Quantification of plaque and detection of high-risk plaque characteristics based on coronary computed tomography angiography: comparison between superresolution deep learning reconstruction and hybrid iterative reconstruction Authors: Limiao Zou, Keting Xu, Xiaohuan Liu, Cheng Xu, Haipeng Liu, Ying Li, Ying Li, Tao Zhou, Lin Yang, Chunli Zhao, Na Li, Yi Lv, Wenli Hong, Bo Wang, Lijun Chen, Min Xu, Yufei Li, Yunxin Zhong, Zhiping Zhong, Jikun Zheng, Zhihua Wu, Hongzhong Mo, Tong Zhang, Demao Deng, Yaliang Liu, Shouqiang Jia, Shaofei He, Changhua Liang, Yining Wang Published: 2026 ABSTRACT AbstractBackgroundIn addition to stenosis evaluation, coronary computed tomography angiography (CCTA) also allows for noninvasive plaque quantification and characterization, which informs risk stratification and management strategies. This study aimed to compare a novel superresolution deep learning reconstruction (SR-DLR) technique with a conventional hybrid iterative reconstruction (HIR) method in quantifying CCTA-derived plaque and detecting high-risk plaque (HRP) characteristics.MethodsA total of 255 consecutive patients (mean age 63.7±9.6 years; 184 men and 71 women) who underwent clinically indicated CCTA for assessment of suspected or known coronary artery disease were prospectively enrolled. The HIR algorithm with the FC43 kernel and the SR-DLR algorithm with the cardiac kernel reconstructed the same raw CCTA data. Total plaque, calcified plaque, noncalcified plaque (NCP), and low-attenuation plaque (LAP) volumes were quantified. Moreover, the ability to detect HRP characteristics (LAP, spotty calcification, positive remodeling, and napkin-ring sign) was evaluated. Paired comparisons were made between the two reconstruction pipelines.ResultsCompared with the HIR pipeline, the SR-DLR pipeline obtained higher median volumes of total plaque {103.4 [interquartile range (IQR) 52.9–201.7] vs. 124.5 (IQR 66.8–236.2) mm3}, NCP [43.9 (IQR 20.1–87.7) vs. 56.4 (IQR 25.2–115.7) mm3], and LAP [0.1 (IQR 0–3.0) vs. 0.9 (IQR 0–8.7) mm3] (all P<0.001); meanwhile, calcified plaque volume did not differ significantly between the pipelines [44.8 (IQR 16.2–100.6) vs. 44.8 (IQR 17.3–98.2) mm3; P=0.87]. SR-DLR detected more LAPs than did HIR (91 vs. 67; P=0.01), but there were no detected differences in spotty calcification (63 vs. 56; P=0.33), positive remodeling (110 vs. 110; P>0.99), or napkin-ring sign (42 vs. 40, P=0.76).ConclusionsCompared with the HIR pipeline, the SR-DLR pipeline indicated higher total plaque, NCP, and LAP volumes, but no difference in calcified plaque volume was observed. Moreover, SR-DLR detected more LAPs than did HIR, while other HRP characteristics did not differ significantly. Further large-scale prospective studies including a reference standard are warranted to investigate the clinical impact of SR-DLR.Keywords: Coronary computed tomography angiography (CCTA), superresolution deep learning reconstruction (SR-DLR), hybrid iterative reconstruction (HIR), coronary plaque quantification, high-risk plaque characteristics Introduction Coronary computed tomography angiography (CCTA) has been well established as the first-line imaging modality for the noninvasive diagnosis of obstructive coronary artery disease (CAD) by virtue of its high sensitivity and negative predictive value compared to invasive coronary angiography (ICA) (1,2). Along with stenosis evaluation, CCTA further allows for the quantitative assessment of plaque equivalent to examination with invasive imaging modalities [i.e., intravascular ultrasound (IVUS) and optical coherence tomography (OCT)], providing powerful cardiovascular risk prediction and informing management strategies (3-5). Moreover, high-risk plaque (HRP) characteristics detected on CCTA are associated with histological and invasive features of plaque vulnerability and may indicate adverse cardiovascular events (6-9). However, there are certain limitations to CCTA in terms of spatial resolution and soft tissue contrast, which reduce its ability to accurately quantify and characterize coronary plaques (3,10-12). Innovations in image reconstruction