Introduction
Non-contrast CT is the mainstay in the initial evaluation of patients with suspicion of acute ischemic stroke (AIS) [
1]. However, the sensitivity of non-contrast CT is limited for the detection of acute brain infarct [
2]. The reference standard for infarct detection is MRI with diffusion-weighted imaging (DWI) [
3,
4]. DWI allows for detection of cytotoxic edema within infarcted tissue with high sensitivity, while CT is limited to detecting subtle changes in water content between infarcted and normal brain parenchyma.
Dual-energy CT (DECT) advances CT imaging by improving upon conventional non-contrast head CT (NCT) by acquiring data with two separate energy spectra, which allows improved contrast resolution, reduced image noise and beam-hardening artifacts, and spectral separation of constituent materials at equivalent dose [
5,
6]. Virtual monochromatic CT images (VMI) can be derived from source DECT data [
7‐
9], and reflect the tissue properties of a scan acquired at a single specific monochromatic energy level. Leveraging different monoenergetic reconstructions may help accentuate differences in energy-dependent attenuation differences between similar materials, e.g., normal brain tissue and ischemic/edematous brain tissue. DECT VMI have demonstrated improved contrast-to-noise profiles and reduced beam-hardening artifact relative to NCT [
10‐
15]. The potential of DECT for the visualization of brain edema has been investigated in earlier studies [
16‐
18]. In these studies, elaborate reconstruction methods were required, whereas VMI is a standard derivative of DECT imaging and broadly applicable to any vendor’s software. The use of VMI in the detection of cerebral infarct has not been established. We sought to identify the non-contrast VMI energy which best differentiates normal from infarcted brain parenchyma, and to determine if this VMI can more sensitively and specifically identify infarct in patients with suspected AIS compared with NCT.
Materials and methods
Patients
This study was approved by the Stanford University Institutional Review Board which waived the need for informed consent, and data collection complied with the Health Insurance Portability and Accountability Act. We retrospectively enrolled consecutive patients between October 13, 2018, and April 18, 2019, with suspected AIS who underwent non-contrast DECT and subsequent MRI with DWI within 48 h. Inclusion criteria were as follows: patient age > 18 years and presentation within 24 h of symptom onset. Exclusion criteria were as follows: technical failure of DECT, significant metal artifact limiting interpretation, or corrupted DWI. Baseline clinical data was collected, including age, sex, presentation National Institutes of Health Stroke Scale (NIHSS), time since last known well, time to initial DECT imaging, and time to subsequent MRI.
Imaging protocol
All patients were imaged using a dual-source Somatom Flash CT scanner (Siemens Healthineers, Erlangen, Germany). Non-contrast dual-energy CT protocols are dose neutral with respect to single-energy acquisitions at our institution. The volume CT dose index (CTDIvol) for each scan was 59.8 mGy, and was based on a 16-cm International Electrochemical Commission (IEC) head dosimetry phantom. The scan parameters were as follows: Tube A, 80 kVp and 640 mAs; Tube B, 140 kVp with a tin filter and 320 mAs, beam collimation of 40 × 0.6 mm, a 1.0-s rotation time, matrix size 512 × 512, and a pitch of 1.0. Images were reconstructed at 3 mm using a medium smoothing Q34s kernel. Non-contrast DECT VMI images were reconstructed using the Monoenergetic+ software module in Syngo Via (Siemens Healthineers, Erlangen, Germany).
Study design
Reference standard MRI (DWI) was obtained within 48 h of each patient’s index CT and was reviewed by a single neurointerventional radiologist with 7 years of experience (JJH). DWI imaging parameters included the following: TR 6000 ms, TE 78.2 ms, b-values 0 and 1000, flip angle 90°, and 5-mm slice thickness. Each MRI was scored on a binary scale for presence of cerebral infarct, defined as focal parenchymal restricted diffusion. Cerebral infarcts were binned by location using a modified Alberta Stroke Program Early CT Score (mASPECTS) [
19,
20], in which additional regions corresponding to the posterior circulation (thalamus, superficial PCA, brainstem, and cerebellum) and anterior cerebral artery territory (A1 anteriorly, and A2 posteriorly) are added. The A1 territory is anatomically bounded as the inferior ACA territory, while the A2 territory was bounded by the superior ACA territory.
Phase 1 of this study was used as an initial selection step to identify the VMI reconstruction energies best suited for infarct detection. Phase 2 compared these identified VMI reconstructions against NCT for the detection of cerebral infarct and qualitative infarct localization to identify VMI energies best for differentiation between normal and infarcted brain tissue.
