Comparison of deep learning reconstruction and iterative reconstruction algorithms for virtual monoenergetic image quality in overweight and obese patients with triple-low scan protocol dual-energy carotid computed tomography angiography

Introduction

The World Health Organization (WHO) projects that by 2030, the number of individuals affected by with obesity will increase to 1.02 billion, with the global adult obesity rate exceeding 13% (1,2). This rising prevalence represents a major public health challenge, as overweight and obesity are well-recognized, modifiable risk factors for cardiovascular diseases, particularly for carotid atherosclerosis in patients with metabolic syndrome and type 2 diabetes mellitus (3,4). Individuals with a body mass index (BMI) >25 kg/m² have a 34–57% higher risk of developing carotid atherosclerosis and are consequently more susceptible to ischemic stroke (5,6). Early and accurate identification of carotid atherosclerotic plaques in this population is therefore essential for risk stratification and prevention of cardiovascular complications.

In clinical practice, carotid computed tomography angiography (CTA) has become a cornerstone imaging modality for evaluating carotid artery disease. It offers high spatial resolution and enables rapid, comprehensive visualization of vascular morphology and plaque characteristics, providing valuable information on luminal narrowing and plaque vulnerability (7). Accordingly, CTA is strongly recommended in major diagnostic guidelines, including those of the European Society of Cardiovascular Radiology (8,9).

Nevertheless, performing CTA in overweight and obese patients remains technically challenging. Elevated BMI leads to increased image noise and beam-hardening artifacts caused by the nonlinear attenuation of adipose tissue, which can obscure vascular details and reduce contrast resolution (10-13). Moreover, these patients frequently present with comorbidities such as hypertension, diabetes, and other cardiovascular conditions that further complicate imaging and dose optimization (14,15). To maintain diagnostic quality, obese patients often require higher radiation exposure and larger contrast-agent volumes, which heightens the risk of contrast-induced nephropathy (CIN), particularly in individuals with impaired renal function (16,17). Hence, there is a pressing need for CTA protocols that can simultaneously minimize radiation and contrast dose while preserving diagnostic image quality.

Several technological strategies have been proposed to address these issues. You et al. demonstrated that the adaptive statistical iterative reconstruction-V (ASIR-V) algorithm can enhance image quality in obese and overweight patients by improving signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) (18). Zhang et al. further introduced a triple-low scan protocol—featuring reduced radiation dose, contrast-agent volume, and injection rate—and achieved improved image quality and lower background noise when combined with the deep learning image reconstruction with high setting (DLIR-H) algorithm (19). While these findings are promising, further studies are needed to clarify the relative performance of ASIR-V and DLIR in optimizing CTA image quality for patients with BMI >25 kg/m².

Building upon these developments, Jiang et al. integrated deep-learning image reconstruction (DLIR) with dual-energy computed tomography angiography (DE-CTA) to generate virtual monoenergetic images (VMIs). Their study showed that DLIR markedly enhanced image quality at 50 keV, reducing noise and improving vascular detail (20). However, the clinical performance of DLIR within a triple-low DE-CTA framework remains insufficiently investigated.

Therefore, the present study aims to compare the image quality of 50 keV VMIs reconstructed using DLIR and ASIR-V algorithms in overweight and obese patients (BMI >25 kg/m²) undergoing CTA with a triple-low scan protocol. Specifically, we seek to assess the influence of these reconstruction techniques on image quality, radiation dose, and contrast-agent utilization, thereby contributing to the optimization of CTA protocols for high-risk populations.

Methods

Patient demographics

This prospective study recruited patients who underwent DE-CTA at the Affiliated Hospital of Xuzhou Medical University between December 2024 and February 2025. Patients were randomly assigned to either the control or experimental group according to a computer-generated randomization schedule (Figure 1).

Figure 1 Overview of study design and image evaluation workflow. The diagram summarizes the study framework, including patient enrollment and grouping (control: ASIR-V 50%; experimental: DLIR-L, DLIR-M, DLIR-H), DE-CTA scanning and reconstruction (80/140 kV, 0.4–0.5 mL/kg contrast, 50 keV), and evaluation procedures. Objective evaluation assessed CT value, SD, CNR, and SNR at four carotid levels, while subjective evaluation used a 5-point Likert scale to rate noise, resolution, texture, and overall quality. This schematic illustrates the methodological workflow and does not include study results. ASIR-V, adaptive statistical iterative reconstruction-V; CNR, contrast-to-noise ratio; CT, computed tomography; DE-CTA, dual-energy computed tomography angiography; DLIR-H, deep learning image reconstruction with high setting; DLIR-L, deep learning image reconstruction with low setting; DLIR-M, deep learning image reconstruction with medium setting; GSI, Gemstone Spectral Imaging mode; HU, Hounsfield unit; ROI, region of interest; SDV, standard deviation of the carotid artery vessels; SNR, signal-to-noise ratio.

Ethical statement

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Ethics Committee of the Affiliated Hospital of Xuzhou Medical University (No. XYFY2024-KL456), and informed consent was obtained from all the patients.

