Introduction: In acute ischemic stroke patients, collateral circulation is an important target for neuroprotectant treatments as it influences tissue survival and clinical outcomes. Collateral status is determined with the expert-rated Tan score on CT Angiography (CTA) or tissue-level perfusion metrics on CT Perfusion (CTP). We developed a deep learning–based vessel segmentation method for CTA to quantify all vessels- and artery-only collateral scores (vesQCS, aQCS). We aim to (1) compare the performance of macrovascular and tissue-level collateral status for predicting inter-hospital infarct growth rate (IGR) and (2) evaluate if a mismatch in macrovascular and tissue-level collateral status influences IGR. Methods: We included patients recruited in a prospective, multicenter study who were transferred from a primary to a comprehensive stroke center for EVT (n=387). vesQCS and aQCS were defined as the ratio of deep learning–segmented vessel or artery volume in the infarcted versus contralateral hemisphere (Figure-1). Tan score was expert-rated, while tissue level collaterals such as the Hypoperfusion Intensity Ratio (HIR: Tmax>10s/Tmax>6s) and the mismatch volume (Tmax>6s-rCBF=5ml/hr. We used interaction analysis to evaluate if the association between HIR and fast IGR was affected by macrovascular collaterals. Results: Median inter-hospital growth rate was 3.9 (1.4-8.2) mL/hour and median transfer time was 3.6 (2.6-4.6) hours. All collateral scores, except mismatch volume, were significantly associated with fast IGR (Figure-2). AUC analysis shows similar performance between collateral scores for imputed and complete case sensitivity analyses (Figure-3). For lower Tan scores, the association of HIR with IGR was lower than for higher Tan scores (p for interaction = 0.03, Figure-3). Conclusion: Automated quantitative (vesQCS/aQCS), expert-rated Tan, and hypoperfusion intensity ratio perform similarly to predict inter-hospital infarct growth rate. Patients with poorer macrovascular collaterals are at higher risk of fast IGR despite favourable tissue-level collaterals.
Prasetya et al. (Thu,) studied this question.