Artificial intelligence iterative reconstruction in lower extremity computed tomography angiography (CTA) of diabetic patients: Improved visualization of distal and collateral arteries

Abstract

Background

The impact of deep learning (DL)-based computed tomography (CT) reconstruction on the visualization of distal and collateral arteries in diabetic lower extremity CT angiography (CTA) remains unclear.

Purpose

To investigate the performance of a novel DL-based CT reconstruction algorithm, artificial intelligence iterative reconstruction (AIIR), in visualizing distal and collateral arteries on lower extremity CTA of diabetic patients, compared to the routine hybrid iterative reconstruction (HIR).

Material and Methods

This retrospective study included 59 diabetic patients who underwent clinically indicated lower extremity CTA. The images were reconstructed with both AIIR and HIR. Distal arterial visualization, collateral circulation depiction, and overall image quality were assessed and compared between two reconstruction methods.

Results

Compared with HIR, AIIR significantly improved the vessel visualization scores in the posterior tibial, dorsalis pedis, medial plantar, dorsal metatarsal, and dorsal digital arteries (all P <0.05). The scores for collateral circulation depiction were also higher with AIIR than those with HIR (all P <0.001). AIIR yielded significantly lower noise as well as higher signal-to-noise ratio and contrast-to-noise ratio compared with HIR (all P <0.001). The subjective score on overall image quality was significantly higher with AIIR than that with HIR (P <0.001).

Conclusion

Compared with HIR, AIIR provides improved visualization of distal and collateral arteries, as well as better overall image quality, in lower extremity CTA of diabetic patients.

Keywords

Lower extremity computed tomography angiography deep learning-based reconstruction diabetic foot ulcer image quality

Introduction

Diabetic foot ulcer (DFU) is one of the most common and severe complications of diabetes, with an estimated lifetime incidence in the range of 10%–25% in affected individuals (1,2). A major pathophysiological contributor to it is peripheral arterial disease (PAD), which impairs blood supply to the lower extremities, particularly to their distal parts (3). Therefore, assessment of peripheral arteries, including small-caliber distal arteries, is essential in diabetic patients, not only for identifying the at-risk foot to enable early intervention but also for guiding the effective management of DFU.

Computed tomography angiography (CTA) is routinely employed for non-invasive assessment of peripheral arteries. However, it is less effective in evaluating the distal small vessels, owing to their inherently reduced contrast. Several efforts have been made to improve the image quality of lower extremity CTA, including the optimization of acquisition parameters (4), the use of dual-energy CT (5 –8), and the enhancement of spatial resolution through increased matrix size and reduced field of view (9). In recent years, deep learning-based reconstruction (DLR) algorithms have shown promise in improving image quality for lower extremity CTA. Qu et al. demonstrated that, under the low-dose setting, DLR not only improved image quality but also yielded higher diagnostic accuracy for arterial stenosis compared to iterative reconstruction (10). Similarly, Zhang et.al showed that DLR offered superior image quality over filtered back projection and iterative reconstruction, with higher signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) as well as improved subjective ratings of image appearance and sharpness (11). Despite the encouraging results, few studies have specifically addressed whether these techniques improve the visualization of distal arteries, and importantly, collateral arteries, which are often suboptimally visualized on CTA due to their small caliber, yet play a critical role in maintaining limb perfusion when arterial occlusions are present.

The aim of the present study was to investigate whether a novel deep learning (DL)-based CT reconstruction algorithm, namely, artificial intelligence iterative reconstruction (AIIR; United Imaging Healthcare), can improve the visualization of distal and collateral arteries in lower extremity CTA of diabetic patients compared to hybrid iterative reconstruction (HIR), a routinely used reconstruction method in clinical practice.

Material and Methods

Study population

This retrospective study was approved by the institutional review board of the Second Clinical Medical College of Henan University of Chinese Medicine (No. 1524-01). The requirement for informed consent was waived. Diabetic patients who underwent clinically indicated lower extremity CTA at our institution between July 2024 and March 2025 and had raw data available were initially included. Patients were excluded if their CTA images showed venous overlay, because AIIR, being a reconstruction algorithm, theoretically does not address venous overlay and such condition would introduce confounding factors into the analysis. Notably, patients with bilateral disease or stenoses were not excluded, as these conditions are common among diabetic patients and represent typical clinical scenarios.

