Exploring double low-dose CT technology for achieving coronary mixed reality radiation dose reduction: a preliminary study in coronary artery disease patients

Introduction

Coronary artery disease (CAD) is a cardiac condition caused by stenosis or occlusion of the coronary arteries, which poses a serious threat to human life and health (1). Although coronary computed tomography (CT) angiography (CCTA) is a preferred imaging modality for CAD, the final results are still rendered on two-dimensional films or planes (2,3). Novel visualisation technology, represented by mixed reality, has been increasingly applied in the medical field. Mixed reality integrates the advantages of virtual reality (VR) and augmented reality (AR); specifically, it organically combines and enables interaction between the virtual world, the real world, and the user (4,5). Previous studies have shown that mixed reality can be used for the diagnosis of CAD and congenital heart disease (CHD). Lien et al. demonstrated that mixed reality facilitates antegrade wiring for coronary chronic total occlusion (6). Lau et al. showed that mixed reality is superior to standard digital imaging and communications in medicine (DICOM) images in the visualisation and management of CHD (7). Ponzoni et al. demonstrated that mixed reality is effective for preoperative planning and intraoperative assistance in the surgical correction of complex congenital heart defects (8).

Currently, mixed reality data mainly originate from DICOM data generated after CT examination (9,10). When mixed reality is employed, the associated radiation dose is often overlooked. In recent years, double low-dose CT technology has been applied in CCTA, chiefly referring to low tube voltage and low contrast media dose. Wu et al. demonstrated that high-resolution double-low CCTA improves image quality and the diagnosis of in-stent restenosis in patients after percutaneous coronary intervention (PCI) (11). Li et al. showed that, compared with the routine-dose protocol, the double-low dose one-stop coronary and carotid-cerebrovascular CTA provides higher image quality with lower radiation and contrast doses (12). Previous investigations into coronary mixed reality have largely utilised standard-dose CT protocols (typically 120 kVp) derived from routine diagnostic workflows (13,14). To the best of our knowledge, the specific optimisation of CT acquisition parameters for the sole purpose of coronary mixed reality has not been investigated. Specifically, it remains unknown whether the double low-dose technique (low tube voltage combined with low contrast volume), which is now established in conventional CCTA, can be effectively translated to coronary mixed reality. Addressing this gap is clinically essential, as the adoption of mixed reality technology should not inadvertently expose patients to unnecessary radiation solely to achieve a more “vivid” three-dimensional (3D) display. Therefore, the hypothesis of this study is that double low-dose CT acquisition can generate DICOM data sufficient for high-fidelity coronary mixed reality while simultaneously minimising patient radiation exposure.

In the present study, we adopted double low-dose CT technology for coronary mixed reality data acquisition. The primary objective was to evaluate whether this optimised protocol could maintain coronary mixed reality model quality whilst reducing radiation and contrast doses. We hypothesised that (I) coronary mixed reality models derived from double low-dose CT data would achieve quality scores not significantly different from those of conventional-dose protocols, with an expected radiation dose reduction exceeding 80%; and (II) coronary mixed reality-assisted coronary angiography (CAG) navigation would significantly reduce intraoperative fluoroscopy time and contrast media consumption compared with standard CAG procedures. To test these hypotheses, we conducted a prospective study comparing radiation metrics, model fidelity, diagnostic performance, and perioperative outcomes. We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0317/rc).

Methods

Study design and participant

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, China (No. 2022DZGZR-075), and all patients provided informed consent before the examination. A total of 192 consecutive patients with suspected CAD defined by the presence of typical or atypical angina, anginal equivalent symptoms, or evidence of ischemia on non-invasive stress testing, and a pre-test probability of obstructive CAD >5% according to European Society of Cardiology (ESC) guidelines (15), were prospectively enrolled in Jinling Hospital, Medical School of Nanjing University, China from October 2022 to March 2024. Patients were prospectively and randomly assigned to one of three groups in a 1:1:1 ratio. The randomization sequence was computer-generated by an independent statistician using SAS 9.4. Allocation concealment was maintained using sequentially numbered, opaque, sealed envelopes, which were opened by the CT technologist only after patient consent and positioning were confirmed. Patients were randomly divided into Group A [double low-dose coronary mixed reality group, 70 kVp, 300 mgI/mL (Ultravist^®, Bayer AG, Berlin, Germany), n=65], Group B [conventional-dose coronary mixed reality group, 120 kVp, 370 mgI/mL (Ultravist^®, Bayer AG), n=62] and Group C (conventional-dose CCTA group, 120 kVp, 370 mgI/mL, n=65). According to the ESC guidelines to select CAG treatment (16).

The inclusion criteria were as follows: (I) underwent CCTA examinations; (II) clinically suspected CAD defined by the presence of typical or atypical angina, anginal equivalent symptoms, or evidence of ischemia on non-invasive stress testing, according to ESC guidelines (15).

The exclusion criteria were as follows: (I) had a body mass index (BMI) ≥30 kg/m²; (II) were pregnant; (III) were <18-year-old; (IV) were allergic to iodinated contrast media; or (V) had received heart surgery or coronary stenting.

The diagnosis of obstructive CAD (≥50% diameter stenosis in any major epicardial vessel or branch) was not an inclusion criterion but rather the endpoint assessed by the reference standard (invasive CAG) in a subset of patients who underwent CAG based on clinical indication. This endpoint was used to evaluate the diagnostic performance of coronary mixed reality.

Data acquisition

A dual-source CT device (SOMATOM Definition Flash, Siemens Healthcare, Germany) was used. Prospective electrocardiogram (ECG)-triggered sequential acquisition was performed, employing the patient’s own ECG signal; a heart rate of 60 beats/min and sinus rhythm were required. Automatic tube current modulation (ATCM) was applied. Dedicated bolus-tracking software was used to automatically initiate the scan once the attenuation within a region of interest (ROI) placed in the descending aorta reached 80 Hounsfield units (HUs). A total of 60 mL of contrast medium was injected at a rate of 4–5 mL/s via a high-pressure syringe (Ulrich Medical, Ulm, Germany). The remaining parameters were as follows: pitch, 3.4; tube rotation time, 0.28 s; and collimation width, 64 × 2 mm × 0.6 mm.

