Effect of ZeeFree algorithm on the reduction of stair-step artifacts for coronary computed tomography angiography Authors: Xing Guo, Shufang Wei, Linfeng Li, Minghua Sun, Yukun Pan, Yinghui Ge, Ruigang Xie Published: 2026 ABSTRACT Stair-step artifacts is a common type of artifacts in the sequential scanning of CCTA and severely affects the image quality of coronary artery scans. This study aims to evaluate a novel algorithm for the removal of stair-step artifacts. 95 patients (52 males and 43 females, average weight of 70.81 ± 11.68 kg, median height of 1.66 [1.72, 1.6], average BMI of 25.57 ± 3.09 kg/m2) who received CCTA scanning on the new generation of dual source CT scanner, SOMATOM Pro.Pulse, were prospectively enrolled for the study. Images were reconstructed with or without ZeeFree to compare the effect of this novel algorithm. The doctors were asked to evaluate the severity of stair-step artifacts, image quality and diagnostic confidence with a 5-Likert scale. The subjective scoring comparison was performed Chi-square test in Python 3.11.10. Subgroups were divided by heartbeat and BMI and the same comparison was performed to evaluate the performance of ZeeFree with different factors. Additionally, a preliminary assessment of the degree of luminal stenosis was conducted and compared with the ICA. There was no significant difference for patients’ basic information between subgroups, except for the weight, contrast medium and radiation dose when divided by BMI. 76.8% (86/112) of the stair-step artifacts were resolved with ZeeFree algorithm. For the subjective scoring results, ZeeFree algorithm always shows a significant improvement for the severity of stair-step artifacts, image quality and diagnostic confidence (all p < 0.001). The differences stay significant for subgroup analysis. The Cohen’ s Kappa value of the ZeeFree reconstruction with ICA was 0.85, which demonstrated a great consistency in stenosis assessment. ZeeFree algorithm could significantly reduce the occurrence of stair-step artifacts, enhance the quality of coronary artery images and diagnostic usability.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-53546-9. Introduction Coronary computed tomography angiography (CCTA) is currently the best non-invasive evaluation of coronary artery anatomy1,2, the reference standard is invasive coronary angiography (ICA). Sequential acquisition mode of CCTA scanning is the potential occurrence of stair-step artifacts with a reported prevalence ranging from 18%3to 78%4. A typical Electrocardiogram (ECG)-triggered sequence acquisition covers the heart in several consecutive sequence steps. The acquisition is performed in an overlapping fashion to accommodate the 3D cone beam nature of a multi-slice CT scanner. While the individual amount of overlap depends on the system geometry, a typical value found in most computed tomography (CT) scanner systems is 10%. Patient breathing during scanning misaligns data stacks and creates vessel discontinuities, whose stair-step artifacts mimic or obscure coronary stenosis, causing false findings, revascularization changes, and repeat imaging or angiography. In traditional ways, the stacked area of data would be averaged for reconstruction, or each primary axial thin slice image is only allowed to be reconstructed from data of a single cardiac cycle, which were called Standard mode. Either way would result in the poor image quality especially for stacked areas. There were two kinds of techniques which provided potential solution for stair-step artifacts. First way is the application of wide detector CT scanners. The SnapShot Freeze(SSF) reconstruction can be used to correct coronary artery motion artifacts, but only on specific machines from certain manufacturers5,6. However, wide detector scanner system didn’t improve the temporal resolution, which plays an important role in improving the image quality of cardiac and other organs with motions7. Dual source CT scanner is currently the system with the highest temporal resolution, capable of providing high-quality images for the heart and coronary arteries8. Second way is neural network with deep learning for image reconstruction and artifact reduction, it mainly has advantages in reducing image noise, improving the spatial resolution and reducing beam-hardening artifacts or motion artifacts9–11. There were also some studies trying to remove the stair-step artifacts with neural network12 but it has not yet been extensively studied. Recently, a new algorithm, ZeeFree, which is a proprietary technology developed by Siemens Healthineers specifically to correct stair-step artifacts that occur in coronary CT