Scan protocol and image reconstruction

All CTA scans were acquired using a 320-row CT scanner (uCT 968; United Imaging Healthcare) in helical mode, with each examination performed in the craniocaudal direction from the third lumbar vertebra to the toes. The acquisition parameters were as follows: longitudinal detector collimation = 40 mm; pitch = 1.0; gantry rotation time = 0.5 s; tube voltage = 100 kVp, with automatic tube current modulation (reference 77 mAs). Iodinated contrast medium (iohexol 350 mg I/mL, Omnipaque; GE Healthcare, Chicago, IL, USA) was administered with a weight-dependent volume of 1.5–2.0 mL/kg at an injection rate of 3.0 mL/s followed by a 50-mL saline flush. The scan acquisition was initiated 20 s after a threshold of 200 Hounsfield Units (HU) was achieved in the abdominal aorta at the level of the third lumbar vertebra. Each scan was reconstructed with both AIIR at a strength level of 3 (range = 1–5) and a routinely available HIR algorithm (Karl 3D; United Imaging Healthcare) at a strength level of 5 (range = 1–10), with a slice thickness of 1 mm, slice interval of 0.5 mm, and pixel matrix of 512 × 512. Notably, the strength levels for each reconstruction algorithm were carefully selected to achieve a reasonable balance between the noise suppression and the preservation of clinically acceptable texture. AIIR is a DL-based iterative reconstruction algorithm that has been proven effective and clinically valuable across a range of applications, including abdominal (12 –15), thoracic (16 –18), and vascular CT imaging (19). AIIR and HIR image datasets were anonymized, randomly ordered, and transferred to a clinical postprocessing workstation for subsequent evaluations.

Distal arterial visualization

Visualization of the distal arteries, including the anterior tibial arteries, posterior tibial arteries, dorsalis pedis arteries, medial plantar arteries, dorsal metatarsal arteries, and dorsal digital arteries, was graded by two senior radiologists in consensus, each with over 10 years of experience in vascular imaging, according to the following schema: 1 = poor, vessel barely discernible; 2 = suboptimal, insufficient opacification with poorly defined margins; 3 = moderate, acceptable opacification with unsharp margins; 4 = good, adequate opacification with mildly blurred margins; and 5 = excellent, optimal opacification with sharp margins. In addition, one of the two readers measured the diameter of each evaluated vessel three times, with the mean value used as the final vessel size to minimize measurement variability. For each patient, if a complete occlusion with no visible flow was identified within a given vessel, that vessel would be excluded from the analysis. Of note is that the consensus interpretation in this study refers to the evaluations performed independently by two radiologists who were blinded to the reconstruction techniques and patient demographics, with any discrepancies subsequently resolved through the negotiation between the two readers. In the radiologic literature, both consensus (20 –24) and independent interpretation (25,26) are commonly used. In this study, although both strategies would have been acceptable, consensus interpretation was chosen as we considered it to better reflect clinical practice, where a single diagnostic conclusion is typically reached.

Collateral circulation depiction

The same two radiologists assessed the depiction of collateral arteries in consensus using a 5-point ordinal scale: 1 = absent, no collateral arteries visible; 2 = suboptimal, faintly discernible collateral arteries with poor continuity; 3 = moderate, recognizable collateral arteries with interrupted course; 4 = good, adequately visualized collateral arteries with generally continuous course; and 5 = excellent, well-opacified collateral network with clearly visualized course.