Inclusion criteria

Patients were eligible for inclusion if they met the following criteria:

• Age ≥18 years, as DE-CTA is routinely used for evaluating carotid artery disease in adults;
• Clinical suspicion of carotid artery disease, based on symptoms or examination findings suggesting carotid pathology;
• Body mass index (BMI) ≥25 kg/m², representing overweight or obese individuals at increased risk for carotid atherosclerosis.

Exclusion criteria

The exclusion criteria were as follows:

• Allergy to iodinated contrast agents;
• Cardiac, hepatic, or renal dysfunction that could interfere with contrast metabolism or clearance;
• Hyperthyroidism;
• History of vascular surgery or stent implantation, to avoid confounding from prior interventions;
• Inability to cooperate with scanning or the presence of severe motion artifacts leading to non-diagnostic image quality.

Clinical data

Demographic characteristics, including sex, age, and BMI, were recorded for all participants. The contrast medium volume and injection rate were also documented, as both parameters can influence image quality and patient safety.

DE-CTA scanning protocol

All examinations were performed using a 256-slice dual-energy CT scanner (Revolution APEX Expert, GE Healthcare, Chicago, IL, USA) in spectral imaging mode (80/140 kV). The scanning range extended from the tracheal bifurcation to the cranial vertex, covering both carotid and intracranial arteries for comprehensive vascular evaluation.

Bolus tracking was used to trigger scanning. A region of interest (ROI) was placed in the ascending aorta, and image acquisition began automatically two seconds after the attenuation reached 100 Hounsfield unit (HU).

For contrast administration, iopromide (350 mg·I/mL; Hengrui Medicine, Lianyungang, China) was injected via an antecubital vein using a dual-head power injector (CT Motion, Oerlikon Medical, Erlangen, Germany).

Control group: 0.5 mL/kg at 3.7 mL/s.

Experimental group: 0.4 mL/kg at 3.2 mL/s.

Each injection was followed by 40 mL of saline at the same rate.

In routine clinical practice at our institution, the standard contrast dose for carotid CTA is 0.5 mL/kg (approximately 35–45 mL total). The experimental protocol used a reduced dose of 0.4 mL/kg (approximately 28–36 mL), corresponding to about a 20% decrease in contrast volume. This reduction was selected to evaluate whether the triple-low DE-CTA protocol combined with DLIR could maintain diagnostic image quality while minimizing iodine exposure.

Scanning was performed in spectral dual-energy mode with alternating tube voltages of 80 and 140 kV. The noise index (NI) was set to 4 for the control group and 13 for the experimental group. In our institution, an NI of 4 is used for the standard carotid CTA protocol. For the triple-low protocol, the NI was increased to 13 based on preliminary phantom and patient tests, which demonstrated that this setting allows meaningful radiation dose reduction while maintaining diagnostic image quality when combined with DLIR.

Other scanning parameters were: pitch =0.992, rotation time =0.5 s, detector width =128×0.625 mm, slice thickness =1.25 mm, interslice interval =0.625 mm, and matrix =512×512. In traditional CT reconstruction methods such as FBP or IR, kernels (i.e., convolution filters) are used to adjust the image sharpness or smoothness, which helps optimize the balance between noise reduction and spatial resolution. In contrast, the DLIR algorithm used in this study employs a deep neural network model, which directly generates high-quality images from the raw data, bypassing the need for traditional reconstruction kernels. These parameters provided uniform image resolution and diagnostic quality between groups (Table 1).

Image processing

Raw scan data were transferred to a GE Advantage Workstation (AW Server 4.7, software version 3.2) for image reconstruction. Spectral data were processed using GSI Volume Viewer 3.2 (GE Healthcare, Chicago, USA) to generate 50 keV VMIs, which enhance vascular contrast and reduce image noise.

In routine clinical practice at our institution, the standard contrast dose for carotid CTA is 0.5 mL/kg (approximately 35–40 mL total). In this study, the experimental protocol used a reduced dose of 0.4 mL/kg (approximately 28–32 mL), representing about a 20% decrease in contrast volume. This reduction was selected to evaluate whether the triple-low DE-CTA protocol combined with DLIR could maintain diagnostic image quality while minimizing iodine exposure.

Patients were assigned to either the control or experimental group according to acquisition protocols. The control group (NI =4) was reconstructed using the ASIR-V 50% algorithm (Group 1), while the experimental group (NI =13) was reconstructed using four algorithms: ASIR-V 50% (Group 2), DLIR-L (Group 3), DLIR-M (Group 4), and DLIR-H (Group 5).

Algorithm comparisons within the experimental group were performed as intra-individual analyses based on identical raw data, whereas comparisons between the control and experimental groups represented inter-group analyses reflecting different acquisition protocols. All enrolled patients were overweight or obese (BMI ≥25 kg/m²); therefore, no additional BMI stratification was performed.

Objective image quality evaluation

Objective image analysis was conducted by a cardiovascular imaging resident (W.X.) using the AW Server 3.2 with GSI Viewer software. Measurements were obtained at four anatomical levels: the aortic arch (AA), common carotid artery (CCA) origin, carotid bifurcation (CB), and internal carotid artery (ICA) origin (Figures S1,S2).