angiography. Between 2024 and 2025, studies by Moser LJ13, Lisi C14, Bonner B15 and Haag NP16 validated the efficacy of the ZeeFree algorithm in reducing stair-step artifacts in CCTA using photon-counting CT. Furthermore, Lisi C14 extended the evaluation to include CT-FFR and plaque analysis. However, the practical performance of the ZeeFree algorithm on dual-source CT remains unknown, and previous studies have not made comparisons with ICA. In this study, we aim to validate the effectiveness of the ZeeFree algorithm- a fixed, proprietary reconstruction option provided by the scanner manufacturer- in reducing stair-step artifacts in CCTA images acquired with dual-source CT, especially focusing on the reduction of occurrence and severity of stair-step artifacts and image quality improvement, which leads to a higher diagnostic confidence on coronary-artery-related diseases for doctors, rather than assessing machine learning model performance. Additionally, a preliminary assessment of the degree of luminal stenosis was conducted and exploratorily compared with ICA to verify the correction effect of ZeeFree. Materials and methods Study population This study was performed at a tertiary hospital, all study participants provided informed consent, and the study design was approved by the local ethics review board【2023(62)】. All methods were performed in accordance with the guidelines and regulations of the Declaration of Helsinki. 117 patients in total received CCTA scanning from December 2023 to April 2024. Inclusion criteria: (1) Meet the indications and have no contraindications for CCTA; (2) age ≥ 18 years old; (3) non-pregnancy, non-lactation period. Exclusion criteria: (1) severe arrhythmia; (2) There are metal implants and metal artifacts are severe in the heart. 12 patients were excluded because of severe arrhythmia. 10 patients were excluded because of severe metal artifacts in the heart. There were 95 patients involved in the end (Fig. 1). The patients were divided into subgroups with heart rate (heart rate > 75 bpm and ≤ 75 bpm) and body mass index (BMI) (BMI > 25 kg/m2 and ≤ 25 kg/m2) in order to compare the effects of ZeeFree algorithm.Fig. 1Patient flowchart. Patient flowchart. CT image acquisition All the cardiovascular CT scanning and reconstruction parameters employed in this study were recommended by authoritative clinical guidelines for cardiovascular CT imaging17 and set as default parameters based on extensive experience in routine clinical diagnostic practice. All the CCTA examinations were performed on the new generation of dual-source CT scanner, SOMATOM Pro.Pulse (Siemens Healthineers, Forchheim, Germany), with the automatic tube voltage modulation (CARE kV) and tube current modulation (CARE Dose4D). Rotation time was set as 0.33 s, collimation as 64 × 0.6 mm and image quality level of 75 in the scanning protocol. Images were reconstructed with the kernel of Bv40, slice thickness of 0.8 mm, interval of 0.5 mm and advanced modeled iterative reconstruction (ADMIRE) level of 4. In each reconstruction method, the image parameters were set to the following: field of view (FOV), 17–20 cm, matrix size, 512 × 512. The radiation dose of each patient’s scanning was recorded with CT dose index (CTDI), dose length product (DLP) and size-specific dose estimate (SSDE)18 and compared in the following. A triphasic contrast media protocol was used (0.8 mL/kg iopromide, Ultravist 370; Bayer, Germany) with injection rates from 4.0 mL/s to 5.0 mL/s, according to the patients’ body mass index. ZeeFree algorithm is a commercial reconstruction method which has been integrated in the scanner’s system. No new machine learning model was trained in this study; the algorithm was applied as a fixed reconstruction option, and the reconstruction with ZeeFree was simply implemented by selecting the corresponding option on the scanner operation interface. Each patient’s data was performed with the aforementioned reconstruction with or without ZeeFree (Pro.pulse_cn, syngo CT VB20, Siemens Healthineers, Forchheim, Germany), described as Group Standard and Group ZeeFree, which could be an option during the reconstruction parameters’ selection, to enable a head-to-head comparison. All the images were then imported into the post-processing workstation (Syngo.Via VB40, Siemens Healthineers, Forchheim, Germany) for further evaluation and measurement. ZeeFree algorithm The ZeeFree algorithm can be broken into three individual steps13: A non-rigid registration applied separately to each stack transition followed by a vector field interpolation and resampling and a final “user” reconstruction vector field application step. The first