Overall image quality

Overall image quality was evaluated with both qualitative and quantitative metrics. Qualitatively, the same two radiologists who performed the preceding evaluations, graded the image quality in consensus using a 5-point scale as follows: 1 = unacceptable, extensive noises or artifacts, or both; 2 = inferior, substantial noises or artifacts, or both; 3 = adequate, moderate but acceptable noises or artifacts, or both; 4 = good, mild noises or artifacts, or both; and 5 = excellent, negligible noises or artifacts, or both. It is worth noting that the image texture, though not explicitly incorporated into the scoring criteria, was in fact considered during the study design, where the reconstruction strength levels of both algorithms were specifically selected to avoid the waxy appearance. This preselection ensured all evaluated images maintained clinically acceptable texture, enabling a fair comparison of both reconstruction methods under their respective optimal clinical settings. Quantitatively, a board-certified radiologist with 7 years of experience in vascular imaging measured the mean and the standard deviation (SD) of attenuation within circular regions of interest (ROIs) placed on the popliteal artery, anterior tibial artery, and posterior tibial artery, with the corresponding size set as large as possible, carefully avoiding inclusion of vessel edges and plaques. The SD of attenuation within the ROI was defined as image noise. The SNR and CNR were calculated as $S N R_{R O I} = \frac{H U_{R O I}}{S D_{R O I}}$ and $C N R_{R O I} = \frac{H U_{R O I} - H U_{b a c k g r o u n d}}{S D_{b a c k g r o u n d}}$ , respectively, where $H U_{R O I}$ and $S D_{R O I}$ are the mean and the SD of attenuation within the ROI, respectively. $H U_{b a c k g r o u n d}$ and $S D_{b a c k g r o u n d}$ denote the mean and the SD of attenuation within the ROI positioned on the gastrocnemius, respectively. Of note is that foot arteries were not selected for ROI placement due to their relatively small caliber, which leads to substantial measurement variability. The ROIs were initially positioned on the HIR image and then copy-pasted to the AIIR image to ensure the ROI consistency throughout the image datasets.

Statistical analysis

All the statistical analyses were conducted using SPSS version 26 (IBM Corp., Armonk, NY, USA) and GraphPad Prism version 9 (GraphPad Corp., San Diego, CA, USA). The Kolmogorov–Smirnov test was used to assess the distribution of the data. Depending on the normality of the data, either the paired t-test or the Wilcoxon signed-rank test was applied for comparison. A confusion matrix was used to present a detailed comparison of grading results for collateral circulation depiction between two reconstruction methods. P <0.05 was considered statistically significant.

Results

Patient characteristics and radiation dose

A total of 59 patients (47 men, 12 women; mean age = 72 ± 10 years; age range = 42–89 years; mean body mass index [BMI] = 25.4 ± 1.7 kg/m²; range = 22.1–32.8 kg/m²) with type II diabetes were included in the final study cohort. All patients had bilateral arterial stenoses, which occurred naturally within the study cohort rather than through selection. The mean volumetric CT dose index (CTDI_vol) and dose-length product (DLP) were 2.8 ± 0.2 mGy and 363.9 ± 49.1 mGy·cm, respectively.

Distal arterial visualization

As shown in Table 1, there was no significant difference between HIR and AIIR in the visualization scores of the anterior tibial artery on either side. In contrast, AIIR yielded significantly higher scores than HIR in all other arteries (all P <0.05). Fig. 1 illustrates the comparison of vessel visualization scores between HIR and AIIR across vessel diameters. Visualization scores showed an increasing trend as vessel diameter increased for both reconstruction methods. Specifically, the score differences between HIR and AIIR were more pronounced in smaller distal arteries. Fig. 2 illustrates an example of distal arteries displayed with HIR and AIIR, where vessel delineation is noticeably improved on the AIIR image compared to the HIR image.

Fig. 1.

Comparison of vessel visualization scores between HIR and AIIR across vessel diameters. Of note, the scores were averaged for vessels of the same diameter. Logarithmic regression curves were applied to illustrate the relationship between vessel diameter and vessel visualization score. AIIR, artificial intelligence iterative reconstruction; HIR, hybrid iterative reconstruction.

Fig. 2.

Lower extremity CTA in a 70-year-old man with a 10-year history of diabetes mellitus. Images reconstructed with HIR and AIIR are presented as maximum intensity projections and axial views at three anatomical levels: (1) mid-calf, (2) midfoot, and (3) forefoot. Compared with HIR, AIIR demonstrates improved visualization of distal arteries (yellow arrows) and clearer delineation of small arterial branches (white arrows). A calcified plaque in the left anterior tibial artery is more sharply visualized on the AIIR image, whereas it appears blurred on the HIR image. AIIR, artificial intelligence iterative reconstruction; CTA, computed tomography angiography; HIR, hybrid iterative reconstruction.

