Carotid atherosclerotic plaque characterization using dual-energy computed tomography angiography and its correlation with cerebral ischemic stroke Authors: Zhiyan Ji, Xueping He, Xingzhe Wei, Xianjie Chen, Yaoguo Li, Yufeng Ye Published: 2026 ABSTRACT BackgroundCarotid plaque vulnerability is a key risk factor for cerebral ischemic stroke (CIS). This study aimed to investigate the value of dual-energy computed tomography angiography (CTA) for carotid plaque diagnosis and its associated CIS occurrence.MethodsIn this retrospective study, 103 patients underwent the initial dual-energy CTA for carotid plaque identification and the following brain magnetic resonance imaging (MRI) for CIS detection. The types and components of carotid plaques were qualitatively and quantitatively characterized using spectral parameters including virtual monoenergetic imaging, spectral curve with slope determined as (CT90keV − CT40keV) HU/(90−40) keV, and effective atomic number. These parameters were further correlated with CIS occurrence.ResultsAmong 232 carotid plaques detected, the number of lipid, lipid-rich necrotic core, mixed lipid–fibrous, fibrous, hemorrhagic, and calcified plaques was 39, 25, 49, 44, 17, and 58, respectively. The corresponding spectral curve slope was 0.31±0.17, 0.27±0.12, 0.25±0.21, –0.99±0.35, –1.78±0.43, and –29.75±2.68 Hounsfield units (HU)/keV, respectively, with statistically significant differences among lipid-containing, fibrous, hemorrhagic, and calcified plaques (P<0.01). The corresponding effective atomic number was 6.85±0.87, 7.01±0.64, 7.18±1.05, 8.15±0.62, 8.31±0.73, and 13.54±1.25, respectively, with statistically significant differences among lipid-containing, fibrous or hemorrhagic, and calcified plaques (P<0.01). The corresponding CIS proportion was 52.63% (10/19), 45.45% (5/11), 33.33% (8/24), 85.71% (6/7), 11.76% (2/17), and 0% (0/25), respectively, indicating that plaque composition was closely related to CIS risk (i.e., the more hemorrhagic and lipid content in plaques, the higher stroke risk).ConclusionsDual-energy CTA can be employed to analyze components and types of carotid plaques using multiple spectral parameters (especially the spectral curve slope), thus offering a promising clinical tool for CIS prevention. Introduction Among non-communicable disorders, stroke remains the second leading cause of death and the third leading cause of death and disability combined worldwide, with a disproportionately high burden residing in low- and middle-income countries (1-3). Driven by the huge population, accelerating demographic aging, and the suboptimal control of modifiable risk factors, China carries the heaviest global stroke burden and further increases are projected in the coming decades (4-6). Therefore, rigorous and sustained efforts are required to strengthen comprehensive stroke care across the continuum—from primary prevention through acute treatment to long-term management—especially for older adults (7-9). Carotid atherosclerotic plaque, which develops from the lipid accumulation (i.e., fatty deposits) within the carotid arterial wall and progresses through the accretion of multiple components—including necrotic cellular debris, fibrous tissue, and calcium—is a key risk factor for cerebral ischemic stroke (CIS) (10-12). Carotid plaques can produce critical luminal stenosis with consequent hypoperfusion, and unstable plaques may rupture or shed debris to embolize intracranial vessels, both mechanisms capable of triggering CIS (13,14). Growing evidence has shifted the focus from plaque burden/luminal stenosis toward plaque composition and vulnerability: it is increasingly accepted that the CIS risk is driven less by how narrow the artery becomes and more by the composition of the plaque—particularly lipid-rich necrotic core and intraplaque hemorrhage—which indicate instability (15-17). Unstable lipid and hemorrhagic plaques are high-risk factors for CIS and thus require proactive lipid-lowering therapy, whereas stable fibrous and calcified plaques pose a lower CIS risk and thus require only standard risk factor management. Therefore, non-invasive imaging that can accurately characterize carotid plaque composition and stability holds significant clinical value for guiding CIS prevention and treatment strategies (18-20). Carotid duplex ultrasonography remains