For each level, the ROI was placed at the center of the vessel lumen, with a diameter approximately one-half to two-thirds of the vessel width to minimize partial-volume effects. CT values (CTV) and their standard deviation of the carotid artery vessels (SDV, representing image noise) were recorded. Additional ROIs were placed in the pectoralis major (PM) and sternocleidomastoid (SC) muscles to assess background noise.

All ROI placements were performed according to consistent anatomical landmarks. In cases of uncertainty, measurements were verified by a senior radiologist to ensure accuracy and reproducibility.

SNR and CNR

To quantitatively assess image quality, the SNR and CNR were calculated as follows:

SNR=CTVSDV[1]

CNR=CTV−PM_CT/SC_CTPM_SD/SC_SD[2]

where CTV is the mean CT value of the target vessel, SDV is the standard deviation of the vessel ROI representing image noise, PM_CT/SC_CT is the mean CT attenuation of the adjacent muscle (pectoralis major or sternocleidomastoid), and PM_SD/SC_SD is the corresponding background noise.

The background reference muscle was selected according to the anatomical level: the pectoralis major muscle was used for the AA and CCA origin, whereas the sternocleidomastoid muscle was used for the CB and ICA origin. This anatomical correspondence ensured consistent tissue density and minimized attenuation variability, following the approach commonly adopted in previous carotid CTA studies.

These measurements were obtained at all four anatomical levels for each reconstruction algorithm to ensure comparability between groups.

Subjective image quality evaluation

Two senior radiologists with 20 and 30 years of cardiovascular imaging experience (Y.M. and K.X.) independently evaluated image quality using a five-point Likert scale. Both readers were blinded to patient information and reconstruction algorithms. Assessments were performed in a dedicated reading room using a diagnostic display monitor (JUSHA-C42E, 2,560×1,440 pixels; Nanjing Jusha Display Technology Co., Ltd.).

The Likert scale included four aspects: image noise, spatial resolution, noise texture, and overall image quality. The radiologists discussed discrepancies and reached a consensus score, which was used for subsequent statistical analysis.

Radiation dose

Radiation exposure data were obtained from the CT dose report by a cardiovascular imaging resident (W.X.). The recorded parameters included CT dose index volume (CTDI_vol), dose-length product (DLP), and effective dose (ED). The ED was calculated according to the following equation:

ED=DLP×k[3]

Where k=0.0031 mSv/mGy·cm

Independent and dependent variables

Independent variables were as follows:

• Type of reconstruction algorithm (ASIR-V 50%, DLIR-L, DLIR-M, DLIR-H);
• Acquisition protocol (control vs. experimental), representing the standard and triple-low protocols;
• Contrast medium dose (0.5 mL/kg in the control group and 0.4 mL/kg in the experimental group).

Dependent variables were as follows:

• Objective image-quality parameters: CTV, SDV, CNR, and SNR;
• Subjective image-quality scores assessed by two senior radiologists using a five-point Likert scale;
• Radiation-dose parameters: CTDI_vol, DLP, and ED.

Statistical analysis

All statistical analyses were performed using SPSS version 26.0 (IBM Corp., Armonk, NY, USA). The normality of continuous variables was assessed with the Shapiro-Wilk test, and the Levene test was used to examine the homogeneity of variances.

Continuous variables with a normal distribution are expressed as mean ± standard deviation, and non-normally distributed variables as median (interquartile range, IQR). Categorical variables are presented as percentages.

For data that were normally distributed with equal variances, one-way analysis of variance (ANOVA) was applied, followed by least significant difference (LSD) tests for pairwise comparisons. For data with non-normal distributions or unequal variances, the Kruskal-Wallis test was used, and post-hoc pairwise comparisons were conducted with the Dunn test. Post-hoc analyses were performed only when overall group differences were statistically significant.

A P value <0.05 was considered to indicate statistical significance.

Results

Patient demographics

The study involved 62 participants, comprising 40 males (64.52%) and 22 females (35.48%). The mean age of the participants was 58 years, with a median BMI of 27.7 kg/m². Patient demographics are summarized in Table 2.

Radiation dose and contrast agent usage

Building on the patient demographics, the study also assessed the ED and contrast medium usage in relation to the triple-low scan protocol. In the experimental group, the ED was 49.4% lower than in the control group, with mean values of 0.78 and 1.54 mSv, respectively (P<0.001). CTDI_vol was also significantly reduced in the experimental group, with a mean value of 5.97±0.62 mGy, compared with 12.21±0.20 mGy in the control group, representing a 51.1% reduction (P<0.001). Additionally, the contrast medium volume was reduced by 13.5% in the experimental group (P<0.001) (Table 2, Figure S3).

BMI and its impact on imaging parameters

Moreover, the study further investigated the role of BMI in influencing imaging parameters. Among the participants, 5 patients in the control group had a BMI >30 kg/m², while 6 patients in the experimental group exceeded this threshold. As BMI increased, both CTDI_vol and ED decreased in the control group, while the contrast agent dose and injection rate increased.