Data reconstruction

All CT DICOM data were reconstructed on an image postprocessing workstation (Syngo.via, Siemens Healthineers, Erlangen, Germany). The sinogram-affirmed iterative reconstruction (SAFIRE) strength levels 1–5 (S1–S5) were selected for Group A. In accordance with previous studies, S3 was selected for the conventional-dose groups (Group B and Group C) (14,15). The reconstruction slice thickness was 0.75 mm, the reconstruction interval was 0.4 mm, and the B46f soft-tissue convolution kernel was used.

Data quality evaluation

The quality of the DICOM data was evaluated by two radiologists (with 10 and 20 years of experience in cardiovascular imaging, respectively). The ROIs were measured at the following locations: (I) the aortic root; (II) the left main (LM); (III) the left anterior descending (LAD); (IV) the left circumflex (LCX); (V) the right coronary artery (RCA); and (VI) the pectoralis major (PM). The standard deviation (SD) of the CT value of the PM was defined as the noise level. The ROIs were selected to avoid the vascular wall, plaques or calcifications as consistently as possible. The size of the ROI was recorded as the maximum diameter near the lumen (Figure 1).

Figure 1 Schematic illustration of ROIs selection on DICOM data for different coronary arteries. (A) Gross anatomy of the coronary arteries; (B) left coronary artery and its branches; (C) right coronary artery. AO, aorta; DICOM, digital imaging and communications in medicine; LAD, left anterior descending artery; LCX, left circumflex artery; LM, left main; RCA, right coronary artery; ROI, region of interest.

The signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) were calculated as:

SNR=CTtargetvesselSDpectoralismajor[1]

CNR=CTtargetvessel−CTpectoralismajorSDpectoralismajor[2]

The quality of the DICOM data was evaluated using a 4-point scale (16): 1= poor (severe artefacts, non-diagnostic); 2= fair (moderate artefacts, limited diagnostic value); 3= good (slight artefacts, sufficient for clinical diagnosis); 4= excellent (no artefacts, clear anatomy). A score of ≥3 points was considered to meet the requirements for clinical diagnosis.

Radiation dose estimation

The dose length product (DLP) and volume CT dose index (CTDI_vol) for all enrolled patients were recorded and used to calculate the effective dose (ED) according to the following formula: ED = DLP × k, where k is the dose conversion factor, equal to 0.014 mSv/(mGy·cm) (17,18).

Coronary 3D modelling and mixed reality display

The CT DICOM data were imported into holographic visual 3D modelling software (VISUAL Co., Ltd., Beijing, China). The automatic segmentation and visual edge detection functions were selected and applied. The extracted target organs and tissues were post-processed using morphological dilation (expansion) to fill small gaps, morphological erosion (corrosion) to remove fine spurious connections, and Gaussian smoothing (smoothing) to reduce surface irregularity, then saved in the visual 3D (V3D) format. The 3D modelling software may also be used to evaluate the degree of coronary stenosis; by utilising multiplanar reformation and a volume rendering reconstruction function, the degree of coronary stenosis can be clearly displayed. In this study, the identification and grading of coronary artery stenosis (≥50% diameter reduction) were performed by uAI-Coronary CTA (United Imaging Intelligence, Beijing, China), a deep-learning-based artificial intelligence (AI) software that automatically analyses CCTA DICOM data (Figure 2). The AI algorithm generates a stenosis severity label for each major coronary segment (19,20). This AI output is then displayed and visualised within the mixed reality environment (HoloLens 2, Microsoft, Redmond, USA), allowing the operator to interact with the 3D model and verify the AI findings. Thus, the reported diagnostic performance (AUC, sensitivity, specificity) reflects the combined performance of the AI detection system plus the mixed reality visualisation interface, not mixed reality alone. The role of mixed reality in the diagnostic process is to provide immersive 3D spatial context, enable 360° inspection, and facilitate operator confirmation; stenosis classification itself is initially performed by the AI algorithm.

Figure 2 Coronary stenosis AI analysis and holographic visual 3D modelling interface. (A) Coronary stenosis AI analysis. By multi-angle imaging and vascular annotation, visually display the anatomical structure of cardiac blood vessels and possible lesions. The red arrow represents the right coronary artery; the yellow arrow represents the measurement of the degree of coronary artery stenosis. (B) Coronary 3D modelling interface. The software automatically segments the coronary tree from DICOM data. (a) 3D modelling result; (b) axial view; (c) coronal view; (d) sagittal view. 3D, three-dimensional; AI, artificial intelligence; DICOM, digital imaging and communications in medicine.

Coronary mixed reality quality evaluation

Blinded to the experimental protocol, two radiologists independently assessed the quality of the coronary mixed reality models. In accordance with previous studies (9,10), the quality of coronary mixed reality was evaluated using the following four indices: artefacts, completeness, clarity, and accuracy (Figure 3):

• 1 point (poor), which involved severe artefacts, mostly missing structures, extremely poor clarity, and unevaluable accuracy;
• 2 points (general), which involved moderate artefacts, moderate partial loss of structures, poor clarity, and poor accuracy;
• 3 points (good), which involved slight artefacts, few missing structures, and good clarity and accuracy;
• 4 points (excellent), which involved almost no artefacts, no missing structures, good clarity, and high accuracy.

Figure 3 Coronary mixed reality quality evaluation on a four-point scale. (A) Score 1 (poor): the model exhibits severe stairstep artefacts along the vessel wall and fragmentation of the distal RCA branches (arrow), resulting in a non-diagnostic reconstruction. (B) Score 2 (moderate): the model shows moderate surface irregularity and partial loss of small-caliber diagonal branches (arrow). The overall lumen contour is preserved but remains blurry. (C) Score 3 (good): the model demonstrates smooth vessel contours with minimal artefacts (arrow). Only a few minor gaps are noted at the very distal segments of the LCX artery, without affecting diagnostic interpretability. (D) Score 4 (excellent): the model presents sharp, well-defined margins of the LAD and LCX arteries. There is no discernible missing structure or noise-induced artefact (arrow). LAD, left anterior descending; LCX, left circumflex; RCA, right coronary artery.

