Clinical value of dual low-dose computed tomography angiography incorporating artificial intelligence iterative reconstruction for assessing arteriovenous fistulas/grafts among hemodialysis patients

Introduction

Arteriovenous fistula or arteriovenous graft (AVF/AVG) in the upper extremities (1,2) serves as essential vascular access for patients with chronic kidney disease (CKD) requiring hemodialysis (3-6). However, AVF/AVG is prone to complications such as stenosis, occlusion, thrombosis, and aneurysmal degeneration, often precipitated by mechanical stress, trauma, or repeated cannulation (1,7). These complications impair access patency, reduce dialysis efficiency, and necessitate frequent interventions, thereby underscoring the critical need for vascular monitoring and assessment of AVF/AVG access (7).

Computed tomography angiography (CTA) offers superior diagnostic performance over ultrasound for detecting AVF/AVG stenosis, with Baz et al. (8) reporting a 10% higher stenosis detection rate. Heye et al. (9) further validated CTA’s diagnostic equivalence to digital subtraction angiography (DSA; accuracy: 92.0%, sensitivity: 90.2%, specificity: 92.8%) for grading >50% stenosis. However, conventional CTA protocols are associated with relatively high radiation exposure and contrast agent dose, which is of particular concern in hemodialysis patients requiring repeated examinations (5–10 mSv, up to 17 mSv, 80–100 mL contrast) (10-12). Therefore, this dual challenge underscores the critical need for dose-reduction strategies that maintain image quality while also ensuring diagnostic accuracy for stenosis detection.

In this context, dual low-dose CTA strategies—combining reduced tube voltage and contrast dosage—represent a promising solution. Lowering tube voltage from 100 to 70–90 kVp can increase vascular attenuation by approximately 40–60% per 10 kVp reduction due to the photoelectric effect near iodine’s K-edge (33.2 keV), thereby enabling contrast dose reduction while maintaining vascular enhancement. This is particularly important for upper-extremity CTA with long scan coverage (60–80 cm) (8,9,13-16). Additionally, the lower contrast dosage used in patients with renal insufficiency may lower the risk for the loss of residual renal function (13). However, this dual low-dose CTA technique especially at low tube voltage introduces photon-starved challenges, increasing image noise and artifacts that may compromise the observation of slender, tortuous peripheral vessels (10).

Recent studies have highlighted the noise reduction, artifacts suppression and structural delineation of artificial intelligence (AI)-based reconstruction algorithms, especially under low-dose conditions (17-26). Li et al. (22) demonstrated that AI-based reconstruction not only achieves 62% image noise reduction in low-dose aortic CTA compared with hybrid iterative reconstruction (HIR), but also enhances the visualization of small vessels. Furthermore, Li et al. (20) reported that using an AI-based algorithm can effectively reduce artifacts in abdominal imaging, even for patients with suboptimal positioning. However, these studies have mainly targeted high-contrast, central vascular regions, whereas upper-extremity CTA for hemodialysis access poses distinct challenges—including long coverage (60–80 cm), tortuous peripheral vessels, and metallic cannulation artifacts that amplify image noise at low tube voltages. To date, the feasibility of AI-based algorithm applied to dual low-dose CTA protocols that simultaneously reduce radiation and contrast doses in this specific clinical setting has not been systematically investigated. Given the cumulative radiation exposure and nephrotoxic risk faced by dialysis patients, establishing such a tailored low-dose imaging strategy is clinically important.

This study aims to investigate the clinical value of dual low-dose upper extremity CTA for hemodialysis AVF/AVG, with the use of an AI-based algorithm, specifically artificial intelligence iterative reconstruction (AIIR). We hypothesized that dual low-dose CTA with AIIR could maintain diagnostic accuracy while substantially reducing radiation and contrast doses compared with routine-dose CTA. We present this article in accordance with the STARD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0109/rc).

Methods

Study participants enrollment

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This prospective study was approved by the Institutional Review Board of Sir Run Run Shaw Hospital (approval No. 2024-YAN-1096), and written informed consent was obtained from all participants. The initial study cohort consisted of 119 patients from Sir Run Run Shaw Hospital who were suspected with AVF/AVG dysfunction and scheduled to receive upper extremity CTA from January 2024 to April 2025. AVF/AVG dysfunction was defined by any of the following criteria: abnormally low arterial flow or high venous pressure during hemodialysis; difficulty in cannulating the AVF/AVG with inadequate blood flow despite multiple attempts; suspicion of aneurysm formation; suspicion of graft thrombosis or occlusion; upper extremity swelling with suspected central vein lesion. Patients were excluded if they had malignant tumors, neurological or psychiatric disorders, a history of allergy to iodinated contrast agent, or an expected survival period of less than 6 months. A priori sample size estimation was performed using G*Power software (version 3.1, Heinrich Heine University, Dusseldorf, Germany), based on a two-sided test with α =0.05 and power =0.80. The analysis indicated that a minimum of 51 patients per group was required. Ultimately, 102 eligible patients were randomly assigned into one of the two protocols using a computer-generated randomization sequence, with 51 patients in the routine-dose CTA group and 51 in the low-dose CTA group, as shown in Figure 1.