Objective image quality evaluation

Subsequent analysis showed significant differences between the experimental and control groups. Specifically, except for CTV at the AA level and SNR at the CCA level, all other variables demonstrated statistically significant differences (P<0.05) (Table 3). Group 5 exhibited the lowest SDV, reflecting reduced image noise and more consistent image quality. Conversely, Group 2 showed the lowest CNR and SNR, indicating reduced image clarity. Notably, the CNR and SNR values increased progressively from Group 2 to Group 5, indicating the beneficial impact of the triple-low scan protocol (Figure S4).

Post-hoc comparison of objective image quality between experimental and control groups

Further post-hoc analysis showed that at the AA level Group 5 had a significantly lower CTV than Group 1 (P<0.05). At the CCA level, Group 5 demonstrated a significantly lower SDV than Group 1; however, no significant differences were observed at the other levels (P>0.05). Furthermore, no significant differences in CNR or SNR were observed between Group 5 and Group 1 (P>0.05) (Table 4, Figure 2).

Figure 2 Violin plots showing quantitative comparisons of image quality among groups. Violin plots illustrate the distribution of quantitative parameters—including CTV, SDV, CNR, and SNR—at four anatomical levels: (A) AA, (B) CCA, (C) CB, and (D) ICA. Red dashed lines represent median reference values of the control group. Group 5 achieved lower SDV and comparable CNR/SNR relative to Group 1, indicating reduced image noise while maintaining diagnostic quality. Group 1: ASIR-V 50% (NI=4, contrast 0.5 mL/kg); Group 2: ASIR-V 50% (NI=13, contrast 0.4 mL/kg); Group 3: DLIR-L (NI=13, contrast 0.4 mL/kg); Group 4: DLIR-M (NI=13, contrast 0.4 mL/kg); Group 5: DLIR-H (NI=13, contrast 0.4 mL/kg). AA, aortic arch; ASIR-V, Adaptive Statistical Iterative Reconstruction-V; CB, carotid bifurcation; CCA, common carotid artery; CNR, contrast-to-noise ratio; CTV, computed tomography value of the carotid artery vessels; DLIR-H, deep learning image reconstruction with high setting; DLIR-L, deep learning image reconstruction with low setting; DLIR-M, deep learning image reconstruction with medium setting; ICA, internal carotid artery; NI, noise index; SDV, standard deviation of the carotid artery vessels; SNR, signal-to-noise ratio.

Post-hoc comparison of objective image quality within the experimental group

When Group 5 was examined in isolation, no significant differences in SDV were observed between Groups 4 and 5 at the CCA level. However, at other levels, Group 5 demonstrated significantly lower SDV compared with Groups 2, 3, and 4 (P<0.05). Additionally, Group 5 exhibited higher CNR and SNR than Groups 2, 3, and 4 at several levels. These findings reinforce the ability of the triple-low scan protocol to improve image quality while maintaining lower radiation exposure and reduced contrast medium usage (Table 4).

Subjective image quality evaluation

The two radiologists independently evaluated the images and subsequently reached consensus on all scores. Therefore, consensus values were used for statistical analysis. The subjective evaluation of image quality revealed that Group 5, under the triple-low scan protocol with the DLIR-H algorithm, achieved significantly higher scores across all evaluated parameters, namely image noise, spatial resolution, noise texture, and overall image quality. This finding was consistent across all regions of interest (P<0.001) (Table 5, Figure S5).

Post-hoc comparison of subjective image quality between experimental and control groups

Post-hoc comparisons further confirmed significant differences in subjective image quality between Group 5 and the other groups. Specifically, Group 5 demonstrated superior scores for image noise, spatial resolution, noise texture, and overall image quality compared with Group 1 (P<0.001). Importantly, Group 1 also outperformed Groups 2, 3, and 4 across all parameters (P<0.001) (Table 6, Figure S6).

Post-hoc comparison of subjective image quality within the experimental group

Within the experimental group, Group 5 demonstrated consistently higher scores in image noise, spatial resolution, noise texture, and overall image quality compared with Groups 2, 3, and 4 (P<0.001). Furthermore, Group 4 achieved significantly higher scores than Groups 2 and 3 in all subjective quality parameters (P<0.05) (Table 6, Figure S6).

For image noise, spatial resolution, noise texture, and overall image quality, the proportion of subjective scores greater than 4 in Group 5 was significantly higher than in the other groups (Figure 3).

Figure 3 Percentage distribution of subjective image quality scores across groups. Stacked area plots illustrate the proportion of cases within different score ranges (≤3, ≤4, and >4) for image noise, spatial resolution, noise texture, and OIQ. Groups 1 and 5 demonstrated higher proportions of scores >4, indicating superior subjective image quality and lower image noise compared with Groups 2–4. Group 1: ASIR-V 50% (NI=4, contrast 0.5 mL/kg); Group 2: ASIR-V 50% (NI=13, contrast 0.4 mL/kg); Group 3: DLIR-L (NI=13, contrast 0.4 mL/kg); Group 4: DLIR-M (NI=13, contrast 0.4 mL/kg); Group 5: DLIR-H (NI=13, contrast 0.4 mL/kg). ASIR-V, Adaptive Statistical Iterative Reconstruction-V; DLIR-H, deep learning image reconstruction with high setting; DLIR-L, deep learning image reconstruction with low setting; DLIR-M, deep learning image reconstruction with medium setting; NI, noise index; OIQ, overall image quality.

