Acute infarction after mechanical thrombectomy is better delineable in virtual non-contrast compared to conventional images using a dual-layer spectral CT

The aim was to evaluate Virtual Non-Contrast (VNC)-CT images for the detection of acute infarcts in the brain after mechanical thrombectomy using a dual-layer spectral CT. 29 patients between September 2016 and February 2017 with unenhanced head spectral-CT after mechanical thrombectomy and available follow-up images (MRI, n:26; CT, n:3) were included. VNC-CT and conventional CT (CT) images were reconstructed using dedicated software. Based on those, contrast-to-noise ratio (CNR), and the volume of infarction were measured semi-automatically in VNC-CT, CT and MRI. Furthermore, two readers independently assessed the VNC-CT and CT images in a randomized order by using the ASPECT score, and inter-rater reliability, sensitivity and specificity were calculated. CNR was significantly higher in VNC-CT compared to CT (3.1 ± 1.5 versus 1.1 ± 1.1, p < 0.001). The mean estimated volume of infarction was significantly higher in VNC-CT compared to CT (72% versus 55% of the volume measured in MRI, p < 0.005). Inter-rater reliability was higher in VNC-CT compared to CT (0.751 versus 0.625) and sensitivity was higher in VNC-CT compared to CT (73% versus 55%). In conclusion, acute ischemic lesions after mechanical thrombectomy are better definable in VNC-CT compared to CT images using a dual-layer spectral CT system.

Sensitivity and specificity. Table 2 lists the sensitivity and specificity values for the detection of acute infarction by both readers analyzing VNC-CT or CT images, separately for the ASPECT score and pc-ASPECT

Discussion
The results of our study show that VNC-CT images produced by a dual-layer spectral CT enable a better delineation of acute infarction after mechanical thrombectomy compared to conventional CT images with significantly higher CNR, volume estimation compared to the gold standard DWI-MRI, and higher sensitivity and better inter-rater reliability.
The results of our study are generally in conformance with two previous studies analyzing VNC-CT images of dual-energy CTs 10,11 . They concluded that VNC-CT images allow better detection of acute ischemia than conventional CTs. The novelty of our study is the first use of a dual-layer spectral CT that has the advantage to acquire spectral data without additional radiation doses or alteration of x-ray tube parameters. The radiation dose of the head CT scan in our study was relatively low with 1.9 mSv, compared to the other studies where the radiation dose was 2.3 mSv 10 . A major difference which could have an influence of stroke detection in CT is the exact time point of the CT scan. In our study, the control CT scans were performed around 12 h after the interventional procedure (13 ± 6 h, according to the standard operation procedure of our hospital), whereas other studies evaluated stroke delineation in DECT 24 h 11 and 1h 10 after mechanical thrombectomy. In our study, we have seen a very good agreement for the estimation of ischemic brain volume using DECT and MRI as a gold standard. Additionally, we also included patients with an occlusion of the basilar artery and added the pc-ASPECT Score for the analysis of the posterior circulation.
The better delineation of infarction in VNC-CT images compared to CT images could be explained due to the disrupted blood brain barrier of ischemic lesions with extravasation of iodinated contrast agent, which is injected intra-arterially during the recanalization procedure. The contrast agent could mask the soft hypodensity of ischemic lesions in the conventional CT in the early stages whereas the hyperdense signal of the iodinated contrast agent disappears in VNC-CT images and reveal the developing infarction. As ischemic lesions appear more hypodense in VNC-CT images compared to conventional CT images, the contrast between ischemic lesions and non-affected tissue is higher in VNC-CT images. The noise of the non-affected tissue is comparable in conventional and VNC-VT images. This effect is desirable and improves delineation of infarction. In accordance to this consideration, the calculated CNR was higher in VNC-CT images compared to conventional CT images.
One group 12 studying the contrast enhancement and contrast extravasation in ischemic stroke patients after intraarterial thrombolysis described a persistence of the extravasated contrast medium in the 24 h control CT. Therefore, it seems assumable that the iodinated contrast agent rests for some time in the brain tissue and is not resorbed within the next 12 h -the timepoint of our CT scans. This seems to be a good explanation for the pronounced hypodensity of ischemic lesions in the VNC-CT images.
With this technique, developing ischemic lesions can be depicted earlier compared to conventional CT and may improve therapy management and prediction of outcome. Additionally, in the future, imaging might become even more sensitive with photon-counting CT. First results of photon-counting CT scans of the brain  have recently been published 13 showing greater gray-white matter contrast compared with conventional CT. Furthermore, the improvement of the delineation of ischemic lesions might have an impact on studies comparing the outcome of different recanalization techniques as the estimated infarction volume is closer to the reality using VNC-CT compared to conventional CT. Limitations: This study has a retrospective design and the total number of patients was relatively small. Another limitation was the temporal delay of 2.4 days between CT and follow-up MRI as gold standard. During this delay, the ischemic areas could have expanded.
To conclude, acute ischemic lesions after mechanical thrombectomy are better definable in VNC-CT images compared to CT images as produced by a dual-layer spectral CT with significantly higher CNR, sensitivity, volume estimation and higher inter-rater reliability.

