Evaluation of optimal ClearInfinity artificial intelligence deep learning reconstruction algorithm weights with a 70-kVp tube voltage combined with “dual-low” abdominal computed tomography angiography

Introduction

Abdominal computed tomography angiography (CTA) is an imaging technique that is critical for evaluating the abdominal vascular structures and plays a key role in the precise localization and dynamic monitoring of vascular lesions. It has been widely applied in the diagnosis of abdominal vascular diseases, assessment of unexplained abdominal pain, and postoperative follow-up evaluations (1). However, conventional abdominal CTA typically requires contrast-enhanced acquisitions and a relatively wide scan coverage, and, in certain clinical scenarios, may also involve multiphase imaging, thereby resulting in higher radiation exposure. For patients who require repeated imaging follow-up, the increased cumulative radiation dose may be associated with potential long-term stochastic risks, including radiation-related malignancy (2). Therefore, reducing radiation exposure is particularly important for patients undergoing repeated abdominal CTA examinations (3). Lowering the tube voltage is an effective means to reducing the radiation dose, but it often leads to increased image noise and decreased contrast resolution (4). In addition, the injection rate and total volume of the contrast medium substantially affect image quality. Hence, developing a low-dose abdominal CTA protocol that maintains diagnostic image quality under conditions of a low tube voltage and low contrast dose is of significant clinical value.

In recent years, deep learning-based image reconstruction (DLR) algorithms have been increasingly incorporated into routine clinical practice (5-9). Compared with traditional iterative reconstruction techniques (e.g., adaptive statistical iterative reconstruction-V or ClearView), DLR effectively reduces image noise and suppresses artifacts, thereby enhancing overall image quality while preserving the authenticity of image texture (10,11). ClearInfinity (Neusoft Medical Systems Co., Ltd.) represents a new generation of DLR algorithms. Based on a convolutional neural network, it is trained with a large number of paired high- and low-dose CT datasets through supervised learning, allowing the network to learn the distribution patterns of noise and artifacts and to effectively separate structural information from interfering signals (12).

Previous studies have demonstrated that ClearInfinity can significantly improve image quality and reduce noise in head and neck CTA and aortic CTA examinations (13,14). However, in abdominal CTA, the large variations in organ density and prominent respiratory motion artifacts make low-dose imaging particularly challenging, with a more pronounced decline in the signal-to-noise ratio (SNR). These issues place higher demands on the algorithm’s capability for noise suppression and texture preservation. Thus far, no study has systematically evaluated or optimized the reconstruction weight of ClearInfinity for a low-kilovolt peak, low-contrast medium protocol in abdominal CTA. Therefore, this study systematically evaluated the impact of different ClearInfinity reconstruction weights on image quality and radiation dose under a 70-kVp dual-low scanning protocol, aiming to determine the optimal reconstruction weight that achieves a balance between diagnostic image quality and patient safety. We present this article in accordance with the TREND reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2589/rc).

Methods

Study participants

This prospective single-center study was approved by the Ethics Committee of Shiyan Taihe Hospital (approval No. 2024KS15), and written informed consent was obtained from all participants. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. A total of 100 consecutive patients [59 males and 41 females; age range, 27–80 years; body mass index (BMI) 17.36–27.97 kg/m²] who underwent abdominal CTA at Shiyan Taihe Hospital between November 2023 and June 2024 for suspected abdominal vascular diseases were prospectively enrolled. All patients were consecutively included from departmental examinations and were assigned to two different imaging protocols according to the order of their examinations (Figure 1). Patients with a known allergy to iodinated contrast medium or other severe allergies, cardiac insufficiency, renal dysfunction, or inability to cooperate with breath-holding were excluded.

Figure 1 Study flowchart. BMI, body mass index; CTA, computed tomography angiography; DLIR-CI, deep learning image reconstruction-ClearInfinity; HIR-CV, hybrid iterative reconstruction-ClearView.