Discussion

This study found that experimental group, which reduced the ED and contrast agent usage by 49.4% and 13.5%, respectively, while achieving subjective and objective image quality comparable to that of ASIR-V. These findings suggest that DLIR-H provides a viable solution for improving image quality in ultra-low-dose CTA for overweight and obese patients, especially in clinical settings that require high image quality with minimal exposure to radiation and contrast agents.

In further detail, the DLIR-H algorithm processes low-dose CT data through deep neural networks, optimizing parameters through training on large datasets (21). This method effectively suppresses image noise and artifacts while maintaining the integrity of fine image details (22,23). The algorithm’s ability to map high-noise input data to high-quality images results in substantial improvements in both SNR and CNR. Importantly, the DLIR-H algorithm not only outperforms iterative reconstruction techniques but also offers higher computational efficiency, making it especially suited for emergency settings, where both speed and accuracy are critical.

When compared to previous single-energy CTA studies, this study demonstrated a reduction in both radiation dose and contrast medium volume, while successfully acquiring dual-energy data, as supported by the findings from established research on dual-energy CT’s efficacy in dose and contrast reduction (24,25). This is significant, as obese patients typically require higher doses of both radiation and contrast agents. The low-dose protocol employed in this study has two primary advantages: it reduces patient radiation exposure and simultaneously mitigates the risk of CIN. By incorporating the DLIR algorithm, we were able to obtain high-quality images under these low-dose conditions. This is especially advantageous for patients who require continuous or multiple scans, such as those in chronic disease management, because it reduces both radiation burden and contrast-related medication load, thus contributing to overall healthcare cost savings.

In addition to the reductions observed in DLP and ED, this study also demonstrated a significant decrease in CTDI_vol in the experimental group compared with the control group (P<0.001). This reduction can be attributed primarily to adjustment of the NI, which directly influences automatic tube current modulation (ATCM). By setting a higher NI value (13 vs. 4), the tube current output was reduced, thereby lowering CTDI_vol. Optimized scan parameters—including alternating 80/140 kV tube voltages, a pitch of 0.992, and a shortened rotation time—further contributed to dose reduction without compromising vascular enhancement. Importantly, the application of DLIR algorithms effectively suppressed image noise and preserved spatial resolution under these lower-dose conditions, supporting the clinical feasibility of the triple-low protocol for overweight and obese patients (26,27).

Regarding to VMI, this technique significantly enhances image quality at specific energy levels, particularly by improving contrast (28,29). For instance, 50 keV VMI provides improved visualization of small vascular structures, enabling more precise assessment of atherosclerotic plaque morphology. However, VMI is typically associated with higher image noise, especially when using low radiation and contrast agent doses. The DLIR algorithm mitigates this issue, reducing noise while improving image quality. The integration of DLIR with VMI enhances the diagnostic value of the images, allowing clinicians to obtain more detailed information, thereby improving diagnostic accuracy and patient care.

The generalizability of our findings is limited by the inclusion of only overweight and obese patients. These patients tend to generate more image noise and require higher doses of radiation and contrast agents, which could affect the performance of the reconstruction algorithms. While the results demonstrate the effectiveness of the triple-low scan protocol in reducing radiation dose and contrast usage, it is important to note that the findings may not directly apply to normal-BMI populations. Future studies involving normal-BMI patients are necessary to validate whether the same improvements can be achieved under lower-noise imaging conditions.

Furthermore, the simultaneous variation in several parameters, including the reconstruction algorithm, radiation dose, and contrast medium injection dose, may introduce confounding effects. Specifically, the differences in radiation settings (e.g., tube voltage and tube current), contrast agent dosing (0.5 vs. 0.4 mL/kg), and reconstruction algorithms (ASIR-V, DLIR) complicate the isolation of the effect of each factor on image quality and radiation dose reduction. As a result, future studies with more controlled experimental designs are necessary to isolate the effects of these independent variables more strictly.

The difference in contrast medium dosage between groups was intentional and reflects a clinically relevant optimization strategy. While ideally contrast injection parameters should remain constant to isolate the effect of reconstruction algorithms, our design aimed to simulate a real-world scenario where both radiation and contrast dose reduction are pursued simultaneously. We acknowledge that this introduces a potential confounding factor; however, supporting evidence from previous studies and our own results demonstrate that DLIR algorithms can preserve vascular enhancement and diagnostic confidence even at reduced contrast doses (30,31).

In this study, the grouping was based on acquisition protocols rather than BMI categories. While algorithm comparisons were conducted as intra-individual analyses using the same raw data, the control versus experimental groups reflected different acquisition settings. Because all patients were overweight or obese, the findings may not be directly generalizable to normal-BMI populations, and further studies including patients across BMI ranges are needed.

