Optimizing carotid plaque visualization: the advantage of subtraction imaging from dual-layer spectral CT

Abstract

Background

Computed tomography angiography (CTA) has become the primary method for detecting carotid atherosclerosis. However, the visibility of the arterial wall on the CTA is poor because of the high attenuation lumen filled with contrast agent. To optimize the display of the carotid arterial walls, this study proposed a black blood technique by generating subtraction images (CTA_sub) based on dual-layer spectral CT, and explored its role in the assessment of carotid atherosclerotic plaques.

Materials and methods

CTA_sub were generated by subtracting virtual monochrome images (VMI) from virtual non-contrast images (VNC) obtained from SDCT scans. Two senior radiologists and two junior radiologists read conventional CTA images (CTA_con) with and without CTA_sub respectively. The inter-reader agreement for plaque classification of each extracranial carotid artery and diagnostic confidence were compared. The extracranial carotid wall visibility, outer margin conspicuity of non-calcified components, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of the non-calcified components as well as the stenosis rate at the narrowest level were compared between CTA_sub and CTA_con. Using high-resolution magnetic resonance imaging (hrMRI) as the standard, the accuracy in plaque burden estimation on CTA_con and CTA_sub were compared. The paired t-test and the Wilcoxon signed-rank test were used for the comparison between CTA_sub and CTA_con. The Cohen κ value and the intraclass correlation coefficient (ICC) were used to evaluate the consistency.

Results

A total of 119 patients were included in the comparisons between CTA_sub and CTA_con and a total of 31 patients were enrolled in the plaque burden comparison between CTA and hrMRI. CTA_sub improved the consistency of plaque classification between radiologists and increased their diagnostic confidence. CTA_sub also provide better arterial wall visibility and clearer outer margin conspicuity of non-calcified components. The SNR and CNRs of non-calcified components were obviously elevated on CTA_sub (SNR: 12.86 ± 6.51 vs. 7.77 ± 4.30; CNR_{non−calcified/fat}: 33.13 ± 2.65 vs. 24.96 ± 7.90; CNR_{non−calcified/muscle}: 8.58 ± 4.27 vs. 6.64 ± 4.11, all p < 0.001). Moreover, CTA_sub displayed good to excellent consistency with CTA_con in stenosis rate measurement (NASCET: ICC = 0.919; ECST: ICC = 0.889, both p < 0.001) and a more closer estimation of plaque burden to hrMRI than CTA_con(ICC = 0.938 vs. 0.842, p < 0.001).

Conclusions

As a complement to CTA_con, CTA_sub not only improves visualization of the extracranial carotid wall and non-calcified plaque components but also enables accurate quantification of stenosis rate and plaque burden. Howerer, CTA_sub can not be used alone when calcified components are evaluated.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12880-026-02596-6.

Background

Carotid atherosclerotic disease is the main cause of ischemic stroke [1]. High-resolution magnetic resonance imaging (hrMRI) and PET-CT have better performance in revealing vulnerable plaques, but long-time procedure, high fee as well as numerous contraindications pose challenges to their widespread adoption [1–4]. Computed tomography angiography (CTA) has become the primary method for detecting atherosclerotic diseases because of its exceptional resolution and rapid scanning speed [5, 6]. Lumen filled by contrast agent on the CTA is conducive to the accurate measurement of stenosis. However, the display of the arterial wall on the CTA is not satisfactory because of the “vague” effect of the high attenuation lumen to the inner boundary of the wall and the lack of contrast between the outer boundary and its surrounding tissues, resulting in a limited ability to identify early lesions of the arterial wall.

In recent years, scholars have successively explored the dark blood technology based on CT and its role in plaque assessment. For example, Li et al. [7] found that the contrast enhancement boost (CE boost)-derived dark-blood imaging can improve the visibility for both calcified and non-calcified plaques. Lu et al. [8] obtained a more accurate evaluation of the carotid stenosis rate and plaque burden on black-blood CT generated by subtracting arterial-phase images from delayed-phase images. Zhang et al. [9] proved that the black-blood technique of spectral CT can increase the detectability of intracranial plaques with an accuracy similar to that of hrMRI.