techniques have been stimulated by the demand for high-quality computed tomography (CT) imaging. Filtered back projection and iterative reconstruction (IR) are the two predominant image reconstruction algorithms (13), yet in recent years, deep learning reconstruction (DLR) algorithms that incorporate deep convolutional neural networks into image reconstruction process have been increasingly used (14). A novel superresolution DLR (SR-DLR) algorithm, trained on data acquired on true ultrahigh-resolution CT (UHRCT) (Aquilion Precision, Canon Medical Systems), has been recently introduced (15). SR-DLR exhibits several advantages over conventional image reconstruction methods, including lower image noise, higher spatial resolution, and better edge sharpness (16-19). Preliminary studies have confirmed that the superiority of SR-DLR translates into improved coronary stenosis assessment and stent evaluation (20,21). However, there is a lack of evidence regarding the possible impact of SR-DLR on the assessment of plaque volumes and HRP characteristics. Therefore, this study aimed to compare SR-DLR with hybrid IR (HIR) in their ability to quantitatively assess plaque and detect HRP characteristics under a conventional fixed Hounsfield unit (HU) threshold. We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0199/rc). Methods Study participants The purpose of this prospective, multicenter study was to clarify the utility of SR-DLR. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The research protocol was approved by the Institutional Review Board of Peking Union Medical College Hospital (approval No. I-25PJ1387) and was registered in the Chinese Clinical Trial Registry (ChiCTR2300075364). All participating hospitals were informed of and agreed to this study. Written informed consent was obtained from all participants. In brief, from September 2023 to November 2024, a total of 330 patients with suspected or known CAD who underwent clinically indicated CCTA and subsequent ICA within a 2-month interval were consecutively enrolled from 11 Chinese medical centers: (I) Peking Union Medical College Hospital; (II) The First Affiliated Hospital of Henan Medical University; (III) People’s Hospital of Mengzi City; (IV) People’s Hospital Affiliated to Shandong First Medical University; (V) Hanzhong Central Hospital; (VI) The People’s Hospital of Guangxi Zhuang Autonomous Region; (VII) The Fourth Affiliated Hospital of Harbin Medical University; (VIII) The First People’s Hospital of Yulin; (VIIII) The Fourth Affiliated Hospital, Guangzhou Medical University; (X) Yunnan Provincial Hospital of Traditional Chinese Medicine; and (XI) Guangdong Provincial Hospital of Chinese Medicine. ICA served only as an inclusion criterion of the parent study and had no analytic role in this substudy on plaque volumes and characteristics. In this substudy, patients were excluded due to (I) no visible plaque except for the stented segment, (II) poor image quality (e.g., severe motion and misalignment artifacts) that would preclude accurate analysis, and (III) missing raw data needed for image reconstruction. CT acquisition and image reconstruction All patients received sublingual nitrates before examinations, and β-blockers were administered additionally for patients with a heart rate ≥70 bpm. CT scans were performed on a 320-row detector CT scanner (Aquilion ONE GENESIS, Canon Medical Systems, Otawara, Japan) in all participating centers. Contrast agent with at least 350 mg of iodine per milliliter of contrast was injected at a rate of 3–5.5 mL/s followed by a flush of saline solution, and the automatic bolus triggering technique was used for initiating image acquisition. The scan parameters were as follows: a tube voltage of 100–120 kVp, a gantry rotation time of 0.275 s, a collimation of 320 mm × 0.5 mm, and a tube current automatically adjusted with a noise index standard deviation (SD) of 23–33. Images were reconstructed with both HIR [adaptive iterative dose reduction 3D (AIDR 3D), FC43 kernel, standard strength] and SR-DLR [Precise IQ Engine (Canon Medical Systems), cardiac kernel, standard strength], with