The quantitative analysis of SNR and CNR was performed using GE commercial software, which, while widely used in clinical practice, has limitations in transparency and reproducibility because the internal algorithms are not fully disclosed and results may vary with software versions or operator interaction. Future studies may incorporate open-source platforms such as MATLAB or Python to provide fully transparent, reproducible, and shareable analysis pipelines.

Several studies have demonstrated the effectiveness of spectral CT imaging in reducing contrast medium and radiation doses while maintaining diagnostic quality. For instance, Vargas et al. achieved a 50% reduction in contrast dose in coronary CTA using motion-compensated VMI with spectral dual-layer CT (32). Xu et al. similarly showed significant reductions in both radiation and contrast media doses with spectral photon-counting detector coronary CTA, while preserving diagnostic accuracy (33). In a study of pulmonary embolism CTPA, Stammen et al. demonstrated that reduced contrast doses with PCD-CT maintained image quality (34), while Klambauer et al. optimized contrast and radiation dose through task-based automatic keV selection in PCD-CT without sacrificing image quality (35). Our study, utilizing the “triple-low” protocol combined with DLIR, aligns with these findings, showing that reduced doses can still provide high-quality diagnostic images. However, further validation in larger, multi-center trials is needed to confirm its broader applicability.

However, there are several limitations in this study. First, the relatively small sample size (N=62) was adequate for two-group comparisons but underpowered for multi-group analyses; larger multi-center studies with BMI stratification are needed to strengthen statistical rigor and generalizability. Second, this study only compared 50 keV VMI with ASIR-V images at 50% weighting, and it did not explore CTA image quality at different low keV levels or with varying ASIR-V weightings. Third, restricting the study population to overweight and obese patients may introduce bias, as body habitus influences CT image quality and radiation dose; future studies should include normal-BMI patients to verify whether the observed benefits apply more broadly. A further limitation is that subjective evaluation was performed by only two radiologists who reached consensus. Future studies should include a larger panel of readers and incorporate quantitative agreement metrics to enhance reliability. This study did not apply formal multiple comparison corrections (e.g., Bonferroni) because it was exploratory and involved a limited number of correlated image quality parameters. Unadjusted P values were used to preserve statistical sensitivity, and future larger confirmatory studies will include multiple testing corrections to validate these results. Lastly, the study did not assess the comparative value of different reconstruction algorithms in evaluating plaque morphology, which may provide further insights into the utility of the DLIR-H algorithm in atherosclerotic plaque evaluation.

Conclusions

The DLIR-H algorithm improves image quality in 50 keV VMI with the triple-low scan protocol DE-CTA, reducing radiation exposure and contrast agent usage in overweight and obese patients.

Acknowledgments

None.

Footnote

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-856/dss

Funding: This study was funded by the Key Research and Development Program of Xuzhou Science and Technology Bureau (No. KC23279), Xuzhou Health Commission Technology Projects (No. XWKYHT20240086), and the Affiliated Hospital of Xuzhou Medical University Institutional Research Project (No. 2024ZY08). Additional support was provided by the Special Fund for Overseas Training of Employees of the Affiliated Hospital of Xuzhou Medical University.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-856/coif). A.S. and S.Z. are employees of GE Healthcare. The other authors have no conflicts of interest to declare.