the most widely used first-line modality for carotid plaque detection and surveillance because it is non-invasive, relatively low-cost, repeatable, and broadly available (21). It can delineate plaque location, size, and surface morphology (e.g., ulceration) and provides a semi-quantitative impression of composition based on echogenicity. However, its quantitative accuracy is limited by acoustic shadowing from dense calcification and by operator-dependent factors—including probe positioning/angulation, Doppler angle of insonation, color-gain settings, selected imaging planes, and interpretative experience—so its role is strongest in initial screening and follow-up rather than definitive tissue characterization (22). Traditional contrast-enhanced computed tomography angiography (CTA) excels at depicting calcification and defining luminal geometry. Nevertheless, because conventional CTA yields only polyenergetic computed tomography (CT) numbers (Hounsfield units, HU), the substantial overlap in attenuation among lipid-rich, fibrous, and hemorrhagic components limits its reliability for differentiating non-calcified plaque composition and vulnerability, especially when heavy calcification causes beam-hardening/blooming artifacts (23). High-resolution vessel-wall magnetic resonance imaging (MRI) using multiple weightings (black-blood techniques) can directly characterize plaque internal structure and key components (lipid-rich necrotic core, fibrous cap, intraplaque hemorrhage, and calcification). However, it involves longer scan times, higher costs, and is not feasible in patients with non-MRI-conditional ferromagnetic implants (24). As an important CT commercial development since 2006, dual-energy (spectral) CT has been developed through several novel technologies including single-tube rapid kVp switching, dual-source, dual-layer detector, twin-beam split-filter, and photon-counting detector (25-29). Unlike conventional polyenergetic (single-energy) CT, which provides only a mixed-energy attenuation value (HU), dual-energy CT acquires data at two distinct energy spectra, enabling material decomposition imaging, virtual monoenergetic imaging to reduce beam-hardening and blooming artifacts, spectral attenuation-curve analysis, and effective atomic number mapping—capabilities that collectively permit quantitative tissue characterization beyond simple density measurement (30,31). A limited but growing body of studies has employed spectral CTA validated against ex vivo histopathology (carotid endarterectomy specimens) to differentiate plaque components—calcified, fibrous, lipid-rich necrotic core, and intraplaque hemorrhage—supporting its feasibility for non-invasive plaque typing (32,33). However, further evidence is still required regarding its performance, particularly its differentiation capability for unstable plaques, before spectral CTA can be translated into routine clinical decision-making. This single-center retrospective observational study evaluated the utility of dual-energy CTA for characterizing carotid plaque composition and quantifying plaque phenotypic features (spectral curve slope and effective atomic number) in patients with carotid atherosclerotic disease. The correlation between the initial carotid plaque characterization and the following CIS occurrence in all patients was further analyzed. The aim was to generate a clinically annotated imaging dataset and to test the association between spectral CTA-derived plaque parameters and CIS occurrence, thereby informing whether spectral CTA can support non-invasive plaque risk stratification beyond luminal stenosis alone. We present this article in accordance with the STROBE reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2789/rc). Methods General patient data and study design This was a retrospective single-center study conducted from September 2023 to August 2025 at The Affiliated Panyu Central Hospital of Guangzhou Medical University, in Guangzhou, South China. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Institutional Ethics Committee of The Affiliated Panyu Central Hospital of Guangzhou Medical University (No. PY-LL-2023-0457). The requirement for written informed consent was waived by the Institutional Ethics Committee, because all patient data were recorded and stored anonymously. The eligibility criteria were as follows: (I) patients had at least one carotid atherosclerotic plaque at the bulb or proximal internal carotid artery demonstrated on dual-energy CTA, with no evidence of acute cerebral infarction on a baseline diffusion-weighted imaging (DWI)-MRI obtained at the same presentation; (II) patients underwent a brain MRI (including DWI) within following 6 months to capture interval CIS events; (III) CTA acquisitions had to cover common carotid arteries, internal carotid arteries, and vertebral arteries, provide adequate contrast enhancement and analyzable image quality, show no congenital arch/vessel anomaly, and correspond to luminal stenosis <70% by North American Symptomatic Carotid Endarterectomy Trial (NASCET) criteria; (IV) patients presented with a focal neurological symptom referable to the carotid territory (e.g., unilateral limb weakness/numbness, dysarthria, facial asymmetry, amaurosis fugax). Under this design, any new DWI lesion appearing on the interval MRI could be interpreted as an incident carotid-related CIS event. CTA and MRI methodology Head and neck CTA was performed using a gemstone dual-energy CT scanner (GE Discovery CT750 HD; GE Healthcare, Chicago, IL, USA), with patients in the supine position, and the scanning range starting from the aortic arch to the top of the skull. Non-contrast CT scan was performed first, followed by Gemstone Spectral Imaging (GSI) fast kVp switching enhanced scan. The scanning parameters were as follows: tube voltage switching between 80 and 140 kVp with a switching interval of ≤0.5 ms, tube current 630 mA, 0.5 seconds per rack rotation, detector width 64×0.625 mm, scan field of view (FOV) 25.0 cm, and helical acquisition pitch 0.984. The contrast agent was iopamidol with dosage of 1.5 mL/kg and injected via the right elbow vein using a high-pressure syringe at a speed of 4.5 mL/s. The middle of the ascending aorta was selected as the region of interest (ROI), with a threshold set at 100 HU. After the scan reached the peak, a delay of 5 seconds was applied, and scanning was performed using GSI enhanced mode. Brain MRI was acquired on a 1.5‑T Siemens Magnetom Avanto system (Siemens, Erlangen, Germany) and included DWI [with apparent diffusion coefficient (ADC) maps], fluid-attenuated inversion recovery (FLAIR), T2-weighted imaging (T2WI), and T1-weighted imaging (T1WI) sequences. The presence of acute infarction was determined by DWI‑restricted diffusion (DWI ↑/ADC ↓); the pattern of DWI‑FLAIR mismatch (DWI‑positive, FLAIR‑negative/equivocal) was additionally noted when onset time was uncertain. Chronic infarctions were identified on T2WI/T1WI. Image analysis and processing Raw spectral CTA data were reconstructed at a section (slice) thickness of 1.25 mm and reconstruction interval (slice spacing) of 0.625 mm, then transferred to a GE Advantage Workstation 4.6 for post-processing and vascular analysis with quantitative measurements performed using the GSI Viewer software. For each identified plaque, an ROI (≥2 mm2) was manually drawn on the axial slice showing the largest plaque cross-sectional area and on the immediate adjacent slice above and below (three slices in total), carefully avoiding gross calcification and plaque margins to minimize partial-volume effects and covering the key plaque components as comprehensively as possible. For each ROI, the average CT attenuation values across the 40–140 keV virtual monoenergetic series (typically at 10-keV increments) were exported from the workstation to generate a spectral attenuation (HU–keV) curve, and the effective atomic number (Zeff, dimensionless) was recorded from the Zeff histogram (frequency vs. Zeff). Separate small reference ROIs were placed within adjacent subcutaneous fat, ipsilateral neck musculature, and cortical bone to provide internal attenuation baselines. The plaque composition was determined by comparing its spectral curve with the above-mentioned known tissues (32,34,35): (I) if the plaque curve followed a similar trend to fat (showing an convex shape) and CT value was below 0 HU, it was a lipid plaque (shown in Figure 1); (II) if the plaque curve followed a similar trend to fat and CT value was distributed above and below 0 HU, it was a mixed plaque containing lipid and fibrous components, where the plaque was mainly composed of lipid-rich necrotic core with the mean CT