Skip to main content
Have a personal or library account? Click to login
External Validation of BrainScan AI for Detection of Hemorrhagic Stroke on Non-Contrast Head CT Cover

External Validation of BrainScan AI for Detection of Hemorrhagic Stroke on Non-Contrast Head CT

Open Access
|Jul 2026

Abstract

Background and Purpose

Rapid and accurate detection of intracranial hemorrhage on non-contrast head CT (NCCT) is critical for acute stroke triage and safe thrombolysis. AI tools may support high-volume emergency workflows, but independent real-world validation remains limited. This study externally validated BrainScan CT (v1.4) for hemorrhage detection, with secondary evaluation of ischemia detection.

Methods

In this retrospective single-centre study at UMHAT “St. George” (Plovdiv, Bulgaria), 1,419 consecutive NCCT scans from 1,260 adults (January–December 2025) were analysed. The reference standard was routine radiology reports. Hemorrhage was present in 120 scans (8.5%) and ischemia in 201 (14.2%). Performance metrics included sensitivity, specificity, predictive values, likelihood ratios, AUC, and McNemar testing. Subgroup analyses examined hemorrhage subtype (ECASS) and anatomical location. Sensitivity analyses assessed threshold variation and repeated scans.

Results

BrainScan CT showed high accuracy for hemorrhage detection: sensitivity 96.7% (95% CI 91.7–99.1), specificity 99.8% (99.4–100.0), PPV 98.3% (94.0–100.0%), NPV 99.7% (99.3–100.0%) and AUC 0.983 (0.967–0.999). Likelihood ratios were 627.9 (positive) and 0.03 (negative), with no systematic disagreement versus the reference standard (p=0.68). Sensitivity reached 100% for parenchymal hemorrhage, while false negatives occurred only in hemorrhagic infarction (HI-1 86.7%; HI-2 87.5%). Performance was consistent across regions, lowest for occipital hemorrhage (84.6%). Results were invariant across thresholds and unchanged when restricted to first scans (AUC 1.000). For ischemia, sensitivity was lower (59.2%) despite high AUC (0.930), significantly inferior to hemorrhage detection (p<0.001).

Conclusions

BrainScan CT demonstrated excellent real-world performance for hemorrhage detection on NCCT, supporting its role in triage, while expert oversight remains essential, particularly for subtle hemorrhagic infarction.

Language: English
Page range: 258 - 269
Published on: Jul 23, 2026
In partnership with: Paradigm Publishing Services
Publication frequency: 4 issues per year

© 2026 Ivan Sitnikov, Eray Halil, Neli Atanasova, Petra Popova, Kostadin Kostadinov, Fares Ezeldin, Edisson Judge, Penka Atanassova, published by European Biotechnology Thematic Network Association
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 3.0 License.