Detection and localization of posterior circulation large vessel occlusions using perfusion imaging

BACKGROUND AND PURPOSE: Cerebral perfusion imaging may help detect the presence of, and infer vessel occlusion location in, anterior circulation stroke, but its value in posterior circulation large vessel occlusion (PC-LVO) is less well defined. We evaluated the accuracy and inter-reader reliability of CT and MR perfusion imaging for occlusion classification and posterior circulation vessel localization. METHODS: We performed a retrospective multicenter diagnostic accuracy study from 11 centers that underwent perfusion imaging for stroke evaluation. The cohort was enriched for posterior circulation stroke. Four blinded readers assigned each case to one of five categories: no occlusion, anterior circulation large vessel occlusion (AC-LVO), basilar artery, posterior cerebral artery (PCA), or vertebral/posterior inferior cerebellar artery (vert/PICA) occlusion. Final classification was determined by majority vote. RESULTS: 184 patients were included. Perfusion imaging correctly classified 150/184 cases on the 5-class task (81.5%, 95% CI 75.1% to 86.9%). Across the 3-class task of LVO(-), AC-LVO, and PC-LVO, accuracy was 163/184 (88.6%, 95% CI 83.1% to 92.8%), with sensitivity of 100% for LVO(-), 100% for AC-LVO, and 84.4% for PC-LVO. Within the posterior circulation subgroup, perfusion imaging correctly localized basilar, PCA, and vert/PICA occlusions in 101/135 cases (74.8%, 95% CI 66.6% to 81.9%), including 82/107 basilar (76.6%), 10/12 PCA (83.3%), and 9/16 vert/PICA (56.3%). Fleiss κ was 0.889 for the 3-class task, 0.742 for the 5-class task, and 0.604 for posterior localization. CONCLUSION: Perfusion imaging is moderately accurate for the detection and localization of PC-LVO, but performs less well compared with the detection of AC-LVO.