A score of ≥3 points was considered sufficient to meet clinical needs.

Indications for referral to invasive CAG

Not all enrolled patients underwent invasive CAG, owing to ethical and clinical considerations. The decision to proceed with CAG was made by the referring cardiologist on the basis of pre-specified clinical criteria, in accordance with ESC guidelines (15,16):

• persistent typical angina despite optimal medical therapy;
• high pre-test probability of obstructive CAD (≥15%) with positive non-invasive stress test (e.g., exercise ECG, myocardial perfusion imaging, or stress echocardiography);
• symptoms suggestive of acute coronary syndrome;
• inconclusive CCTA findings requiring invasive confirmation.

All CAG procedures were performed within 30 days after CCTA. Patients who did not meet any of the above criteria were managed conservatively and did not undergo CAG.

CAG

The severity of coronary stenosis was assessed using the diameter measurement method on digital subtraction angiography (DSA) cine fluoroscopy by two cardiologists (with 15 and 25 years of CAG experience, respectively). The degree of stenosis was calculated using the formula: coronary stenosis (%) = (a − b)/a × 100%, where a is the normal lumen diameter proximal to the stenosis, and b is the minimum lumen diameter at the site of stenosis (21).

Coronary mixed reality-assisted CAG

The coronary mixed reality model was matched to the actual anatomy under fluoroscopy and cine acquisition sequences in two standard postures. Automatic tracking and registration software (EPVision, China) was employed for real-time fusion of the coronary mixed reality images. The following perioperative parameters were recorded: (I) time to locate CAD (min), defined as the total time from the start of the search for the first lesion to the completion of screening for all CAD; (II) total procedural time (min), defined as the entire intervention time from catheter insertion to catheter removal; (III) total radiation time (min) and (IV) total radiation exposure (Gy·cm²), both provided and automatically recorded by the DSA system; and (V) contrast media dose (mL), as documented in the postoperative records.

Statistical analysis

SPSS version 24.0 (IBM Corp., Armonk, NY, USA) was used for all statistical analyses. The Kolmogorov-Smirnov test was employed to assess the normality of continuous data distribution. Quantitative variables are presented as mean ± SD, while categorical variables are expressed as frequencies or percentages. Baseline demographic characteristics were compared across the three groups using one-way analysis of variance (ANOVA) for continuous variables and the Chi-squared test for categorical variables.

For the objective quality metrics (SNR, CNR, CT attenuation values) obtained at different SAFIRE reconstruction strengths (S1–S5) in Group A, a one-way repeated-measures ANOVA was conducted with SAFIRE strength as the within-subject factor. Mauchly’s test of sphericity was applied; when sphericity was violated, the Greenhouse-Geisser correction was used. Post-hoc pairwise comparisons between SAFIRE levels were performed using Bonferroni correction. Comparisons of objective quality metrics between Group A (S5) and Group B (S3) were performed using a linear mixed-effects model with Group as a fixed effect and subject as a random effect, specifying an unstructured covariance structure to account for within-patient correlation in Group A. Subjective image quality scores (DICOM quality and coronary 3D modelling quality, ordinal data) within Group A were compared across SAFIRE levels using the Friedman test; post-hoc pairwise comparisons were conducted with Wilcoxon signed-rank tests and Bonferroni correction. Between-group comparisons of subjective scores (Group A, S5 vs. Group B, S3) were performed using the Mann-Whitney U test. Interobserver agreement for qualitative assessments was evaluated using weighted Cohen’s kappa (κ). Diagnostic performance of coronary mixed reality was assessed using receiver operating characteristic (ROC) curve analysis, with the area under the curve (AUC), sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) calculated using CAG as the reference standard. A two-tailed P<0.05 was considered statistically significant, except where Bonferroni correction was applied to adjust for multiple comparisons. Of note, the study was not designed as a formal non-inferiority trial; no non-inferiority margin was pre-specified, and all comparisons were performed as two-sided tests for differences. Absence of statistical significance should not be interpreted as proof of non-inferiority.

The primary endpoint for sample size determination was the difference in ED between Group A (double low-dose) and Group B (conventional dose). Based on preliminary data from our institution (12 patients per group, acquired during protocol development and not included in the final study), the expected ED was 1.15±0.28 mSv for Group A and 6.10±1.25 mSv for Group B, corresponding to a Cohen’s *d* of 5.46. Using an independent-samples *t*-test with α=0.05 and power =0.90, the minimum required sample size was 6 patients per group (calculated with G*Power 3.1.9.7). For the secondary endpoint of coronary mixed reality diagnostic performance (ROC analysis), we assumed an AUC of 0.92; with α=0.05 and power =0.90, a minimum of 56 patients undergoing CAG was required. To ensure robust comparisons and allow for potential exclusions, we prospectively enrolled 192 patients, with 65, 62, and 65 finally allocated to Groups A, B, and C, respectively. No post hoc power analyses were performed.

Results

Participant characteristics

The final cohort included 65 patients in Group A, 62 patients in Group B, and 65 patients in Group C (Figure 4). The mean age was (54.5±13.0) years in Group A, (53.0±11.8) years in Group B, and (56.3±12.1) years in Group C (P=0.421). BMI was (25.2±3.5) kg/m² in Group A, (24.7±3.0) kg/m² in Group B, and (23.8±3.3) kg/m² in Group C (P=0.388). Heart rate, systolic and diastolic blood pressure, sex distribution, height, and weight also showed no significant differences among the three groups (all P>0.05) (Table 1).

Figure 4 Patient flow diagram. Group A, the double low-dose mixed reality group; Group B, the conventional-dose mixed reality group; Group C, the conventional-dose computed tomography group. CAD, coronary artery disease; CAG, coronary angiography; CCTA, coronary computed tomography angiography.

Comparison of data quality

The subjective DICOM image quality scores within Group A were significantly different across the five SAFIRE levels [Friedman test, χ²(4)=158.6, P<0.001]. Post-hoc Wilcoxon signed-rank tests with Bonferroni correction showed that S5 [median 4, interquartile range (IQR), 3–4] was rated significantly higher than S4 (median 3, IQR, 3–4; P=0.012), S3 (median 3, IQR, 2–3; P<0.001), S2 (median 2, IQR, 2–3; P<0.001), and S1 (median 2, IQR, 1–2; P<0.001) (Table 2). The subjective evaluation scores of Group A (S5, 3.8±0.5) was also greater than that of Group A (S1–4, 2.9±0.4, 3.0±0.3, 3.2±0.6, 3.6±0.6) and Group B (S3, 3.5±0.7) (P<0.001). The interobserver agreement was good (κ=0.666).

