AI-Deep Learning Framework for Predicting Neuropsychiatric Outcomes Following Toxic Effects of Drugs on The Brain

Introduction: drug-induced neurotoxicity represents a significant clinical challenge, with neuropsychiatric complications affecting treatment outcomes and patient quality of life. Current predictive tools lack both accuracy and interpretability, limiting their clinical utility. Methods: We developed a hybrid CNN-LSTM deep learning framework with attention mechanisms, trained on multimodal clinical data including electronic health records, neuroimaging, and biomarker profiles. Model interpretability was achieved through SHAP value analysis, with performance evaluated via 5-fold cross-validation.Results: The model achieved 92 % accuracy (AUC-ROC 0,93), significantly outperforming traditional approaches. Key predictors included drug dosage (SHAP=0,15), treatment duration (SHAP=0,12), and age. High-risk subgroups (patients >60 years) showed 2,5× increased risk of cognitive decline (p<0,01).Conclusions: This interpretable AI framework enables precise, clinically actionable prediction of neuropsychiatric outcomes following drug-induced neurotoxicity, supporting personalized treatment decisions and risk mitigation strategies.