tsta

فارسی

Home / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

EEG-Based Classification of Schizophrenia and Healthy Controls Subjects Using Statistical and Nonlinear Features with Emphasis on Fuzzy Entropy

Authors :

Mahdiyeh Tofighi Milani¹ Sina Shamekhi² Asghar Zarei³

1- دانشگاه صنعتی تبریز(سهند) 2- دانشگاه صنعتی تبریز(سهند) 3- دانشگاه صنعتی تبریز(سهند)

Keywords :

Schizophrenia،Electroencephalogram،Machine Learning،Fuzzy Entropy

Abstract :

Schizophrenia is a severe mental disorder that frequently causes the patient to have numerous problems with normal daily activities, and still, doctors struggle to accurately diagnose it in the early stages. Brain imaging and clinical tests, even if they are sometimes capable of achieving the goal, are often a lengthy procedure, expensive, and can also be somewhat uncomfortable for patients. New scientific work seeks to come up with a less intrusive and cheaper method, which will include the use of the EEG signal and the ML algorithm in identifying abnormalities of the schizophrenic patients as compared with the healthy ones. At first, the Fast Fourier Transform (FFT) was used to decompose the EEG signal into multiple sub-bands of frequency, and it was decided to extract a set of features from each sub-band, where the features included the statistical and nonlinear features - kurtosis, skewness, Shannon entropy, fuzzy entropy, mobility, and complexity. Subsequently, the ReliefF algorithm was utilized for the selection of features, and the significant features thus extracted were used as input for a number of classifiers, including the k-nearest neighbors (KNN), linear support vector machine (SVM), and the random forest (RF), to name but a few. The functional capabilities of the designed system were verified on a genuine EEG dataset that contains recorded signals from teenage schizophrenia patients as well as from healthy subjects. Random forest was identified as the most effective one among the various implemented classifiers, as it achieved the highest performance with an average accuracy of 97.69%. Also, fuzzy entropy was identified to be a constantly discriminative feature, implying it could serve as a sound biomarker for the differentiation of schizophrenia from healthy subjects by utilizing EEG signals.