Scanning protocol

All eligible participants underwent abdominal CTA on a 128-slice CT scanner (NeuViz ACE 128; Neusoft Medical Systems Co., Ltd., Shenyang, China). For the experimental group (group A), the tube voltage was set at 70 kVp with ODose automatic tube current modulation (200–450 mA). The contrast injection protocol consisted of a flow rate of 2.0–2.5 mL/s and a total volume of 40–45 mL. Other parameters were as follows: gantry rotation time, 0.48 s; collimation, 0.75 mm × 128; pitch, 1.0; matrix, 512×512; and a scan range from the diaphragm to the ischial tuberosities. Group A was further reconstructed into four subgroups with different ClearInfinity weights: A1 (30%), A2 (50%), A3 (70%), and A4 (90%). According to preliminary experimental results and discussions with technicians and two senior radiologists, the most marked variation in image noise and texture characteristics was observed between 30% and 90% weights. Thus, four representative levels were selected to capture the full range of noise reduction intensity. This stepwise design facilitated a systematic evaluation of the balance between noise reduction and structural fidelity, providing a basis for determining the optimal reconstruction weight for abdominal CTA. During the pre-experimental phase, we considered using the combination of 70 kVp and ClearView for the control group. However, we found that the image quality of this combination did not meet the clinical diagnostic standards, particularly in terms of image contrast and clarity, and thus did not satisfy the expected requirements. As a result, this combination was not used for the formal experimental group. For the control group (group B), images were reconstructed with the routinely applied 50% ClearView iterative reconstruction algorithm under a 100-kVp tube voltage, a contrast injection rate of 3.0–3.5 mL/s, and a total contrast volume of 60–70 mL. The other scanning parameters were identical to those of group A. For both groups, ioversol (350 mg I/mL; Aisuxian Pharmaceutical Co., Ltd., Beijing, China) was used as the contrast medium. The region of interest (ROI) for bolus tracking was placed in the abdominal aorta at the level of the diaphragm, and the scan was automatically triggered when the CT attenuation reached 150 Hounsfield units (HU).

Image processing

All raw data were transferred to an Advanced Visualization Workspace (AVW; Neusoft Medical Systems Co., Ltd., China) and reconstructed into 1-mm-thick axial images with a 1-mm interval. Postprocessing included volume rendering (VR) and maximum intensity projection (MIP) for visualization of the vascular anatomy.

Radiation dose analysis

For both groups, computed tomography dose index volume (CTDIvol), dose-length product (DLP), and effective dose (ED) were recorded. The conversion factor for the abdomen was k=0.015 mSv/(mGy·cm) (15).

Objective image evaluation

On 1-mm reconstructed axial images, the mean CT attenuation value and standard deviation (SD) were measured in five sets of images at the following vascular sites: the abdominal aorta, superior mesenteric artery, mid-segment of the left renal artery, and origin of the left common iliac artery. The erector spinae muscle on the same slice was used as the reference tissue. For both group A and group B, the ROIs were placed at standardized levels: at the celiac trunk, origin of the superior mesenteric artery, mid-left renal artery, and origin of the left common iliac artery. In subgroups A1–A4, measurements were performed at identical axial levels. The ROI was a circular region that covered one-half to two-thirds of the vessel’s diameter, and measurements are performed three times to obtain an average value. In the placement of the ROI, its center was positioned within the lumen of the vessel, and the vessel wall, calcified plaques, and surrounding fat tissue were avoided to minimize measurement bias.

The abdominal aorta and superior mesenteric artery were selected as primary vessels because they represent major abdominal arteries and reflect overall vascular opacification, while the left renal and left common iliac arteries were selected for their consistent positions, straight courses, and homogeneous surrounding tissues, which are less affected by hepatic or bowel gas artifacts as compared to their right-side counterparts. To ensure cross-group consistency, the left-sided vessels were uniformly selected as measurement targets. The SNR and contrast-to-noise ratio (CNR) were calculated with the following formulae:

SNR=HUvesselSDvessel[1]

CNR=HUvessel−HUmuscleSDmuscle[2]