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 study was approved by the Institutional Ethics Committee of the Affiliated Hospital of Xuzhou Medical University (No. XYFY2024-KL456), and informed consent was obtained from all the patients.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Worldwide trends in underweight and obesity from 1990 to 2022: a pooled analysis of 3663 population-representative studies with 222 million children, adolescents, and adults. Lancet 2024;403:1027-50. [Crossref] [PubMed]
2. Lingvay I, Cohen RV, Roux CWL, Sumithran P. Obesity in adults. Lancet 2024;404:972-87. [Crossref] [PubMed]
3. Botvin Moshe C, Haratz S, Ravona-Springer R, Heymann A, Hung-Mo L, Schnaider Beeri M, Tanne D. Long-term trajectories of BMI predict carotid stiffness and plaque volume in type 2 diabetes older adults: a cohort study. Cardiovasc Diabetol 2020;19:138. [Crossref] [PubMed]
4. Huang Q, Liu Z, Wei M, Feng J, Huang Q, Liu Y, Liu Z, Li X, Yin L, Xia J. Metabolically healthy obesity, transition from metabolic healthy to unhealthy status, and carotid atherosclerosis. Diabetes Metab Res Rev 2024;40:e3766. [Crossref] [PubMed]
5. Global Burden of Metabolic Risk Factors for Chronic Diseases Collaboration (BMI Mediated Effects). Lu Y, Hajifathalian K, Ezzati M, Woodward M, Rimm EB, Danaei G. Metabolic mediators of the effects of body-mass index, overweight, and obesity on coronary heart disease and stroke: a pooled analysis of 97 prospective cohorts with 1·8 million participants. Lancet 2014;383:970-83. [Crossref] [PubMed]
6. Itoh H, Kaneko H, Kiriyama H, Yoshida Y, Nakanishi K, Mizuno Y, Daimon M, Morita H, Yatomi Y, Yamamichi N, Komuro I. Effect of Metabolically Healthy Obesity on the Development of Carotid Plaque in the General Population: A Community-Based Cohort Study. J Atheroscler Thromb 2020;27:155-63. [Crossref] [PubMed]
7. Ospel JM, Singh N, Marko M, Almekhlafi M, Dowlatshahi D, Puig J, Demchuk A, Coutts SB, Hill MD, Menon BK, Goyal M. Prevalence of Ipsilateral Nonstenotic Carotid Plaques on Computed Tomography Angiography in Embolic Stroke of Undetermined Source. Stroke 2020;51:1743-9. [Crossref] [PubMed]
8. Saba L, Loewe C, Weikert T, Williams MC, Galea N, Budde RPJ, Vliegenthart R, Velthuis BK, Francone M, Bremerich J, Natale L, Nikolaou K, Dacher JN, Peebles C, Caobelli F, Redheuil A, Dewey M, Kreitner KF, Salgado R. Correction to: State-of-the-art CT and MR imaging and assessment of atherosclerotic carotid artery disease: standardization of scanning protocols and measurements-a consensus document by the European Society of Cardiovascular Radiology (ESCR). Eur Radiol 2023;33:1497-8. [Crossref] [PubMed]
9. Lupi A, Suchá D, Cundari G, Fink N, Alkadhi H, Budde RPJ, et al. Standards for conducting and reporting consensus and recommendation documents: European Society of Cardiovascular Radiology policy from the Guidelines Committee. Insights Imaging 2024;15:207. [Crossref] [PubMed]
10. Faggioni L, Gabelloni M. Iodine Concentration and Optimization in Computed Tomography Angiography: Current Issues. Invest Radiol 2016;51:816-22. [Crossref] [PubMed]
11. Fleischmann D, Chin AS, Molvin L, Wang J, Hallett R. Computed Tomography Angiography: A Review and Technical Update. Radiol Clin North Am 2016;54:1-12. [Crossref] [PubMed]
12. Murphy DJ, Aghayev A, Steigner ML. Vascular CT and MRI: a practical guide to imaging protocols. Insights Imaging 2018;9:215-36. [Crossref] [PubMed]
13. Fink MA, Stoll S, Melzig C, Steuwe A, Partovi S, Böckler D, Kauczor HU, Rengier F. Prospective Study of Low-Radiation and Low-Iodine Dose Aortic CT Angiography in Obese and Non-Obese Patients: Image Quality and Impact of Patient Characteristics. Diagnostics (Basel) 2022.
14. Sohn W, Lee HW, Lee S, Lim JH, Lee MW, Park CH, Yoon SK. Obesity and the risk of primary liver cancer: A systematic review and meta-analysis. Clin Mol Hepatol 2021;27:157-74. [Crossref] [PubMed]
15. Raphael H, Klang E, Konen E, Inbar Y, Leibowitz A, Frenkel-Nir Y, Apter S, Grossman E. Obesity Is Associated with Fatty Liver and Fat Changes in the Kidneys in Humans as Assessed by MRI. Nutrients 2024;16:1387. [Crossref] [PubMed]
16. Fursevich DM. LiMarzi GM, O'Dell MC, Hernandez MA, Sensakovic WF. Bariatric CT Imaging: Challenges and Solutions. Radiographics 2016;36:1076-86. [Crossref] [PubMed]
17. Berezina TA, Berezin OO, Lichtenauer M, Berezin AE. Predictors for Irreversibility of Contrast-Induced Acute Kidney Injury in Patients with Obesity After Contrast-Enhanced Computed Tomography Coronary Angiography. Adv Ther 2025;42:293-309. [Crossref] [PubMed]
18. You J, Dai Y, Huang N, Li JJ, Cheng L, Zhang XL, Liu Q, Liu Y, Xu K. Low-Dose Computed Tomography With Adaptive Statistical Iterative Reconstruction and Low Tube Voltage in Craniocervical Computed Tomographic Angiography: Impact of Body Mass Index. J Comput Assist Tomogr 2015;39:774-80. [Crossref] [PubMed]