Figure 1 Flowchart of participant enrollment. AVF, arteriovenous fistula; AVG, arteriovenous graft; CKD, chronic kidney disease; CTA, computed tomography angiography; DSA, digital subtraction angiography; w/, with; w/o, without.

Computed tomography (CT) scanning protocol and image reconstruction

All CTAs were performed using a 320-row scanner (uCT 960+; United Imaging Healthcare, Shanghai, China). Patients were positioned supine on the scanning table, with the AVF/AVG arm placed alongside the body and a small gap maintained between the arm and body to avoid vein compression. The contralateral arm was elevated above the head, with the catheter for contrast agent injection placed in it, to minimize artifacts. The scanning range extended from the upper margin of the shoulder to the wrist.

For the routine-dose protocol, the tube voltage was set at 100 kVp, while for the low-dose protocol, it was reduced to 80 kVp. All other parameters remained consistent: 130 mAs reference tube current, 40 mm collimation, 0.79 pitch, and 0.5 s rotation time. Both protocols administered the same nonionic iodinated contrast agent (Iohexol, 350 mgI/mL; Omnipaque; GE, Shanghai, China), which is a low-osmolar contrast medium, at an injection rate of 3.5 mL/s using an automatic injector system. The total contrast volume was calculated based on body weight: 1.0 mL/kg for the routine-dose protocol and 0.6 mL/kg for the low-dose protocol, corresponding to iodine loads of 350 and 210 mgI/kg, respectively. Therefore, while the iodine concentration and osmolarity of the contrast agent were identical between the two groups, the total iodine load (mgI/kg) differed due to the reduced contrast volume in the low-dose protocol. A region of interest (ROI) for bolus tracking was placed on the descending aorta, with a triggering threshold set at 100 Hounsfield unit (HU) and a delay time of 8.0 s. Following contrast injection, 40 mL of saline was injected at a rate of 4.0 mL/s. The volume CT dose index (CTDI_vol) and the dose length product (DLP) were retrieved from the dose reports stored in the picture archiving and communication system. The effective dose (ED) was estimated using the formula ED = DLP × k, where a conversion factor of k =0.0145 mSv·mGy⁻¹·cm⁻¹ was applied based on the average of chest (0.014) and abdominopelvic (0.015) coefficients, considering the extended scan coverage of upper extremity CTA.

CT data acquired using the routine-dose protocol were reconstructed with the HIR algorithm (Karl 3D; reconstruction kernel: B_SOFT_A; United Imaging Healthcare), named Group A. Data acquired using the low-dose protocol were reconstructed using the AIIR algorithm (reconstruction kernel: BodyStandard, Group B; United Imaging Healthcare). Specifically, the AIIR algorithm integrates a dedicated convolutional neural network into the model-based iterative reconstruction to replace the traditional regularization term for reducing noises and artifacts, thereby combining the advantage of model-based iterative reconstruction in characterizing image detail and the ability of the convolutional neural network in proper handling of image noise and texture (21-23). All reconstructions were performed with a 512×512-pixel matrix and a slice thickness/interval of 1.0 mm. The reconstructed datasets were transferred to a dedicated workstation for post-processing, including maximum intensity projection, curved planar reformation, and hyper-realistic rendering (HRR; uOmnispace.CT; United Imaging Healthcare). HRR images, a volume-rendering-based technique with enhanced photorealistic visualization, were used as a complementary tool to improve the depiction of vascular anatomy, but were not used as the primary basis for quantitative or diagnostic evaluation.

Qualitative image quality

Qualitative image quality was evaluated by two radiologists with 20 and 18 years of experience in vascular CT imaging, respectively. Ratings were assigned using two 5-point Likert scales, one for overall image quality and the other for diagnostic confidence, as detailed in Table 1. All image sets were anonymized and randomly mixed using a computer-generated randomization sequence before evaluation. Both radiologists were blinded to the scanning protocol, reconstruction algorithm, and all patient clinical information. Each reader independently reviewed the images on a dedicated workstation in a randomized order, and no consensus reading or discussion was performed prior to analysis. After independent grading, images with scores ≥3 by both readers were considered clinically acceptable.