non-rigid registration step is independently performed for all transitions of the entire acquisition based on a pair of stack images in each case. The overlapping data allows for the selection of a single image from two stacks at the same image position in the middle of the overlapping area. The two stacks are then subjected to a demon-type registration algorithm19,20 based on this vector field model, which describes the image transformation as a joint vector field with opposing signs for the images from two stacks, minimizing the root mean squared of the difference of the images. The above-described registration step is derived from images that were independently reconstructed in the background with a set of fixed parameters to exclude a potential influence of user-specific settings (e.g., reconstruction kernel, slice thickness, matrix size, iterative strength, etc.) on the registration result. Therefore, a resampling to the final matrix size and pixel spacing of the “user” reconstruction is performed. The third step for the final ZeeFree images is derived from hidden “user” reconstructions, which are resampled by means of the final vector field as described above. Image quality evaluation Two doctors with 6 years and 7 years of diagnostic expertise were involved in the evaluation of the images. They were blind to the reconstruction algorithms and asked to assess the images for the severity of stair-step artifacts, image quality and diagnostic confidence. Each indicator would be individually assessed with a 5-Likert scale as below. Subjective score of the severity of the stair-step artifacts level: 1 = the stair-step artifacts was very serious, the vessel staggered distance was relatively large or connection less than 25% of the vessel diameter, and the image could not be diagnosed; 2 = the stair-step artifacts is serious, with 25%-49% of the vessel diameter connected, the image diagnosis is affected; 3 = part of the stair-step artifacts, with 50%-74% of the vessel diameter connected, the image can basically diagnose; 4 = a few stair-step artifacts, with 75%-99% of the vessel diameter connected, can meet the diagnosis; 5 = the stair-step artifacts are basically absent, image diagnosis is not affected. A score of ≥ 3 is defined as usable for diagnosis. Subjective score of image quality level: 1 = image quality is very poor, there is no way to diagnose; 2 = the image quality is relatively poor, only a small part of image can diagnose; 3 = the image quality is average, but basically meet the diagnostic requirements; 4 = the image quality is better, can meet the diagnostic requirements; 5 = the image quality is very good. A score of ≥ 3 is defined as usable for diagnosis. Subjective score of doctors’ diagnostic confidence level: 1 = not sure about the diagnostic results; 2 = with a 20% confidence in the diagnosis; 3 = with 40% certainty of diagnosis; 4 = with 60% certainty of diagnosis; 5 = with 80% certainty of diagnosis. A score of ≥ 4 is defined as usable for diagnosis. The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were calculated as objective evaluation of image quality for Group Standard and Group ZeeFree. The regions of interest (ROIs) were placed in the ascending aorta at the level of the origin of left main artery with an area of 200 mm2, and in the epicardial adipose tissue (EAT) surrounding the proximal of right coronary artery (RCA) with an area of 50 mm2. To measure mean CT value (hounsfield unit, HU) and standard deviation (SD) respectively, each parameter was measured twice, with values averaged. HUaorta and SDaorta refer to the mean CT value and the SD of the ROI placed in the ascending aorta, HUEAT refers to the mean CT value of the ROI placed in the EAT. The SNR and CNR were calculated as below21:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$SNR= {HU}_{aorta}/{SD}_{aorta}$$\end{document}\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$CNR= \left({HU}_{aorta}-{HU}_{EAT}\right)/{SD}_{aorta}$$\end{document} Stenosis evaluation ICA results of the followed-up patients within 6 months before and after the current CCTA examination were collected, and stenosis comparisons were performed based on the official reports issued by cardiologists. A total of 18 patients underwent ICA, corresponding to 48 lesions with stair-step artifacts, and vascular stenosis assessment was conducted for these 48 lesions. The calculation of coronary artery stenosis was performed using Shukun 1.0 software (Beijing ShuKun Technology, China). Stenosis detection was categorized as ≥ 50% or < 50%22. A diagnostic radiologist with 7 years of experience, blinded to the grouping, measured the degree of stenosis