Classification of Motor Imagery Tasks Using EEG Signal Analysis and Machine Learning Methods

Abstract:

Brain-Computer Interfaces (BCIs) play a crucial role in establishing a connection between the brain and external systems and can be recognized as a key tool in neurorehabilitation. These technologies are particularly significant for patients with amyotrophic lateral sclerosis (ALS), spinal cord injuries, strokes, or other neuro-motor disorders. Such patients are unable to perform voluntary movements due to disrupted communication pathways between the brain and limbs. BCIs can improve their motor abilities and enhance their quality of life by creating alternative pathways. Electroencephalogram (EEG) signals have gained popularity in the development of such systems due to their ease of acquisition and low cost. One of the major challenges in EEG signal analysis is the cognitive and individual differences among participants, which can affect the quality of motor imagery (MI) tasks and often reduce the accuracy and efficiency of BCI systems. This research aims to address this challenge and improve MI task classification outcomes. To this end, the CHO-2017 database, comprising 52 participants, was selected. The large number of participants makes this dataset particularly suitable for investigating individual cognitive differences. In this study, a hybrid approach combining Common Spatial-Spectral Pattern (CSSP) filters and the Tunable-Q Wavelet Transform (TQWT) was employed. Using only the top 10 features, this approach enhanced the classification results of MI tasks and extracted highly discriminative features between the two classes of left and right-hand imagery tasks. This method was effective for 99% of participants, providing a solution to the aforementioned challenge. Using the extracted features, the K-Nearest Neighbors (KNN) classifier achieved an accuracy of 98.84%, significantly improving upon recent studies. The findings of this research could pave the way for the development of more accurate BCI systems capable of extracting optimal commands for MI tasks.