In this retrospective study using prospectively collected data, we included patients who were admitted to Seoul National University Bundang Hospital between January 2021 and May 2023 and underwent CTP scans within 24 h of symptom onset. We excluded patients with (1) severe motion artifacts on CTP or poor contrast bolus arrival and (2) failed automated perfusion calculation by RAPID. A subset of patients, who had available follow-up diffusion-weighted images (DWI) taken within 3 h of the CTP, was used to compare the baseline ischemic core volumes predicted by the CTP with the early follow-up infarct volumes. The study protocol was approved by the institutional review board of Seoul National University Bundang hospital [IRB# B-1710-429-102], and patients or their legal representatives provided a written informed consent.

Using a standardized protocol (Kim et al., 2015), we prospectively collected demographic data, prior medication history, and the presence of vascular risk factors including hypertension, diabetes mellitus, hyperlipidemia, coronary artery disease, atrial fibrillation, and smoking history. Stroke subtypes were determined by an experienced vascular neurologist, using a validated MRI-based classification system built on the TOAST (Trial of ORG 10172 in Acute Stroke Treatment) criteria (Ko et al., 2014).

All CTP scans were performed using a 256-Slice CT scanner (Brilliance iCT 256, Philips Healthcare, Best, the Netherlands). The imaging parameters for CTP scans were as follows: 80 kVp, 150 mAs, Beam collimation 6 × 1.25 mm, rotation time 0.45 s. In the arterial phase scan, 30 cycles were captured every 2 s, resulting in a total duration of 60 s. Following a delay of 9 s, the delayed phase scan acquires 2 cycles at intervals of 10 s, amounting to a total duration of 20 s. A total of 50 mL of iodinated contrast agent (Iomeprol 400; Bracco Imaging, Milan, Italy) followed by 30 mL of saline flush was administered intravenously at a rate of 5 mL/s. The image matrix was 512 × 512 and the images were reconstructed with the slice thickness of 10 mm and the increment of 10 mm covering 8 cm in z-axis (Biesbroek et al., 2013).

JLK-CTP is a fully automated CTP software package to visualize and quantify ischemic core and hypoperfused tissue, which relies on a block-circulant singular value decomposition method (Kudo et al., 2010). We compared the volumes of ischemic core and hypoperfused tissue volumes estimated by RAPID and JLK-CTP, both of which are automated CTP analysis packages using a delay-insensitive algorithm. A default setting of rCBF <30% was used for defining the ischemic core. For assessing of hypoperfused tissue volume, we used the default setting of both packages (T_max > 6 s). Both software packages carry out automated registration, segmentation, and motion correction, employing a delay-insensitive method, and conducting post-processing (Figure 1). The RAPID software package has been routinely used for CTP in our hospital setting. Restored CTP source images were analyzed with the research version of JLK-CTP within the hospital for this study.

We collected DWI (3.0 Tesla, Ingenia CX, Phillips Medical Systems [n = 192] or MAGNETOM Vida, Siemens Healthcare system [n = 27]) taken within 3 h of the CTP scans. Areas of infarction were automatically segmented using a validated deep learning algorithm (JLK-DWI, JLK Inc., Seoul, Korea) (Ryu et al., 2023). Automatically segmented infarct areas underwent visual inspection by a stroke neurologist (W-SR) and were manually corrected as necessary.

Data were presented as mean ± standard deviation, median (interquartile range [IQR]), and number (percentage), as appropriate. To compare the volumes of ischemic core and hypoperfused tissue using two software packages, we utilized the concordance correlation coefficient (ρ) with 95% confidence intervals (CIs) and further investigated the data using reduced major axis regression (Lin, 1989). The magnitude of agreement was classified as follows: values from 0.0 to 0.2 indicating poor agreement; 0.21 to 0.40 indicating fair agreement; 0.41 to 0.60 indicating moderate agreement; 0.61 to 0.80 indicating substantial agreement; and 0.81 to 1.0 indicating excellent agreement. Additionally, Bland–Altman plots were used to assess the agreement of ischemic core volumes and hypoperfused areas as determined by the two packages. For the comparison between follow-up infarct volumes and ischemic core volumes analyzed by the two packages, both concordance correlation coefficients and Bland–Altman plots were employed. In the subgroup analysis, concordance correlation coefficients and Bland–Altman plots were utilized after stratifying patients by EVT and early infarct volume, arbitrarily divided at 20 mL. The agreement between ischemic core volume and follow-up infarct volume, categorized according to the DAWN (DWI or CTP Assessment with Clinical Mismatch in the Triage of Wake-Up and Late Presenting Strokes Undergoing Neurointervention with Trevo) clinical trial’s criteria (Jovin et al., 2018), was assessed using Cohen’s kappa. A two-tailed p value of <0.05 was considered statistically significant. All statistical analyses were performed using STATA (version 16.0, StataCorp LP, College Station, TX).