Phase 1 design
In the first phase of the study, VMI were reconstructed at 10 keV increments from 40 to 140 keV. VMI reconstructions were randomized and reviewed by a neuroradiologist and a neurologist (JWD and GZ, with 12 and 19 years of experience in radiology, respectively, and both more than 10 years of experience in stroke imaging) who were blinded to the reconstruction type, but did have CT indication information. Each reader reviewed a full axial image stack for each series with fixed window level and width settings. As attenuation values (HU) vary with tube voltage [
21], the fixed window level (WL) for each reconstruction energy was visually identified by a neuroradiologist (MW) and set to maintain similar attenuation uniformity between series. The settings were as follows: 40–50 keV (WL 50), 60–70 keV (WL 45), 80–110 keV and NCT (WL 40), and 120–140 keV (WL 35). Window width (WW) was set to 50 HU for all reconstructions. Reviewers evaluated for the presence of acute cerebral infarct, and localized infarct foci using the mASPECTS system. Acute infarct was defined as a focal loss of gray-white matter differentiation, focal edema, or hypoattenuation without volume loss. Discrepant reviews were resolved by consensus. A number of VMI reconstructions, who were the most sensitive and specific for the detection of cerebral infarct relative to the reference standard MRI with DWI, were selected and used for comparison with NCT and used to identify the VMI energies that best differentiated normal from infarcted tissue.
Phase 2 design
In the second phase, two different independent reviewers scored a separate, larger cohort of patients presenting with suspicion for AIS. Each examination was reconstructed as a NCT (WL/WW 40/50) and the most sensitive and specific VMI energies identified in phase 1 (60–90 keV), and were presented in a randomized, blinded fashion using the same fixed window and level settings as described in phase 1. An experienced neuroradiologist (JWD) and a neuroradiology fellow (DNW) scored each examination for evidence of acute infarct. Subgroup analyses of patients imaged in the early window (≤ 4.5 h after last-seen-well) and those presenting in the late window (> 4.5 h after last-seen-well) were performed to assess for time-dependent differences in infarct detection. We expect that infarcts in the early window require different improvement in differentiation between similar attenuating tissue (fat, water, and soft tissue), then infarcts in the late window, and might have an influence on the required VMI energy for infarct detection. The volume of cerebral edema increases over time resulting in increased hypoattenuation on CT. At early time points, the contrast between the edema and normal brain tissue is therefore less conspicuous than at later time points. Increasing the contrast between edema and normal brain tissue at different time points may therefore require different energies.
Statistical analysis
Baseline patient characteristics were compared between phase 1 and 2 using a Mann-Whitney signed rank test for continuous variables and a chi-squared test for discrete variables. In phase 1, the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for the detection of cerebral infarct for each scored VMI reconstruction relative to the reference standard MRI with DWI. In phase 2, the sensitivity, specificity, PPV, NPV, and accuracy for the detection of cerebral infarct were calculated for the selected VMI reconstruction and the NCT relative to the reference DWI. Sensitivity and specificity were compared between NCT and VMI using the chi-squared test. Regional infarct localization between DWI and CT was calculated as the infarct location match (ILM), which is defined as the percentage of true positive detections of infarct on CT compared with the total of positive detections for infarct on DWI in an anatomical region. ILM was compared between NCT and VMI for the anterior cerebral artery (ACA; A1 and A2), basal ganglia and insular cortex (BG; caudate, lentiform nucleus, internal capsule and insular cortex), posterior circulation (PCA; thalamus, superficial PCA, cerebellum and brainstem), middle cerebral artery at the level of the basal ganglia (sub-MCA; M1, M2 and M3), middle cerebral artery at the level of the ventricles immediately above the basal ganglia (sup-MCA; M4, M5 and M6), and total of all regions combined using McNemar’s test. Statistical significance was set at α = 0.05. To correct for multiple comparisons, a Bonferroni correction was applied. Inter-rater reliability was evaluated using Cohen’s Kappa, for which the following interpretations were used: slight (0.01–0.20), fair (0.21–0.40), moderate (0.41–0.60), good (0.61–0.80), near perfect (0.81–0.99), and perfect agreement (1). Subgroup analyses of early and the late window patients were performed using the same methods. Statistical analyses were performed in SPSS (version 25.0, IBM, New York).