Subjective image evaluation

Two radiologists with over 10 years of CT diagnostic experience independently evaluated the images of groups A1–A4 and B on the AVW in a double-blinded manner. The window width and level were fixed at 350 and 50 HU (abdominal soft tissue window), respectively, to ensure standardized and reproducible image evaluation across all datasets. This fixed window setting was applied consistently to minimize observer-dependent variability during subjective image quality assessment and to enable consistent visualization of vascular structures and adjacent abdominal soft tissues across reconstruction protocols. All images were rated (16) based on contrast, noise, vessel sharpness, and overall diagnostic acceptability via the following 5-point scheme: 1 point was defined as very low contrast, poor image quality, and severe artifacts and noise, with only the abdominal aorta being visible; 2 points was defined as low contrast, an overall blurry image, and unclear vessel edges, with the abdominal aorta and other visceral artery trunks being visible; 3 points was defined as good contrast, less noise, and an overall image quality meeting diagnostic criteria, with the second-order branches of the renal artery being clearly visible; 4 points was defined as high contrast, clear vessel edges, and image quality providing good diagnostic confidence, with the third-order branches of the renal artery being visible; and 5 points was defined as high contrast, a clear and sharp overall image, very few noise artifacts, and an image quality favorable for diagnosis, with the third-order and higher-order branches of the renal artery visible.

Statistical analysis

All statistical analyses were performed with SPSS software version 27.0 (IBM Corp., Armonk, NY, USA). The normality of continuous variables was assessed with the Shapiro-Wilk test, and categorical variables were analyzed with the chi-squared test. For the analysis of objective image quality parameters across the five image groups, paired-sample analysis of variance was applied to normally distributed data, while the Friedman correlation test was used for nonnormally distributed data. The interobserver consistency for subjective image scores was evaluated according to the Cohen kappa coefficient: κ<0.20 indicated poor agreement, 0.21–0.40 fair, 0.41–0.60 moderate, 0.61–0.80 good, and 0.81–1.00 excellent. A P value <0.05 was considered statistically significant.

Results

Clinical characteristics and radiation dose

There were no statistically significant differences in age, sex, or BMI between groups A and B (P>0.05). The CTDIvol, DLP, and ED in group A were significantly lower than those in group B (P<0.05). Compared with group B, the 70-kVp dual-low protocol reduced the mean radiation dose by approximately 64.08% and the contrast medium volume by about 34.61% while maintaining comparable vascular enhancement (Table 1).

Objective evaluation

The objective measurements of image quality for the abdominal aorta at different anatomical levels are summarized in Table 2. No significant differences were found in the mean CT attenuation values of the abdominal aorta, superior mesenteric artery, left renal artery, or left common iliac artery between the groups (all P values >0.05). As the ClearInfinity reconstruction weight increased from A1 to A4, image noise (SD) decreased in succession, whereas both the SNR and CNR progressively improved (all P values <0.05).

Groups A1 to A4 were compared with the iterative reconstruction control group (B). Group A1 exhibited higher SD values and lower SNR and CNR values across all vessels as compared to group B (P<0.05); group A2, exhibited higher SD values of the abdominal aorta, superior mesenteric artery, and renal artery were higher but a lower SNR for all vessels and a lower CNR for the common iliac artery (P<0.05); group A3 had lower SD values of the abdominal aorta, superior mesenteric artery, and common iliac artery but a higher SNR and CNR across all vessels (P<0.05); finally, group A4 demonstrated the most pronounced improvements, showing significantly lower SD values and a markedly higher SNR and CNR in all vessels compared with group B (P<0.05).

Overall, with group B serving as the reference, groups A3 and A4 achieved image quality that was comparable to or superior to that of the control group in terms of the SNR and CNR. Group A4 demonstrated the best performance, with the mean SNR and CNR increased by approximately 45.62% and 79.68%, respectively, confirming that at a 90% weighting level, the ClearInfinity deep learning reconstruction algorithm can substantially enhance image quality.

Subjective evaluation

Figure 2 illustrates the stacked bar chart of subjective image quality scores. For groups A1, A2, A3, A4, and B, interobserver agreement was good, with kappa coefficients of 0.839, 0.880, 0.848, 0.845, and 0.819, respectively. The mean scores from the two radiologists were used for subsequent analyses. Statistical testing revealed that the image quality score in group A1 differed significantly from those of the other groups (A2, A3, A4, and B; P<0.05), whereas no significant differences were observed among groups A2, A3, A4, and B (P>0.05).

