Contribution of an Automatic Algorithm for Quantifying the Volume of Aneurysmal Subarachnoid Hemorrhage to the Evaluation of the Risk of Occurrence of Delayed Cerebral Ischemia: A Cohort Study.

BACKGROUND

This study focuses on aneurysmal subarachnoid hemorrhage (aSAH) with a high risk of delayed cerebral ischemia (DCI) and acute hydrocephalus (AH). The aim was to compare the performance of an automatic algorithm for quantifying the volume of intracranial blood with the reference radiological scales to predict DCI, AH, and neurological outcome.

METHODS

This was a single-center retrospective observational study of a cohort of patients with aSAH. We developed an automated blood detection algorithm based on the specific density of the blood clot. The blood clot was segmented on the first brain scan (total, supratentorial, cisternal, intraventricular). The predictive value of our model was compared, using the area under the receiver operating characteristic curve (ROC), to eight radiological scales: Fisher, modified Fisher, Claassen, Barrow Neurological Institute, Hijdra, Graeb, LeRoux scales, and intraventricular hemorrhage score.

RESULTS

We analyzed the scans of 145 patients with aSAH. In our cohort, 51 patients (43%) had DCI and 70 patients (54%) had AH. At 3 months, 22% of patients had died and 19% had poor outcome (Glasgow Outcome Scale extended 2-4). Cisternal blood volume was significantly correlated with cisternal Hijdra scale (R = 0.79; P < 0.001). The ROC of cisternal blood volume was comparable to the ROC of the Hijdra scale in predicting the occurrence of DCI (ROC = 0.83 [95% confidence interval {CI} 0.75-0.89] vs. 0.86 [95% CI 0.79-0.9]; P = 0.23). The ROC of intraventricular blood volume was not significantly different from the intraventricular hemorrhage score in predicting the occurrence of AH (ROC = 0.78 [95% CI 0.70-0.84] vs. 0.79 [95% CI 0.72-0.85]; P = 0.28). The ROC and supratentorial blood volumes were not significantly different from the Simplified Acute Physiology Score II in predicting the occurrence of poor neurological outcome at 3 months (ROC = 0.75 [95% CI 0.67-0.82] vs. 0.81 [95% CI 0.74-0.87]; P = 0.073).

CONCLUSIONS

With no manual intervention, our algorithm performed as well as the best radiological scores in predicting the occurrence of DCI, AH, and neurological outcome.