at stair-step artifact sites using this software. Statistical analysis All statistical analysis was performed with Python 3.11.10. Quantitative data was showed with mean and standard deviation if it conforms to normal distributions, and medians and interquartile ranges if not. Categorical variables were presented as percentages and frequencies. The student’s t-test or Wilcoxon-Mann–Whitney test was used to determine quantitative data’s discrepancy. The consistency of doctors’ evaluation results and vascular stenosis assessment results was assessed with Cohen’s Kappa, 95% confidence intervals (95% CIs) were reported to quantify the precision of agreement estimates. A good consistency was determined with a Kappa greater than 0.8. The difference between two reconstruction methods was tested with Chi-square test. The same statistical analysis was performed for each subgroup to evaluate the effect of ZeeFree algorithm on different cohorts. Robustness analysis was performed using interaction tests. A p value smaller than 0.05 was considered as significant. Post-hoc power analysis was performed for the primary outcome. With a two-sided α of 0.05, a moderate effect size (Cohen’s d = 0.5), and the achieved sample size of 95 patients, the power exceeded 0.92, indicating sufficient sensitivity to detect the observed differences. Results Study participants Basic information of patients and the comparison for subgroups were shown in Tables 1 and 2. There were 112 stair-step artifacts were found, distributed in 46 patients for standard group, and the number decreased to 26 stair-step artifacts in 19 patients as Fig. 2 showed. 86 stair-step artifacts were resolved with ZeeFree algorithm, representing a paired per-lesion reduction of 76.8% (86/112). As we compare the discrepancy between different subgroups, all the baseline information between subgroups was not significantly different if divided by heart rate (all p > 0.05) including the radiation dose. However, for subgroups divided by BMI, significant difference was always found between basic information since these parameters were clearly affected by weight or BMI, especially the radiation dose parameters. The radiation doses were significantly higher for patients with higher BMI (p = 0.0002 for CTDI, 0.0007 for DLP and 0.0015 for SSDE).Table 1Basic information of involved patients.CharacteristicValue(n = 95)Patient Number with Stair-step Artifacts46Number of Stair-step Artifacts112GenderMale 52 (54.7%)Female 43 (45.3%)Weight/kg70.81 ± 11.68Height/m1.66 [1.72, 1.6]Body mass index(BMI)/(kg/m2)25.57 ± 3.09Heart rate/bpm68.68 ± 10.62Contrast media/ml65.0 [70.0, 65.0]CTDI/mGy12.3 [19.8, 7.76]DLP/mGy*cm174.0 [295.5, 113.5]SSDE/mGy16.80 [27.30, 10.85]CTDI: CT dose index.DLP: dose length product.SSDE: size-specific dose estimate.Note: Descriptive statistics for the same variable may differ across tables because they are derived from different patient subgroups. Mean ± SD is used for normally distributed data, while medians and interquartile ranges is used for non-normally distributed data, consistent with the statistical methods section.Table 2Comparison of basic information for subgroups.Heart rate > 75Heart rate ≤ 75p valueBMI > 25BMI ≤ 25p valuePatient Number2669NA5243NAPatient Number with Stair-step Artifacts1234NA2224NANumber of Stair-step Artifacts2785NA5755NAGenderMale 15 (57.7%)Female 11 (42.3%)Male 37 (53.6%)Female 32 (46.4%)0.9012Male 37 (71.2%)Female 15 (28.8%)Male 28 (65.1%)Female 11 (34.9%)0.6830Weight/kg70.69 ± 12.2370.86 ± 11.470.952477.0 [82.0, 73.0]60.0 [67.0, 58.0] < 0.0001Height/m1.66 ± 0.081.66 [1.72, 1.60]0.60611.70 [1.73, 1.62]1.65 [1.70, 1.60]0.0882Body mass index(BMI)/(kg/m2)25.49 ± 2.9925.59 ± 3.120.8870NANANAHeart rate/bpmNANANA68.56 ± 10.5268.84 ± 10.740.8997Contrast media/ml65.0 [70.0, 65.0]65.0 [70.0, 64.5]0.702366.2 [70.0, 65.0]65.0 [67.5, 62.5]0.0088CTDI/mGy12.1 [14.775, 9.4625]12.4 [22.6, 7.01]0.930215.3 [25.325, 10.385]9.31 [12.75, 6.045]0.0002DLP/mGy*cm172.5 [229.25, 125.75]174.0 [312.0, 104.0]0.8087232.0 [354.5, 131.5]136.0 [180.5, 82.85]0.0007SSDE/mGy16.80 [20.9, 13.5]16.80 [31.20, 9.81]0.893823.51 ± 12.4513.90 [17.80, 9.05]0.0015CTDI: CT dose index.DLP: dose length product.SSDE: size-specific dose estimate.NA: not available.Note: Descriptive statistics for the same variable may differ across tables because they are derived from different patient subgroups. Mean ± SD is used for normally distributed data, while medians and interquartile ranges is used for non-normally distributed data, consistent with the statistical methods section.Fig. 2Comparison of the distribution of stair-step artifacts in different branches of coronary arteries. Table 