A novel visual ranking system based on arterial spin labeling perfusion imaging for evaluating perfusion disturbance in patients with ischemic stroke

Sangjoon Lee, Dong Woo Park, Tae Yoon Kim, Dong Sun Kim, Ji Young Lee, Young Jun Lee, Chun Ki Kim

Research output: Contribution to journalArticle

Abstract

We developed a visual ranking system by combining the parenchymal perfusion deficits (PPD) and hyperintense vessel signals (HVS) on arterial spin labeling (ASL) imaging. This study aimed to assess the performance of this ranking system by correlating with subtypes classified based on dynamic susceptibility contrast (DSC) imaging for evaluating the perfusion disturbance observed in patients with ischemic stroke. 32 patients with acute or subacute infarcts detected by DSC imaging were reviewed. Each patient's brain was divided into 12 areas. ASL ranks were defined by the presence (+) or absence (-) of PPD/HVS as follows; I:-/-, II:-/+, III: +/+, and IV: +/-. DSC imaging findings were categorized based on cerebral blood flow (CBF) and time to peak (TTP) as normal (normal CBF/TTP), mismatched (normal CBF/delayed TTP), and matched (decreased CBF/delayed TTP). Two reviewers rated perfusion abnormalities in the total of 384 areas. The four ASL ranks correlated well with the DSC subtypes (Spearman's r = 0.82). The performance of ASL ranking system was excellent as indicated by the area under the curve value of 0.94 using either matched or mismatched DSC subtype as the gold standard and 0.97 using only the matched DSC subtype as the gold standard. The two methods were in good-to-excellent agreement (maximum ê-values, 0.86). Inter-observer agreement was excellent (ê-value, 0.98). Although the number of patients was small and the number of dropouts was high, our proposed, ASL-based visual ranking system represented by PPD and HVS provides good, graded estimates of perfusion disturbance that agree well with those obtained by DSC perfusion imaging.

Original languageEnglish
Article numbere0227747
JournalPLoS ONE
Volume15
Issue number1
DOIs
StatePublished - 2020 Jan 1

    Fingerprint

Cite this