a 0.5-mm slice thickness and increments of 0.25 mm. Quantitative plaque assessment The reconstructed image datasets were transferred to a dedicated workstation (Vitrea version 4.0.693, Vital Images, Inc., Minnetonka, MN, USA) for analysis, and observers performed the analysis blinded to the reconstruction methods and clinical information. Through use of an 18-coronary-segment model, coronary segments with a diameter ≥2.0 mm, except for the stented segments, were evaluated on axial images, multiplanar reformation images, and curved multiplanar reformation images. Window-setting adjustments were made according to personal preference. For plaque quantification, a board-certified radiologist, with 5 years of experience in cardiovascular imaging, placed reference points in areas immediately proximal and distal to the plaques. Luminal and outer vessel boundaries were generated automatically and adjusted manually if not correctly delineated. Total plaque volume was quantified between the luminal and outer vessel boundaries, and based on the predefined HU thresholds, total plaque volume was further divided into calcified plaque (>350 HU), noncalcified plaque (NCP) (30–350 HU), and low-attenuation plaque (LAP) (<30 HU) volumes (12,22). To assess the reproducibility, 50 patients were randomly selected and reanalyzed by a cardiac radiologist, with 3 years of experience in cardiovascular imaging who followed the same standard procedure and was blinded to the results of the first reader. The radiologists reviewed cases in two sessions, separated by a 2-week washout period. Each session included one of the two image sets (HIR or SR-DLR) for each patient in a fully randomized order. The locations and screenshots of the analyzed lesions in the first session were provided to guarantee that the identical coronary lesions were analyzed across different reconstruction datasets, especially for difficult or borderline lesions. Importantly, screenshots contained no quantitative measurement results from prior measurements. Identification of HRP characteristics Four HRP characteristics on CCTA are well-established, including LAP (a focal central area of plaque with an attenuation density <30 HU), spotty calcification (calcification <3 mm in size), positive remodeling (a remodeling index >1.1 and an outer vessel diameter 10% greater than the mean diameter of the segments immediately proximal and distal to the plaque), and napkin-ring sign (a ringlike peripheral higher attenuation with central low CT attenuation) (7,23). All identified plaques were evaluated for the presence of HRP characteristics by two above-mentioned radiologists in a joint reading, and any disagreements were resolved via discussion. A third radiologist (with 8 years of experience in cardiovascular imaging) who was blinded to the previous results performed the qualitative HRP analysis independently. Statistical analysis Statistical analyses were performed with statistical software, including R version 3.6.1 (The R Foundation for Statistical Computing, Vienna, Austria) and Origin 2024 (OriginLab Corporation, Northampton, MA, USA). The Shapiro-Wilk test was used to evaluate the normality of the data distribution, and continuous variables are presented either as the mean ± SD or as the median and interquartile range (IQR). Categorical variables are reported as counts and percentages. Normally distributed continuous data were compared with the paired t-test, while nonnormally distributed variables were analyzed with the Wilcoxon signed-rank test. Mixed effects models were employed for lesion-level analyses to account for within-patient clustering effects, with the R package lme4 used for model fitting. Quantitative plaque assessment was analyzed with a linear mixed-effects model and estimated via restricted maximum likelihood, with denominator degrees of freedom approximated with the Satterthwaite method. Binary HRP characteristics were analyzed and estimated via a mixed-effects logistic regression model and Laplace approximation, respectively. The random structure included a patient-specific random intercept, within-patient