Figure 2 Stacked bar chart of subjective image quality scores. Image quality was evaluated with a 5-point Likert scale (1= very poor and 5= excellent) based on contrast, image noise, vessel sharpness, and overall diagnostic acceptability. The left and right panels show the score distributions assigned by reader 1 and reader 2, respectively. The respective mean subjective image quality scores for reader 1 and reader 2 were as follows: A1 (30% CI), 3.18 and 3.08; A2 (50% CI), 4.16 and 4.10; A3 (70% CI), 3.98 and 3.96; A4 (90% CI), 3.80 and 3.74; and B (50% CV), 4.14 and 4.10. CI, ClearInfinity; CV, ClearView.

In addition, compared with the ClearView iterative reconstruction algorithm, the ClearInfinity algorithm demonstrated superior noise-suppression performance and provided subjectively clearer visualization of the first- and second-order branches of the renal artery. Figure 3 presents a representative case reconstructed with various ClearInfinity weights at 70 kVp, and Figure 4 shows the morphology and internal structure of the abdominal aorta reconstructed with the 50% ClearView iterative algorithm at 100 kVp. The corresponding reader evaluations of these images are provided in the Table S1. Under relatively noisy imaging conditions, higher ClearInfinity blending weights (e.g., A3–A4) were associated with improved perceived noise suppression and delineation of small-caliber vessels.

Figure 3 The first three columns represent the volume rendering images, maximum intensity projection images, and axial images showing the location, morphology, and internal condition of the abdominal aorta. (A-L) Follow-up CTA of the abdominal aorta at 70 kVp in a 44-year-old male (BMI, 23.46 kg/m²; contrast medium volume, 45 mL; injection rate, 2.5 mL/s). The abdominal aorta (below the renal arteries) was clearly visualized without evidence of contrast extravasation. The images were reconstructed with different ClearInfinity deep learning reconstruction algorithms. (A-C) Images reconstructed with A1 (30% CI), (D-F) images reconstructed with A2 (50% CI), (G-I) images reconstructed with A3 (70% CI), and (J-L) images reconstructed with A4 (90% CI). Compared with A1 reconstruction, A2–A4 reconstructions showed reduced image noise and improved vessel sharpness; no obvious visual differences were observed between A2, A3, and A4 in this representative case. BMI, body mass index; CI, ClearInfinity; CTA, computed tomography angiography.

Figure 4 Abdominal aortic CTA in a 57-year-old male (BMI, 26.6 kg/m²; contrast medium volume, 60 mL; injection rate, 3.2 mL/s) acquired at 100 kVp and reconstructed with the 50% ClearView iterative algorithm. (A) MIP image. (B) VR image. (C) Axial image (1-mm thickness). BMI, body mass index; CTA, computed tomography angiography; MIP, maximum intensity projection; VR, volume rendering.

Discussion

This study aimed to determine the optimal blending weight of the ClearInfinity deep learning reconstruction algorithm under a 70-kVp dual-low abdominal CTA protocol and to assess its ability to maintain diagnostic image quality while reducing both radiation dose and contrast medium usage. The results showed that using the 70-kVp protocol combined with 90% ClearInfinity blending and a low-flow, low-volume contrast protocol significantly reduced patient radiation exposure and contrast dose while still providing images that met diagnostic requirements. Compared with the conventional 100-kVp protocol reconstructed with 50% ClearView, radiation dose was reduced by 64.08% and contrast volume by 34.61%, whereas the SNR and CNR increased by 45.62% and 79.68%, respectively. This suggests that this strategy has the potential to achieve an optimized low-dose imaging paradigm with artificial intelligence reconstruction.

Although recent studies have examined low-dose CT protocols combined with artificial intelligence-based reconstruction (17), the clinical scenario and optimization focus of our study differ from those of previous work. Specifically, our investigation targeted abdominal CTA, in which diagnostic confidence depends primarily on vascular sharpness, intraluminal contrast stability, and noise-related effects on vessel delineation rather than on parenchymal enhancement. Moreover, instead of identifying a single optimal reconstruction parameter, we systematically characterized the relationship between ClearInfinity blending weight and both objective and subjective image quality across multiple reconstruction levels. This task-specific evidence provides practical guidance for CTA protocol optimization under aggressive dose- and contrast-reduction conditions.