Discussion
In this study, VMI reconstructions at 90 keV showed higher sensitivity (36.3%) and specificity (94.4%) for the detection of acute cerebral infarct in AIS patients than NCT (32.4% and 91.1%, respectively). This VMI energy is concordant with prior reports suggesting that optimal VMI for maximum contrast resolution is between 60 and 100 keV in adults depending on the DECT approach [
10,
14,
22]. Pomerantz et al. [
10] suggested 65–75 keV for overall parenchymal contrast resolution, with agreement by Neuhaus et al. [
14] at 65 keV for gray-white differentiation, whereas Yoshida et al. [
22] suggested that 99 keV offered superior evaluation of supratentorial acute infarct. Yoshida et al., however, only used confirmed infarct patients; we included AIS patients. By comparing our CT findings with DWI, we were able to identify the best VMI energies for differentiating between normal and infarcted tissue. Additionally, VMI reconstructions at 90 keV also had the highest NPV (56.6 vs 54.2%). As CT is often used as a screening modality in cases where there is concern for AIS, improvements in the sensitivity and NPV offered by VMI may lend confidence to the neuroradiologist in excluding cerebral infarct, particularly if MRI is contraindicated or unavailable.
Subgroup analyses of patients imaged ≤ 4.5 h and > 4.5 h from last-seen-well were performed as infarct conspicuity changes with increasing edema in the maturing lesion. Analysis of patients presenting ≤ 4.5 h from last known well demonstrated a small improvement in sensitivity (26.0 vs 18.0) and NPV (54.8 vs 52.3%), and similar specificity (90.0%) for the detection of acute cerebral infarct using VMI at 80 keV as compared with NCT. Given that patients presenting within this window are considered eligible for intravenous thrombolytic therapy, even small improvements in infarct detection may increase clinical and therapeutic certainty [
23].
Infarct localization relative to reference standard DWI was assessed using ILM for each pre-defined cerebral regions. Within the middle cerebral artery territories (sub-MCA and sup-MCA), a non-significant trend towards superior VMI-ILM at 80–90 keV was observed relative to NCT-ILM. However, overall VMI-ILM showed a significant improvement with 80–90 keV in comparison with total NCT-ILM (28.7, 27.4 vs 19.5%), which suggests that VMI may significantly increase the accuracy of infarct localization. Given regional improvement in infarct detection of VMI seen at 80 keV in the early window within the anterior circulation, assessment of VMI in an anterior large-vessel occlusion population is warranted. To determine if infarct core may be better estimated with VMI than with NCT and could be used for early window treatment decisions [
24], in our study, we see a clear trend towards a significant increase in ILM using VMI; however, our study was underpowered given the number of territories surveyed, and further testing in a larger cohort is warranted.
Relative to NCT, VMI reconstructions at 80–90 keV showed a significant improvement of acute infarct detection. This improvement is most likely due to the qualitatively demonstrated improved soft tissue contrast. This improved contrast therefore increases small inherent attenuation differences between gray matter, white matter, and edematous tissue compared with NCT. We hypothesize that the increase in contrast is due to increased differences between attenuation of gray and white matter and to the lower noise levels in 80 and 90 keV VMI [
15,
22]. The range of best VMI energies is 80–90 keV, and we recommend individual institutions to test their own optima, but 90 keV generally has a slightly better sensitivity and specificity for infarct detection compared with 80 keV. Infarct location detection is slightly better with 80 keV. However, if dual-energy CT is unavailable, similar results are unlikely to be obtained by modifying single-energy CT protocols to an average energy of 90 keV, as this will involve an increase in peak tube voltage and a substantial decrease in exposure to maintain a similar dose level. This increase in kVp will result in a decrease in contrast, because there will be a large number of photons with an energy higher than 90 keV which will not attenuate in tissue.
Our study has several limitations. First, our retrospective study design may introduce bias. The small sample size may limit our statistical analysis, particularly for analysis of ILM. Second, our sensitivity analysis may be artificially increased as the imaging reviewers were aware that the testing population was enriched for patients with cerebral infarct. Third, a recognition bias can be assumed, resulting in an underestimation of the differences between the reconstruction types. We, however, blinded our reviewers to reconstruction type, CT reconstructions were presented in a random fashion, and the large amount of CT reconstructions in the comparison ensured that no single energy level was favored over the other. Lastly, we investigated the applicability of VMI using a single type of DECT acquisition (dual-source CT scanner). DECT acquisitions, VMI techniques, and reconstruction algorithms vary between manufacturers and types of DECT scanners. As a result, monochromatic energy levels cannot be reproduced between manufacturers or post-processing algorithms, thereby limiting study generalizability and indicating that different institutions may need to determine VMI optima independently [
25].
In conclusion, use of virtual monochromatic images at 80–90 keV with non-contrast dual-energy CT results in a modest improvement of sensitivity and specificity for the detection of acute infarcts in suspected ischemic stroke patients relative to conventional non-contrast CT.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit
http://creativecommons.org/licenses/by/4.0/.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.