In World neurosurgery ; h5-index 47.0
OBJECTIVE : The role of hemorrhage volume in risk of vasospasm, delayed cerebral ischemia (DCI) and poor outcomes after aneurysmal subarachnoid hemorrhage (SAH) is well-established. However, the relative contribution of blood within individual compartments is unclear. We present an automated technique for measuring not only total but also volumes of blood in each major compartment after SAH.
METHODS : We trained convolutional neural networks to identify compartmental blood (cisterns, sulci, and ventricles) from baseline CT scans of patients with SAH. We compared automated blood volumes against traditional markers of bleeding (modified Fisher score [mFS], Hijdra sum score [HSS]) in 190 SAH patients for prediction of vasospasm, DCI, and functional status (mRS) at hospital discharge.
RESULTS : Combined cisternal and sulcal volume was better correlated with mFS and HSS than cisternal volume alone (ρ=0.63 vs. 0.58 and 0.75 vs. 0.70, p<0.001). Only blood volume in combined cisternal plus sulcal compartments was independently associated with DCI (OR 1.023 per ml, 95% CI 1.002-1.048), after adjusting for clinical factors while ventricular blood volume was not. Total and specifically sulcal blood volume was strongly associated with poor outcome (OR 1.03 per ml, 1.01-1.06, p=0.006 and OR 1.04, 1.00-1.08 - for sulcal) as was HSS (OR 1.06 per point, 1.00-1.12, p=0.04), while mFS was not (p=0.24).
CONCLUSIONS : An automated imaging algorithm can measure the volume of bleeding after SAH within individual compartments, demonstrating cisternal plus sulcal (and not ventricular) blood contributes to risk of DCI/vasospasm. Automated blood volume was independently associated with outcome, while qualitative grading was not.
Yuan Jane, Chen Yasheng, Jayaraman Keshav, Kumar Atul, Zlepper Zach, Allen Michelle L, Athiraman Umeshkumar, Osbun Joshua, Zipfel Gregory, Dhar Rajat
2022-Oct-30
cerebral vasospasm, deep learning, image segmentation, intracranial aneurysm, subarachnoid hemorrhage
