Spatial accuracy of computed tomography perfusion to estimate the follow-up infarct on diffusion-weighted imaging after successful mechanical thrombectomy

This was a retrospective cohort study of patients with AIS who underwent MT between October 2019 and December 2021. The study protocol was approved by the Ethics Committee of the First Afliated Hospital of Nanjing Medical University. The ethics committee waived the written informed consent because of that present study was retrospective. The inclusion criteria were as follows: 1) diagnosis of AIS; 2) anterior circulation occlusion (internal carotid artery and/or middle cerebral artery); 3) baseline CTP and follow-up DWI were performed; 4) successful recanalization was achieved after MT; 5) the imaging quality was adequate for subsequent analyses. We excluded the patients with a ischemic core volume of 0 mL on baseline CTP. Successful recanalization was defined as a modified Treatment in Cerebral Infarction (mTICI) score ≥ 2b [16]. Besides that, sensitivity analysis was performed in patients achieving mTICI 2c or 3.

We collected the following demographic and clinical information from the electronic medical records: gender, age, stroke related risk factors (hypertension, hyperlipidemia, diabetes, atrial fibrillation, smoker), baseline National Institutes of Health Stroke Scale (NIHSS) score, intravenous thrombolysis was performed or not, the time interval from stroke onset to CTP, the time interval from CTP to recanalization, and the modified Rankin Scale (mRS) score at 3 months after treatment. The clinical outcome was assessed based on the mRS score which was a standardized 6-point score that measures the functional independency of patients with acute stroke at 90 days after treatment. Good functional outcome was defined as a mRS ≤ 2 [17].

All the patients with AIS underwent baseline CT scan on a 128-section CT scanner (Optima CT 660; GE Healthcare), including noncontrast CT (120 kV, 100–350 auto-mAs, contiguous 5-mm axial sections) and whole-brain volumetric CTP (four-dimensional adaptive spiral mode, periodic spiral approach, 80 mm in z-axis, 2-second delay after the start of contrast medium injection, temporal resolution of 1.7 sec, total CTP duration of 53 sec, 100 kVp, 200 mAs, rotation time of 0.4 sec, 0.984 maximum pitch). CTP acquisition was started after initiation of intravenous injection of 50 mL of iopromide (Iopromide, Ultravist 370, Bayer Schering Pharma) at a rate of 5 mL/sec. Contrast was followed by a 30 mL saline flush at the same rate. Simulated CT angiography (CTA) images (section thickness, 0.625 mm) were reconstructed from the peak arterial phase of CTP source images.

Follow-up MRI scan was performed with a 3 T scanner (Magnetom Skyra, Siemens Healthcare) equipped with a 20-channel head-neck coil, usually within 7 days after treatment. The MRI protocol included DWI, FLAIR, susceptibility-weighted imaging (SWI), MR angiography (MRA) and arterial spin labeling (ASL). DWI were scanned using the following parameters: repetition time (TR) = 6400 msec; echo time = 98 msec; b value = 0 and 1000 mm²/s; field-of-view (FOV) = 220 × 220 mm²; matrix = 192 × 192; section thickness = 4 mm; number of slices = 20. The time interval from CTP scan to follow-up MRI scan was also recorded.

The location of intracranial LVO was evaluated based on the simulated CTA, was defined as intracranial ICA or not [16]. CTP data was post-processed with a commercial software named CTPdoc AI (SHUKUN, Beijing, China), by using the oscillation index singular value decomposition (oSVD) deconvolution method which was widely applied in congeneric automated perfusion analysis softwares (e.g., RAPID) [18]. Predefined relative CBF threshold value of < 30% at CTP was used to calculate the volume of baseline ischemic core [18]. We assessed the occurrence and degree of leukoaraiosis based on the pre-treatment CT images according to the method proposed in previous study [19]. The degree of leukoaraiosis was dichotomized into absent or mild (Fazekas scores, 0–1) versus moderate or severe (Fazekas scores, 2–3) [19]. The volume of cerebral infarct was measured by a neuro-radiologist (with 10 years of experience), based on the follow-up DWI. After the bordline of infarct areas was manually outlined section by section based on the ITK-SNAP software (version 3.8, http://www.itksnap.org), the volume could be aumatically obtained. Hemorrhage transformation (HT) was assessed based on the post-treatment CT. If uncertain, SWI was referred. We included the HT area that was located within the infarct area in the delineation. The area of infarct on each section could be obtained after manual delineation. Then, the volume of infarct on follow-up DWI could be calculated by multiplying the sum of infarction areas by the slice thickness. The registration between CTP and DWI images is done by an automated image analysis toolkit (SHUKUN, Beijing, China), which is based on a Python library named ANTsPy [20]. CTP and DWI lesion overlap was calculated using SciPy [21] and spatial agreement assessed using SciPy and scikit-image [22]. The Dice similarity coefficient was calculated to assess spatial agreement between CTP and DWI lesions [20]. We also used the average Hausdorff distance (the average of all minimum distances between two segmentations) to quantify the spatial agreement [23]. The average Hausdorff distance was calculated using Formula 2 given by Aydin OU et al. [24].

Normality test of quantitative variables were performed using the Shapiro-Wilk test. If normally distributed, the variables were expressed as mean ± standard deviation (SD). If not, they were reported as median with interquartile range (IQR). The categorical variables were reported as numbers and proportions. According to the median Dice similarity coefficient, we dichotomized the study population into high and low Dice groups. According to the median Hausdorff distance, we also dichotomized the study population into high and low Hausdorff distance groups. For categorical variables, Chi-square tests were used to compare the differences. For quantitative variables in normal distribution, independent sample t tests were used to compared the differences. For those not in normal distribution, Mann-Whitney U tests were used to compare the differences. Spearman correlation analyses were applied to assess the correlations between the Dice similarity coefficient and the significantly different variables. The variables with P values less than 0.1 in univariable analysis were included into the multivariable logistic regression to assess the independent factor associated with the Dice similarity coefficient. Bland-Altman plots were used to illustrate mean differences and limit of agreement (LoA), and the intraclass correlation coefficient (ICC) was used to assess the agreement between the volume of ischemic core on baseline CTP and the volume of infarct area on follow-up DWI. All statistical analyses were performed using SPSS software (version 23.0) and MedCalc software (version 20.1). Two-sided P value less than 0.05 was considered statistically significant.