There are two main theories for generating black blood images. One theory is subtraction technique [7, 8], which requires multiple scans and results in an additional radiation dose and a complex reconstruction process. Another approach is material decomposition technique based on dual-energy CT, which uses the differences in the attenuation of various substances at high- and low-energy levels to project all voxels onto a coordinate system derived from water and iodine, ultimately achieving the enhancement of the arterial wall [9, 10]. However, this method requires substantial calculations and cannot be implemented at commercially available workstations. The dual-layer spectral detector CT (SDCT) based on detectors adopts a single X-ray source-detector system to achieve spectral separation at the detector level. Virtual non-contrast (VNC) and virtual monochrome (VMI) images generated from the same dataset have complete spatial and temporal alignment. Therefore, a major advantage is the complete elimination of mismatched artifacts [11]. Inspired by the subtraction technique, we propose a spectral CT-based subtraction method (CTA_sub) by subtracting VMI from VNC images. Previous studies have explored its role in stenosis rate assessment and stent subtraction in coronary arteries [12, 13]. However, to the best of our knowledge, no research has explored its application in carotid artery plaques. This study aimed to explore the ability of CTA_sub to evaluate carotid plaques and compare it with that of conventional CTA (CTA_con) and hrMRI.

Methods

Study population

A total of 145 consecutive patients who underwent spectral craniocervical CTA from February to April 2025 with suspected cerebral or carotid artery diseases were initially enrolled in this study. The exclusion criteria included: underwent carotid stent implantation (n = 9), severe motion artifacts (n = 6), carotid artery occlusion (n = 5), carotid artery dissection (n = 4), and carotid artery malformation (n = 2).

A total of 65 patients underwent both craniocervical CTA and hrMRI examinations from June 2024 to December 2025 were initially enrolled for plaque burden analysis. The exclusion criteria included: no plaques or only calcified plaques in extracranial carotid arteries within the scanning range (n = 15), interval between the two examinations exceeded two months (n = 7), surgical or interventional treatment was performed during the interval between the two examinations (n = 7), poor image quality (n = 5). The flowchart illustrates patient inclusion and exclusion (Fig. 1).

Fig. 1.

Flowchart of patient inclusion and exclusion

SDCT protocols and CTA_sub generation

All patients underwent craniocervical CTA scanning on a dual-layer spectral detector CT system (IQon, Philips Healthcare, Netherlands). Nonionic iodine contrast agent (iopromide 370 mgI/ml, Bayer, Germany) was injected at a dose of 0.8 ml/kg via the antecubital vein by a high-pressure syringe, and then followed by a saline flush of 30 mL (both 4.5 mL/s). The acquisition was triggered by the bolus tracking technique with a trigger threshold of 120 HU in the aortic arch. Data acquisition started 4.5 s after triggering and covered from the aortic arch to the vertex. The following scan presets were used: tube voltage, 120 kVp; pitch, 0.992; rotation time, 0.5 s; and collimation, 128 × 0.625 mm. Images were reconstructed with a slice thickness of 0.9 mm and a section increment of 0.7 mm.

Conventional images, VMI at 70 kilo electron-volts (VMI_{70 keV}) and VNC images were extracted from one spectral-based image dataset via a dedicated postprocessing workstation (Intellispace Portal 9.0, Philips Healthcare, Netherlands). When the images were subtracted, two sets of images were sorted by Z position. First, the VNC images was subtracted from the VMI_{70 keV} images to obtain a set of vessel images (Vessel = VMI-VNC). Then, CTA_sub was generated by subtracting the vessel images from the VNC images (CTA_sub=VNC-Vessel). (Additional file 1: Fig. S1). Similar to Qin’s [12] method, bias (HU) was used to compensate for the loss of CT values resulting from the subtraction technique. CTA_con was observed with a window level (WL) of 300 HU and a window width (WW) of 1000 HU, whereas the WL and WW of CTA_sub were set to 40 HU and 400 HU, respectively, to optimize the demonstration of the vessel wall.