Radiation dose and contrast media dose

Compared with Group B, Group A achieved a substantial reduction in radiation and contrast media doses. Specifically, CTDI_vol decreased from (30.8±5.6) to (5.9±1.4) mGy (about 81% reduction), DLP from (427.4±90.5) to (75.9±21.7) mGy·cm (about 82% reduction), and ED from (6.0±1.3) to (1.1±0.3) mSv (about 82% reduction). The contrast media dose was reduced by 47%, from (70.6±2.5) mL in Group B to (37.2±2.9) mL in Group A, and the injection rate was also significantly lower (3.5±0.1 vs. 4.9±0.3 mL/s). All differences were statistically significant (all P<0.001) (Table 3).

Comparison of coronary 3D modelling quality

The subjective quality scores for coronary 3D modelling (Figure 5) within Group A also differed significantly across SAFIRE levels [Friedman test, χ²(4)=133.5, P<0.001]. Pairwise comparisons revealed that S5 (3.8±0.5) was superior to S4 (3.6±0.6, P=0.04), S3 (3.2±0.6, P<0.001), S2 (3.0±0.3, P<0.001), and S1 (2.9±0.4, P<0.001). There was no statistically significant difference between Group A S5 and Group B S3 (3.5±0.7; Mann-Whitney U, P=0.15). All groups had mean scores ≥3 points, indicating clinical acceptability. Interobserver agreement ranged from moderate to good (κ=0.603–0.792).

Figure 5 Comparison of coronary 3D modelling quality. (A) The subjective scoring results of coronary 3D modelling quality between Group A and Group B. (B) The comparison of coronary 3D modelling results between Group A and Group B. (a) Group A (S1), (b) Group A (S2), (c) Group A (S3), (d) Group A (S4), (e) Group A (S5) and (f) Group B (S3). The yellow box indicates the details of the 3D modelling quality of the right coronary artery [(e) was the best performance]. Group A, the double low-dose mixed reality group; Group B, the conventional-dose mixed reality group. S1–S5, levels 1–5 of SAFIRE iterative reconstruction. 3D, three-dimensional.

Comparison of coronary mixed reality quality

Based on the above-mentioned comparison results, we selected Group A (S5) and Group B (S3) for third-party perspective comparisons. No statistically significant differences were observed between Group A (S5) and Group B (S3) in any of the qualitative indices—artefacts, integrity, clarity, or accuracy (all P>0.05, Mann-Whitney U test; Figures 5,6). At the same time, Group A was associated with significantly lower radiation and contrast agent doses than Group B.

Figure 6 Coronary mixed reality displays, diagnostic efficacy and clinical application. (A) Comparison of coronary mixed reality displays viewed from a third-party perspective. (a,b) Male, 56-year-old, adopting the Group A (S5) protocol. (c,d) Male, 67-year-old, according to the Group B protocol; (B) ROC curve analysis for coronary mixed reality diagnostic efficacy. Group A, the double low-dose mixed reality group; Group B, the conventional-dose mixed reality group. CT, computed tomography; ROC, receiver operating characteristic.

Diagnostic efficacy of coronary mixed reality

Among the 192 enrolled patients with suspected CAD, a total of 89 patients (Group A, n=33; Group B, n=29; Group C, n=27) proceeded to CAG based on clinical indication (such as persistent symptoms or high pre-test probability). As expected, patients referred to CAG had significantly higher pre-test probability, more frequent typical angina, and more positive stress tests, reflecting appropriate clinical decision-making for invasive confirmation (Table 4). This subset was used to evaluate the diagnostic performance of coronary mixed reality against the gold standard of CAG. In the clinically referred subgroup of patients who underwent CAG (n=62, representing 32.3% of the total cohort), the diagnostic performance of coronary mixed reality for detecting ≥50% coronary stenosis was evaluated using CAG as the reference standard. Given the potential for verification bias due to selective CAG referral, these estimates should be interpreted as preliminary and may overestimate real-world performance. Using the AI-assisted mixed reality workflow (where stenosis detection was performed by uAI-CCTA and the results were displayed via mixed reality), the diagnostic performance for detecting ≥50% coronary stenosis (with CAG as the reference standard) was: AUC =0.92 [95% confidence interval (CI): 0.88–0.96], sensitivity =0.94 (95% CI: 0.90–0.98), specificity =0.90 (95% CI: 0.86–0.94), accuracy =0.92 (95% CI: 0.89–0.95), PPV =0.89 (95% CI: 0.84–0.93), and NPV =0.96 (95% CI: 0.92–0.98). These values are likely optimistic due to disease enrichment in the CAG-referred subgroup. Future studies employing alternative reference standards [e.g., CT-fractional flow reserve (CT-FFR)] in unselected patients are needed to mitigate this bias.

Perioperative results of coronary mixed reality-assisted CAG

Table 5 shows that Group A or Group B had less time to search for CAD, less total surgical time, less total radiation time, less total radiation exposure and less contrast media dose than Group C (all P<0.017). Coronary mixed reality not only assisted with CAG navigation but also provided all-round 3D real-time imaging.

Discussion

Coronary mixed reality is a novel digital holographic visual technology that is increasingly influencing clinical diagnosis and treatment. In our study, we proved that double low-dose CT technology can be effectively applied to coronary mixed reality, reducing the radiation dose (CTDIvol, DLP, and ED) by approximately 81%, 82%, and 82%, respectively, and the contrast media dose by approximately 47%, without a statistically significant degradation in coronary mixed reality quality. Using CAG as the gold standard, the diagnostic efficiency of coronary mixed reality for coronary stenosis, with an AUC of 0.92, a specificity of 0.90, a sensitivity of 0.94 and an accuracy of 0.92. Finally, we found that coronary mixed reality‑assisted CAG can effectively reduce the time to diagnosis of CAD, total surgical time, total radiation time, total radiation exposure and contrast media dose by approximately 36%, 39%, 45%, 28%, and 40%, respectively.

