Multimodal Artificial Intelligence-Driven Clinical Decision Support for TBI Severity Assessment, Post-Traumatic Seizure Prediction, and Personalized Management

Traumatic brain injury (TBI) is a dynamic and diverse clinical phenomenon characterized by neurologic damage, post-traumatic seizures, impairment of functioning, and mortality. Conventional diagnostics may fail to provide a complete picture of the complicated interrelationship between factors relating to the clinical, imaging, neurophysiological, and treatment aspects. This study introduces a multimodal artificial intelligence clinical decision-support system (AI-CDSS) for TBI severity assessment, post-traumatic seizure occurrence forecasting, risk stratification, and individualized management aid. The model utilizes information on demographic and trauma-related factors, as well as medical history, computed tomography, electroencephalography data, treatment, and outcomes. The logistic regression, random forests, support vector machines, and extreme gradient boosting models were tested, and with the help of stratified cross-validation, the best model XGBoost demonstrated 92% accuracy, sensitivity, and specificity, as well as 92% precision, recall, and F1-score with the area under the ROC curve equal to 95% and Brier score equal to 82%. SHapley Additive exPlanations determined admission Glasgow coma scale, cortical contusion, intracranial bleeding, EEG abnormalities, type of Marshall CT class, pre-hospital seizures, age, and skull fractures as key predictors. The predicted probabilities of seizures were grouped into low (<20%), moderate (20–49%), and high (≥50%) risk categories, and the resulting probabilities were linked to evidence-based recommendations for follow-up, EEG evaluation, specialist referral, and re-evaluation. In spite of the fact that the proposed framework shows promise for personalized TBI management, the data that is currently reported still needs to be validated against patient data, needs to be verified externally, needs to be calibrated, evaluated prospectively, and assessed clinically before they can be employed in practice.