Material and Methods
Patients. Between September 2016 and February 2017, 56 patients underwent unenhanced computed tomography (CT) of the head within the standard clinical protocol as a matter of routine. Within this group, 29 patients had been examined after mechanical thrombectomy (13 ± 6 h) (15 female, 14 male, age range 42-90 y, mean age 72 ± 14 y), having available follow-up imaging (MRI, n: 26; CT, n: 3; 57 ± 46 h after spectral CT). The latter group of 29 patients were included into our study, data were anonymized and analyzed retrospectively with the approval of the local ethics committee. The study was conducted in accordance with the 2013 revised Declaration of Helsinki. The pattern of vessel occlusion was in 24 cases in the anterior circulation (ICA or terminal carotid segment, n: 7; M1, n: 13; M2, n: 4) and in 5 cases in the posterior circulation (basilar artery) with TICI score "3/2b" in 28 cases, and "0" in 1 case.

Imaging acquisition.
CT was performed on a dual-layer spectral CT (IQon spectral CT, Philips Healthcare, USA) with 120 kVp and 260 mAs. Post-processing was performed using a dedicated software (IntelliSpace Portal v6.5.0.02901, Philips, Healthcare, USA). The VNC-CT images without iodine and the conventional CT images were generated from spectral data sets, reformatted with 5 mm slice thickness and reoriented in a standard manner in the anterior commissure -posterior commissure plane. For both, the conventional CT and VNC-CT images, we used a sharp filter kernel for the brain (UC). The conventional CT images were reconstructed using iterative reconstruction (iDose4), whereas the VNC-CT images were reconstructed using a special spectral reconstruction mode. In both cases, level 2 was used, therefore the images were considered comparable and image analysis was performed. Follow up images were available of a 3 T MRI (Philips, Achieva) or a conventional CT (Brilliance, Philips Healthcare, USA; Somatom, Siemens Healthineers, Germany). The effective radiation dose was calculated by multiplying the dose-length product (840 [mGy*cm]) by a conversion coefficient (0.0023 mSv mG-1cm-1) for the head according to European guidelines (Quality criteria for computed tomography, EU report 16262. Luxembourg: Commission of the European Communities, 1999, available from http://www.drs.dk/guidelines/ct/quality). The effective radiation dose was 1.9 mSv. Imaging analysis. The imaging data were analyzed retrospectively on a standard PACS workstation (Sectra Workstation IDS7, Version 17.1.18.3596, Sectra Healthcare). Small circular region of interest (ROI) (8-12 mm²) were drawn in the centre of the ischemic area in CT and VNC-CT images consulting additionally diffusion-weighted images of follow-up MRI or follow-up CT and in the corresponding contralateral unaffected brain tissue in order to measure mean densities of the Hounsfield units (HU). We carefully tried to avoid areas of haemorrhagic transformations or iodine extravasation. Furthermore, contrast-to-noise ratio (CNR) was calculated according to following formula: where S lesion and S non-affected tissue represent the mean signal in a ROI in the ischemic lesion and normal-appearing brain tissue in the corresponding contralateral side, respectively. SD non-affected tissue is the standard deviation of the normal-appearing brain tissue in the contralateral side. Furthermore, the volume of infarction was estimated by one reader in VNC-CT and CT images and MRI or follow-up CT in a randomized order, semi-automatically using the IS Portal (IntelliSpace Portal v. 5.02.40009, 11. Sep 14, Philips Healthcare Netherlands). Additionally, two experienced readers independently assessed the VNC-CT and CT images in a randomized order regarding the presence of acute ischemic lesions using the ASPECT score 14 for infarction in the area of the medial cerebral artery or the posterior circulation (pc)-ASPECT score 15 for the territory of the basilar artery. Sensitivity and specificity were calculated. Statistical analysis. Paired student's t-test was used for the comparison of the CNR and differences between estimated volume measurements. Significant differences were defined by a p < 0.05. Cohen's kappa was calculated to determine inter-observer variability of the qualitative evaluation using the (pc-) ASPECT scores, with values < 0 indicating no agreement; 0-0.20: slight; 0.21-0.40: fair; 0.41-0.60: moderate; 0.61-0.80: substantial and 0.81-1.0: almost perfect agreement. Data Availability. The datasets analyzed during the current study are available from the corresponding author on reasonable request.