In recent years, double low‑dose CT technology has been effectively applied in CCTA. Chen et al. demonstrated that a 7 kVp protocol with SAFIRE iterative reconstruction in patients with a BMI of 26–28 kg/m² achieved an ED of 1.10±0.10 mSv, representing an approximately 76% reduction compared with conventional 120 kVp protocols, while maintaining diagnostic image quality (22). Similarly, Ren et al. recently reported that combining 70 kVp with deep learning image reconstruction (DLIR) in slender patients (BMI ≤25 kg/m²) reduced radiation dose by 75.6% and contrast dose by 32.9%, with DLIR-H providing the highest image quality (23). A recent review on radiation dose reduction in CCTA emphasized that standardized, patientspecific protocols substantially reduce radiation exposure while maintaining diagnostic accuracy (24). Furthermore, Fahrni et al. explored the trade-off between iodine and radiation dose in coronary CT, highlighting the importance of optimizing both parameters simultaneously (25). By shifting the tube potential from 120 to 70 kVp, the mean photon energy aligns more closely with the iodine absorption peak (26). This exponential increase in the linear attenuation coefficient of iodine compensates for the reduction in contrast concentration (300 vs. 370 mgI/mL), resulting in an equivalent or even enhanced luminal gradient for DICOM quality (27). Our research builds upon these findings and applies dual low-dose CCTA technology to coronary mixed reality, thereby extending the utility of low-dose CT from conventional diagnostic imaging to novel holographic visualisation techniques.

The use of AI in CCTA analysis has gained substantial momentum. A large multicenter study employing a deep learning-based CCTA analysis system demonstrated excellent diagnostic performance for ≥50% coronary stenosis, with an accuracy of 92.8%, a sensitivity of 95.3%, and a specificity of 91.4% at the patient level, and reduced reading time from 5.94 to 2.01 min (P<0.001) (19). A systematic review of deep learning enabled CCTA further confirmed the high diagnostic performance and predictive capabilities of various deep learning models for plaque and stenosis quantification, highlighting strong correlations with intravascular ultrasound findings (28). These findings align with our use of the software’s built-in AI automatic identification function to further determine coronary stenosis, and support the integration of AI-assisted analysis into coronary mixed reality workflows.

The application of mixed reality in cardiac interventions has emerged as a transformative frontier. A recent feasibility study demonstrated that HoloLens^® 2 can serve as an effective intraoperative guide during minimally invasive cardiac surgery, including coronary artery bypass grafting via left anterior small thoracotomy, with the holographic image superimposed on the patient providing real-time anatomical guidance (29). Lien et al. reported the first pilot study of mixed realityguided antegrade wiring for chronic total occlusion PCI, showing that mixed reality technology enhanced operator confidence and safety in achieving intraluminal wiring (6). Kundzierewicz et al. provided a comprehensive review of extended reality applications in cardiac catheterization laboratories, covering current applications, benefits, drawbacks, and limitations of virtual, augmented, and mixed reality technologies in cardiac interventions (30). A review further explored the transformative applications of mixed reality in interventional cardiology, spanning transcatheter aortic valve implantation, PCI, and electrophysiology procedures (31). These emerging studies support our finding that coronary mixed reality-assisted CAG can significantly improve perioperative outcomes, including reduced radiation exposure and contrast media dose.

In our study, we selected the double low-dose CCTA protocol (70 kVp, 300 mg/mL) to acquire DICOM data for coronary mixed reality, and the final results showed no negative impact on mixed reality quality. The key enabling factor was the application of SAFIRE iterative reconstruction. SAFIRE is a second-generation iterative reconstruction algorithm that reduces image noise and artefacts by iteratively comparing measured projection data with forward-projected data in the raw data domain (22). Higher SAFIRE strength levels (S3–S5) provide progressively greater noise reduction and improved spatial resolution, which explains why Group A (S5) achieved the highest objective SNR and CNR values. However, SAFIRE also has several limitations. First, higher strength levels can produce an overly smooth (32). Second, SAFIRE are computationally more intensive than filtered back projection, leading to longer reconstruction times. Third, SAFIRE does not incorporate system optics models or neural network-based denoising. Despite these drawbacks, SAFIRE remains a robust and well-validated algorithm for low-dose CCTA, and our results confirm that S5 provides excellent DICOM quality for coronary mixed reality without clinically significant oversmoothing.

In our prior study, the use of 80 kVp and DLIR achieved a 42% reduction in radiation dose and a 31% reduction in contrast media dose for coronary mixed reality. The present study extends this work in several important respects. First, we employed a more aggressive 70 kVp protocol with SAFIRE iterative reconstruction, achieving substantially greater reductions—approximately 82% for radiation and 47% for contrast media, nearly double the previous savings. Second, this study is the first to systematically compare multiple iterative reconstruction strengths (SAFIRE S1–S5) to optimise noise suppression and model fidelity for mixed reality, whereas the prior work utilised only a single DLIR setting. Third, we evaluated the clinical utility of low-dose coronary mixed reality in assisting real-world CAG procedures, demonstrating significant perioperative improvements: reduced fluoroscopy time (45%), radiation exposure (28%), and contrast consumption (40%). Fourth, beyond the prior study’s focus on technical feasibility (‘data source optimisation’), our work provides comprehensive clinical validation, including diagnostic performance assessment against invasive CAG and perioperative outcome analysis. To our knowledge, no previous study has demonstrated the successful use of 70 kVp double low-dose CT for coronary mixed reality with validation in CAG-assisted procedures.