value below 0 HU (shown in Figure 2), and the plaque was a mixed lipid and fibrous plaque with the mean CT value above 0 HU (shown in Figure 3); (III) if the plaque curve followed a similar trend to musculature (showing a mild concave shape) and CT value was above 0 HU, the plaque was a fibrous plaque (shown in Figure 4); (IV) if the plaque curve followed a similar trend to soft tissue (showing a moderate concave shape) and CT value was above 0 HU, it was a hemorrhagic plaque (shown in Figure 5); (V) if the plaque curve followed a similar trend to bone (showing a pronounced concave shape) and CT value was above 130 HU, it was a calcified plaque (shown in Figure 6). The slope of the spectral curve was calculated using the two-point method on the curve corresponding to 40 and 90 keV [i.e., (CT90keV − CT40keV) HU/(90−40) keV], which captured most changes of CT value with full mono-energy from 40 to 140 keV and would increase the sensitivity of slope calculation (36). According to the plaque classification of the American Heart Association (37), lipid plaques, lipid/fibrous mixed plaques and hemorrhagic plaques were classified as unstable plaque groups, whereas fibrous plaques and calcified plaques were classified as stable plaque groups. Representative dual-energy CTA work-up of a lipid carotid plaque in a 50-year-old man. (A) ROI was delineated on the left carotid plaque [1] and adjacent fat tissue [2]; (B) the spectral HU curve of the carotid plaque with CT value <0 HU showed a similar convex trend to the adjacent fat tissue, identifying it as a lipid plaque with the slope of 0.24 HU/keV determined by two-point (40 and 90 keV) method; (C) the histogram showed that the carotid plaque had the effective atomic number with the mean value of 7.38 and the range of 6.85–8.01, very similar to the adjacent fat tissue. CT, computed tomography; CTA, computed tomography angiography; HU, Hounsfield units; ROI, region of interest. Representative dual-energy CTA work-up of a lipid-rich necrotic core carotid plaque in a 67-year-old man. (A) ROI was delineated on the right carotid plaque [1] and adjacent fat tissue [2]; (B) the spectral HU curve of the carotid plaque with mean CT value <0 HU showed a similar convex trend to the adjacent fat tissue, identifying it as a lipid-rich necrotic core plaque with the slope of 0.32 HU/keV determined by two-point (40 and 90 keV) method; (C) the histogram showed that the carotid plaque had the effective atomic number with the mean value of 7.45 and the range of 7.24–7.70, similar to the adjacent fat tissue. CT, computed tomography; CTA, computed tomography angiography; HU, Hounsfield units; ROI, region of interest. Representative dual-energy CTA work-up of a mixed lipid and fibrous carotid plaque in a 60-year-old woman. (A) ROI was delineated on the right carotid plaque [1] and adjacent fat tissue [2]; (B) the spectral HU curve of the carotid plaque with mean CT value >0 HU showed a similar convex trend to the adjacent fat tissue, identifying it as a mixed lipid and fibrous plaque with the slope of 0.26 HU/keV determined by two-point (40 and 90 keV) method; (C) the histogram showed that the carotid plaque had the effective atomic number with the mean value of 7.25 and the range of 6.48–8.15, similar to the adjacent fat tissue. CT, computed tomography; CTA, computed tomography angiography; HU, Hounsfield units; ROI, region of interest. Representative dual-energy CTA work-up of a fibrous carotid plaque in a 62-year-old woman. (A) ROI was delineated on the right carotid plaque [1] and adjacent fiber tissue [2]; (B) the spectral HU curve of the carotid plaque with CT value >0 HU showed a similar concave trend to the adjacent musculature tissue, identifying it as a fibrous plaque with the slope of –1.06 HU/keV determined by two-point (40 and 90 keV) method; (C) the histogram showed that the carotid plaque had the effective atomic number with the mean value of 8.15 and the range of 8.01–8.27, very similar to the adjacent musculature tissue. CT, computed tomography; CTA, computed tomography angiography; HU, Hounsfield units; ROI, region of interest. Representative dual-energy CTA work-up of a hemorrhagic carotid plaque in a 71-year-old man. (A) ROI was delineated on the left carotid plaque [1] and adjacent soft tissue [2]; (B) the spectral HU curve of the carotid plaque with CT