Machine Learning-Based Brain Disease Classification Using EEG and MEG Signals

Abstract

Electroencephalography (EEG) and Magneto-encephalography (MEG) are important tools for assessing brain activity that are being developed scientifically. EEG enables better clinical and healthcare services to meet the rising need for the early diagnosis of brain disease at cheap prices. In this paper, EEG and MEG signals are used as an input to detect brain cancer and strokes in its early stages. A single-trial channel data (STD), averaged channel data (ACT), and time–frequency data organisation of the channel data is required for the EEG/MEG signal projection from the channel to the source space (TFD). The signals are pre-processed using discrete wavelet transform to adaptive time–frequency resolution of analysis on nonstationary signals. Then, the signals are given as an input to the Fast Fourier Transform to get subsets of typical in-class “invariant” coefficients from wavelet coefficients (time–frequency information). Finally, the multi-class SVM is employed for classifying normal, cancer, and stroke cases using EEG and MEG signals. A quantitative analysis of the proposed method is conducted using parameters like accuracy, specificity, and precision. The proposed Dual signal classification model achieved higher accuracy is 92.59%. According to the proposed Dual signal classification model, Bootstrap models, SVM, XGBoost, and Fast Fourier Transform improve the overall accuracy by 4.61%, 0.16%, 15.77%, and 9.63%, respectively.