Diagnostic Accuracy of Artificial Intelligence for Detection of Intracranial Haemorrhage on Non-contrast CT Head: A Systematic Review

Intracranial haemorrhage (ICH) is a life-threatening condition requiring rapid diagnosis to improve clinical outcomes. Non-contrast computed tomography (CT) is the primary imaging modality; however, increasing workload and diagnostic variability may lead to delays. Artificial intelligence (AI) has emerged as a potential tool to enhance detection. This systematic review evaluates the diagnostic accuracy of AI algorithms for detecting ICH on non-contrast CT. This systematic review was conducted in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) guidelines, and the research protocol was registered in PROSPERO (CRD420261320521). A comprehensive search of PubMed, Cochrane Library, CENTRAL, ScienceDirect, and Google Scholar was performed. Diagnostic accuracy studies assessing AI-based detection of ICH on CT were included. Data extraction and study selection were conducted independently. Validated tools, including Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2) and Grading of Recommendations Assessment, Development and Evaluation (GRADE), were additionally utilised to assess risk of bias and certainty of evidence. A total of 25,528 records were identified through database searching (PubMed: 392; Google Scholar: 17,900; ScienceDirect: 7,218; Cochrane Library: 18). A total of eight studies, which met the inclusion criteria, were included, demonstrating variability in AI models, datasets, and validation methods. Overall, AI algorithms showed high diagnostic performance, with sensitivities ranging from 0.73 to 0.95 and specificities from 0.80 to 0.98. Studies using larger datasets reported higher accuracy. However, heterogeneity in study design and reference standards was significant. AI demonstrates promising diagnostic accuracy for ICH detection, particularly as a triage tool. However, variability and low-quality evidence limit generalizability. AI should be used alongside radiologists, with further prospective studies required before widespread clinical implementation.