value >0 HU showed a similar concave trend to the adjacent soft tissue, identifying it as a hemorrhagic plaque with the slope of –2.60 HU/keV determined by two-point (40 and 90 keV) method; (C) the histogram showed that the carotid plaque had the effective atomic number with the mean value of 8.77 and the range of 8.51–8.93, very similar to the adjacent soft tissue. CT, computed tomography; CTA, computed tomography angiography; HU, Hounsfield units; ROI, region of interest. Representative dual-energy CTA work-up of a calcified carotid plaque in a 60-year-old woman. (A) ROI was delineated on the left carotid plaque [1] and adjacent bone tissue [2]; (B) the spectral HU curve of the carotid plaque with CT value >130 HU showed a similar convex trend to the adjacent bone tissue, identifying it as a calcified plaque with the slope of –31.32 HU/keV determined by two-point (40 and 90 keV) method; (C) the histogram showed that the carotid plaque had the effective atomic number with the mean value of 13.05 and the range of 12.21–13.55, very similar to the adjacent bone tissue. CT, computed tomography; CTA, computed tomography angiography; HU, Hounsfield units; ROI, region of interest. Statistical analysis Both spectral curve slope and effective atomic number as continuous variables were tested for normality using the Shapiro-Wilk test. Normally distributed parameters were presented as mean ± standard deviation (SD). Comparisons of these parameters among all plaque composition groups were performed using one-way analysis of variance (ANOVA), followed by post hoc pairwise comparisons. When only two prespecified subgroups were compared, an independent-samples t-test was used. All analyses were performed using the software SPSS 21.0 (IBM Corp., Armonk, NY, USA), and two-sided P<0.05 was considered statistically significant. Results Carotid plaques analysis According to the above-mentioned eligibility criteria, 103 patients were successfully enrolled with carotid luminal stenosis of 35.3–68.5% with mean value of 52.3±10.5%, including 55 males and 48 females, aged 35–78 years with a median age of 61±11 years. Carotid plaque detection results by GSI dual-energy CTA are shown in Table 1. There were a total of 232 carotid plaques detected among the 103 patients with 1–4 plaques per patient, which were classified into 39 lipid plaques, 25 lipid-rich necrotic core plaques, 49 mixed lipid and fibrous plaques, 44 fibrous plaques, 17 hemorrhagic plaques, and 58 calcified plaques. The corresponding spectral curve slopes were 0.31±0.17, 0.27±0.12, 0.25±0.21, −0.99±0.35, −1.78±0.43, and −29.75±2.68 HU/keV, respectively. There was no statistically significant difference in the slopes among lipid-containing plaques (lipid plaques, lipid-rich necrotic core plaques, mixed lipid and fibrous plaques) (P=0.21–0.61), but there was a statistically significant difference in the slopes among lipid-containing plaques, fibrous plaques, hemorrhagic plaques, and calcified plaques (P<0.01). The corresponding effective atomic numbers were 6.85±0.87, 7.01±0.64, 7.18±1.05, 8.15±0.62, 8.31±0.73, and 13.54±1.25, respectively. There was no statistically significant difference in the effective atomic numbers among lipid-containing plaques (lipid plaques, lipid-rich necrotic core plaques, mixed lipid and fibrous plaques) (P=0.18–0.72). There was also no statistically significant difference in the effective atomic numbers between fibrous plaques and hemorrhagic plaques (P=0.35). However, there was a statistically significant difference in the effective atomic numbers among lipid-containing plaques, fibrous or hemorrhagic plaques, and calcified plaques (P<0.01). Spectral curve slope (P<0.01) showed better differentiation capability between fibrous and hemorrhagic plaques than effective atomic number (P=0.35) in this study. Table 1Summary of carotid plaques detected by GSI spectral CTA among all patientsItemLipid plaqueLipid-rich necrotic core plaqueMixed lipid and fibrous plaqueFibrous plaqueHemorrhagic plaqueCalcified plaqueNumber of detection392549441758Spectral curve slope (HU/keV)0.31±0.170.27±0.120.25±0.21−0.99±0.35−1.78±0.43−29.75±2.68Effective atomic number6.85±0.877.01±0.647.18±1.058.15±0.628.31±0.7313.54±1.25P for pairwise comparison (spectral curve slope/effective atomic number) Lipid plaque–0.42/0.530.21/0.18<0.01/<0.01<0.01/<0.01<0.01/<0.01 