19. Zhang Q, Lin Y, Zhang H, Ding J, Pan J, Zhang S. The application value of a vendor-specific deep learning image reconstruction algorithm in "triple low" head and neck computed tomography angiography. Quant Imaging Med Surg 2024;14:2955-67. [Crossref] [PubMed]
20. Jiang C, Jin D, Liu Z, Zhang Y, Ni M, Yuan H. Deep learning image reconstruction algorithm for carotid dual-energy computed tomography angiography: evaluation of image quality and diagnostic performance. Insights Imaging 2022;13:182. [Crossref] [PubMed]
21. Available online: https://www.gehealthcare.com/-/jssmedia/gehc/us/images/products/computed-tomography/truefidelity/related-content/new/truefidelity-for-gsi-whitepaper_digital_jb19879xx_final.pdf?rev=-1
22. Long J, Wang C, Yu M, Liu X, Xu W, Liu Z, Wang C, Wu Y, Sun A, Zhang S, Hu C, Xu K, Meng Y. Comparison of DLIR and ASIR-V algorithms for virtual monoenergetic imaging in carotid CTA under a triple-low protocol. Jpn J Radiol 2026;44:43-54. [Crossref] [PubMed]
23. Zhang H, Xu X, Long J, Wang C, Liu X, Xu W, Sun X, Dou P, Zhou D, Cao W, Xu K, Meng Y. Acute stroke risk prediction model based on dual-energy CTA-derived carotid plaque, perivascular adipose tissue characteristics, and serum lipid parameters: a dual-center study. Neuroradiology 2025;67:3157-71. [Crossref] [PubMed]
24. Guerrini S, Zanoni M, Sica C, Bagnacci G, Mancianti N, Galzerano G, Garosi G, Cacioppa LM, Cellina M, Zamboni GA, Minetti G, Floridi C, Mazzei MA. Dual-Energy CT as a Well-Established CT Modality to Reduce Contrast Media Amount: A Systematic Review from the Computed Tomography Subspecialty Section of the Italian Society of Radiology. J Clin Med 2024;13:6345. [Crossref] [PubMed]
25. Shuman WP, Mileto A, Busey JM, Desai N, Koprowicz KM, Dual-Energy CT. Urography With 50% Reduced Iodine Dose Versus Single-Energy CT Urography With Standard Iodine Dose. AJR Am J Roentgenol 2019;212:117-23. [Crossref] [PubMed]
26. Kim JH, Yoon HJ, Lee E, Kim I, Cha YK, Bak SH. Validation of Deep-Learning Image Reconstruction for Low-Dose Chest Computed Tomography Scan: Emphasis on Image Quality and Noise. Korean J Radiol 2021;22:131-8. [Crossref] [PubMed]
27. Jo GD, Ahn C, Hong JH, Kim DS, Park J, Kim H, Kim JH, Goo JM, Nam JG. 75% radiation dose reduction using deep learning reconstruction on low-dose chest CT. BMC Med Imaging 2023;23:121. [Crossref] [PubMed]
28. Cester D, Eberhard M, Alkadhi H, Euler A. Virtual monoenergetic images from dual-energy CT: systematic assessment of task-based image quality performance. Quant Imaging Med Surg 2022;12:726-41. [Crossref] [PubMed]
29. D'Angelo T, Cicero G, Mazziotti S, Ascenti G, Albrecht MH, Martin SS, Othman AE, Vogl TJ, Wichmann JL. Dual energy computed tomography virtual monoenergetic imaging: technique and clinical applications. Br J Radiol 2019;92:20180546. [Crossref] [PubMed]
30. Otgonbaatar C, Ryu JK, Shin J, Kim HM, Seo JW, Shim H, Hwang DH. Deep learning reconstruction allows for usage of contrast agent of lower concentration for coronary CTA than filtered back projection and hybrid iterative reconstruction. Acta Radiol 2023;64:1007-17. [Crossref] [PubMed]
31. Tang Y, Zhang H, Chen L, Yu M, Zhang H, Zhang D, Wu Y, Liu Z, Sun A, Meng Y, Xu K. Comparison of image quality of 40 keV virtual monoenergetic images of vertebral arteries using DLIR and ASIR-V algorithms under a dual-low scanning protocol. Eur J Radiol 2025;191:112276. [Crossref] [PubMed]
32. Vargas EE, Tetteroo PM, Kristiansen CH, Dobrolinska MM, Greuter MJW, Vembar M, Grass M, Leiner T, Velthuis BK, van der Werf NR, Suchá D. Optimizing coronary CT angiography with spectral dual-layer CT: motion-compensated virtual monoenergetic imaging achieves 50% contrast dose reduction in a phantom study. Int J Cardiovasc Imaging 2025; Epub ahead of print. [Crossref]
33. Xu C, Sun Y, Wang L, Zou L, Wang M, Lin L, Wang Y, Sun X, Liu X, Yu X, Leidecker C, Wang L, Liu Y, Qian H, Tian R, Vliegenthart R, Liu Z, Wang Y. Spectral Photon-Counting Detector Coronary CTA With Reduced Radiation and Contrast Medium Dose in Inflammation-Associated Coronary Artery Disease: A Prospective Study. AJR Am J Roentgenol 2025;225:e2533493. [Crossref] [PubMed]
34. Stammen L, Hoeijmakers EJI, Flohr TG, Gietema HA, Vandewall J, Wildberger JE, Jeukens CRLPN, Martens B. Reducing contrast media dosage for pulmonary embolism CTPA in PCD-CT: a comparative study of EID-CT and PCD-CT in the era of individualized protocolling. Eur Radiol 2025; Epub ahead of print. [Crossref]
35. Klambauer K, Flohr T, Moser LJ, Mergen V, Eberhard M, Prokein A, Alkadhi H, Jost G, Pietsch H. Contrast media and radiation dose optimization with task-based automatic keV selection: a proof-of-concept study with photon-counting detector CT. Eur Radiol 2025;35:7975-84. [Crossref] [PubMed]