correlation across repeated measurements being accounted for. Differences in plaque quantification between SR-DLR and HIR were compared via Bland-Altman analysis, with results presented as the mean bias and limits of agreement (LoA). Interobserver agreements were assessed according to the intraclass correlation coefficient (ICC) for quantitative plaque measurements and the Cohen’s kappa for binary HRP characteristics. A P<0.05 was considered statistically significant. Results Baseline characteristics Based on the predefined criteria, 18 patients without visible plaque except for the stented segment were excluded; another 49 patients were excluded due to poor image quality resulting from motion artifacts, severe coronary calcifications, or inadequate opacification and 8 due to loss of raw data. Ultimately, 255 patients (184 males and 71 females; mean age 63.7±9.6 years) were enrolled for the final analysis. Figure 1 shows the flowchart of the study, and the details of the patients’ baseline characteristics are summarized in Table 1. Flowchart of the study. CCTA, coronary computed tomography angiography; ICA, invasive coronary angiography. Table 1Patient baseline characteristicsCharacteristicValue (n=255)Sex Male184 (72.0) Female71 (28.0)Age (years)63.7±9.6BMI (kg/m2)24.9±3.4Cardiac risk factors Diabetes75 (29.0) Hypertension186 (73.0) Hypercholesterolemia118 (46.0) Smoking104 (41.0) Family history of CAD70 (27.0)Data are the n (%) or the mean ± SD. BMI, body mass index; CAD, coronary artery disease; SD, standard deviation. Patient baseline characteristics Data are the n (%) or the mean ± SD. BMI, body mass index; CAD, coronary artery disease; SD, standard deviation. Quantitative plaque assessment A total of 690 plaques were identified, 22 (3%) of which were in the left main trunk, 285 (41%) in the left anterior descending artery, 95 (14%) in the left circumflex artery, and 288 (42%) in the right coronary artery. On a per-plaque basis, the median quantified total plaque, calcified plaque, NCP, and LAP volumes for HIR were 103.4 (IQR 52.9–201.7), 44.8 (IQR 16.2–100.6), 43.9 (IQR 20.1–87.7), and 0.1 (IQR 0–3.0) mm3, respectively, while those for SR-DLR were 124.5 (IQR 66.8–236.2), 44.8 (IQR 17.3–98.2), 56.4 (IQR 25.2–115.7), and 0.9 (IQR 0–8.7) mm3, respectively. The mean bias between SR-DLR and HIR was 23.2 (LoA: −59.2 to 105.5) mm3 for total plaque volume, −0.6 (LoA: −31.2 to 30.0) mm3 for calcified plaque volume, 19.5 (LoA: −42.5 to 81.4) mm3 for NCP volume, and 4.3 (LoA: −14.1 to 22.7) mm3 for LAP volume. Total plaque, NCP, and LAP volumes were significantly higher with SR-DLR than with HIR (all P<0.001), whereas calcified plaque volume did not differ significantly between the two reconstructions (P=0.87). A summary of the quantitative results is shown in Table 2 and Figure 2, while a representative example is provided in Figure 3. Table 2Quantitative plaque assessmentParameterHIRSR-DLRBiasLoAFixed effectsRandom effectsPβSEPatient varianceResidual varianceTPV (mm3)103.4 (52.9, 201.7)124.5 (66.8, 236.2)23.2−59.2 and 105.523.2 (11.7, 34.6)5.85,25111,699<0.001LAPV (mm3)0.1 (0, 3.0)0.9 (0, 8.7)4.3−14.1 and 22.74.3 (3.2, 5.4)0.671.1111.1<0.001NCPV (mm3)43.9 (20.1, 87.7)56.4 (25.2, 115.7)19.5−42.5 and 81.419.5 (13.4, 25.6)3.12,9243,309<0.001CPV (mm3)44.8 (16.2, 100.6)44.8 (17.3, 98.2)−0.6−31.2 and 30.0−0.6 (−7.9, 6.7)3.72,6154,7800.87Values are median (interquartile range) unless otherwise indicated. CPV, calcified plaque volume; HIR, hybrid iterative reconstruction; LAPV, low-attenuation plaque volume; LoA, limits of agreement; NCPV, noncalcified plaque volume; SE, standard error; SR-DLR, superresolution deep learning reconstruction; TPV, total plaque volume. Quantitative plaque assessment Values are median (interquartile range) unless otherwise indicated. CPV, calcified plaque volume; HIR, hybrid iterative reconstruction; LAPV, low-attenuation plaque volume; LoA, limits of agreement; NCPV, noncalcified plaque volume; SE, standard error; SR-DLR, superresolution deep learning reconstruction; TPV, total plaque volume. Quantitative plaque assessment. Data are presented as median (interquartile range). HIR, hybrid iterative reconstruction; SR-DLR, superresolution deep learning reconstruction. Representative case of plaque quantification. Lumen area, LAP, NCP, and calcified plaque are represented by green, red, blue, and yellow, respectively. CPV, calcified plaque volume; HIR, hybrid iterative reconstruction; LAP, low-attenuation plaque; LAPV, low-attenuation plaque volume; NCP, noncalcified plaque; NCPV, noncalcified plaque volume; SR-DLR, superresolution deep learning reconstruction; TPV, total plaque volume. Interobserver agreements for quantitative measurement of total and compositional plaques were excellent for both HIR and SR-DLR (all ICCs ≥0.90). Identification of HRP characteristics Of the 690 plaques identified on HIR images, 114 had ≥1 HRP characteristics, including 9 plaques with 1 characteristic, 58 plaques with 2 characteristics, 40 plaques with 3 characteristics, and 7 plaques with all the 4 characteristics. Spotty calcification was detected in 56 plaques, positive remodeling in 110 plaques, LAP in 67 plaques, and napkin-ring sign in 40 plaques. Meanwhile, on SR-DLR images, 115 plaques had ≥1 HRP characteristics, including 7 plaques with 1 characteristic, 40 plaques with 2 characteristics, 53 plaques with 3 characteristics, and 15 plaques with all the four characteristics. Moreover, the number of plaques showing spotty calcification, positive remodeling, LAP, and napkin-ring sign was 63, 110, 91, and 42, respectively. More LAPs were detected with SR-DLR than with HIR (91 vs. 67; P=0.01), whereas no evidence of a difference was observed in the presentation of spotty calcification (63 vs. 56; P=0.33), napkin-ring sign (42 vs. 40; P=0.76), or positive remodeling (110 vs. 110; P>0.99). Interobserver agreement for the detection of HRP characteristics was excellent for both HIR [0.96, 95% confidence interval (CI): 0.95–0.98] and SR-DLR (0.97, 95% CI: 0.96–0.99). The HRP characteristics in coronary plaques are summarized in Table 3, while a representative example is shown in Figure 4. Table 3Detection of high-risk plaque characteristicsCharacteristicHIRSR-DLRHIR presentHIR absentOR (95% CI)PSR-DLR presentSR-DLR absentSR-DLR presentSR-DLR absentSpotty calcification566356076271.31 (0.76–2.27)0.33Positive remodeling110110110005801.00 (0.67–1.49)>0.99Low-attenuation plaque6791670245991.68 (1.11–2.55)0.01Napkin-ring sign404240026481.09 (0.61–1.96)0.76CI, confidence interval; HIR, hybrid iterative reconstruction; OR, odds ratio; SR-DLR, superresolution deep learning reconstruction. Detection of high-risk plaque characteristics CI, confidence interval; HIR, hybrid iterative reconstruction; OR, odds ratio; SR-DLR, superresolution deep learning reconstruction. Representative case of the detection of LAP. Paired images of HIR (A) and SR-DLR (B) demonstrate that SR-DLR detected more LAPs than HIR (yellow arrows). HIR, hybrid iterative reconstruction; LAP, low-attenuation plaque; SR-DLR, superresolution deep learning reconstruction. Discussion To the best of our knowledge, this study is the first to compare the novel SR-DLR with conventional HIR in quantitative plaque assessment and HRP characteristics detection. The results demonstrated that compared with the HIR pipeline, the SR-DLR pipeline obtained significantly higher median total plaque [103.4 (IQR 52.9–201.7) vs. 124.5 (IQR 66.8–236.2) mm3], NCP [43.9 (IQR 20.1–87.7) vs. 56.4 (IQR 25.2–115.7) mm3], and LAP [0.1 (IQR 0–3.0) vs. 0.9 (IQR 0–8.7) mm3] volumes (all P<0.001), whereas calcified plaque volume did not differ significantly (P=0.87). Moreover, SR-DLR detected more LAPs than did HIR (91 vs. 67; P=0.01), but there was no difference in the presentation of other HRP characteristics (all P>0.05). CCTA is a robust noninvasive method for the comprehensive assessment of CAD, enabling stenosis evaluation and the quantification and characterization of plaque, and is equivalent to the current gold standards for plaque analysis (i.e., ICA and OCT) (1,6,24,25). However, concerns regarding the limited spatial resolution and soft tissue contrast of CCTA remain, as this can impair its capacity to accurately