HrMRI scan techniques

HrMRI scans were performed on a 3T scanner (Prisma, Siemens Healthineers) with a 64-channel coil. All sequences were performed at identical scan planes covering 2 cm below the bifurcation to the top of the skull. The imaging sequences included T1-weighted imaging (TIWI) and contrast-enhanced T1-weighted imaging (CE-T1WI). T1WI using different flip angle evolutions (SPACE) sequence: TR 800 ms, TE 16 ms, FOV 250 mm164 mm, voxel size 0.6 mm 0.6 mm0.6 mm. T1WI imaging was repeated 5 min after contrast administration.

Image reading

All analyses were conducted at the narrowest level of each extracranial carotid artery (including common carotid artery and extracranial internal carotid artery). All readers were blinded to the clinical information and the study images were reviewed in random order. Readers with the same seniority discussed and reached consensuses when the diagnosis or subjective scores were different. The consensus of senior radiologists was denoted as ‘R’, and that of the junior ones was represented as ‘r’.

Carotid artery plaque recognition and diagnostic confidence

CTA_con images were first read independently by two senior radiologists (R1 and R2, with 12 and 15 years of experience in cardiovascular imaging, respectively) and two junior radiologists (r1 and r2, with 3 and 2 years of experience in cardiovascular imaging). After 8 weeks, four radiologists reviewed CTA_con combined with CTA_sub side by side respectively. Each reading recorded each carotid artery’s type and diagnostic confidence. Carotid arteries can be classified into 4 types [14]: no plaque (Type 0), calcified plaque (Type 1), non-calcified plaque (Type 2) and mixed plaque (Type 3) (containing both calcified and non-calcified components). Mixed plaques are further classified into 3 categories according to the proportion of two components: mainly calcified (Type 3a: >1/2 calcified component); mainly non-calcified (Type 3b: <1/2 calcified component); and equivalent (Type 3c: approximately 1/2 calcified component). Diagnostic confidence was scored by a 3-point scale, where 1 was insufficient confidence to diagnose, 2 was sufficient confidence to diagnose but cannot ensure, and 3 was full confidence to make rapid diagnosis.

Carotid wall visibility and assessment of non-calcified components

Carotid wall visibility and the assessment of non-calcified components were evaluated via the side-by-side review of CTA_con and CTA_sub. A 3-point scale was used to score the carotid wall visibility (0: invisible; 1: barely visible; 2: clearly visible).

Assessment of non-calcified components were conducted in two aspects. Subjectively, a 3-point scale was used to evaluate the outer margin conspicuity of the non-calcified components, where 0 represented cannot recognize the margin, 1 represented a blurring margin and 2 represented a clear margin which can be clearly differentiated from surrounding tissues. Objectively, regions of interest (ROIs) were placed on non-calcified component, lumen, pericarotid fat, and sternocleidomastoid muscle respectively on three consecutive axial levels by R1. In order to ensure the repeatability, only plaques with a thickness of non-calcified components exceeding 2 mm were included in objective assessment. Selection and delineation criteria of ROIs were as follows: For non-calcified component, ROIs were positioned at the thickest and most homogeneous portion of the lesion. For lumen, ROIs covered about half of the carotid lumen area at the level of the plaque. For pericarotid fat, 5 mm² circular ROIs were placed within the pericarotid or intermuscular adipose tissue at the corresponding level. For sternocleidomastoid muscle, 10 mm² circular ROIs were placed within the muscle at the same anatomical level (Additional file 1: Fig. S2). The average attenuation (HU) and corresponding standard deviation (SD, defined as image noise) within the ROIs were recorded and the mean value of the measurement results on the three consecutive images was used. Signal-to-noise ratio (SNR) of non-calcified components, contrast-to-noise ratios (CNR) between non-calcified component and lumen (CNR_{non−calcified/lumen}), non-calcified component and fat (CNR_{non−calcified/fat}), and non-calcified component and muscle (CNR_{non−calcified/muscle}) were calculated as follows [9, 10]:

In this formula, ‘tissue’ refers to the adjacent structure including the carotid lumen, fat and muscle.