The physical rationale for using 70 kVp in this study is well established: lowering the tube potential shifts the mean photon energy closer to the iodine K-edge (33.2 keV), thereby exponentially increasing the linear attenuation coefficient of iodine. This effect enhances vascular contrast and partially offsets the use of a lower iodine concentration (300 vs. 370 mgI/mL) and a reduced contrast volume. However, this benefit entails a fundamental trade‑off. At 70 kVp, the X-ray photon flux is substantially lower than at 120 kVp, resulting in increased quantum noise. This elevation in noise is particularly problematic in patients with larger body habitus, as greater soft-tissue attenuation further degrades the SNR. Indeed, we excluded patients with BMI ≥30 kg/m² precisely because pilot testing showed that 70 kVp acquisitions in obese individuals produced unacceptably high noise levels, compromising both DICOM image quality and subsequent holographic edge detection for mixed reality modelling. This limitation is not unique to our study; prior investigations have consistently reported that low-kVp CCTA protocols are optimal only for patients with low-to-average body habitus (typically BMI ≤25–27 kg/m²). For patients with a BMI between 25 and 30 kg/m², a moderate tube potential (80 or 100 kVp) combined with higher iterative reconstruction strength may offer a better balance between dose reduction and image noise. For patients with obesity (BMI ≥30 kg/m²), standard-dose protocols (100–120 kVp) remain necessary, and alternative low-dose strategies, such as dual-energy CT or photon-counting CT, should be explored for mixed reality applications. Thus, the generalizability of our 70 kVp protocol is explicitly limited to non-obese patients, and readers should not extrapolate these dose savings to individuals with a larger body habitus.

The coronary mixed reality in this study achieved an AUC of 0.92, with sensitivity and specificity of 0.94 and 0.90, respectively. When contextualized within the existing literature, this diagnostic performance is comparable to the pooled estimates of AI-assisted CCTA from a recent meta-analysis (patient-level AUC 0.932, sensitivity 0.89, specificity 0.80), with slight numerical advantages in both sensitivity and specificity (33). When contextualizing the diagnostic performance of our AI‑assisted coronary mixed reality workflow (AUC =0.92, sensitivity =0.94, specificity =0.90), it is informative to compare with established benchmarks for conventional CCTA. Xiong et al. recently reported that standard CCTA for detecting ≥50% coronary stenosis achieved an AUC of 0.826, a sensitivity of 87.36% and a specificity of 93.48% using invasive CAG as the reference standard (34). These figures represent the typical diagnostic performance of unenhanced CCTA interpretation without AI assistance. Our workflow demonstrated a numerically higher AUC (0.92 vs. 0.83) and sensitivity (0.94 vs. 0.87) with comparable specificity (0.90 vs. 0.93). This improvement likely reflects the synergistic benefit of AI-assisted detection combined with mixed reality 3D visualization, although direct head-to-head comparisons are needed to confirm this advantage.

The high accuracy of mixed reality may be attributed to the following three mechanisms: (I) 360° inspection eliminates the projection-related blind spots of 2D reading, improving assessment of eccentric and ostial stenoses; (II) binocular stereopsis enhances depth perception, aiding interpretation of tortuous anatomy and heavy calcification; (III) integrated AI assistance creates a collaborative interpretation paradigm that may synergistically elevate sensitivity and specificity. Furthermore, recent advances in image-guided navigation techniques for cardiovascular interventions have demonstrated that integration of AR systems and patientspecific modelling can significantly improve procedural accuracy (35). Our findings confirm that coronary mixed reality‑assisted CAG can improve perioperative indices of CAG, consistent with the broader trend toward extended reality integration in cardiac procedures. The specific mechanisms can be summarized: (I) coronary mixed reality enables objective calculation of optimal C-arm projections prior to catheterization, directly reducing the iterative fluoroscopic search required for ostial and bifurcation lesions. (II) By functioning as a 3D roadmap, mixed reality provides real-time spatial orientation for anomalous or complex vessel origins (such as shepherd’s crook RCA), thereby shortening the time to search for CAD. (III) Immersive 3D assessment of bifurcation anatomy facilitates prospective selection of orthogonal working views, minimizing branch overlap and foreshortening during intervention.

We recognize several limitations. First, this single-center study, conducted at a tertiary hospital with experienced specialists, may not reflect the learning curve or demographic heterogeneity of broader community settings. Multicenter validation is warranted to confirm generalizability. Second, patients with BMI ≥30 kg/m² were excluded. As 70 kVp is optimal only for low-to-average body habitus, increased soft-tissue attenuation in obesity severely reduces photon flux, generating noise that impedes holographic edge detection. Given the high prevalence of obesity in CAD, findings are not generalizable to this population, and alternative low-dose strategies for mixed reality require exploration. Third, significant partial verification bias is inherent to this study. Invasive CAG was performed in only 62/192 (32.3%) patients, and referral was based on pre-specified clinical criteria (persistent angina, high pre-test probability, positive stress test, or inconclusive CCTA). Consequently, the CAG-referred subgroup had significantly higher pre-test probability (42.3% vs. 18.5%, P<0.001) and more frequent typical angina (45.2% vs. 23.8%, P=0.002) compared with non-referred patients. This disease-enriched spectrum inevitably leads to overestimation of sensitivity and specificity, a well-characterized form of verification bias. The reported AUC of 0.92, sensitivity of 0.94, and specificity of 0.90 are therefore likely optimistic and should not be generalized to unselected screening populations. Future studies should employ alternative reference standards (such as CT-FFR) to mitigate this bias. Fourth, the impact of coronary calcification on mixed reality model fidelity was not systematically evaluated. Severe calcification produces CT blooming artifacts, risking stenosis overestimation or segmentation inaccuracy. Although AI-assisted algorithms aim to mitigate calcium-related errors, dedicated evaluation in heavily calcified vessels, stratified by calcium score, is needed. Fifth, data were acquired on a single manufacturer’s dual-source scanner with a SAFIRE; generalizability to other platforms or newer techniques (such as ADMIRE, DLIR) remains unknown. Sixth, while the overall sample size was adequate for the primary radiation dose comparison, the diagnostic sub-analysis (n=62) was modest for subgroup evaluations. Larger multicenter studies with dedicated power calculations are required to establish diagnostic non-inferiority. Seventh, the commercial 3D modelling software may harbor inherent segmentation and registration limitations affecting accuracy. Finally, the observed perioperative improvements in mixed reality‑assisted CAG (Groups A/B) compared with conventional CAG (Group C) could be confounded by differences in lesion complexity, operator experience, access route, or procedural indication. While our subgroup analysis demonstrated no significant differences in these variables among the three CAG groups (all P>0.05), residual confounding cannot be completely excluded. We will further investigate these potential confounding factors in our future studies, ideally through a randomized controlled trial with blinded outcome assessment and standardized operator training.