Quantitative image quality

Quantitative image quality was assessed via CT number, image noise, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR). Measurements were performed by an experienced radiologist with over 15 years of experience in vascular CT imaging, who was blinded to the scanning protocol and reconstruction algorithm. For each case, three ROIs were placed at specific anatomical locations: the subclavian artery near the first rib, the brachial artery at the level of the ulnar olecranon, and the radial artery 2 cm above the radial-cephalic anastomosis, while carefully avoiding the vessel wall, calcification, or plaque (Figure 2). The ROI size was set to approximately 2/3 of the lumen cross-sectional area. The CT attenuation and image noise were measured within three ROIs. The image noise was defined as the standard deviation (SD) of CT attenuation within the ROI. An additional ROI was placed on the surrounding fat tissue as the background to calculate the CNR. The SNR and CNR were calculated as:

SNR=μartery/σartery[1]

CNR=(μartery−μfat)/σfat[2]

Figure 2 Placement of ROIs for quantitative image quality evaluation. The HRR image (right) illustrates the CTA scan range, extending from the upper margin of the shoulder to the wrist. The red circles on the axial CT images (left) denote the locations of ROIs placed at the subclavian artery (I), the brachial artery (II), and the radial artery (III). CT, computed tomography; CTA, computed tomography angiography; HRR, hyper-realistic rendering; ROI, region of interest.

where 𝜇 and 𝜎 represent the CT attenuation and SD within the ROI, respectively.

AVF/AVG stenosis assessment

Among the enrolled patients, DSA was performed either on the same day as CTA (n=15) or within 7 days thereafter (n=29), during which no major clinical interventions affecting vascular status were undertaken, as shown in Figure 1. All CTA and fistulography procedures were deliberately scheduled before routine hemodialysis, ensuring prompt clearance of the contrast agents. DSA was performed only when interventional evaluation or therapeutic angioplasty was clinically indicated, such as suspected significant (>50%) stenosis on CTA or persistent hemodialysis dysfunction after imaging. This selective use of DSA reflects real-world clinical practice and minimizes unnecessary invasive procedures.

The diagnostic performance for AVF/AVG stenosis was assessed and compared between the two image groups on both per-segment and per-patient basis, with DSA serving as the reference standard. DSA examinations were predominantly conducted through direct retrograde puncture of the AVF/AVG. In select cases, where deemed as optimal for accessing the stenosis, punctures were alternatively performed on the efferent vein in the upper arm or the radial artery in the forearm. These punctures were executed either freehand or under ultrasound guidance, utilizing a mini puncture set.

An interventional radiologist with two decades of experience assessed the stenosis in four distinct vessel segments based on DSA findings. These segments included the inflow artery in the upper arm and forearm (Segment I), the anastomosis (Segment II), the outflow vein in the forearm and upper arm (Segment III), and the subclavian and central veins (Segment IV). Stenosis grades were divided into five grades: grade 0, normal patency; grade 1, less than 50% stenosis; grade 2, 50–75% stenosis; grade 3, greater than 75% stenosis; and grade 4, total occlusion. A radiologist with 20 years of experience, blinded to DSA findings, evaluated the CTA images, classifying each vessel segment using the same methodology.

When multiple stenoses were present within a single segment, the most severe stenosis was recorded. Vessel segments were considered non-diagnostic and excluded from subsequent analysis if not encompassed within the CTA or DSA imaging fields or inadequately enhanced for clear visualization. Stenosis severity was categorized based on a threshold of ≥50%, which is commonly used in clinical practice and prior literature to define hemodynamically significant lesions.

Statistical analysis

All the statistical analyses were performed with SPSS version 22 (IBM Corp, Armonk, NY, USA). Normality of data was tested using the Kolmogorov-Smirnov test. Normally distributed data were compared using Student t-test, while non-normally distributed data were compared using the Mann-Whitney U test between two samples. Clinical characteristics, qualitative and quantitative results between the routine-dose and low-dose CTA were compared using student t-test or the Mann-Whitney U. Categorical variables were expressed as frequencies or percentages and compared using the Chi-squared test or Fisher’s exact test, as appropriate. A P<0.05 was considered statistically significant. The Cohen’s kappa test (unweighted) was used to evaluate the inter-reader consistency: κ of 0.81–1.0, 0.61–0.80, 0.41–0.60, 0.21–0.40, and ≤0.20 were considered excellent, good, moderate, fair, and poor agreement, respectively.

Diagnostic performance analysis was limited to patients who underwent both CTA and DSA, and results were interpreted in this context. The concordance between DSA-derived and CTA-derived quantitative measurements was evaluated using the intraclass correlation coefficient (ICC) based on a two-way random-effects model with absolute agreement. ICC values were interpreted as poor (0–0.50), moderate (0.51–0.75), good (0.76–0.90), and excellent (0.91–1.00). Receiver operating characteristic (ROC) analysis was performed to evaluate the ability of CTA to detect significant stenosis (≥50%), using DSA as the reference standard. The area under the curve (AUC), sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) were calculated with corresponding 95% confidence intervals at both the per-segment and per-patient levels. Generalized estimating equations (GEEs) with a logit link function and exchangeable correlation structure were used to account for within-patient correlation at the segment level. The results were compared between the two groups using the Chi-squared test or Fisher test. A P<0.05 was considered statistically significant.

