AI-Based Approaches for the Identification and Quantification of Traumatic Brain Injury in Computed Tomography Imaging: Systematic Review.

Authors

Abstract

Traumatic brain injury (TBI) is a leading cause of global disability and mortality, requiring timely diagnosis to prevent secondary injury. Manual computed tomographic (CT) evaluation often causes diagnostic delays, especially in smaller hospitals with limited radiological expertise. AI methods have been increasingly proposed to automate CT-based TBI assessment. The aim of this review is to present a comprehensive review of AI techniques applied to CT scans for TBI assessment. This study presents a systematized review conducted in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 reporting guidelines. Titles, abstracts, and full texts were screened against predefined eligibility criteria. Data on datasets, model architectures, preprocessing, and performance metrics were extracted using a structured form. Risk of bias was not formally assessed using a standardized tool. Instead, we qualitatively evaluated common sources of bias, such as dataset size, clinical evaluation, and external validation. Due to substantial heterogeneity in tasks, datasets, and outcome measures, we performed narrative synthesis only. We searched PubMed, Scopus, Web of Science, and IEEE Xplore from inception to December 31, 2025, for English-language peer-reviewed studies. We included original research papers that used AI or machine learning methods applied to human noncontrast head CT for TBI-related assessment. Studies had to report quantitative performance metrics. Nonoriginal papers, non-CT modalities, pediatric-only cohorts, and non-TBI applications were excluded. We restricted inclusion to English-language, peer-reviewed studies and did not perform a formal risk-of-bias or publication-bias assessment, which may overestimate the strength of the evidence. By screening the 674 publications found, we identified 101 studies for evaluation. We grouped these 101 studies into 4 categories: binary TBI classification (TBI or non-TBI), hemorrhage detection (classification, localization, segmentation, and quantization), midline shift measurement, and increased intracranial pressure estimation. Most work focused on hemorrhage detection, with fewer studies on TBI severity classification, midline shift estimation, and increased intracranial pressure assessment. Reported performance was frequently high, but many studies relied on small, single-center datasets, limited annotation detail, and internal validation only. External validation and prospective clinical evaluation were rare. AI methods for CT-based TBI assessment demonstrate strong performance in retrospective evaluations under controlled conditions, particularly for hemorrhage detection and segmentation; however, the current evidence base is constrained by limited dataset diversity, incomplete reporting, and lack of clinical validation. Accordingly, these findings should be interpreted as evidence of methodological feasibility rather than clinical readiness.

Tags

Topics