Carotid stenosis rate and plaque burden comparison

Comparison of the carotid stenosis rate was conducted between CTA_con and CTA_sub. Curved planar reconstruction (CPR) images were generated manually. Carotid arteries with plaque maximum length greater than 5 mm on CPR images were selected for stenosis rate calculation. The inner diameters of the residual (Dr) and original (Ds) lumens at the narrowest site, as well as the normal lumen diameter distal to the stenosed segment (Dn) were measured on three consecutive CPR images and the average measurements were used for calculation (Additional file 1: Fig. S3). Stenosis rates were calculated according to the North American Symptomatic Carotid Endarterectomy Trial (NASCET) and the European Carotid Surgery Trial (ECST) respectively.

Cross-sectional images perpendicular to the lumen central line were used for plaque burden evaluation. Comparison of the plaque burden was conducted on plaques mainly composed of non-calcified components between CTA and hrMRI. The vascular area and the residual lumen area where the plaques were shown best were measured on three consecutive images and the average measurements were used for calculation (Additional file 1: Fig. S4). The plaque burden was calculated via the following formula: plaque burden (%) = [(vascular area − lumen area)/vascular area]×100%. Taking hrMRI as the standard, the accuracy of measuring plaque burden greater than 50% on CTA_con and CTA_sub was calculated respectively.

Statistical analysis

IBM SPSS 27.0 was used for statistical analysis. Numerical variables are presented as the means ± standard deviations (SD) and categorical variables are expressed as values or percentages. The Kolmogorov-Smirnov test was used to examine the normality of the data distribution. The paired t-test was used to compare the SNR and CNRs of non-calcified components and the Wilcoxon signed-rank test was used to compare all ordered variables. P values of less than 0.05 were considered statistically significant. The simple Kappa test was used for the consistency of unordered categorical variables, while the weighted Kappa test was used for the consistency of ordered hierarchical variables. The Cohen κ value was used to interpret the strength of the inter-observer agreement with values ≤ 0 indicating no agreement, 0.01–0.20 indicating none to slight agreement, 0.21–0.40 indicating fair agreement, 0.41–0.60 indicating moderate agreement, 0.61–0.80 indicating substantial agreement, and 0.81–1.00 indicating almost perfect agreement. The consistency of the stenosis rates and plaque burden was evaluated via the intraclass correlation coefficient (ICC). The level of reliability was determined by following general guidelines: values less than 0.50 are indicative of poor reliability, values between 0.50 and 0.75 indicate moderate reliability, values between 0.75 and 0.90 indicate good reliability, and values greater than 0.90 indicate excellent reliability. The chi-square test was used to compare the accuracy of CTA_con and CTA_sub in diagnosing plaque burden > 50%.

Results

Patient characteristics

A total of 119 patients were ultimately evaluated for this study, including 69 males (58.0%) and 50 females (42.0%), with a mean age of 63.1 ± 14.4 years.

Plaque classification and diagnostic confidence

Inter-observer agreement on plaque classification and diagnostic confidence improved when reading were combined with CTA_sub (Table 1). The results of plaque classification are listed in Figs. 2 and 3. Compared with reading CTA_con alone, the combination of CTA_con and CTA_sub increased diagnostic confidence, especially for junior readers (Fig. 4a; Table 2). Non-calcified components missed or underestimated on CTA_con can be identified with the assistance of CTA_sub (Fig. 5).

Table 1.