In the objective evaluation, no significant differences were observed in CT attenuation values for any of the vessels in the five image groups; meanwhile, the SD decreased and SNR and CNR increased progressively from A1 to A4 as the ClearInfinity weight increased, indicating that the algorithm primarily improves image quality through noise reduction without altering CT attenuation values (18). Notably, although group A4 achieved the best objective metrics, its subjective scores did not differ significantly from those of groups A2, A3, and B. This may be because subjective image quality depends not only on noise level but also on texture characteristics, edge sharpness, and structural contrast (19). Higher reconstruction weights, although further reducing noise, may attenuate high-frequency components, leading to a blunting of vessel wall and small-branch edge gradients and generating an overly smoothed or “plastic-like” appearance that diminishes perceived fine detail (20). In addition, interreader variability in adaptation to the texture characteristics of deep learning-reconstructed images may contribute to scoring discrepancies. Adjustments in window width, window level, or vascular-focused display presets may further improve subjective perception, but this is speculative and remains to be verified.

In the comparison of different tube voltages, although group A was scanned at 70 kVp, CT attenuation values of major vessels did not differ significantly from those of the conventional 100-kVp group, indicating that low-kilovolt peak imaging can exploit the iodine K-edge absorption effect to provide sufficient signal compensation and maintain adequate vascular contrast (21). With the aid of deep learning reconstruction, this energy-dependent contrast enhancement effect is better preserved, allowing the SNR and CNR to remain high even at reduced dose levels. Similarly, a recent study on liver metastases demonstrated that high-weight deep learning reconstruction significantly improved lesion conspicuity and contour definition compared with iterative reconstruction and lower deep learning image reconstruction weights, which is consistent with our findings (22). Collectively, these results suggest that a low tube voltage combined with deep learning reconstruction can substantially reduce radiation dose while preserving image quality, which may hold considerable value for optimizing radiation protection strategies.

To leverage low-kilovolt peak imaging and improve iodine utilization efficiency, we employed a novel contrast medium injection protocol, with both the iodine dose and injection flow rate in group A being reduced by nearly 50% compared to those of group B. This adjustment was based on the enhancement of iodine contrast medium utilization efficiency with low-kilovolt peak imaging, as well as the findings from animal experiments by Lell et al. (23). In their study, adjusting the iodine delivery rate and injection time under low-kilovolt peak, low-dose, and low-flow rate conditions could yield time-density curves comparable to those obtained with a 120-kVp protocol and simultaneously reduce the contrast medium dose by approximately 50%. Building on this theoretical foundation, we incorporated the empirical formula from CTA imaging guidelines, in which for every 10-kVp reduction in tube voltage, the iodine dose can be reduced by 10% and the injection flow rate by 30%. For example, reducing the tube voltage from 120 to 70 kVp can reduce the iodine dose by 50% and the flow rate by 40%. Preliminary clinical studies also confirmed that under artificial intelligence reconstruction optimization conditions, reducing both the contrast medium dose and flow rate by 50% can still yield satisfactory image quality. Therefore, in our study, this optimized protocol was applied in group A to further examine the optimal weight of the ClearInfinity reconstruction algorithm.

This study involved several limitations that should be acknowledged. First, the sample size was relatively small, and patients with a BMI >28 kg/m² were not included; therefore, the performance of low-kilovolt peak scanning in overweight or obese populations requires further validation. Second, image quality assessment was mainly based on objective metrics and subjective scores, and no specific clinical diagnostic tasks or lesion-based analyses were included. Third, the control group employed a 100-kVp protocol with 50% ClearView iterative reconstruction, and a comparison arm consisting of a 70-kVp protocol and ClearView was not included; thus, the independent contributions of tube voltage reduction and deep learning reconstruction to image quality improvement could not be fully separated. A 70-kVp plus ClearView arm was not included because preliminary testing showed that excessive noise at 70 kVp with hybrid iterative reconstruction resulted in nondiagnostic images, making such a comparison clinically irrelevant. Nevertheless, the primary aim of this study was to verify a clinically feasible optimization strategy that achieves simultaneous reductions in radiation dose and contrast load, and the resulting design holds practical relevance. Future studies could adopt a factorial design to further clarify the independent effects of each parameter on image quality and incorporate sharpness metrics and diagnostic performance analyses to establish a more comprehensive evaluation framework.