Table 1.

Comparison of grading results for distal arterial visualization between HIR and AIIR.

	HIR	AIIR	P value
ATA (L) (n = 48)	4.38 ± 0.67/4 (4–5)	4.43 ± 0.65/5(4–5)	0.250
ATA (R) (n = 50)	4.28 ± 0.76/4 (4–5)	4.44 ± 0.64/5(4–5)	0.063
PTA (L) (n = 44)	4.29 ± 0.73/4 (4–5)	4.43 ± 0.59/4(4–5)	0.031
PTA (R) (n = 48)	4.37 ± 0.61/4 (4–5)	4.56 ± 0.54/5(4–5)	0.008
DPA (L) (n = 55)	3.44 ± 0.86/3 (3–4)	3.82 ± 0.75/4(3–4)	<0.001
DPA (R) (n = 55)	3.62 ± 0.71/4 (3–4)	3.89 ± 0.63/4(3–4)	0.001
MPA (L) (n = 52)	2.42 ± 0.78/3 (2–3)	3.54 ± 0.69/4(3–4)	<0.001
MPA (R) (n = 49)	2.63 ± 1.01/3 (2–3)	3.59 ± 0.70/4(3–4)	<0.001
DMA (L) (n = 54)	2.30 ± 0.88/2 (2–3)	3.39 ± 0.49/3(3–4)	<0.001
DMA (R) (n = 55)	2.53 ± 0.88/2 (2–3)	3.34 ± 0.58/3(3–4)	<0.001
DDA (L) (n = 37)	1.86 ± 0.75/2 (1–2)	3.11 ± 0.84/3(3–3.5)	<0.001
DDA (R) (n = 39)	1.92 ± 0.62/2 (2–2)	3.23 ± 0.71/3(3–4)	<0.001

Values are given as mean ± SD/median (interquartile range).

AIIR, artificial intelligence iterative reconstruction; ATA, anterior tibial artery; DDA, dorsal digital artery; DMA, dorsal metatarsal artery; DPA, dorsalis pedis artery; HIR, hybrid iterative reconstruction; L, left; MPA, medial plantar artery; PTA, posterior tibial artery; R, right.

Collateral circulation depiction

The scores for collateral circulation depiction in bilateral lower extremities were significantly higher with AIIR than those with HIR (all P <0.001), with median scores (25th–75th percentiles) as follows: left lower extremity, HIR 0 (0–2) versus AIIR 2 (0–3); right lower extremity, HIR 0 (0–2) versus AIIR 2 (0–3). The details of grading results are presented as confusion matrices in Fig. 3, with corresponding examples shown in Fig. 4. For the left lower extremity, of the 29 patients with collateral arteries detected on both HIR and AIIR images, 8 (27.6%) had identical grading between two reconstructions, while in the remaining cases, AIIR images demonstrated superior visualization of collateral vessels compared to HIR images. Similarly, for the right lower extremity, among the 28 patients in whom collateral circulation was identified on both reconstructions, grading was identical in 5 (17.9%) cases, whereas in the remaining patients, collateral vessels were better visualized on AIIR images compared to HIR images.

Fig. 3.

The detailed comparison of grading results for collateral circulation depiction between HIR and AIIR images. AIIR, artificial intelligence iterative reconstruction; HIR, hybrid iterative reconstruction.

Fig. 4.

Lower extremity CTA in an 80-year-old man with a 15-year history of diabetes mellitus. Maximum intensity projections from HIR and AIIR images are compared side-by-side. AIIR enables better depiction of collateral vessels around the right femoral artery (yellow arrows), showing a richer and more continuous collateral network compared to HIR. AIIR, artificial intelligence iterative reconstruction; CTA, computed tomography angiography; HIR, hybrid iterative reconstruction.