Conclusions

In conclusion, double low-dose CT technology can be effectively applied to coronary mixed reality, achieving a substantial reduction in radiation dose without a significant difference in coronary mixed reality quality compared with the conventional-dose protocol. The AI-assisted coronary mixed reality workflow demonstrates high diagnostic performance for the detection of coronary stenosis. Coronary mixed reality can effectively improve perioperative indicators of CAG.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0317/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0317/dss

Funding: This research was supported by Clinical Trials from the Affiliated Drum Tower Hospital, Medical School of Nanjing University, China (No. 2024-LCYJ-MS-27), and China Hospital Reform and Development Research Institute of Nanjing University, Nanjing Drum Tower Hospital (No. NDYGN2024012).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0317/coif). The 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. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Ethics Committee of Jinling Hospital, Affiliated Hospital of Medical School, Nanjing University, China (No. 2022DZGZR-075), and all patients provided informed consent before the examination.

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. Benjamin EJ, Muntner P, Alonso A, Bittencourt MS, Callaway CW, Carson AP, et al. Heart Disease and Stroke Statistics-2019 Update: A Report From the American Heart Association. Circulation 2019;139:e56-e528. [Crossref] [PubMed]
2. Hulten E, Bittencourt MS, Singh A, O'Leary D, Christman MP, Osmani W, Abbara S, Steigner ML, Truong QA, Nasir K, Rybicki FF, Klein J, Hainer J, Brady TJ, Hoffmann U, Ghoshhajra BB, Hachamovitch R, Di Carli MF, Blankstein R. Coronary artery disease detected by coronary computed tomographic angiography is associated with intensification of preventive medical therapy and lower low-density lipoprotein cholesterol. Circ Cardiovasc Imaging 2014;7:629-38. [Crossref] [PubMed]
3. Han D, Liu J, Sun Z, Cui Y, He Y, Yang Z. Deep learning analysis in coronary computed tomographic angiography imaging for the assessment of patients with coronary artery stenosis. Comput Methods Programs Biomed 2020;196:105651. [Crossref] [PubMed]
4. Verhey JT, Haglin JM, Verhey EM, Hartigan DE. Virtual, augmented, and mixed reality applications in orthopedic surgery. Int J Med Robot 2020;16:e2067. [Crossref] [PubMed]
5. Kang SL, Shkumat N, Dragulescu A, Guerra V, Padfield N, Krutikov K, Chiasson DA, Chaturvedi RR, Yoo SJ, Benson LN. Mixed-reality view of cardiac specimens: a new approach to understanding complex intracardiac congenital lesions. Pediatr Radiol 2020;50:1610-6. [Crossref] [PubMed]
6. Lien CW, Lo HY, Lee WJ, Chen LP, Wu CC, Kao HL, Hsieh MY, Tan CT. Utilizing Mixed Reality Technology to Facilitate Coronary Chronic Total Occlusion Antegrade Wiring: A Pilot Study. JACC Case Rep 2025;30:104997. [Crossref] [PubMed]
7. Lau I, Gupta A, Ihdayhid A, Sun Z. Clinical Applications of Mixed Reality and 3D Printing in Congenital Heart Disease. Biomolecules 2022;12:1548. [Crossref] [PubMed]
8. Ponzoni M, Bertelli F, Yoo SJ, Peel B, Seatle H, Honjo O, Haller C, Barron DJ, Seed M, Lam CZ, Jeewa A, Carretero EG, Ordonez A, Valverde I. Mixed reality for preoperative planning and intraoperative assistance of surgical correction of complex congenital heart defects. J Thorac Cardiovasc Surg 2025;170:327-335.e1. [Crossref] [PubMed]
9. Li G, Cao Z, Wang J, Zhang X, Zhang L, Dong J, Lu G. Mixed reality models based on low-dose computed tomography technology in nephron-sparing surgery are better than models based on normal-dose computed tomography. Quant Imaging Med Surg 2021;11:2658-68. [Crossref] [PubMed]
10. Li G, Zhang L, Liu T, Wang J, Cao Z, Lu G, Zhang L. Exploration of thoracoabdominal aortic mixed reality optimisation and its clinical application value in type A aortic dissection. Eur Radiol 2023;33:4313-22. [Crossref] [PubMed]
11. Wu W, Zhan H, Wang Y, Ma X, Hou J, Ren L, Liu J, Wang L, Zhang Y. Improving image quality and in-stent restenosis diagnosis with high-resolution "double-low" coronary CT angiography in patients after percutaneous coronary intervention. Front Cardiovasc Med 2024;11:1330824. [Crossref] [PubMed]
12. Li W, Huang W, Li P, Wen Y, Shuai T, He Y, You Y, Yu J, Diao K, Song B. Application of deep learning image reconstruction-high algorithm in one-stop coronary and carotid-cerebrovascular CT angiography with low radiation and contrast doses. Quant Imaging Med Surg 2024;14:1860-72. [Crossref] [PubMed]
13. Goo HW, Park SJ, Yoo SJ. Advanced Medical Use of Three-Dimensional Imaging in Congenital Heart Disease: Augmented Reality, Mixed Reality, Virtual Reality, and Three-Dimensional Printing. Korean J Radiol 2020;21:133-45. [Crossref] [PubMed]
14. Li G, Zhou Y, Gao L, Tang Y, Liang Q, Zhang B. Double-low-dose CT combined with deep learning image reconstructions (DLIR) achieves coronary mixed reality data source optimization. Front Virtual Real 2026;7:1756733.
