Diagnostic Accuracy of Artificial Intelligence for Intracranial Hemorrhage Detection on Non-contrast CT: A Systematic Review and Meta-analysis
Meghana Chennupati1, Shradha Kakde2, Harshawardhan Dhanraj Ramteke3, Rakhshanda Khan4, Sharath Chandra Anne5, Ahmed Harb6, Anas Mansour6
1Mamata Academy of Medical Sciences, 2MGM medical college and hospital, 3Anhui medical university, 4Ayaan Institute of medical sciences, 5pinnamaneni siddharta medical college, 6Faculty of Medicine, Al-Azhar University, Cairo, Egypt
Objective:
To evaluate the pooled diagnostic accuracy of artificial intelligence (AI) algorithms for detecting intracranial hemorrhage (ICH) on non-contrast head computed tomography (CT) and to identify design and technical factors associated with performance variability
Background:

ICH requires rapid and accurate diagnosis on non-contrast CT to prevent morbidity and mortality. Artificial intelligence (AI) algorithms offer potential to enhance detection speed and consistency. This meta-analysis evaluates the overall diagnostic accuracy and clinical performance of AI models for automated ICH detection across diverse clinical settings.

Design/Methods:
A systematic review and meta-analysis of diagnostic studies (PubMed, Embase, Web of Science, Scopus; inception–May 2025) pooled AI-based ICH detection accuracy using bivariate random-effects and HSROC models, with QUADAS-2 quality assessment and subgroup analyses by study design, validation type, and algorithm category
Results:

Nineteen studies encompassing 90 514 CT scans were included. AI models comprised convolutional neural networks, hybrid CNN-RNN architectures, and commercial triage systems (Aidoc, Annalise CTB, qER.ai, VeriScout™, Siemens AI). The pooled sensitivity was 0.91 (95 % CI 0.89–0.93) and specificity 0.93 (95 % CI 0.90–0.95), with an overall AUC of 0.95 (95 % CI 0.93–0.96) and DOR ≈ 130, indicating excellent discrimination. Heterogeneity was moderate (I² ≈ 60 %), largely explained by study design and validation type. Externally validated and prospective cohorts showed comparable performance to internal test sets, confirming real-world generalizability. AI integration shortened reporting or notification times by 1–3 minutes per scan in emergency workflows.

Conclusions:

Artificial intelligence achieves radiologist-level accuracy for intracranial hemorrhage detection on non-contrast CT, supporting its safe, rapid, and generalizable integration into clinical triage workflows.

10.1212/WNL.0000000000216765
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.