During the study period, a total of 2,544 patients were diagnosed with ischemic stroke, and 657 patients underwent a CTP scan. We excluded 315 patients who received CTP more than 24 h after their last known well and 15 patients who had severe motion artifacts or whose automated perfusion calculations failed using RAPID. Among the 327 patients included in the analysis, the mean age was 70.7 (SD 13.0), and 41.9% were women (Table 1). The median of the initial National Institute of Health stroke scale score was 9 (IQR, 4 to 17). The most common stroke etiology was identified as cardioembolism. Of the 205 (62.7%) patients who received revascularization therapy, 83 (25.3%) underwent intravenous thrombolysis alone, 59 (18.0%) underwent EVT alone, and 63 (19.3%) received combined therapy. The median interval between the time last known well to CTP scan was 192 min (IQR, 101 to 395 min). The median interval between CTP and DWI was 41 min (IQR, 27 to 100 min).

The mean difference between ischemic core volumes calculated by RAPID and JLK-CTP was −0.54 mL (95% CI, −1.72 to 0.64 mL; p = 0.85). There was excellent agreement in between JLK-CTP and RAPID (ρ = 0.958 [95% CI, 0.949 to 0.966]; Figure 2A). In the Bland–Altman plot, the limits of agreement were −20.74 and 21.83 mL (Figure 2B). We observed excellent agreement between JLK-CTP and RAPID (ρ = 0.942 [95% CI, 0.916 to 0.960]) with the mean difference of −0.94 mL (95% CI, −4.90 to 3.00 mL), when we restricted the analysis to patients whose ischemic core volume was not zero by both JLK-CTP and RAPID (n = 90; Supplementary Figure S1).

The mean difference between hypoperfused tissue volumes calculated by JLK-CTP and RAPID was 13.31 mL (95% CI, 8.22 to 18.39; p = 0.06). There was excellent agreement between JLK-CTP and RAPID (ρ = 0.855 [95% CI, 0.828 to 0.877]; Figure 2C). In the Bland–Altman plot, the limits of agreement were −78.22 and 104.83 mL (Figure 2D). When the analysis was confined to patients receiving EVT (n = 122), there was excellent agreement in ischemic core volumes and substantial agreement in hypoperfused tissue volumes between JLK-CTP and RAPID (Supplementary Figure S2).

In patients with available early follow-up DWI (n = 218), there was substantial agreement between follow-up infarct volumes and ischemic core volumes as determined by JLK-CTP (ρ = 0.753 [95% CI, 0.699 to 0.800]; Figure 3A) and RAPID (ρ = 0.736 [95% CI, 0.685 to 0.780]; Figure 3C) at the default rCBF threshold of <30%. The limits of agreement for the volumes of the ischemic core and the follow-up infarct volume were comparable between the two packages (Figures 3B,D). Nevertheless, at the default rCBF threshold of <30%, both packages tended to overestimate infarct core volumes, as indicated by the slope of the reduced major axis being under one. Furthermore, in patients with visible infarcts on the follow-up DWI (n = 187), the ischemic core volume determined by RAPID was zero in 123 (65.8%) cases, whereas it was zero in 104 (55.6%; p = 0.04) cases as determined by JLK-CTP (see Table 2).

When dividing patients by early follow-up infarct volume (<20 mL [n = 193] vs. ≥20 mL [n = 24]), in patients with follow-up infarct volume < 20 mL, both packages exhibited similarly poor agreement with early follow-up infarct volume (ρ = 0.196 for JLK-CTP and ρ = 0.181 for RAPID, respectively; Supplementary Figure S3). In patients whose early follow-up infarct volume was ≥20 mL, both JLK-CTP (ρ = 0.497; 95% CI, 0.213 to 0.704) and RAPID demonstrated fair agreement (ρ = 0.438; 95% CI, 0.182 to 0.639). Setting the rCBF threshold to <26% improved the correlation between early follow-up infarct volume and ischemic core volumes determined by JLK-CTP, as shown by both the concordance correlation coefficient and Bland–Altman analysis (Supplementary Figure S4).

The median mismatch volumes determined by JLK-CTP and RAPID were 23.26 (IQR 0 to 85.33) and 23 (0 to 101), respectively, and there was substantial agreement between the mismatch volumes determined by JLK-CTP and RAPID (ρ = 0.774; 95% CI 0.735 to 0.808; Figure 4A). The mean difference in mismatch volume between the two software tools was 13.85 mL (95% CI, 8.80 to 18.90 mL; p = 0.01; Figure 4B).

When considering early infarct volume as a standard reference and categorizing ischemic core volumes by JLK-CTP and RAPID according to the DAWN trial’s criteria, similar Cohen’s kappa values were observed for JLK-CTP (0.55) and RAPID (0.51). However, in the medium-sized ischemic core volume groups (20–30 mL and 30–50 mL), the ischemic core volumes determined by both JLK-CTP and RAPID exhibited poor agreement with the categorization by early infarct volumes.