Overall image quality

The subjective image quality score was significantly higher with AIIR than with HIR (P <0.001): mean HIR = 3.92 ± 0.28 versus mean AIIR = 4.90 ± 0.30; median HIR 4 (4–4) versus median AIIR 5 (5–5). As shown in Fig. 5, AIIR significantly suppressed the image noise, resulting in a significantly higher SNR and CNR (all P <0.001).

Fig. 5.

Comparisons of noise, SNR, and CNR between HIR and AIIR images. AIIR, artificial intelligence iterative reconstruction; ATA, anterior tibial artery; CNR, contrast-to-noise ratio; HIR, hybrid iterative reconstruction; L, left; PA, popliteal artery; PTA, posterior tibial artery; R, right; SNR, signal-to-noise ratio.

Discussion

In this study, we thoroughly investigated the performance of a novel DL-based CT image reconstruction algorithm, i.e. AIIR, in visualizing distal and collateral arteries on lower extremity CTA of diabetic patients, which are frequently suboptimally visualized on CTA due to low contrast, yet are critical for the clinical management of DFU.

Our results demonstrated that AIIR substantially improved the visualization of distal arteries compared to HIR. This was not unexpected, given that AIIR showed improvements in SNR and CNR in the range of 64%–80% and 60%–80%, respectively, contributing to clearer vessel margins and enhanced conspicuity of vessels. Specifically, the differences in vessel visualization scores between AIIR and HIR were relatively small for the anterior and posterior tibial arteries (all mean scores >4), but became more pronounced in the smaller distal branches, such as the medial plantar, dorsal metatarsal, and dorsal digital arteries. This may be explained, at least in part, by the limited baseline visibility of these smaller branches on HIR images, allowing greater room for contrast-related enhancement. In addition, AIIR also significantly improved the visualization of collateral arteries, further supporting the capacity of AIIR to improve the depiction of small and low-contrast vascular structures.

Previous studies have shown that DL-based reconstruction techniques can effectively suppress the image noise and improve the SNR and the CNR, thereby enhancing the overall image quality of lower extremity CTA (10,11), which is in line with our findings. However, our study extends beyond these previous works by specifically evaluating the visualization of small-caliber vessels, showing that in addition to improving the overall image quality, AIIR also offers substantial advantages in the depiction of distal and collateral arteries. Apart from DL-based reconstruction methods, dual-energy CT has also been employed to improve the image quality of lower extremity CTA. It has been proven effective in enhancing the visualization of below-the-knee arteries through virtual monoenergetic imaging (8). This acquisition-based approach can be regarded as another technical route for improving CTA image quality, distinct from reconstruction-based optimization achieved with AIIR. A direct comparison between these two strategies remains unexplored, and in practice the choice between them may depend more on clinical preference and specific imaging scenarios. With the development of CT techniques, there is also the possibility that AIIR could be applied to dual-energy CT datasets, which may further enhance the vascular visualization beyond either approach alone.

Lower extremity arterial evaluation is particularly critical in diabetic patients, who have a high prevalence of PAD. It is estimated that up to 50% of patients with DFU exhibit evidence of PAD (27), with a predominant involvement of the below-the-knee arteries such as the anterior tibial, posterior tibial, and peroneal arteries (3,28,29). These occlusive changes are often bilateral, multi-segmental, and predominantly distal, significantly impairing perfusion to the foot. In such cases, accurate visualization of both distal arteries and collateral circulation is essential not only for determining the presence and extent of ischemia, but also for guiding appropriate therapeutic decision-making. The presence of well-developed distal or collateral arteries may obviate the need for intervention, whereas their absence may strengthen the indication for revascularization or amputation planning. In this context, our findings support the clinical value of the AIIR. By significantly enhancing the delineation of the vessels that are often poorly visualized on routine CTA, AIIR may facilitate a more accurate stratification of perfusion status in diabetic limbs. This may in turn refine patient selection for endovascular therapy, improve procedural planning, and ultimately contribute to better limb preservation outcomes.