1Basic information of involved patients.CharacteristicValue(n = 95)Patient Number with Stair-step Artifacts46Number of Stair-step Artifacts112GenderMale 52 (54.7%)Female 43 (45.3%)Weight/kg70.81 ± 11.68Height/m1.66 [1.72, 1.6]Body mass index(BMI)/(kg/m2)25.57 ± 3.09Heart rate/bpm68.68 ± 10.62Contrast media/ml65.0 [70.0, 65.0]CTDI/mGy12.3 [19.8, 7.76]DLP/mGy*cm174.0 [295.5, 113.5]SSDE/mGy16.80 [27.30, 10.85]CTDI: CT dose index.DLP: dose length product.SSDE: size-specific dose estimate.Note: Descriptive statistics for the same variable may differ across tables because they are derived from different patient subgroups. Mean ± SD is used for normally distributed data, while medians and interquartile ranges is used for non-normally distributed data, consistent with the statistical methods section. Basic information of involved patients. Male 52 (54.7%) Female 43 (45.3%) CTDI: CT dose index. DLP: dose length product. SSDE: size-specific dose estimate. Note: Descriptive statistics for the same variable may differ across tables because they are derived from different patient subgroups. Mean ± SD is used for normally distributed data, while medians and interquartile ranges is used for non-normally distributed data, consistent with the statistical methods section. Table 2Comparison of basic information for subgroups.Heart rate > 75Heart rate ≤ 75p valueBMI > 25BMI ≤ 25p valuePatient Number2669NA5243NAPatient Number with Stair-step Artifacts1234NA2224NANumber of Stair-step Artifacts2785NA5755NAGenderMale 15 (57.7%)Female 11 (42.3%)Male 37 (53.6%)Female 32 (46.4%)0.9012Male 37 (71.2%)Female 15 (28.8%)Male 28 (65.1%)Female 11 (34.9%)0.6830Weight/kg70.69 ± 12.2370.86 ± 11.470.952477.0 [82.0, 73.0]60.0 [67.0, 58.0] < 0.0001Height/m1.66 ± 0.081.66 [1.72, 1.60]0.60611.70 [1.73, 1.62]1.65 [1.70, 1.60]0.0882Body mass index(BMI)/(kg/m2)25.49 ± 2.9925.59 ± 3.120.8870NANANAHeart rate/bpmNANANA68.56 ± 10.5268.84 ± 10.740.8997Contrast media/ml65.0 [70.0, 65.0]65.0 [70.0, 64.5]0.702366.2 [70.0, 65.0]65.0 [67.5, 62.5]0.0088CTDI/mGy12.1 [14.775, 9.4625]12.4 [22.6, 7.01]0.930215.3 [25.325, 10.385]9.31 [12.75, 6.045]0.0002DLP/mGy*cm172.5 [229.25, 125.75]174.0 [312.0, 104.0]0.8087232.0 [354.5, 131.5]136.0 [180.5, 82.85]0.0007SSDE/mGy16.80 [20.9, 13.5]16.80 [31.20, 9.81]0.893823.51 ± 12.4513.90 [17.80, 9.05]0.0015CTDI: CT dose index.DLP: dose length product.SSDE: size-specific dose estimate.NA: not available.Note: Descriptive statistics for the same variable may differ across tables because they are derived from different patient subgroups. Mean ± SD is used for normally distributed data, while medians and interquartile ranges is used for non-normally distributed data, consistent with the statistical methods section. Comparison of basic information for subgroups. Male 15 (57.7%) Female 11 (42.3%) Male 37 (53.6%) Female 32 (46.4%) Male 37 (71.2%) Female 15 (28.8%) Male 28 (65.1%) Female 11 (34.9%) CTDI: CT dose index. DLP: dose length product. SSDE: size-specific dose estimate. NA: not available. Note: Descriptive statistics for the same variable may differ across tables because they are derived from different patient subgroups. Mean ± SD is used for normally distributed data, while medians and interquartile ranges is used for non-normally distributed data, consistent with the statistical methods section. Comparison of the distribution of stair-step artifacts in different branches of coronary arteries. Consistency of the doctor’s evaluation results As showed in Table 3, a great consistency was found between two doctors’ evaluation results for no matter the severity of artifacts, image quality or diagnostic confidence, with the Cohen’ s kappa values of 0.9627, 95% CI: [0.9419, 0.9811], 0.9674, 95% CI: [0.9472, 0.9833] and 0.9687, 95% CI: [0.9496, 0.9836] respectively. And the interaction p‑values were 0.754, 0.839 and 0.863. These results indicate no significant interaction between readers, supporting consistent interpretation across different readers.Table 3Consistency of two doctors’ evaluation results.Severity of Stair-step ArtifactsImage QualityDiagnostic ConfidenceCohen’s Kappa0.96270.96740.968795% CI[0.9419, 0.9811][0.9472, 0.9833][0.9496, 0.9836]CI: confidence interval. Table 3Consistency of two doctors’ evaluation results.Severity of Stair-step ArtifactsImage QualityDiagnostic ConfidenceCohen’s Kappa0.96270.96740.968795% CI[0.9419, 0.9811][0.9472, 0.9833][0.9496, 0.9836]CI: confidence interval. Consistency of two doctors’ evaluation results. CI: confidence interval. Objective image quality comparison As showed in Table 4 and Fig. 8, the SNR and CNR didn’t show significant difference between Group Standard and Group