Conclusions

Under a 70-kVp, low-tube voltage abdominal CTA protocol combined with a dual-low contrast strategy, ClearInfinity deep learning reconstruction substantially reduced both radiation dose and contrast medium usage while maintaining a high SNR and diagnostic image quality. Among the tested blending levels, 90% ClearInfinity provided the most favorable balance between objective image quality and clinical feasibility. This protocol may offer a practical and effective approach for safer, lower-dose, and high-quality abdominal CTA imaging.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TREND reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2589/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2025-1-2589/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-2025-1-2589/coif). G.W. is current employee of Neusoft Medical Systems Co., Ltd. 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 prospective single-center study was approved by the Ethics Committee of Shiyan Taihe Hospital (approval number: 2024KS15), and 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. Sun J, Li H, Yang L, Zhou Z, Li M, Peng Y. Application of 70 kVp in abdominal CT angiography to reduce both radiation and contrast dosage and improve patient comfort for children. J Xray Sci Technol 2021;29:813-21. [Crossref] [PubMed]
2. Cao CF, Ma KL, Shan H, Liu TF, Zhao SQ, Wan Y. Jun-Zhang, Wang HQ. CT Scans and Cancer Risks: A Systematic Review and Dose-response Meta-analysis. BMC Cancer 2022;22:1238. [Crossref] [PubMed]
3. Shin HJ, Chung YE, Lee YH, Choi JY, Park MS, Kim MJ, Kim KW. Radiation dose reduction via sinogram affirmed iterative reconstruction and automatic tube voltage modulation (CARE kV) in abdominal CT. Korean J Radiol 2013;14:886-93. [Crossref] [PubMed]
4. Ning P, Zhu S, Shi D, Guo Y, Sun M. X-ray dose reduction in abdominal computed tomography using advanced iterative reconstruction algorithms. PLoS One 2014;9:e92568. [Crossref] [PubMed]
5. Nam JG, Hong JH, Kim DS, Oh J, Goo JM. Deep learning reconstruction for contrast-enhanced CT of the upper abdomen: similar image quality with lower radiation dose in direct comparison with iterative reconstruction. Eur Radiol 2021;31:5533-43. [Crossref] [PubMed]
6. Noda Y, Iritani Y, Kawai N, Miyoshi T, Ishihara T, Hyodo F, Matsuo M. Deep learning image reconstruction for pancreatic low-dose computed tomography: comparison with hybrid iterative reconstruction. Abdom Radiol (NY) 2021;46:4238-44. [Crossref] [PubMed]
7. Park C, Choo KS, Jung Y, Jeong HS, Hwang JY, Yun MS. CT iterative vs deep learning reconstruction: comparison of noise and sharpness. Eur Radiol 2021;31:3156-64. [Crossref] [PubMed]
8. Brady SL, Trout AT, Somasundaram E, Anton CG, Li Y, Dillman JR. Improving Image Quality and Reducing Radiation Dose for Pediatric CT by Using Deep Learning Reconstruction. Radiology 2021;298:180-8. [Crossref] [PubMed]
9. Jensen CT, Liu X, Tamm EP, Chandler AG, Sun J, Morani AC, Javadi S, Wagner-Bartak NA. Image Quality Assessment of Abdominal CT by Use of New Deep Learning Image Reconstruction: Initial Experience. AJR Am J Roentgenol 2020;215:50-7. [Crossref] [PubMed]
10. Koetzier LR, Mastrodicasa D, Szczykutowicz TP, van der Werf NR, Wang AS, Sandfort V, van der Molen AJ, Fleischmann D, Willemink MJ. Deep Learning Image Reconstruction for CT: Technical Principles and Clinical Prospects. Radiology 2023;306:e221257. [Crossref] [PubMed]