In clinical practice, lower-extremity CTA interpretation is often complicated by bilateral disease, stenoses, occlusions, and venous overlay, which can alter runoff scenarios. Among these conditions, bilateral disease and stenoses are common among diabetic patients and were considered in this study. In such situations, distal arteries often showed suboptimal visualization, partly due to reduced contrast related to luminal narrowing or irregular opacification. By effectively suppressing the image noise and improving the CNR, AIIR provides clearer delineation of distal arteries, which may facilitate image interpretation and enhance diagnostic confidence, thereby offering a more reliable imaging basis for clinical decision-making. With respect to venous overlay and complete occlusions, although these conditions were not investigated in this study, it is reasonable to assume that the impact of AIIR is limited in such scenarios, because venous overlap is unrelated to reconstruction performance and complete occlusion lacks intraluminal opacification that is physically present for AIIR to improve.

In this study, the inclusion of patients with type II diabetes was not based on a predefined selection criterion but rather reflected the actual clinical composition of diabetic patients who underwent clinically indicated lower extremity CTA at our institution. The fact that the resulting cohort consisted entirely of patients with type II diabetes is not unexpected, as type II diabetes is the most common subtype of diabetes, accounting for approximately 90%–95% of diabetes cases worldwide (30,31), which thus primarily reflects the cohort's clinical representativeness. Importantly, the image quality improvements achieved by AIIR are intrinsic to the reconstruction process itself and are thus not specifically related to diabetes type. Therefore, the technological benefits of AIIR observed in this clinically representative cohort are expected to be reproducible across all patient populations undergoing lower extremity CTA, regardless of their specific disease subtype.

The present study has some limitations. First, it was performed only in the routine dose setting that we believed to be most applicable to daily clinical workflows, while the performance of AIIR in low-dose settings remains to be investigated. Second, although comparison with analogous reconstruction methods from other vendors would be of interest, it was not included in this study due to limited access to each commercial algorithm. Third, the inherent visual differences between reconstruction algorithms have made the readers’ assessment less likely to be truly blind, potentially introducing a bias that may be unavoidable in real-world imaging evaluations.

In conclusion, the newly proposed DL-based iterative reconstruction algorithm, AIIR, offers a better visualization of distal and collateral arteries as well as an improved overall image quality in lower extremity CTA of diabetic patients, compared to HIR.

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Zhongjie Yang

References

Singh

Armstrong

Lipsky

. Preventing foot ulcers in patients with diabetes. JAMA 2005;293:217–228.

Wild

Roglic

Green

, et al. Global prevalence of diabetes: estimates for the year 2000 and projections for 2030. Diabetes Care 2004;27:1047–1053.

Miranda

Da Ros

Marfella

. Update on prevention of diabetic foot ulcer. Arch Med Sci Atheroscler Dis 2021;6:e123–e131.

Buls

de Brucker

Aerden

, et al. Improving the diagnosis of peripheral arterial disease in below-the-knee arteries by adding time-resolved CT scan series to conventional run-off CT angiography. First experience with a 256-slice CT scanner. Eur J Radiol 2019;110:136–141.

Jia

, et al. Improving diagnostic accuracy for arteries of lower extremities with dual-energy spectral CT imaging. Eur J Radiol 2020;128:109061.

Wichmann

Gillott

Cecco

, et al. Dual-energy computed tomography angiography of the lower extremity runoff: impact of noise-optimized virtual monochromatic imaging on image quality and diagnostic accuracy. Invest Radiol 2016;51:139–146.

Huang

Nelson

Miller

, et al. Assessment of vascular contrast and depiction of stenoses in abdominopelvic and lower extremity vasculature: comparison of dual-energy MDCT with digital subtraction angiography. Acad Radiol 2012;19:1149–1157.

Yan

Zhou

, et al. Improved small vessel visibility in diabetic foot arteriography using dual-energy CT. Clin Radiol 2024;79:e424–e431.

Scheschenja

Konig

Viniol

, et al. CT angiography of the lower limbs: combining small field of view with large matrix size and iterative reconstruction further improves image quality of below-the-knee arteries. Acta Radiol 2024;65:774–783.

10.

Guo

, et al. Iterative reconstruction vs deep learning image reconstruction: comparison of image quality and diagnostic accuracy of arterial stenosis in low-dose lower extremity CT angiography. Br J Radiol 2022;95:20220196.