Results

Participants characteristics

A total of 102 participants were enrolled in this study (Figure 1), with dialysis histories ranging from 6 months to 15 years, specifically with a frequency of 2–3 sessions per week, and a duration of 3–4 hours per session. The demographic and clinical characteristics of the participants are summarized in Table 2. No contrast-induced adverse events were observed during this study period. There were no significant differences in age, sex ratio, weight, body mass index (BMI), dialysis history, access type, and anastomosis site between the two groups (all P>0.05). No significant intergroup differences were observed in scanning coverage. The radiation dose and contrast dosage in low-dose CTA were reduced by 52.2% (6.57±1.1 vs. 3.14±0.5 mSv) and 40.5% (37.1±6.5 vs. 62.4±5.1 mL), respectively, compared to the routine-dose CTA (both P<0.001). Among the included patients, 12 cases of AVF/AVG thrombosis were identified (routine-dose CTA: n=5; low-dose CTA: n=7).

Qualitative image quality results

The proportion of excellent images (score >3) and the mean scores assessed by the two radiologists are summarized in Table 3. No image across the two groups was scored 2 or lower, indicating that all images in this study met the minimum criteria for clinical diagnosis. There were no statistically significant differences between Groups A and B in the proportion of excellent images or in the mean scores for overall image quality and diagnostic confidence (all P>0.05). Notably, the AIIR significantly reduced image noise and improved stent contrast resolution and vessel contrast at low-dose CTA, as shown in Figures 3-5. Inter-reader agreement was good. For image quality assessment, the κ values were 0.788 and 0.705 for Groups A and B, respectively, while for diagnostic confidence, the κ values were 0.764 and 0.688, respectively.

Figure 3 CTA image of AVG under two scan protocols. (A) HRR and enlarged axial images of different sites from a 71-year-old woman (BMI: 25 kg/m²) in Group A (100 kVp, 130 mAs), reconstructed using HIR. (B) HRR and axial images from a 63-year-old man (BMI: 23 kg/m²) in Group B (80 kVp, 130 mAs), reconstructed using AIIR. The three axial images (top to bottom) on the right of each group show the subclavian artery (I), inflow artery-graft anastomosis (II), and graft-outflow vein anastomosis (III), respectively. The display window width/level settings are WW/WL =800/300 HU. Red arrows indicate the anatomical locations in Group A, and blue arrows indicate the corresponding anatomical locations in Group B. Group A: routine-dose CTA with HIR. Group B: low-dose CTA with AIIR. AIIR, artificial intelligence iterative reconstruction; AVG, arteriovenous graft; BMI, body mass index; CTA, computed tomography angiography; HIR, hybrid iterative reconstruction; HRR, hyper-realistic rendering; HU, Hounsfield unit; WL, window level; WW, window width.

Figure 4 Comparative CTA visualization of an AVG with stent placement across two reconstruction protocols. (A) HRR and axial images from a 51-year-old male (BMI: 25 kg/m²) in Group A (routine-dose HIR). (B) HRR and axial images from a 60-year-old male (BMI: 27 kg/m²) in Group B (low-dose AIIR). Zoom-in axial sections (right panels) display anatomical landmarks: subclavian artery (I), inflow artery-graft anastomosis (II), grafts anastomosis-outflow vein (III), and stents. The AIIR algorithm can reduced artifacts and improve the lumen visualization with enhanced contrast resolution. Window settings: WW/WL =800/300 HU for vessel and stent evaluation. Red arrows indicate the anatomical locations in Group A, and blue arrows indicate the corresponding anatomical locations in Group B. Group A: routine-dose CTA with HIR. Group B: low-dose CTA with AIIR. AIIR, artificial intelligence iterative reconstruction; AVG, arteriovenous graft; BMI, body mass index; CTA, computed tomography angiography; HIR, hybrid iterative reconstruction; HRR, hyper-realistic rendering; HU, Hounsfield unit; WL, window level; WW, window width.

Figure 5 CTA images from a 26-year-old female patient who underwent both routine-dose and low-dose protocols at a 2-month interval, demonstrating AVF with stent placement. Magnified axial views (right panels in each group) illustrate stent morphology, subclavian artery (I), anastomotic site (II), and outflow vein (III). The display window level settings are WW/WL =800/300 HU for vessel and WW/WL =800/400 HU for stents. Red arrows indicate the anatomical locations in Group A, and blue arrows indicate the corresponding anatomical locations in Group B. AIIR, artificial intelligence iterative reconstruction; AVF, arteriovenous fistula; AVG, arteriovenous graft; CTA, computed tomography angiography; HIR, hybrid iterative reconstruction; HU, Hounsfield unit; WL, window level; WW, window width.

