Abstract
Introduction and Hypothesis: Obstructive (non-communicating) hydrocephalus (OH) is a time-critical neurological emergency in which delayed recognition can lead to herniation and death. Non-contrast head CT (NCCT) is the first-line test. However, OH may be under-recognized when prior imaging is unavailable and early ventricular changes are subtle, including ventricular disproportion, third-ventricle bowing, and transependymal edema - or when these findings are obscured by concurrent hemorrhage, mass effect, or tumor. In this study, we aimed to validate Rapid OH version 1.0: an AI-powered decision-support triage tool designed to automatically flag suspected OH on adult head NCCT, supporting quicker clinician recognition and prioritization.
Methods: In a retrospective, multicenter study, we evaluated Rapid OH's accuracy against a consensus reference standard of three expert neuroradiologists. Consecutive adult NCCT cases enriched for obstructive hydrocephalus (OH) were identified (n=369; mean patient age 63.5 years ± SD 16.5; 194 male, 166 female, 9 unknown). 49 cases were excluded due to technical incompatibility (4), patient age (1), or being shunt cases (44), yielding a final cohort of 320 scans. Consensus reads identified 153 positive and 167 negative OH cases. We compared Rapid OH's triage output (suspected vs processed/not suspected) to the expert consensus, calculating sensitivity, specificity, 95% confidence intervals, and mean processing time per case.
Results: Rapid OH correctly identified OH in 137/153 cases, resulting in a sensitivity of 89.5% (95% CI: 0.837-0.935). It correctly excluded OH in 163/167 cases, yielding a specificity of 97.6% (95% CI: 0.940-0.991). 16 OH cases were missed by the software, while 4 negative (not OH) cases were incorrectly labeled as OH.
The mean processing time per scan was 30.3 seconds (range 10.5-55.5 seconds), enabling near real-time prioritization.
Conclusion: Rapid OH demonstrates high diagnostic accuracy for detecting obstructive hydrocephalus on NCCT with sub-minute processing times. Integration into neuro-emergency workflows-with real-time alerts and automatic routing to on-call neuroradiology-may reduce time-to-diagnosis, prioritize urgent reviews, and enable faster intervention. When indicated, expedited escalation to CSF diversion or transfer can prevent neurological deterioration in at-risk patients.