral-domain features, different frequency-domain [Fourier transform (Kaya et al, 2018 )] and time-frequency domain [Wavelet transform (Yahya et al, 2019 ), Short-time Fourier transform (Azizah et al, 2022 ), Empirical mode decomposition (Mwata-Velu et al, 2021 )] representation algorithms and their various versions were investigated to classify FM.…”

“…In another related paper Zahra et al ( 2022 ), which was published recently in 2022, the performance of a Convolutional Neural Networks (CNN) was evaluated based on an original study design. In their model, EEG time series were combined with sliding window (Dietterich, 2002 ) and noise enhancement (Mitaim and Kosko, 1998 ) methods to extract the features.…”