Evaluation of Diffusion-Perfusion Mismatch in Acute Ischemic Stroke with a New Automated Perfusion-Weighted Imaging Software: A Retrospective Study

Abstract

Introduction: The aim of this study was to evaluate the accuracy of automated software (iStroke) on magnetic resonance (MR) apparent diffusion coefficient (ADC) and perfusion-weighted imaging (PWI) against ground truth in assessing infarct core, and compare the hypoperfusion volume and mismatch volume on iStroke with those on Food and Drug Administration-approved software (RAPID) in patients with acute ischemic stroke.

Methods: We used the single-volume decomposition method to develop the iStroke (iStroke; Beijing Tiantan Hospital, Beijing, China) software. Patients with ischemic stroke were collected from two educational hospitals in China with MR-PWI performed in the emergency department within 24 h of symptom onset. Infarct core volume was defined as ADC < 620 × 10^-6 mm²/s and hypoperfusion volume was defined as Tmax > 6 s. We compared the accuracy of infarct core volume using iStroke and RAPID (iSchema View Inc, Menlo Park, CA) software with ground truth.

Results: We included 405 patients with acute ischemic stroke with MR ADC and PWI sequences. The infarct core volume on iStroke (median 2.43 ml, interquartile range [IQR] 0.60-10.32 ml) was not significantly different from the ground truth (median 2.89 ml, IQR 0.77-9.17 ml) (P = 0.07); Bland-Altman curves showed that the core volume of iStroke and RAPID software were comparable with each other on individual agreement with ground truth. The hypoperfusion volume and mismatch volume on iStroke were not statistically different from those on the RAPID software, respectively. In patients with large vessel occlusion (n = 74), the agreement between iStroke and RAPID was substantial (kappa = 0.76) according to DEFUSE 3 criteria (infarct core < 70 ml, mismatch volume ≥ 15 ml, and mismatch ratio ≥ 1.8).

Conclusions: The iStroke automatic processing of ADC and PWI is a reliable software for the identification of diffusion-perfusion mismatch in acute ischemic stroke.

Keywords: Infarct; Penumbra; Perfusion-weighted imaging; Software analysis; Stroke.

Fig. 1

Agreement between iStroke, RAPID software, and manual tracing core volume in all patients. A iStroke and manual tracing. B RAPID software and manual tracing

Fig. 2

Agreement between iStroke, RAPID software, and manual tracing core volume in patients with large vessel occlusion. A iStroke and manual tracing. B RAPID software and manual tracing

Fig. 3

A case’s penumbral evaluation on iStroke, RAPID, and DWI source image. A iStroke summary result: infarct core volume 17.3 ml, hypoperfusion volume 137.6 ml. B iStroke artery input function: left middle cerebral artery M1 segment; artery input function and venous output function curves were acceptable. C, D magnetic resonance-diffusion-weighted imaging and apparent diffusion coefficient showed. Centrum semiovale and corona radiata infarct with ground truth 15.8 ml. E RAPID summary result: infarct core volume 13 ml, hypoperfusion volume 141 ml. F RAPID artery input function: left middle cerebral artery M2 segment. G RAPID artery input function and venous output function curves were acceptable as well

Fig. 3

A case’s penumbral evaluation on iStroke, RAPID, and DWI source image. A iStroke summary result: infarct core volume 17.3 ml, hypoperfusion volume 137.6 ml. B iStroke artery input function: left middle cerebral artery M1 segment; artery input function and venous output function curves were acceptable. C, D magnetic resonance-diffusion-weighted imaging and apparent diffusion coefficient showed. Centrum semiovale and corona radiata infarct with ground truth 15.8 ml. E RAPID summary result: infarct core volume 13 ml, hypoperfusion volume 141 ml. F RAPID artery input function: left middle cerebral artery M2 segment. G RAPID artery input function and venous output function curves were acceptable as well