Abstract Details

Title Applications of Artificial Intelligence in Traumatic Brain Injury Research and Practice: A Systematic Umbrella Review

Topic Neuro Trauma and Critical Care

Presentation(s) P2 - Poster Session 2 (11:45 AM-12:45 PM)

Poster/Presentation
Number 3-011

Objective To systematically synthesize the current state meta-research evidence from existing systematic reviews and meta-analyses on the applications of artificial intelligence (AI) technologies in traumatic brain injury (TBI) research and clinical practice.

Background TBI diagnosis and prognosis are complex and critical aspects of patient management, with significant variability in outcomes. In data-driven predictions, AI models have been used to analyze patient data for early detection of adverse outcomes following TBI.

Design/Methods We systematically searched Scopus, Google Scholar, and additional sources. We included articles published from January 1st, 2020 to June 18th, 2025 that met the following criteria: (1) published as full-length article in peer-reviewed journals, (2) focused on the applications of any form of AI technologies in any area of TBI research or practice, (3) used one of the systematic literature review methods, and (4) full text was available in English language.

Results Across the 13 systematic reviews and meta-analyses analyzed, machine learning (ML) models consistently showed high accuracy in predicting mortality, adverse outcomes, and intracranial complications following TBI. The reported AUC values varied between 0.72 and 0.96, with most studies reporting an overall accuracy above 80%. Support Vector Machines (SVM) and Artificial Neural Networks (ANN) were the most utilized techniques and frequently surpassed traditional logistic regression models in performance. Ensemble methods like Random Forest and Gradient Boosting enhanced predictions for intracranial hypertension by as much as 20%. Significant heterogeneity in methods, measures, and outcomes were identified across the included articles.

Conclusions AI and ML methods offer strong potential for improving prognostic predictions in TBI, with several algorithms achieving high diagnostic and predictive accuracy. Nonetheless, existing findings are limited due to methodological and data ethics-related issues and inadequate validation in real-world scenarios. Future studies should prioritize advancing methods and ethics of AI-based predictive analytics, alongside evidence-informed integration of AI in the clinical management of TBI.

Authors/Disclosures
Rolake Feyisetan PRESENTER	Ms. Feyisetan has nothing to disclose.
Sabrina R. Jones	Ms. Jones has nothing to disclose.
Zaara Hossain	Miss Hossain has nothing to disclose.

��ɫ��

��ɫ��