An Extended Computer Aided Diagnosis System for Robust BCI Applications

Electroencephalogram (EEG) signal processing is the pivotal procedure to decipher meaningful information from the lowkey signals to drive practical applications. This paper investigates an EEG signal processing model, which utilizes three EEG electrode combinations, six feature extraction methods, and seven classification algorithms together with an improved empirical Fourier decomposition (IEFD) for motor imagery (MI) EEG signal analysis. The feasibility of IEFD is further validated on a large GigaDB dataset with 52 participants along with the BCI competition III datasets IVa and IVb. Results reveal that IEFD mechanism yields robust classification outcomes when coupled with 18 electrodes combination, welch PSD features, and multilayer perceptron classifier, and the best classification accuracy of 99.52%, 99.35%, 98.89%, 99.52%, 100%, and 93.19% is achieved for dataset IVa and IVb subjects, respectively. Moreover, the GigaDB dataset yields an average classification accuracy, sensitivity, specificity, and fl-score of 83.84%, 83.71%, 83.98%, and 83.80% accordingly. Results compared with previous studies conclude that the proposed model improves the average classification accuracy by 16.6%. Such promising findings conclude that the proposed IEFD method is robust and adaptive for MI EEG signals classification, independent of subject-to-subject variance for multiple datasets.