ZeeFree, with a p value of 0.949 for SNR and 0.931 for CNR, indicating that the perceived improvement in image quality and diagnostic usability was driven by artifact reduction rather than changes in noise or contrast characteristics.Table 4Comparison of objective measurement results.StandardZeeFreep valueSNR19.29 ± 5.2019.34 ± 5.180.949CNR22.81 [25.27, 18.40]22.53 [25.05, 19.30]0.931SNR: signal-to-noise ratio.CNR: contrast-to-noise ratio.Note: Descriptive statistics for the same variable may differ across tables because they are derived from different patient subgroups. Mean ± SD is used for normally distributed data, while medians and interquartile ranges is used for non-normally distributed data, consistent with the statistical methods section. Table 4Comparison of objective measurement results.StandardZeeFreep valueSNR19.29 ± 5.2019.34 ± 5.180.949CNR22.81 [25.27, 18.40]22.53 [25.05, 19.30]0.931SNR: signal-to-noise ratio.CNR: contrast-to-noise ratio.Note: Descriptive statistics for the same variable may differ across tables because they are derived from different patient subgroups. Mean ± SD is used for normally distributed data, while medians and interquartile ranges is used for non-normally distributed data, consistent with the statistical methods section. Comparison of objective measurement results. SNR: signal-to-noise ratio. CNR: contrast-to-noise ratio. Note: Descriptive statistics for the same variable may differ across tables because they are derived from different patient subgroups. Mean ± SD is used for normally distributed data, while medians and interquartile ranges is used for non-normally distributed data, consistent with the statistical methods section. Subjective image quality comparison As the example in Fig. 3(A, B and C), before the ZeeFree algorithm was applied, the severe stair-step artifacts could be found on LAD (A, C) and RCA (B, C) and the vessel could not be properly diagnosed. After the application of ZeeFree, the LAD and RCA can be fully displayed with no more discontinuities in the image.Fig. 3CPR, cVRT images and locally enlarged images of a 55-year-old woman (average heart rate of 62 bpm and BMI of 21.56 kg/m2). (A) CPR images with stair-step artifact on LAD and improvement after ZeeFree’s correction; (B) CPR images with stair-step artifact on RCA and improvement after ZeeFree’s correction; (C) cVRT images showed clearly stair-step artifacts for vessels and myocardium, but corrected after ZeeFree algorithm was used for reconstruction. CPR, cVRT images and locally enlarged images of a 55-year-old woman (average heart rate of 62 bpm and BMI of 21.56 kg/m2). (A) CPR images with stair-step artifact on LAD and improvement after ZeeFree’s correction; (B) CPR images with stair-step artifact on RCA and improvement after ZeeFree’s correction; (C) cVRT images showed clearly stair-step artifacts for vessels and myocardium, but corrected after ZeeFree algorithm was used for reconstruction. Figures 4 and 5 shows the comparison between Group Standard and Group ZeeFree. Based on the comparison of evaluation scores between two groups of reconstructed images by doctors, we could demonstrate that the severity of stair-step artifacts was significantly reduced by ZeeFree, and the image quality and doctors’ diagnostic confidence were significantly improved as well, as indicated by the subjective scoring (p < 0.001).Fig. 4Comparison of image quality score between reconstructed images with standard reconstruction mode and ZeeFree mode of doctor 1.Fig. 5Comparison of image quality score between reconstructed images with standard reconstruction mode and ZeeFree mode of doctor 2. Comparison of image quality score between reconstructed images with standard reconstruction mode and ZeeFree mode of doctor 1. Comparison of image quality score between reconstructed images with standard reconstruction mode and ZeeFree mode of doctor 2. Comparison of subgroups for standard and ZeeFree reconstructions To evaluate whether the improvement of image quality with ZeeFree could be affected by different factors, the patients were divided into subgroups by heart rate and BMI. As Figs. 6 and 7 indicated, the image quality remains quite good even if the heart rate was over 75 bpm comparing the evaluation scores for both doctors. For subgroups divided by BMI, we could see from the bar plot that patients with BMI over 25 kg/m2 showed a slightly better image quality than those with BMI below 25 kg/m2. But images with ZeeFree could always show significantly better image quality, diagnostic confidence, and less artifacts than those without ZeeFree (all p < 0.001). The severity of artifacts, image quality and diagnostic confidence