Multidisciplinar, Journal Year: 2025, Volume and Issue: 3, P. 222 - 222
Published: April 25, 2025
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 interpretability, limiting their utility. Methods: We developed hybrid CNN-LSTM deep learning framework attention mechanisms, trained on multimodal data including electronic health records, neuroimaging, biomarker profiles. Model interpretability was achieved through SHAP value analysis, performance evaluated via 5-fold cross-validation.Results: The model 92 % (AUC-ROC 0,93), significantly outperforming traditional approaches. Key predictors included drug dosage (SHAP=0,15), duration (SHAP=0,12), age. High-risk subgroups (patients >60 years) showed 2,5× increased risk cognitive decline (p<0,01).Conclusions: This interpretable AI enables precise, clinically actionable prediction following neurotoxicity, supporting personalized decisions mitigation strategies.
Language: Английский