Explainable Deep Learning for Schizophrenia Risk Stratification: Interpreting Multimodal Biomarkers through Attention Maps and SHAP Values

Schizophrenia exists as a challenging neurological condition that produces major effects on mental health and intellectual ability. The current diagnostic methods display low accuracy levels combined with poor interpretability so deep learning methods must be developed for proper risk assessment. The proposed model combines CNNs with Transformers and implements SHAP values and attention maps for improving model interpretability. A combination of EEG MRI and genetic biomarkers made up the schizophrenia dataset that served for both model training and testing operations. The experimental findings showed 92.4% accuracy together with 91.2% precision while recall reached 90.7% under tests that produced an AUC-ROC value of 95.8% which exceeded existing benchmark models. Through this framework, healthcare professionals obtain clinically meaningful explanations regarding biomarkers that affect schizophrenia diagnosis determination. The proposed method demonstrates better diagnostic reliability and transparency than standard deep learning techniques operate today. It indicates explainable AI systems help close the gap between deep learning approaches and clinical decisions leading to better understanding and effective application of schizophrenia risk assessment by medical personnel.