Lipid-rich necrotic core plaque0.42/0.53–0.61/0.72<0.01/<0.01<0.01/<0.01<0.01/<0.01 Mixed lipid and fibrous plaque0.21/0.180.61/0.72–<0.01/0.01<0.01/<0.01<0.01/<0.01 Fibrous plaque<0.01/<0.01<0.01/<0.01<0.01/0.01–<0.01/0.35<0.01/<0.01 Hemorrhagic plaque<0.01/<0.01<0.01/<0.01<0.01/<0.01<0.01/0.35–<0.01/<0.01 Calcified plaque<0.01/<0.01<0.01/<0.01<0.01/<0.01<0.01/<0.01<0.01/<0.01–Data are presented as number, mean ± standard deviation, or value. CTA, computed tomography angiography; GSI, Gemstone Spectral Imaging; HU, Hounsfield units. Summary of carotid plaques detected by GSI spectral CTA among all patients Data are presented as number, mean ± standard deviation, or value. CTA, computed tomography angiography; GSI, Gemstone Spectral Imaging; HU, Hounsfield units. Correlation between carotid plaques and CIS CIS occurrence on brain MRI and its correlation with carotid plaque characteristics are shown in Table 2. After excluding plaque cross-references and selecting one index plaque (the most unstable one) per patient, the patient number with lipid plaque, lipid-rich necrotic core plaque, mixed lipid and fibrous plaque, fibrous plaque, hemorrhagic plaque, and calcified plaque was 19, 11, 24, 7, 17, and 25, respectively. The corresponding number of patients with CIS detected by brain MRI was 10, 5, 8, 6, 2, and 0, respectively, thus resulting in that the proportion of CIS in each plaque group was 52.63%, 45.45%, 33.33%, 85.71%, 11.76%, and 0%, respectively, and the proportion of CIS in the combined unstable and stable plaque group was 47.54% and 4.76%, respectively. Table 2Summary of CIS occurrence with different carotid plaques among all patientsItemUnstable plaquesStable plaquesLipid plaqueLipid-rich necrotic core plaqueMixed lipid and fibrous plaqueHemorrhagic plaqueFibrous plaqueCalcified plaquePatient number of plaque detection19112471725Patient number of CIS detection1058620CIS proportion (%)52.6345.4533.3385.7111.760.0047.544.76CIS, cerebral ischemic stroke. Summary of CIS occurrence with different carotid plaques among all patients CIS, cerebral ischemic stroke. Discussion Both effective atomic number and spectral curve slope derived from dual-energy CTA yielded quantitative parameters that enabled non-invasive characterization of carotid plaque composition. A 2025 research study (38) showed that the hemorrhagic plaques, lipid-rich necrotic core plaques, fibrous plaques, and calcified plaques had effective atomic numbers of 6.5–7.5, 7.5–8.75, 8.5–11.3, and 11.5–14.4, respectively, showing overall consistency with this study and another study (32) except for hemorrhagic plaques. Several studies (35,39-41) have found that the carotid plaques in the patients with and without CIS had mean effective atomic numbers of 7.33–8.95 and 7.40–12.02, showing overall consistency on the unstable plaques with lower effective atomic number. This study found a better differentiation performance for carotid plaques by spectral curve slope than effective atomic number in spectral CTA, especially between fibrous and hemorrhagic plaques. Another 2025 study (36) employing the same two-point (40 and 90 keV) method for spectral curve slope determination of carotid plaques also demonstrated that the spectral curve slope in patients with and without CIS was 0.36±0.81 and 1.88±1.27 HU/keV with statistically significant difference, whereas the effective atomic number was 6.78 and 6.70, without statistically significant difference. In addition, the reported spectral curve slope of 0.36±0.81 HU/keV in that study (36) was similar to the 0.31±0.17 HU/keV from the lipid plaque in this study, showing the consistency of this unstable plaque with high risk for CIS. One study employing spectral curve slope determination via a two-point (40 and 110 keV) method found that the mean spectral curve slope of unstable and stable carotid plaques were 0.29 and −6.55 HU/keV (34), showing the consistency with the unstable lipid-containing plaque and stable calcified plaque in this study. One study employing spectral curve slope determination via a two-point (40 and 140 keV) method showed that the lipid-rich necrotic core plaque, hemorrhagic plaque, and fibrous plaque had slopes of 0.1±0.3, −0.42±0.42 and −0.54±0.42 HU/keV (32), respectively, which is in overall agreement with this study. Another study employing spectral curve