Quantitative image quality results

The results of the quantitative evaluation of image quality are summarized in Table 4. No statistically significant intergroup differences were observed in CT attenuation across measured anatomical regions (P>0.05), except for the subclavian artery which demonstrated significant variation (P=0.001). Group B exhibited significantly lower image noise, with reduction of 59% compared to Group A (all P<0.001), alongside superior vascular contrast performance across all three arterial segments. Notably, the implementation of the AIIR algorithm in Group B resulted in remarkable improvements in vascular contrast. The SNR improvements reached 175% (subclavian), 165% (brachial), and 157% (radial artery) compared to Group A, with similar enhancement in CNR (all P<0.001).

Diagnostic performance of AVF/AVG stenosis

Among patients with AVF/AVG who underwent routine-dose CTA, 23 received DSA, yielding a total of 102 evaluable vessel segments. DSA findings revealed that stenosis exceeding 50% was present in five anastomotic sites, two inflow arteries, eight outflow veins, and 15 subclavian veins. In the cohort undergoing low-dose CTA, 21 patients received DSA, providing 84 assessable vessel segments. The DSA results demonstrated that five anastomotic sites, 10 outflow veins, and 12 subclavian veins with stenosis greater than 50%. Representative cases illustrating stenosis are presented in Figure 6.

Figure 6 Comparison of MIP and DSA images. (A,B) MIP images from a 64-year-old woman in Group A with AVF for over 11 years, who had been undergoing maintenance hemodialysis for over 11 years, showed grade 3 stenosis (>75% stenosis) in the outflow vein (red arrows), which was consistent with the findings from DSA (B, red arrow). (C,D) MIP images from a 61-year-old woman with AVG for over 4 years, obtained in Group B, demonstrated grade 3 stenosis (>75% stenosis) in the outflow vein-anastomosis and grade 4 stenosis (occlusion) in the graft (red arrows), results that were consistent with the DSA findings (D, red arrows). Group A: routine-dose CTA with HIR. Group B: low-dose CTA with AIIR. AIIR, artificial intelligence iterative reconstruction; AVF, arteriovenous fistula; AVG, arteriovenous graft; CTA, computed tomography angiography; DSA, digital subtraction angiography; HIR, hybrid iterative reconstruction; MIP, maximum intensity projection.

Table 5 summarizes the agreement in stenosis severity (CTA vs. DSA) and the diagnostic performance in detecting stenosis >50% between two groups, with DSA serving as the reference standard in the subset of patients who underwent the procedure. The agreement between CTA and DSA for stenosis grading was excellent, with ICC values consistently above 0.90 across Segments I–IV in both Groups A and B. At the per-segment level, Group B showed numerically higher AUC (0.99), sensitivity (100%), and NPV (100%) compared to Group A, although these differences were not statistically significant (all P>0.05). After accounting for within-patient correlation using GEEs, no significant difference in the detection of ≥50% stenosis was observed between the two groups (odd ratio =0.99; 95% confidence interval: 0.86–1.14; P=0.899). At the per-patient level, both groups also demonstrated high diagnostic performance. The AUC was 0.98 in Group A and 0.94 in Group B, with no significant difference between the groups (P=0.411). Accuracy, sensitivity, specificity, PPV, and NPV were achieved similar diagnostic performance between the two groups (all P>0.05). Notably, there were two false-negative cases in Group A involving subclavian vein stenosis, which was underestimated due to vascular tortuosity. One false-positive case in Group A and one in Group B were identified in the outflow-vein assessment, both resulting from extrinsic vascular compression rather than true stenosis. Representative false-positive and false-negative cases were identified and analyzed to further evaluate diagnostic discrepancies (Figure 7).

Figure 7 False-positive and false-negative cases in CTA compared with DSA. (A-D) A false-positive case in Group A (routine-dose CTA). Axial, coronal, VR, and DSA images of the subclavian vein are shown. CTA suggested no significant stenosis (<50%, orange arrows), whereas DSA demonstrated significant stenosis requiring balloon dilation. (E-H) A false-negative case in Group B (low-dose CTA with AIIR). Axial, oblique coronal, VR, and DSA images are presented. The basilic vein (blue arrows) appeared compressed by the adjacent axillary artery (green arrows). CTA suggested severe stenosis, whereas DSA showed no significant luminal narrowing. This discrepancy may be related to differences in intraluminal pressure during DSA acquisition. AIIR, artificial intelligence iterative reconstruction; CTA, computed tomography angiography; DSA, digital subtraction angiography; VR, volume-rendered.

Discussion

This study investigated the clinical value of upper extremity AVF/AVG CTA using a dual low-dose scan protocol with the AIIR algorithm. The results demonstrated that the dual low-dose CTA with AIIR provided significantly enhanced image contrast compared to routine-dose with HIR. Moreover, the diagnostic performance of dual low-dose CTA with AIIR was similar to that of DSA, with no significant difference observed, suggesting its potential clinical applicability. Additionally, this approach demonstrates promise for routine upper limb CTA examinations.