quantify and characterize coronary plaque (3,10-12). Advanced image reconstruction techniques have attracted considerable interest in the field of CT imaging (14,26,27). The novel SR-DLR algorithm, recently developed for CCTA, was trained with data acquired from a true UHRCT scanner (Aquilion Precision) that features a small detector element size (0.25 mm × 0.25 mm). Previous studies have demonstrated that SR-DLR outperforms conventional image reconstruction methods in terms of image quality and improves diagnostic accuracy for stenosis assessment (16-18,20,21). Therefore, SR-DLR holds substantial promise in reducing radiation dose, which is related to a nonnegligible lifetime attributable risk of cancer. One study found that compared with HIR, SR-DLR allowed for a 60% reduction in dose (2.01±0.84 vs. 0.80±0.34 mSv; P<0.001) while ensuring high image quality and excellent performance in coronary plaque and stenosis analysis (ICCs >0.8) (18). However, the utility of SR-DLR in plaque quantification and detection of HRP characteristics remains unknown. Volumetric plaque assessment, particularly NCP and LAP volumes, has been extensively associated with adverse cardiovascular outcomes (5,28). In our study, compared with the HIR pipeline, the SR-DLR pipeline yielded a higher median total plaque volume [103.4 (IQR 52.9–201.7) vs. 124.5 (IQR 66.8–236.2) mm3; P<0.001], which was mainly attributed to the higher median NCP volume [43.9 (IQR 20.1–87.7) vs. 56.4 (IQR 25.2–115.7) mm3; P<0.001]. This was in line with a previous research that reported the detectability index of NCP to be higher with SR-DLR than with other reconstruction methods (16,17). Of note, the median LAP volume was also higher with the SR-DLR pipeline than with the HIR pipeline [0.9 (IQR 0–8.7) vs. 0.1(IQR 0–3.0) mm3; P<0.001]. This finding was in line with a previous study demonstrating that the use of UHR photon-counting detector CT led to higher LAP volume compared with energy-integrating detector CT (29). As previous studies using photon-counting CT have reported reduced volumes in calcifications (particularly larger calcifications) (30-32), it was somewhat unexpected that the median calcified plaque volume did not differ significantly between the SR-DLR and HIR pipelines [44.8 (IQR 17.3–98.2) vs. 44.8 (IQR 16.2–100.6) mm3; P=0.87] in our study. We speculate that the SR-DLR pipeline may reduce the calcified plaque volume due to a reduction in calcium blooming artifacts. The reason for this may be that the cohort predominantly included cases with smaller calcifications, causing selection bias. LoA were rather wide for plaque volumes, particularly relative to the observed LAP values, which indicated that plaque quantification using the conventional fixed HU threshold may not be interchangeable across the HIR and SR-DLR pipelines. Moreover, CCTA-derived HRP characteristics correlate with their histological and invasive counterparts and hold substantial predictive value for subsequent adverse cardiovascular events (6,7,9,33,34). The binary HRP in this study exhibited a typical paired and clustered data structure: multiple lesion-level observations were nested within individual patients and therefore lacked statistical independence. To mitigate bias arising from intrapatient correlation, we used a random-intercept mixed-effects logistic regression model for analyses. We found that more LAPs were detected with the SR-DLR pipeline than with the HIR pipeline (91 vs. 67; P=0.01), whereas no difference was observed in the presentation of spotty calcification (63 vs. 56; P=0.33), napkin-ring sign (42 vs. 40; P=0.76), or positive remodeling (110 vs. 110; P>0.99). Despite the increase in HRP characteristic values with SR-DLR, it did not have significantly higher values for plaques carrying at least one HRP characteristic compared to HIR (115 vs. 114). This indicates that the incremental detection of HRP characteristics with SR-DLR is confined to lesions already categorized as high-risk on HIR, with little impact on the total number of distinct HRP plaques identified. It was worth mentioning that under the conventional fixed