Abstract
Schizophrenia is a severe psychiatric disorder that affects thought processes, perception, and emotional responsiveness, often resulting in significant cognitive impairment and social dysfunction. Early detection is crucial, as timely intervention can greatly improve patient outcomes and quality of life. Traditional diagnostic methods for schizophrenia rely on clinical observation and self-reported symptoms, which can be subjective and may delay diagnosis. Electroencephalography (EEG) has emerged as a promising non-invasive tool for objective detection, offering insight into neural dynamics associated with psychiatric disorders.In this project, we propose a deep learning-based framework, SCZ-NET, that leverages a hybrid CNN-Transformer architecture to automatically detect schizophrenia from EEG signals. The EEG dataset used, contains recordings from 16 electrodes, capturing rich spatiotemporal information relevant to brain activity. The CNN module is designed to extract localized spatial features across EEG channels, while the Transformer component captures long-range temporal dependencies using self-attention mechanisms, enabling the model to process subtle changes over time that are characteristic of schizophrenia.
The model is trained using a carefully balanced dataset of healthy and schizophrenia EEG recordings, with extensive regularization strategies such as dropout and batch normalization to prevent overfitting. A learning rate scheduler and early stopping were incorporated to ensure stable convergence. The model performance is evaluated and compared against baseline classifiers including CNN, Vision Transformer (ViT), Multi-Layer Perceptron (MLP), Random Forest, and Logistic Regression. Experimental results demonstrate that the SCZ-NET, a CNN + Transformer hybrid model, significantly outperforms traditional models in terms of classification, accuracy, and generalization, establishing it as a robust and scalable solution for EEG-based schizophrenia detection.