11.

Zhang

, et al. Image quality comparison of lower extremity CTA between CT routine reconstruction algorithms and deep learning reconstruction. BMC Med Imaging 2023;23:33.

12.

Cui

, et al. Image quality assessment of artificial intelligence iterative reconstruction for low dose unenhanced abdomen: comparison with hybrid iterative reconstruction. Abdom Radiol (NY) 2025;50:3353–3362.

13.

Meng

Zhang

, et al. Artificial intelligence iterative reconstruction in abdominal CT of patients with irregular arm positioning: a case-by-case evaluation. Acta Radiol 2024;65:907–912.

14.

You

Zhong

Zhang

, et al. 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;37:2089–2098.

15.

Zhu

Zou

, et al. Diagnostic CT of colorectal cancer with artificial intelligence iterative reconstruction: a clinical evaluation. Eur J Radiol 2024;171:111301.

16.

Yang

Liu

Han

, et al. Ultra-low-dose CT lung screening with artificial intelligence iterative reconstruction: evaluation via automatic nodule-detection software. Clin Radiol 2023;78:525–531.

17.

Zhang

Peng

Zhang

, et al. Artificial intelligence iterative reconstruction for dose reduction in pediatric chest CT:A clinical assessment via below 3 years patients with congenital heart disease. J Thorac Imaging 2025;40:e0827.

18.

Zhuo

Zhang

, et al. Ultralow dose coronary calcium scoring CT at reduced tube voltage and current by using deep learning image reconstruction. Eur J Radiol 2024;181:111742.

19.

Gong

Peng

, et al. Artificial intelligence iterative reconstruction in computed tomography angiography: An evaluation on pulmonary arteries and aorta with routine dose settings. J Comput Assist Tomogr 2024;48:244–250.

20.

Liang

Sun

, et al. Second-generation motion correction algorithm improves diagnostic accuracy of single-beat coronary CT angiography in patients with increased heart rate. Eur Radiol 2019;29:4215–4227.

21.

Benz

Fuchs

Grani

, et al. Head-to-head comparison of adaptive statistical and model-based iterative reconstruction algorithms for submillisievert coronary CT angiography. Eur Heart J Cardiovasc Imaging 2018;19:193–198.

22.

Boehm

Willmann

Hilfiker

, et al.

Thin-section CT of the lung: does electrocardiographic triggering influence diagnosis?

Radiology 2003;229:483–491.

23.

van den Bosch

Westenberg

Caris

, et al. Peripheral arterial occlusive disease: 3.0-T versus 1.5-T MR angiography compared with digital subtraction angiography. Radiology 2013;266:337–346.

24.

Lin

, et al. Feasibility of using deep learning to detect coronary artery disease based on facial photo. Eur Heart J 2020;41:4400–4411.

25.

Arslan

Alis

Erdemli

, et al.

Does deep learning software improve the consistency and performance of radiologists with various levels of experience in assessing bi-parametric prostate MRI?

Insights Imaging 2023;14:48.

26.

Halpern

Levin

Zhang

, et al. Comparison of image quality and arterial enhancement with a dedicated coronary CTA protocol versus a triple rule-out coronary CTA protocol. Acad Radiol 2009;16:1039–1048.

27.

Prompers

Huijberts

Apelqvist

, et al. High prevalence of ischaemia, infection and serious comorbidity in patients with diabetic foot disease in Europe. Baseline results from the Eurodiale study. Diabetologia 2007;50:18–25.

28.

American Diabetes Association. Peripheral arterial disease in people with diabetes. Diabetes Care 2003;26:3333–3341.

29.

Jude

Oyibo

Chalmers

, et al. Peripheral arterial disease in diabetic and nondiabetic patients: a comparison of severity and outcome. Diabetes Care 2001;24:1433–1437.

30.

Zheng

Ley

. Global aetiology and epidemiology of type 2 diabetes mellitus and its complications. Nat Rev Endocrinol 2018;14:88–98.

31.

Ali

Pearson-Stuttard

Selvin

, et al. Interpreting global trends in type 2 diabetes complications and mortality. Diabetologia 2022;65:3–13.