To our knowledge, this study is the first to clinically evaluate dual low-dose upper extremity CTA with AIIR for AVF/AVG assessment. While prior research focused on modality comparisons [e.g., CTA vs. ultrasound/magnetic resonance angiography/DSA (8,27,28)], radiation/contrast optimization remains understudied. The low-kVp dynamic CTA (DLP 1,246 mGy·cm, 80 kVp/180 mAs) surpassed our radiation levels, and required >80 mL contrast (10,12). Our protocol reduces radiation exposure by >50% versus routine CTA, while combining AIIR and low-dose CT with 36 mL contrast achieves superior contrast resolution compared to HIR, without compromising diagnostic accuracy—a critical advance for long-term hemodialysis patients.

This study found that although the AIIR algorithm significantly improved quantitative parameters (SNR/CNR) in the low-dose group, no significant differences were observed in qualitative assessments or diagnostic performance compared with the routine-dose group. This may indicate that, within the studied acquisition settings, improvements in quantitative image quality do not necessarily translate into measurable diagnostic gains. The observed imaging characteristics are likely multifactorial, related to reduced tube voltage and AIIR-based noise suppression, which together increase iodine attenuation and enhance vascular conspicuity. While higher attenuation may facilitate visualization of small or tortuous vessels (29), it may also theoretically reduce conspicuity of subtle low-contrast intraluminal abnormalities such as partial thrombosis. However, no deterioration in diagnostic performance was observed in this cohort, suggesting that image quality remained sufficient for clinical interpretation under both protocols. These findings may reflect a plateau effect in diagnostic performance once adequate image quality is achieved, although this interpretation remains speculative and should be validated in future studies. Overall, these results underscore that optimization of CT protocols should consider both quantitative metrics and clinically relevant diagnostic performance rather than relying solely on objective image quality improvements (30). In addition, reducing the iodine load may provide a practical approach to modulating vascular attenuation, thereby helping to avoid excessive enhancement while maintaining diagnostic adequacy.

The study further emphasizes the clinical advantages of CT over alternative imaging modalities such as ultrasound or DSA. Baz et al. noted ultrasound’s limitations in detecting subtle/complex lesions [e.g., missed >33% subclavian vein stenoses (8)], while CT showed comparable diagnostic performance to DSA [sensitivity 90.2%, specificity 92.8% for >50% stenosis; n=36 (9)]. Notably, CT allows for the concurrent detection of underlying diseases, including aortic calcifications, coronary artery calcifications, and subclavian vascular stenosis (rating consistency 88%) during CT examinations (Tables 2,5). However, the interpretation of the very high diagnostic performance metrics, including AUC values, should be made with caution given the relatively limited sample size and potential spectrum bias in available studies. The selection of diagnostic thresholds may also influence performance estimates. Overall, these findings support the role of CTA as a non-invasive imaging modality for comprehensive vascular assessment, particularly in patients with complex AVF/AVG anatomy or in settings with limited access to invasive angiography (31).

This study has several limitations. First, the BMI of participants ranged from 15.82 to 31.71 kg/m², with only a very limited number of patients having BMI >30 kg/m², which may limit the generalizability of the results to obese populations. Further studies are warranted to evaluate the applicability of this protocol in higher-BMI populations. Second, the AIIR algorithm used in this study is specific to a single manufacturer. Although such vendor dependence is common in imaging algorithm research, the dual low-dose CTA protocol itself may be adaptable to other reconstruction platforms and CT systems. Third, although previous studies have explored 70 kVp AIIR protocols for aortic CTA, the present study adopted 80 kVp as a conservative threshold to ensure diagnostic reliability for upper-extremity peripheral vessels, for which low-dose validation data remain limited. Future work should assess ultra-low-dose settings (60–70 kVp) combined with AIIR in this anatomical region. Finally, the DSA-validated cohort was limited in size because DSA was performed selectively based on clinical indications rather than in all patients, reflecting real-world practice. This may introduce partial verification bias and potentially lead to overestimation of diagnostic performance. Therefore, the diagnostic results should be interpreted with caution, and larger studies with more comprehensive DSA validation are warranted to provide more robust evidence.

Conclusions

Dual low-dose upper extremity CTA with AIIR substantially reduces radiation and contrast doses while preserving diagnostic confidence for evaluating AVF/AVG stenosis. This low-dose protocol demonstrates feasibility for clinical assessment of vascular access in hemodialysis patients and may serve as a promising alternative to routine-dose CTA. Further validation in larger DSA-confirmed cohorts is warranted before broader clinical adoption.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0109/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0109/dss

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0109/coif). J.L. and Y.Z. are scientific researchers with the United Imaging Healthcare Co., Ltd., and the company has no conflicts of interest related to this study. 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. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This study was approved by the Institutional Review Board of Sir Run Run Shaw Hospital (approval No. 2024-YAN-1096). Written informed consent was obtained from all participants.