Inter-reader agreement on plaque classification and diagnostic confidence

		CTAcon		CTAcon+CTAsub
		κ value	95% CI	κ value	95% CI
Plaque classification	R1 vs. R2	0.897	-	0.977	-
	r1 vs. r2	0.838	-	0.943	-
	R vs. r	0.850	-	0.989	-
Diagnostic confidence	R1 vs. R2	0.851	0.735–0.967	0.896	0.772–1.020
	r1 vs. r2	0.792	0.644–0.939	0.864	0.727–1.001
	R vs. r	0.715	0.574–0.855	0.873	0.756–0.989

CI = confidence interval; CTA_con = conventional CTA images; CTA_sub = subtraction images

Fig. 2.

Plaque classification results of senior and junior radiologists by reading CTA_con alone. CTA_con = conventional CTA images; CTA_sub = subtraction images; Type 0 = no plaque; Type 1 = calcified plaque; Type 2 = non-calcified plaque; Type 3a = mixed plaque of mainly calcified component; Type 3b = mixed plaque of mainly non-calcified component; Type 3c = mixed plaque of equivalent calcified and non-calcified components

Fig. 3.

Plaque recognition and classification results of radiologists by reading CTA_con combined with CTA_sub. CTA_con = conventional CTA images; CTA_sub = subtraction images; Type 0 = no plaque; Type 1 = calcified plaque; Type 2 = non-calcified plaque; Type 3a = mixed plaque of mainly calcified component; Type 3b = mixed plaque of mainly non-calcified component; Type 3c = mixed plaque of equivalent calcified and non-calcified components

Fig. 4.

Subjective and objective evaluation of CTA_con and CTA_sub. (a) Subjective scores of diagnostic confidence with and without the help of CTA_sub from senior and junior radiologists; (b, c) Subjective scores of extracranial carotid wall visibility and outer margin conspicuity of non-calcified components on CTA_con and CTA_sub by radiologists with different seniority. (d) Objective evaluation of non-calcified components on CTA_con and CTA_sub. SNR = signal-to-noise ratio; CNR = contrast-to-noise ratio

Table 2.

Diagnostic confidence of radiologists when reading CTA_con alone and reading CTA_con combined with CTA_sub

	r	R	p
CTAcon	2.66 ± 0.62	2.71 ± 0.65	0.007*
CTAcon+CTAsub	2.82 ± 0.50	2.87 ± 0.45	0.025*
p	<0.001*	<0.001*

* p<0.05; CTA_con = conventional CTA images; CTA_sub = subtraction images

Fig. 5.

Comparison of CTA_sub with CTA_con for displaying non-calcified plaque. (a, b) A non-calcified plaque in the right common carotid artery (arrow). The outer margin of the plaque on CTA_sub was clear, whereas it was blurred on CTA_con. (c, d) A slightly eccentric thickening of the left common carotid wall (arrow), which was recognized as a non-calcified plaque by senior radiologists on CTA_con but ignored by junior ones. Fortunately, it was detected by junior radiologists with the help of CTA_sub

Carotid wall visibility and assessment of non-calcified components

The scores of both carotid wall visibility and outer margin conspicuity of non-calcified components were obviously greater on CTA_sub than on CTA_con (Figs. 4b and c and 5). The carotid wall visibility scores were 0.43 ± 0.53 (senior) and 0.37 ± 0.49 (junior) on CTA_con and 1.39 ± 0.57 (senior) and 1.39.±0.61 (junior) on CTA_sub. The outer margin conspicuity scores of the non-calcified components were 0.62 ± 0.70 (senior) and 0.56 ± 0.69 (junior) on CTA_con and 1.66 ± 0.55 (senior) and 1.67 ± 0.50 (junior) on CTA_sub. Inter-observer agreements of carotid wall visibility scores and conspicuity scores of non-calcified components were both obviously improved on CTA_sub (Table 3).

Table 3.

