Akademik Veri Yönetim Sistemi
NEURAL COMPUTING & APPLICATIONS, cilt.35, sa.32, ss.23617-23630, 2023 (SCI-Expanded)
Psychiatric diseases are very common worldwide, including schizophrenia (SZ) and Alzheimer's disease (AD). Early diagnosis is very important for medical intervention. Proper diagnosis of these diseases requires long-term observations and extensive testing to recognize their clinical features. These observations and tests may be subjective, expensive, incomplete, or inaccurate. EEG is a strong candidate for the diagnosis of diseases with the advantages of being noninvasive, based on findings, less costly and getting results in a short time. In this paper, we proposed an EEG-based solution for the diagnosis of psychiatric diseases using a fast Walsh-Hadamard transform (FWHT) and deep learning approach. Two different publicly available datasets were combined. SZ, AD and healthy controls (HC) were automatically multi-classified in experiment 1 and SZ and AD groups were classified in experiment 2. Features were extracted using the FWHT. The performances of support vector machine, decision tree, k-nearest neighbour, and bidirectional long-short-term memory (BLSTM) algorithms were compared for each experiment using the extracted features. In the experiments, the BLSTM algorithm displayed the best performance. The BLSTM algorithm demonstrated promising results with 91.05% overall accuracy, 0.86 kappa statistics in experiment 1, and with 98.38% overall accuracy, 0.96 kappa statistics in experiment 2. The results of the experiment provided evidence that people with different psychiatric diseases have different EEG signals. Moreover, it is a rare attempt that SZ, AD and HC groups can be classified automatically and effectively from EEG signals using deep learning model and accuracy was higher than related literature studies.