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. Labriola L, Crott R, Desmet C, André G, Jadoul M. Infectious complications following conversion to buttonhole cannulation of native arteriovenous fistulas: a quality improvement report. Am J Kidney Dis 2011;57:442-8. [Crossref] [PubMed]
2. Harms JC, Rangarajan S, Young CJ, Barker-Finkel J, Allon M. Outcomes of arteriovenous fistulas and grafts with or without intervention before successful use. J Vasc Surg 2016;64:155-62. [Crossref] [PubMed]
3. Chandrashekar A, Ramakrishnan S, Rangarajan D. Survival analysis of patients on maintenance hemodialysis. Indian J Nephrol 2014;24:206-13. [Crossref] [PubMed]
4. Jha V, Garcia-Garcia G, Iseki K, Li Z, Naicker S, Plattner B, Saran R, Wang AY, Yang CW. Chronic kidney disease: global dimension and perspectives. Lancet 2013;382:260-72. [Crossref] [PubMed]
5. Romagnani P, Agarwal R, Chan JCN, Levin A, Kalyesubula R, Karam S, Nangaku M, Rodríguez-Iturbe B, Anders HJ. Chronic kidney disease. Nat Rev Dis Primers 2025;11:8. [Crossref] [PubMed]
6. Macdougall IC, White C, Anker SD, Bhandari S, Farrington K, Kalra PA, McMurray JJV, Murray H, Tomson CRV, Wheeler DC, Winearls CG, Ford IPIVOTAL Investigators and Committees. Intravenous Iron in Patients Undergoing Maintenance Hemodialysis. N Engl J Med 2019;380:447-58. [Crossref] [PubMed]
7. Asif A, Gadalean FN, Merrill D, Cherla G, Cipleu CD, Epstein DL, Roth D. Inflow stenosis in arteriovenous fistulas and grafts: a multicenter, prospective study. Kidney Int 2005;67:1986-92. [Crossref] [PubMed]
8. Baz AAERM, Naguib MM, Kamel AI, El-Khashab SO. Multi-slice CT angiography versus duplex ultrasound in detection of stenosis of access arteriovenous fistulas and grafts in dysfunctional hemodialysis. The Egyptian Journal of Radiology and Nuclear Medicine 2016;47:1459-66.
9. Heye S, Maleux G, Claes K, Kuypers D, Oyen R. Stenosis detection in native hemodialysis fistulas with MDCT angiography. AJR Am J Roentgenol 2009;192:1079-84. [Crossref] [PubMed]
10. Eleti S, Hickman S, Wilson A. Upper limb computed tomography (CT) angiography in the emergency department. Clin Radiol 2024;79:657-64. [Crossref] [PubMed]
11. Friedman DD, Ponkowski MJ, Shetty AS, Hoegger MJ, Itani M, Rajput MZ, Mellnick VM, Raptis CA, Northrup BE, Ballard D, Cabrera Lebron JA, Tsai R. CT Angiography of the Upper Extremities: Review of Acute Arterial Entities. Radiographics 2025;45:e240077. [Crossref] [PubMed]
12. Meyer M, Geiger N, Benck U, Rose D, Sudarski S, Ong MM, Schoenberg SO, Henzler T. Imaging of Patients with Complex Hemodialysis Arterio-Venous Fistulas using Time-Resolved Dynamic CT Angiography: Comparison with Duplex Ultrasound. Sci Rep 2017;7:12563. [Crossref] [PubMed]
13. Chen MC, Tsai WL, Tsai IC, Chan SW, Liao WC, Lin PC, Yang SJ. Arteriovenous fistula and graft evaluation in hemodialysis patients using MDCT: a primer. AJR Am J Roentgenol 2010;194:838-47. [Crossref] [PubMed]
14. Karadeli E, Tarhan NC, Ulu EM, Tutar NU, Basaran O, Coskun M, Niron EA. Evaluation of failing hemodialysis fistulas with multidetector CT angiography: comparison of different 3D planes. Eur J Radiol 2009;69:184-92. [Crossref] [PubMed]
15. McCollough CH, Primak AN, Braun N, Kofler J, Yu L, Christner J. Strategies for reducing radiation dose in CT. Radiol Clin North Am 2009;47:27-40. [Crossref] [PubMed]
16. Sigal-Cinqualbre AB, Hennequin R, Abada HT, Chen X, Paul JF. Low-kilovoltage multi-detector row chest CT in adults: feasibility and effect on image quality and iodine dose. Radiology 2004;231:169-74. [Crossref] [PubMed]
17. Otgonbaatar C, Ryu JK, Shin J, Woo JY, Seo JW, Shim H, Hwang DH. Improvement in Image Quality and Visibility of Coronary Arteries, Stents, and Valve Structures on CT Angiography by Deep Learning Reconstruction. Korean J Radiol 2022;23:1044-54. [Crossref] [PubMed]