Inter-reader agreement in extracranial carotid wall visibility and outer margin conspicuity of non-calcified components

		CTAcon		CTAsub
		κ value	95% CI	κ value	95% CI
Carotid wall visibility	R1 vs. R2	0.741	0.628–0.866	0.812	0.719–0.906
	r1 vs. r2	0.518	0.366–0.670	0.758	0.654–0.861
	R vs. r	0.611	0.468–0.754	0.873	0.795–0.952
Outer margin conspicuity	R1 vs. R2	0.747	0.671–0.904	0.853	0.748–0.957
	r1 vs. r2	0.694	0.568–0.820	0.829	0.702–0.957
	R vs. r	0.700	0.570–0.831	0.871	0.767–0.974

CI = confidence interval; CTA_con = conventional CTA images; CTA_sub = subtraction images

A total of 76 non-calcified components with maximum thickness over 2 mm were included in the objective evaluation. The SNR of non-calcified components on CTA_sub (12.86 ± 6.51) was higher than that on CTA_con (7.77 ± 4.30) (p < 0.001). The CNR_{non−calcified/fat} and CNR_{non−calcified/muscle} on CTA_sub were both higher than those on CTA_con (33.13 ± 2.65 vs. 24.96 ± 7.90; 8.58 ± 4.27 vs. 6.64 ± 4.11, both p < 0.001). However, the CNR_{non−calcified/lumen} was higher on CTA_con (70.88 ± 30.89) than on CTA_sub (37.16 ± 14.01) (p < 0.001) (Fig. 4d).

Carotid stenosis rate and plaque burden comparison

A total of 82 carotid arteries with plaque maximum length greater than 5 mm were selected for stenosis rate calculation. For each diameter, the measurements on CTA_sub were slightly larger than those on CTA_con (Dr: 5.11± 1.30 mm vs. 4.90 ± 1.26 mm; Ds: 9.08 ± 1.81 mm vs. 8.72 ± 1.78 mm; Dn: 6.21 ± 1.53 mm vs. 6.01 ± 1.45 mm, all p < 0.001). There were good to excellent consistency between CTA_sub and CTA_con in displaying carotid stenosis (NASCET: ICC = 0.919, 95%CI: 0.878–0.947, p < 0.001; ECST: ICC = 0.889, 95%CI: 0.833–0.927, p < 0.001). Bland-Altman plots were used to show the difference distribution of the stenosis rate (Fig. 6).

Fig. 6.

Bland-Altman plots used to show the difference distribution of stenosis rate and plaque burden. Bold continuous lines indicate bias and dotted lines indicate limits of agreement (1.96 × SD). (a, b) The differences of stenosis rates measured on CTA_sub and CTA_con via NASCET and ESCT methods were both small. (c, d) Plaque burden measured on CTA_sub was closer to hrMRI than CTA_con with a smaller bias and more narrower limits of agreement. NASCET = North American Symptomatic Carotid Endarterectomy Trial; ECST = European Carotid Surgery Trial; SD = standard deviation

Among the 31 patients with both CTA and hrMRI examinations, a total of 42 plaques, including 25 non-calcified plaques and 17 mixed plaques mainly composed of non-calcified components were identified. The ICC value of plaque burden between CTA_sub and hrMRI was 0.938 (95%CI: 0.805–0.974, p < 0.001), greater than 0.842 (95%CI: 0.0.127–0.951, p < 0.001) between CTA_con and hrMRI. Bland-Altman plots were used to show the difference distribution of the palque burden (Fig. 6). Taking hrMRI as the standard, the diagnostic accuracy of assessing plaque burden > 50% was 76.2% (32/42) for CTA_con and 88.1% (37/42) for CTA_sub respectively, and no significant difference was found (p > 0.05).

Discussion

Our results demonstrated that CTA_sub helped the detection of non-calcified components and increased the diagnostic confidence of junior radiologists to a level similar to that of senior ones. CTA_sub resulted in better visibility of the extracranial carotid wall and non-calcified components than CTA_con with higher inter-reader agreement. Moreover, it exhibited good consistency in the arterial stenosis rate quantification with CTA_con and more closer estimation in plaque burden to hrMRI.

