FFT-based CNN Classification for Schizophrenia Detection in EEG Recordings

Machine learning enhances computer-aided medical diagnosis by enabling accurate and swift decision-making. This study proposes a method for detecting schizophrenia (SZ) using electroencephalography, which measures brain electrical activity to diagnose neurological disorders. Schizophrenia is characterized by complex neural patterns, challenging to identify with traditional methods. This research employs deep learning algorithms to analyze EEG signals for schizophrenia detection, aiming to improve classification accuracy. The methodology involves preprocessing Electroencephalography (EEG) time series to extract spectral power features using Fast Fourier Transformation (FFT), which transforms time-domain signals into the frequency domain, revealing brain oscillatory activity. These features are converted into RGB images representing brain activity's spatial information. A convolutional neural network (CNN) is then used to classify these images. The proposed method achieved an average accuracy of 95.97% with FFT, indicating that FFT-based features are highly effective for classification in this context. The results underscore the importance of data representation when using CNN models for EEG signal analysis.