18. Nagayama Y, Emoto T, Hayashi H, Kidoh M, Oda S, Nakaura T, Sakabe D, Funama Y, Tabata N, Ishii M, Yamanaga K, Fujisue K, Takashio S, Yamamoto E, Tsujita K, Hirai T. Coronary Stent Evaluation by CTA: Image Quality Comparison Between Super-Resolution Deep Learning Reconstruction and Other Reconstruction Algorithms. AJR Am J Roentgenol 2023;221:599-610. [Crossref] [PubMed]
19. Kawai H, Motoyama S, Sarai M, Sato Y, Matsuyama T, Matsumoto R, Takahashi H, Katagata A, Kataoka Y, Ida Y, Muramatsu T, Ohno Y, Ozaki Y, Toyama H, Narula J, Izawa H. Coronary computed tomography angiographic detection of in-stent restenosis via deep learning reconstruction: a feasibility study. Eur Radiol 2024;34:2647-57. [Crossref] [PubMed]
20. Li J, Meng T, Zhang G, Yu X, Lu Z, Zhang W. Artificial intelligence iterative reconstruction in abdominal CT of patients with irregular arm positioning: a case-by-case evaluation. Acta Radiol 2024;65:907-12. [Crossref] [PubMed]
21. Li J, Zhu J, Zou Y, Zhang G, Zhu P, Wang N, Xie P. Diagnostic CT of colorectal cancer with artificial intelligence iterative reconstruction: A clinical evaluation. Eur J Radiol 2024;171:111301. [Crossref] [PubMed]
22. Li W, You Y, Zhong S, Shuai T, Liao K, Yu J, Zhao J, Li Z, Lu C. Image quality assessment of artificial intelligence iterative reconstruction for low dose aortic CTA: A feasibility study of 70 kVp and reduced contrast medium volume. Eur J Radiol 2022;149:110221. [Crossref] [PubMed]
23. Yang L, Liu H, Han J, Xu S, Zhang G, Wang Q, Du Y, Yang F, Zhao X, Shi G. Ultra-low-dose CT lung screening with artificial intelligence iterative reconstruction: evaluation via automatic nodule-detection software. Clin Radiol 2023;78:525-31. [Crossref] [PubMed]
24. You Y, Zhong S, Zhang G, Wen Y, Guo D, Li W, Li Z. Exploring the Low-Dose Limit for Focal Hepatic Lesion Detection with a Deep Learning-Based CT Reconstruction Algorithm: A Simulation Study on Patient Images. J Imaging Inform Med 2024;37:2089-98. [Crossref] [PubMed]
25. Gong H, Peng L, Du X, An J, Peng R, Guo R, Ma X, Xiong S, Ma Q, Zhang G, Ma J. Artificial Intelligence Iterative Reconstruction in Computed Tomography Angiography: An Evaluation on Pulmonary Arteries and Aorta With Routine Dose Settings. J Comput Assist Tomogr 2024;48:244-50. [Crossref] [PubMed]
26. 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]
27. Cavagna E, D'Andrea P, Schiavon F, Tarroni G. Failing hemodialysis arteriovenous fistula and percutaneous treatment: imaging with CT, MRI and digital subtraction angiography. Cardiovasc Intervent Radiol 2000;23:262-5. [Crossref] [PubMed]
28. Jin WT, Zhang GF, Liu HC, Zhang H, Li B, Zhu XQ. Non-contrast-enhanced MR angiography for detecting arteriovenous fistula dysfunction in haemodialysis patients. Clin Radiol 2015;70:852-7. [Crossref] [PubMed]
29. Zhang D, Zhang L, Long J, Wu Y, Zhang H, Wang C, Sun B, Wang C, Zhang H, Sun X, Sun A, Meng Y, Hu C, Xu K. Comparison of image quality in 40 keV virtual monoenergetic images of dual-energy CT pulmonary angiography using deep learning and iterative reconstruction algorithms under optimized low dose scanning protocols. Quant Imaging Med Surg 2025;15:12372-85. [Crossref] [PubMed]
30. Benz DC, Benetos G, Rampidis G, von Felten E, Bakula A, Sustar A, Kudura K, Messerli M, Fuchs TA, Gebhard C, Pazhenkottil AP, Kaufmann PA, Buechel RR. Validation of deep-learning image reconstruction for coronary computed tomography angiography: Impact on noise, image quality and diagnostic accuracy. J Cardiovasc Comput Tomogr 2020;14:444-51. [Crossref] [PubMed]
31. Gonzalez TV, Bookwalter CA, Foley TA, Rajiah PS. Multimodality imaging evaluation of arteriovenous fistulas and grafts: a clinical practice review. Cardiovasc Diagn Ther 2023;13:196-211. [Crossref] [PubMed]