In terms of plaque recognition, the detection of non-calcified components was elevated with the help of CTA_sub, especially for junior radiologists, whereas that of calcified components remained similar. On CTA_con, although the contrast-filled lumen contributed to a relatively high CNR_plaque/lumen of non-calcified plaque, the visibility was still limited because of the halo between the contrast agent and the plaque border caused by the partial volume effect [15]. CTA_sub overcome this limitation by suppressing intraluminal attenuation. In addition, the visibility of the extracranial carotid wall and the outer margin conspiuity of the non-calcified components were significantly enhanced, which was objectively reflected in the CNRs between the non-calcified components and surrounding tissues. This resulted from a slight decrease in attenuation of adipose tissue and a relative increase in that of the vessel wall compared with conventional images. All the visual improvements provided by CTA_sub can enhance diagnostic consistency and confidence among radiologists, with particular benefit for junior or less experienced ones.

The evaluation of calcified components was not included in our study because it was previously reported that spectral black blood could suppress calcium, leading to the misclassification of superficial calcified nodules as a continuation of the lumen or an arterial ulceration [9]. This defect was also observed in our subtraction images. This finding may be related to the calcium-suppressing effect inherent to the VNC algorithm. Although the VNC algorithm is designed to remove iodine contrast agents, it inevitably alters the CT attenuation values of high-density substances such as calcification. Such interference may lead to insufficient attenuation compensation during subtraction. The effect appears to be particularly evident in calcifications that exhibit relatively low original attenuation on CTAcon. Because of this, we only regarded CTA_sub as a supplement to conventional images rather than for independent reading.

The stenosis rate and plaque burden are both key indicators for the clinical assessment of the severity of atherosclerotic diseases [16, 17]. In our study, the measurement of each inner diameter on CTA_sub was slightly greater than that on CTA_con, which was consistent with the results of the phantom study part of Zhang et al. [9]. The blooming artifact produced by the contrast agent obscure the vessel wall and lead to the underestimation of the lumen. Fortunately, the performance of CTA_sub and CTA_con in the assessment of the vascular stenosis rate is basically the same, whether using NASCET or ECST method. Compared with hrMRI, the estimation of plaque burden on CTA_sub is more accurate than on CTA_con, which was similar to the findings of Lu et al. [8]. This study included only non-calcified plaques and mixed plaques with predominantly non-calcified components, in order to ensure an adequate sample size while avoiding potential misestimation of calcified components by hrMRI. This makes it possible for CTA_sub to replace the role of hrMRI in plaque burden assessment. However, comparison with hrMRI was limited to plaque burden measurements in a relatively small subgroup, and hrMRI provides unique information regarding plaque composition and vulnerability that cannot be assessed by CTA_sub. Therefore, it should be emphasized that the present findings support improved burden estimation rather than replacement of hrMRI.

The misregistration artifacts reported in the subtraction process usually stem from the imperfect alignment of two different phases [18, 19]. Research has revealed a significant drawback of existing methods, with misregistration frequencies ranging from 8% to 53% [20]. We generated subtraction images by subtracting VNC datasets from VMI datasets, both of which were obtained from the same scan, thus eliminating the misregistration. We selected 70-keV VMI datasets because they match the energy level of VNC datasets while best approximating conventional CT image quality. This technology, as demonstrated by Qin et al. [12] for coronary stent subtraction, produced artifact-free images that enhanced in-stent lumen visualization and diagnostic accuracy. Similar findings reported by Li et al. [13]. These findings support the clinical utility of this technology for accurate stenosis quantification in coronary arteries. Material decomposition, another technique based on SDCT, was also once used to generate ‘black blood’ images. For example, Rotzinger et al. [10] applied this technique to thoracic aortic wall assessment, significantly improving arterial wall conspicuity. Si et al. [21] facilitating reliable assessment of vascular involvement in pancreatic ductal adenocarcinoma by obtaining superior lumen-to-wall contrast. However, this ‘black blood’ technique requires extra processing algorithms that are incompatible with conventional workstations, currently limiting its clinical applicability. Additionally, projection maps obtained by material decomposition do not have any meaningful physical interpretation and thus cannot be used for quantitative analysis. Our subtraction method for carotid arteries offers superior accessibility for clinical practice. Owing to the principles of CT imaging, more complicated components, such as intraplaque hemorrhage cannot be identified. Nevertheless, our subtraction images enable clear visibility of the vessel wall, especially making up for the deficiency of conventional CTA in showing less obvious non-calcified plaques, Earlier intervention such as dietary modifications or pharmacological therapy facilitated by improved diagnostic sensitivity may help reduce the incidence of cerebrovascular events [22].