11. Arndt C, Güttler F, Heinrich A, Bürckenmeyer F, Diamantis I, Teichgräber U. Deep Learning CT Image Reconstruction in Clinical Practice. Rofo 2021;193:252-61. [Crossref] [PubMed]
12. Chen H, Zhang Y, Zhang W, Liao P, Li K, Zhou J, Wang G. Low-dose CT via convolutional neural network. Biomed Opt Express 2017;8:679-94. [Crossref] [PubMed]
13. Qi K, Xu C, Yuan D, Zhang Y, Zhang M, Zhang W, Zhang J, You B, Gao J, Liu J. Feasibility of Ultra-low Radiation and Contrast Medium Dosage in Aortic CTA Using Deep Learning Reconstruction at 60 kVp: An Image Quality Assessment. Acad Radiol 2025;32:1506-16. [Crossref] [PubMed]
14. Wang SF, Li Z, Dai LH, Liu H, Zhang YQ, Huang Y, Zha XY, Zhang J, Wang QX. Image Quality Optimization in 60 kVp Head-Neck CTA: A Comparative Study of FBP, ClearView, and ClearInfinity Reconstruction Algorithms. Curr Med Sci 2025;45:1504-12. [Crossref] [PubMed]
15. Siegel MJ, Schmidt B, Bradley D, Suess C, Hildebolt C. Radiation dose and image quality in pediatric CT: effect of technical factors and phantom size and shape. Radiology 2004;233:515-22. [Crossref] [PubMed]
16. Lenfant M, Chevallier O, Comby PO, Secco G, Haioun K, Ricolfi F, Lemogne B, Loffroy R. Deep Learning Versus Iterative Reconstruction for CT Pulmonary Angiography in the Emergency Setting: Improved Image Quality and Reduced Radiation Dose. Diagnostics (Basel) 2020;10:558. [Crossref] [PubMed]
17. Ji MT, Wang RR, Wang Q, Li HS, Zhao YX. Feasibility study of "double-low" scanning protocol combined with artificial intelligence iterative reconstruction algorithm for abdominal computed tomography enhancement in patients with obesity. BMC Med Imaging 2025;25:276. [Crossref] [PubMed]
18. Tamura A, Mukaida E, Ota Y, Nakamura I, Arakita K, Yoshioka K. Deep learning reconstruction allows low-dose imaging while maintaining image quality: comparison of deep learning reconstruction and hybrid iterative reconstruction in contrast-enhanced abdominal CT. Quant Imaging Med Surg 2022;12:2977-84. [Crossref] [PubMed]
19. Singh R, Digumarthy SR, Muse VV, Kambadakone AR, Blake MA, Tabari A, Hoi Y, Akino N, Angel E, Madan R, Kalra MK. Image Quality and Lesion Detection on Deep Learning Reconstruction and Iterative Reconstruction of Submillisievert Chest and Abdominal CT. AJR Am J Roentgenol 2020;214:566-73. [Crossref] [PubMed]
20. Kim JH, Yoon HJ, Lee E, Kim I, Cha YK, Bak SH. Validation of Deep-Learning Image Reconstruction for Low-Dose Chest Computed Tomography Scan: Emphasis on Image Quality and Noise. Korean J Radiol 2021;22:131-8. [Crossref] [PubMed]
21. Ippolito D, Talei Franzesi C, Fior D, Bonaffini PA, Minutolo O, Sironi S. Low kV settings CT angiography (CTA) with low dose contrast medium volume protocol in the assessment of thoracic and abdominal aorta disease: a feasibility study. Br J Radiol 2015;88:20140140. [Crossref] [PubMed]
22. Kanan A, Pereira B, Hordonneau C, Cassagnes L, Pouget E, Tianhoun LA, Chauveau B, Magnin B. Deep learning CT reconstruction improves liver metastases detection. Insights Imaging 2024;15:167. [Crossref] [PubMed]
23. Lell MM, Jost G, Korporaal JG, Mahnken AH, Flohr TG, Uder M, Pietsch H. Optimizing contrast media injection protocols in state-of-the art computed tomographic angiography. Invest Radiol 2015;50:161-7. [Crossref] [PubMed]