HU threshold, attenuation-based plaque classification and LAP counts differed significantly between the HIR and SR-DLR pipelines, whereas attenuation-independent HRP characteristics did not. As different reconstruction methods modify pixel CT attenuation values through differential noise suppression, partial-volume correction, and spatial resolution adjustment, these results may be attributable to the threshold-shift effect of SR-DLR, and further studies with external reference standards are needed. Limitations The following limitations to our study merit consideration. First, we only enrolled participants who underwent ICA, introducing a selection bias, and the exclusion of poor-quality scans rendered the cohort less representative. Second, due to the lack of a comparison to IVUS or OCT, the study reflected reconstruction-dependent differences between SR-DLR and HIR in plaque volumes and HRP characteristics, and whether the differences reflect the actual superiority of SR-DLR remains unclear. Studies with external reference standards are being conducted to address this. Moreover, due to the kernel mismatch between reconstructions (FC43 for HIR and cardiac kernel for SR-DLR), the observed differences may reflect a combined algorithm-plus-kernel effect rather than the reconstruction algorithm alone. Another limitation is that intraobserver reproducibility testing was not performed for qualitative HRP detection. Third, although they were blinded, the observers might have been able to visually distinguish between HIR and SR-DLR methods due to the considerable difference in the image characteristics, which might have introduced a bias. Additionally, the use of window settings according to personal preference was a potential source of measurement variability. Moreover, as the image reconstruction algorithms were from a single vendor, it is unclear to what extent these findings can be generalized to other methods. Lastly, due to the lack of patient-level reclassification and outcome analysis, the potential clinical implications of SR-DLR remain unknown, and further large-scale prospective studies are warranted. Conclusions Compared with the conventional HIR pipeline, the novel SR-DLR pipeline yielded higher total plaque, NCP, and LAP volumes, but there was no evidence of a difference in calcified plaque volume. Moreover, more LAPs were identified with the SR-DLR pipeline than with the HIR pipeline, while other HRP characteristics did not differ significantly. Without external validation, the accuracy and clinical implications of SR-DLR remain uncertain, and further large-scale prospective studies involving a reference standard are necessary. Supplementary The article’s supplementary files as Acknowledgments None. Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The research protocol was approved by the Institutional Review Board of Peking Union Medical College Hospital (approval No. I-25PJ1387). All participating hospitals were informed of and agreed to this study. Written informed consent was obtained from all participants. Footnotes Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0199/rc Funding: This study was supported by the National Science Fund for Distinguished Young Scholars (grant No. 82325026), Noncommunicable Chronic Diseases-National Science and Technology Major Project (grant No. 2024ZD0537706), and the National High Level Hospital Clinical Research Funding (grant No. 2025-PUMCH-C-011), the CAMS Innovation Fund for Medical Science (grant Nos. 2023-I2M-C&T-A-004 and 2025-I2M-XHJC-001), and Peking Union Medical College Hospital Talent Cultivation Program (Category A, grant No. ULJ06588). Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0199/coif). M.X., Yufei Li, and Y.Z. report that they are currently employed by Canon Medical Systems China. The other authors have no conflicts of interest to declare. Data Sharing Statement Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0199/dss References Associated Data Supplementary Materials The article’s supplementary files as Data Availability Statement Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0199/dss