Our study had several limitations. First, all patients were form one single institution and the sample size of hrMRI validation cohort was relatively small. It is necessary to prove the universality of this technique in larger patient cohorts. Second, non-calcified plaques were not all validated by hrMRI and characteristics of vulnerable plaques were not involved. Besides, vessel wall visibility was evaluated using only subjective reader scores and was not validated against hrMRI or histopathology. Studies comparing CTA_sub findings with histopathological results or vulnerable plaque markers will be further carried out. Finally, we did not involve the research on intracranial plaques as the display of most intracranial arterial wall on CTAsub is still poor. This may be explained by the anatomical surroundings of intracranial vessel walls, which are frequently adjacent to bone or brain parenchyma with attenuation values higher than or close to the vessel walls, thereby hindering clear delineation of the outer wall boundary. Optimization strategies such as adjusting VMI energy levels, window settings, or post-processing parameters will be further researched for better intracranial arterial wall imaging.

Conclusion

Generally, our study indicates that the subtraction images derived from dual-layer spectral detector CT offer better subjective visibility of the extracranial carotid wall. The combination of subtraction with conventional images can provide a more comprehensive visualization of plaque morphology, which is beneficial for radiologists making more concordant assessments. It should be mentioned that CTA_sub should not be used as a standalone technique, but should be interpreted together with conventional CTA, especially when calcified plaques are present.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

CE boost

Contrast-enhancement-boost

CI

Confidence interval

CNR

Contrast-to-noise ratio

CTA

Computed tomography angiography

ECST

European Carotid Surgery Trial

hrMRI

High-resolution magnetic resonance imaging

HU

Hounsfield unit

NASCET

North American Symptomatic Carotid Endarterectomy Trial

ROI

Region of interest

SD

Standard deviation

SDCT

Dual-layer spectral detector CT

SNR

Signal-to-noise ratio

VNC

Virtual non-contrast images

VMI

Virtual monochrome images

Author contributions

YW: Conceptualization, Methodology, Investigation, Writing – original draft, Data curation; XZ: Investigation, Writing – original draft; ZC: Formal analysis, Data curation; XS: Formal analysis; SH: Data curation; DH: Data curation; YS: Validation; MC: Validation; DZ: Supervision, Writing – Review & editing; XC: Software, Writing – Review & editing; SJ: Supervision; WC: Funding acquisition, Writing – Review & editing, Project administration.

Funding

This study was supported by the Project of State Key Laboratory of Radiation Medicine and Protection, Soochow University (GZK12024025), Scientifc Research Project of Jiangsu Provincial Health Commission (M2024102), Suzhou Clinical Key Disease Diagnosis and Treatment Technology Special Project (LCZX202309), “National Tutor System” Training Program for Health Youth Key Talents in Suzhou (Qngg2021006) and the Subject Construction Support Project of the Second Affiliated Hospital of Soochow University (XKTJ-HTD2025003).

Data availability

The datasets used and analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The institutional ethics committee of The Second Affiliated Hospital of Soochow University approved this single-center retrospective study and waived informed consent. All work was conducted in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.