were assessed, the interaction p‑values for heart rate subgroups were 0.178, 0.885 and 0.535, The interaction p‑values for BMI subgroups were 0.426, 0.996 and 0.693. All interaction p-values were non-significant, indicating that the effect of the intervention on the outcomes did not differ significantly across heart rate or BMI subgroups.Fig. 6Comparison of image quality score between reconstructed images with standard reconstruction mode and ZeeFree mode for different subgroups of doctor 1. (A) Heart rate > 75 bpm; (B) Heart rate ≤ 75 bpm; (C) BMI > 25 kg/m2; (D) BMI ≤ 25 kg/m2.Fig. 7Comparison of image quality score between reconstructed images with standard reconstruction mode and ZeeFree mode for different subgroups of doctor 2. (A) Heart rate > 75 bpm; (B) Heart rate ≤ 75 bpm; (C) BMI > 25 kg/m2; (D) BMI ≤ 25 kg/m2.Fig. 8Comparison of SNR and CNR between reconstructed images with standard reconstruction mode and ZeeFree mode. Comparison of image quality score between reconstructed images with standard reconstruction mode and ZeeFree mode for different subgroups of doctor 1. (A) Heart rate > 75 bpm; (B) Heart rate ≤ 75 bpm; (C) BMI > 25 kg/m2; (D) BMI ≤ 25 kg/m2. Comparison of image quality score between reconstructed images with standard reconstruction mode and ZeeFree mode for different subgroups of doctor 2. (A) Heart rate > 75 bpm; (B) Heart rate ≤ 75 bpm; (C) BMI > 25 kg/m2; (D) BMI ≤ 25 kg/m2. Comparison of SNR and CNR between reconstructed images with standard reconstruction mode and ZeeFree mode. Consistency of vascular stenosis results The results showed that the non-evaluable (stair-step artifact score < 3) rate of CCTA images in the standard reconstruction group reached 51% (57/112), while that in the ZeeFree reconstruction group was merely 4.5% (5/112). We adopted a three-category classification framework for agreement analysis, treating non-evaluable, mild stenosis (< 50%), and significant stenosis (≥ 50%) as three distinct and clinically meaningful outcomes, the Cohen’ s Kappa value between the standard reconstruction and ICA was 0.05, indicating poor consistency, in contrast, the Cohen’ s Kappa value of the ZeeFree reconstruction with ICA was 0.85, which demonstrated a good consistency. (Table 5, Fig. 9).Table 5Consistency of stenosis assessment.StandardZeeFreeICANon-evaluable2410Evaluable < 50%114445 ≥ 50%1333Cohen’s Kappa 1 = 0.05, consistency of Standard vs. ICA.Cohen’s Kappa 2 = 0.85, consistency of ZeeFree vs. ICA.ICA: invasive coronary angiography.Fig. 9CPR images of CCTA and ICA images of a 60-year-old man (average heart rate of 64 bpm and BMI of 25.61 kg/m2). CPR images with stair-step artifact (stenosis rate of 50%) on LAD and improvement after ZeeFree’s correction, meanwhile, ICA images confirmed the absence of stenosis at this site. Table 5Consistency of stenosis assessment.StandardZeeFreeICANon-evaluable2410Evaluable < 50%114445 ≥ 50%1333Cohen’s Kappa 1 = 0.05, consistency of Standard vs. ICA.Cohen’s Kappa 2 = 0.85, consistency of ZeeFree vs. ICA.ICA: invasive coronary angiography. Consistency of stenosis assessment. Cohen’s Kappa 1 = 0.05, consistency of Standard vs. ICA. Cohen’s Kappa 2 = 0.85, consistency of ZeeFree vs. ICA. ICA: invasive coronary angiography. CPR images of CCTA and ICA images of a 60-year-old man (average heart rate of 64 bpm and BMI of 25.61 kg/m2). CPR images with stair-step artifact (stenosis rate of 50%) on LAD and improvement after ZeeFree’s correction, meanwhile, ICA images confirmed the absence of stenosis at this site. Discussion In our study, we explored the effectiveness of the new ZeeFree algorithm in eliminating stair-step artifacts. The images were assessed in terms of the severity of stair-step artifacts, image quality and diagnostic confidence. We also investigated the differences in the effectiveness of the ZeeFree algorithm among various subgroups of the population. If divided by heart rate, there were no significant difference for basic information between subgroups. But there were significant difference in body weight, contrast agent dosage and radiation dose (p < 0.05), which were directly affected by BMI23,24, between subgroups divided by BMI. The interobserver agreement for stair-step artifact scoring was excellent. This may be attributed to intensive training of cardiovascular radiology readers on clear, standardized scoring criteria, alongside the inherent ease of identifying stair-step artifacts. From our comparison results, we could conclude that ZeeFree algorithm could effectively help solve the stair-step artifacts with the total number of artifacts decreasing from 112 to 26, reduced by 76.8%. The significant improvement