slope determination via a two-point (40 and 140 keV) method found that the mean spectral curve slopes of unstable and stable carotid plaques were −0.91 and −1.92 HU/keV (41), which are similar to unstable hemorrhagic plaque and stable calcified plaque in this study. One study showed that the mean spectral curve slope of carotid plaque in patients with and without CIS was 3.86 and 19.72 HU/keV, but no calculation method was provided, making it difficult to compare these values directly with the corresponding results in this study. From the above-mentioned analysis, it is critical to calculate the spectral curve slope data via the same method for better interpretation, and the two-point (40 and 90 keV) method could be suitable considering that most changes of CT value with mono-energy occurs at the low energy zone. The composition of carotid plaques was found to be a more important risk factor triggering CIS than its size (15-17). This study provides further evidence that although the carotid artery stenosis degree of all 103 patients did not exceed 70%, the overall proportion of CIS was still as high as 30.10%. Among different carotid plaques in this study, hemorrhagic plaque resulted in the highest CIS proportion of 85.71% and thus was the highest risking plaque, showing the consistency with previous reports (10,11,17,23). Lipid-containing plaques including lipid plaque, lipid-rich necrotic core plaque, and mixed lipid and fibrous plaque, resulted in the moderate CIS proportion of 52.63%, 45.45%, 33.33%, respectively, and showed a positive correlation with lipid content in this study, which was also in agreement with previous studies, especially for lipid-rich necrotic core plaque (11,15,32,38). Fibrous plaque and calcified plaque resulted in the very low CIS proportion of 11.76% and 0% in this study, matching the well-known facts on both plaques as stable plaques with low risk of CIS (9,12). Thus, plaque composition was closely related to the risk of CIS; the more hemorrhagic and lipid content in the plaques, the worse the stability and the higher the CIS risk. The combined unstable plaques (including lipid-containing and hemorrhagic plaque) and stable plaques (including fibrous and calcified plaque) resulted in CIS proportions of 47.54% and 4.76% in this study, indicating the benefits of carotid plaque identification via non-invasion spectral CTA for healthcare management. This study demonstrated that there was statistically significant difference in the spectral curve slope from dual-energy CTA among lipid-containing plaque, hemorrhagic plaque, fibrous plaque, and calcified plaque. However, both spectral curve slope and effective atomic number failed to distinguish the subgroups of lipid-containing plaque effectively in this study. Besides spectral curve slope and effective atomic number, other parameters (e.g., electron density, virtual non-contrast, CT value at mono-energy) available from dual-energy CTA can be combined into the carotid plaque analysis in future studies with pathologic comparison. In addition, this study only focused on carotid plaque characteristics and the correlation with CIS occurrence. However, there are other parameters (e.g., blood pressure, blood biochemical indices, habits) besides carotid plaques characteristics affecting the occurrence of CIS, which can be integrated into future studies. Conclusions This retrospective study evaluated carotid plaque characterization by dual-energy CTA and its association with CIS occurrence confirmed by brain MRI. Effective atomic number could differentiate among the lipid-containing, fibrous or hemorrhagic, and calcified plaques effectively. Spectral curve slope could further differentiate fibrous from hemorrhagic plaques effectively. However, both effective atomic number and spectral curve slope failed to distinguish effectively among the subclasses of lipid-containing plaques (i.e., lipid, lipid-rich necrotic core, mixed lipid-fibrous plaques). The hemorrhagic and lipid-containing plaques resulted in significantly higher CIS occurrence than the fibrous and calcified plaques, demonstrating the correlation between unstable plaques and CIS. Thus, dual-energy CTA with multiple parameters for carotid plaque diagnosis would be a promising risk assessment tool for CIS prevention. Supplementary The article’s supplementary files as