Integration of Artificial Intelligence (AI) in Computed Tomography for the Management of Intracranial Hemorrhages in Congolese Hospitals: A Systematic Review of the Literature

Abstract

Context and objective. Artificial Intelligence (AI) applied to computed tomography (CT) image analysis is progressively emerging as a promising tool to enhance the detection of neuroradiological emergencies. The present study aimed to assess the diagnostic performance of AI in detecting intracranial hemorrhages (ICH) on CT scans. Methods. This systematic review followed PRISMA guidelines. Included studies reported at least one diagnostic performance indicator (sensitivity, specificity, PPV, NPV). Literature searches were conducted in PubMed, EMBASE, Scopus, and Web of Science, complemented by manual screening (articles published between 2015–2025). Two independent reviewers selected studies, extracted data, and assessed risk of bias using the QUADAS-2 tool. Metrics were analyzed using R Studio and MetaBaye DTA. The protocol was registered in PROSPERO to ensure methodological transparency. Results. Seventeen mainly retrospective studies were included, mostly involving at least two reference radiologists. The pooled sensitivity was 0.945 (95% CI: 0.889–0.974) and specificity 0.937 (95% CI: 0.871–0.971). The positive (PLR = 16.6) and negative (NLR = 0.07) likelihood ratios demonstrated the strong ability of AI to confirm or exclude ICH. The diagnostic odds ratio (DOR) reached 280, with substantial heterogeneity across all measures (I² > 95%). Conclusion. Convolutional neural network–based models show high diagnostic accuracy for ICH detection, supporting their potential to enhance neuroradiological diagnosis. However, their implementation in Sub-Saharan Africa, particularly in the Democratic Republic of Congo (DRC), requires a systemic approach integrating infrastructure development, ethical governance, algorithmic epublic of contextualization, and inter-institutional partnerships. Received: May 30th, 2025 Accepted: October 13th, 2025 https://dx.doi.org/10.4314/aamed.v19i1.17

Ähnliche Arbeiten