Integrating perfusion imaging derived venous outflow and tissue-level collateral parameters in a comprehensive clinical model enhances prognostication in large vessel occlusion stroke

Abstract

Arterial inflow restoration and collateral status have been significantly correlated with functional outcomes in AIS-LVO patients undergoing mechanical thrombectomy (MT). CT perfusion imaging biomarkers, including prolonged venous transit (PVT), cerebral blood volume (CBV) index, and hypoperfusion intensity ratio (HIR), have emerged as reliable pretreatment adjunct parameters of comprehensive flow assessment. However, their absolute and comparative effectiveness in improving prognostic prediction remains unclear when used in conjunction with clinical and arterial inflow parameters. In our prospectively maintained database, we retrospectively reviewed and analyzed 149 patients with anterior circulation AIS-LVO who underwent MT. PVT was defined as Tmax ≥10 ​s timing within the superior sagittal sinus, torcula, or both, where PVT-was considered favorable. CBV index and HIR were derived from automated CTP software and analyzed in both continuous and dichotomized forms (HIR <0.4 and CBV index ≥0.8 represented favorable collaterals). A baseline logistic regression model incorporating significant clinical parameters and arterial inflow information was built first. PVT, CBV index, and HIR were subsequently incorporated individually and then in combination. Model performance was assessed using receiver operating characteristic analysis and compared by Delong's tests.PVT+ was associated with a significantly higher likelihood of unfavorable 90-day modified Rankin Scale outcomes (47.9 ​% vs. 16.7 ​%, p ​< ​0.01). Incorporating PVT into a baseline model comprised of significant clinical and arterial inflow parameters (age, hypertension, NIHSS, and mTICI score) improved outcome prediction (AUC: 0.821 [95%CI 0.749–0.879]), outperforming models incorporating CBV index (AUC: 0.792 [95%CI 0.718–0.854] and 0.799 [95%CI 0.725–0.860] in continuous and dichotomized forms, respectively) or HIR (AUC: 0.789 [95%CI 0.715–0.852] and 0.789 [95%CI 0.714–0.851] in continuous and dichotomized forms, respectively). The highest predictive accuracy was achieved by combining PVT with dichotomized CBV index, significantly outperforming the baseline model (AUC: 0.831 [95%CI 0.761–0.887] vs. 0.780 [95%CI 0.705–0.843], p ​= ​0.04).The combination of PVT and CBV index in conjunction with well-established clinical and interventional parameters significantly enhances predictive accuracy. This comprehensive imaging and clinical model offers potential utility for outcome stratification and clinical decision-making. Furthermore, PVT is a stronger predictor of functional outcomes in AIS-LVO patients than CBV index or HIR, highlighting the importance of VO assessment in stroke prognosis. However, prospective studies are necessary for further evaluation of the strength of these findings.