15. Knuuti J, Wijns W, Saraste A, Capodanno D, Barbato E, Funck-Brentano C, et al. 2019 ESC Guidelines for the diagnosis and management of chronic coronary syndromes. Eur Heart J 2020;41:407-77. [Crossref] [PubMed]
16. Collet JP, Thiele H, Barbato E, Barthélémy O, Bauersachs J, Bhatt DL, et al. 2020 ESC Guidelines for the management of acute coronary syndromes in patients presenting without persistent ST-segment elevation. Eur Heart J 2021;42:1289-367. [Crossref] [PubMed]
17. Fuchs TA, Stehli J, Bull S, Dougoud S, Clerc OF, Herzog BA, Buechel RR, Gaemperli O, Kaufmann PA. Coronary computed tomography angiography with model-based iterative reconstruction using a radiation exposure similar to chest X-ray examination. Eur Heart J 2014;35:1131-6. [Crossref] [PubMed]
18. Matveeva A, Schmitt RR, Edtinger K, Wagner M, Kerber S, Deneke T, Uder M, Barth S. Coronary CT angiography in patients with atrial fibrillation: Standard-dose and low-dose imaging with a high-resolution whole-heart CT scanner. Eur Radiol 2018;28:3432-40. [Crossref] [PubMed]
19. Chen Y, Yu H, Fan B, Wang Y, Wen Z, Hou Z, Yu J, Wang H, Tang Z, Li N, Jiang P, Wang Y, Yin W, Lu B. Diagnostic performance of deep learning-based coronary computed tomography angiography in detecting coronary artery stenosis. Int J Cardiovasc Imaging 2025;41:979-89. [Crossref] [PubMed]
20. Irannejad K, Mafi M, Krishnan S, Budoff MJ. Artificial intelligence in coronary CT angiography: transforming the diagnosis and risk stratification of atherosclerosis. Int J Cardiovasc Imaging 2025;41:1643-56. [Crossref] [PubMed]
21. Collet C, Onuma Y, Sonck J, Asano T, Vandeloo B, Kornowski R, et al. Diagnostic performance of angiography-derived fractional flow reserve: a systematic review and Bayesian meta-analysis. Eur Heart J 2018;39:3314-21. [Crossref] [PubMed]
22. Chen L, Ge S, Chen Y, Zhang TT, Zhu ZH. A feasibility study of 70 kV double low-dose coronary imaging technique in abdomen-fatty patients using dual-source CT. Afr Health Sci 2023;23:599-606. [Crossref] [PubMed]
23. Ren Z, Shen L, Zhang X, He T, Yu N, Zhang M. Radiation and contrast dose reduction in coronary CT angiography for slender patients with 70 kV tube voltage and deep learning image reconstruction. Br J Radiol 2025;98:1046-52. [Crossref] [PubMed]
24. Odigie-Okon E, Ogunleye O, Saxena R, Ganiyu S, Hassib M, Swanson D, Gaalema D. Radiation Dose Reduction in Coronary Computed Tomography Angiography. JACC Case Rep 2025;30:105719. [Crossref] [PubMed]
25. Fahrni G, Saliba T, Racine D, Gulizia M, Tzimas G, Pozzessere C, Rotzinger DC. Trading off Iodine and Radiation Dose in Coronary Computed Tomography. J Cardiovasc Dev Dis 2025;12:195. [Crossref] [PubMed]
26. Willner M, Bech M, Herzen J, Zanette I, Hahn D, Kenntner J, Mohr J, Rack A, Weitkamp T, Pfeiffer F. Quantitative X-ray phase-contrast computed tomography at 82 keV. Opt Express 2013;21:4155-66. [Crossref] [PubMed]
27. Martens B, Hendriks BMF, Mihl C, Wildberger JE. Tailoring Contrast Media Protocols to Varying Tube Voltages in Vascular and Parenchymal CT Imaging: The 10-to-10 Rule. Invest Radiol 2020;55:673-6. [Crossref] [PubMed]
28. Shrivastava P, Kashikar S, Parihar PH, Kasat P, Bhangale P, Shrivastava P. A systematic review on deep learning-enabled coronary CT angiography for plaque and stenosis quantification and cardiac risk prediction. Eur J Radiol Open 2025;14:100652. [Crossref] [PubMed]
29. Aye WMM, Kiraly L, Kumar SS, Kasivishvanaath A, Gao Y, Kofidis T. Mixed Reality (Holography)-Guided Minimally Invasive Cardiac Surgery-A Novel Comparative Feasibility Study. J Cardiovasc Dev Dis 2025;12:49. [Crossref] [PubMed]
30. Kundzierewicz M, Kołodziej K, Khokhar A, Tsung-Ying T, Leśniak A, Zakrzewski P, Borecki H, Bohn E, Hecko J, Januska J, Precek D, Stanuch M, Skalski A, Onuma Y, Serruys P, Bruining N, Złahoda-Huzior A, Dudek D. Catheterization laboratories open the doors for Extended Realities-review of clinical applications in cardiology. Eur Heart J Digit Health 2025;6:1055-68. [Crossref] [PubMed]
31. Annabestani M, Olyanasab A, Mosadegh B. Application of Mixed/Augmented Reality in Interventional Cardiology. J Clin Med 2024;13:4368. [Crossref] [PubMed]
32. Chen GZ, Zhang LJ, Schoepf UJ, Wichmann JL, Milliken CM, Zhou CS, Qi L, Luo S, Lu GM. Radiation dose and image quality of 70 kVp cerebral CT angiography with optimized sinogram-affirmed iterative reconstruction: comparison with 120 kVp cerebral CT angiography. Eur Radiol 2015;25:1453-63. [Crossref] [PubMed]
33. Hafez A, Sobhy A, Ashour M, Aiash K, Aldemerdash A, Radi AM, Elissawy A, Zahrawi K, Besheya T, Riccelli R, Arora A, Al-Mahmoud D, Jishu J, Toraih E, Aiash H. Diagnostic Performance of AI-Assisted Coronary CT Angiography: A Systematic Review and Meta-Analysis. Eur J Radiol 2026;195:112612. [Crossref] [PubMed]
34. Xiong QF, Fu XR, Ku LZ, Zhou D, Guo SP, Zhang WS. Diagnostic performance of coronary computed tomography angiography stenosis score for coronary stenosis. BMC Med Imaging 2024;24:39. [Crossref] [PubMed]
35. Roshanfar M, Salimi M, Jang SJ, Sinusas AJ, Kim J, Mosadegh B. Emerging Image-Guided Navigation Techniques for Cardiovascular Interventions: A Scoping Review. Bioengineering (Basel) 2025;12:488. [Crossref] [PubMed]