Design and Implementation of a Novel Neural Framework for Schizophrenia Detection Using EEG Data

Schizophrenia is a chronic psychiatric disorder that has no definitive diagnostic tests and clinical diagnosis mostly depends on clinical judgment. It introduces a new neural approach for schizophrenia identification based on EEG data, which is more effective than current approaches due to noise, artifacts, and low interpretability. The framework also utilizes artifact removal and normalization of the WT and EMD as hybrid data features. A temporal and spectral feature-based attention map is employed in a custom neural network, which selectively focuses on important EEG areas to improve interpretability. The proposed model produces the highest accuracy of 93.8%, sensitivity of 92.3%, and specificity of 95.2% and surpasses prior works from the year 2021 to 2023. The model shows computational effectiveness with an inference time of 12ms per EEG segment, appropriate for real-time clinical application. This work lays the groundwork for developing explainable AI solutions that can be scaled for use in psychiatric diagnosis, specifically for the early identification of schizophrenia.