of image quality and diagnostic confidence, and the reduction of artifacts’ severity were found as well after the ZeeFree algorithm was applied, the observed improvement in structural continuity from ZeeFree is achieved without altering the underlying noise statistics. Since ZeeFree algorithm was applied during reconstruction step, it didn’t require the higher radiation dose for the scanning protocol which prevent the patients from receiving more radiation exposure than the standard examination process. Meanwhile, ZeeFree is fully integrated into the routine post-processing workflow, delivering a reconstruction speed of 90 images per second and enabling standard image acquisition without additional secondary post-processing. There have been four publications of ZeeFree algorithm with the new photon-counting CT13–16, all of them emphasize the effectiveness of the ZeeFree reconstruction algorithm in photon-counting CT on CCTA scanning. However, our study is the experience of evaluating the effectiveness of ZeeFree algorithm on dual-source CT scanners. As far as we know, dual-source CT offers greater cost-effectiveness, this is particularly advantageous in resource-limited healthcare settings. Improved image quality and diagnostic usability were observed across our and previous studies. Compared with Moser LJ, our patient-level step artifact incidence decreased from 48% (46/95) to 20% (19/95), while theirs fell from 46% (23/50) to 16% (8/50); our lesion-level reduction was marginally higher (76.8% vs. 73%). Such subtle differences may stem from variations in sample size, scanning settings, or patient baseline features; further multicenter, large-sample studies are needed to validate general applicability. Notably, our core objective is to deliver independent confirmatory evidence, with the present findings further validating and strengthening the cross-platform robustness of ZeeFree. Although Lisi C and the group verified the effectiveness of the ZeeFree algorithm in removing stair-step artifacts and utilized the post-processed images for the computation of CT-fractional flow reserve (CT-FFR) and plaque analysis14. Unlike previous studies, our study was a preliminary exploratory analysis against ICA, not intended as definitive clinical validation. In the assessment of vascular stenosis, the ZeeFree group exhibited good consistency with ICA, which demonstrates the potential of ZeeFree to improve diagnostic accuracy and provides valuable implications for future research. We acknowledge, however, that the ICA sample size was relatively small, which may limit the generalizability of these findings, and we aim to address this limitation in subsequent studies. In subgroup analysis, we found that there was no difference of image quality between the groups of average heart rate > 75 bpm and the average heart rate ≤ 75 bpm, which indicating that ZeeFree algorithm may have little relationship with heart rate in correcting stair-step artifacts. In the BMI > 25 kg/m2 group and the BMI ≤ 25 kg/m2 group, the correction effect of ZeeFree in the BMI > 25 kg/m2 group was slightly better than that in the BMI ≤ 25 kg/m2 group, indicating a better performance for ZeeFree algorithm on obese patients. There were still limitations remaining in our study. First, the sample size is relatively small, and future studies should include larger patient populations with additional clinical endpoint investigations. Second, the study did not supplement data with fewer occurrences of stair-step artifacts, which may introduce some bias to the results. Third, the reconstruction algorithm is vendor-specific and cannot be generalized to scanners from different manufacturers. Last but not the least, the data comes from a single center and may not fully describe the applicability to a larger population. For example, more obese patients could be collected for future analysis to prove the better performance of ZeeFree on patients with bigger BMI, and further subgroup analyses based on heart rate could be conducted to verify the heart rate threshold for effective stair-step artifacts correction by ZeeFree, which could be explored in the future study. Conclusion The ZeeFree algorithm can effectively eliminate the stair-step artifacts that may occur during the sequence scanning in CCTA. From our results, the reconstructed images after incorporating the ZeeFree algorithm show a significant reduction in stair-step artifacts, and both the image quality and the diagnostic confidence of doctors are markedly improved. Supplementary Information Supplementary Information 1. Supplementary Information 2. Supplementary Information 1. Supplementary Information 2.