Application of deep learning-based precise image reconstruction algorithm in non-contrast abdominal CT scanning

Introduction

Computed tomography (CT) technology plays a crucial role in medical imaging diagnosis, especially in emergency, intensive care, and rapid diagnostic settings, where its fast-scanning capability provides key support for clinical decision-making. In the evaluation of abdominal diseases, non-contrast CT is widely regarded as the preferred modality for assessing parenchymal organs such as the liver and pancreas due to its rapid and non-invasive nature (1). However, the high radiation dose associated with CT examinations remains a major concern in clinical practice. For patients undergoing repeated or long-term CT scans, cumulative radiation exposure significantly increases, thereby elevating the risk of radiation-induced malignancies (2,3). Consequently, reducing radiation dose while maintaining diagnostic image quality has become a key focus in the optimization of abdominal CT imaging.

Among current strategies for low-dose CT scanning, reducing tube voltage (4) and applying iterative reconstruction techniques (5,6) have proven effective in minimizing radiation exposure. Optimizing image reconstruction algorithms has emerged as another crucial approach (7). The filtered back projection (FBP) algorithm, one of the earliest CT reconstruction methods, offers relatively accurate noise texture but is prone to image distortion due to discretization (8) and is sensitive to external interference (9). In contrast, the iDose4 algorithm, as a representative of iterative reconstruction techniques, employs a unique noise reduction approach to enhance CT image quality by effectively suppressing noise and artifacts. Over the past decade, iDose4 has been widely adopted in abdominal CT imaging (10-12), with studies confirming its ability to reduce radiation dose compared to FBP while improving diagnostic confidence (12-14). With the rapid development of deep learning, image reconstruction algorithms based on this technology have become a focal point in low-dose CT research (15). Compared to traditional iterative reconstruction methods, deep learning-based approaches demonstrate superior performance in detecting small lesions (16-18) and can significantly enhance image quality under low-dose conditions through model training on large datasets (19). It is important to note, however, that FBP algorithm remains a standard and widely used reconstruction technique in clinical CT and is effective for visualizing a wide range of lesions.

The precise image (PI) reconstruction algorithm is a deep learning-based method trained on a large volume of high-quality CT data. It can generate high-quality images under low-dose conditions by significantly improving the signal-to-noise ratio (SNR) and reducing noise levels (20). The CT scans were performed using a Philips Incisive CT scanner (Philips Healthcare, Amsterdam, the Netherlands) and the following image reconstruction algorithms provided by Philips were evaluated: (I) FBP: a conventional analytical method serving as the baseline for comparison. (II) iDose4: an iterative reconstruction algorithm designed to reduce noise and artifacts while preserving image texture. (III) PI standard: a deep learning-based algorithm trained on large datasets to enhance SNR and spatial resolution under low-dose conditions. (IV) PI-smooth: a reconstruction mode of the PI algorithm that provides stronger smoothing and noise suppression compared to the PI-standard mode. This study aims to investigate the clinical value of the PI reconstruction algorithm in non-contrast abdominal CT scans under low-dose conditions by comparing image quality across different radiation doses, providing scientific evidence to support low-dose CT protocols in clinical practice. We present this article in accordance with the STROBE reporting checklist (available at https://amj.amegroups.com/article/view/10.21037/amj-25-57/rc).

Methods

Ethical approval

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Ethical approval was not required for this retrospective study by the Huazhong University of Science and Technology Institutional Review Board/Research Ethics Committee. This determination was made because the study involved the analysis of existing, fully anonymised imaging and clinical data collected as part of routine clinical care. All patient identifiers were permanently removed prior to analysis, rendering the data non-identifiable. Individual consent for this retrospective analysis was waived.

General information

A total of 120 patients who underwent non-contrast abdominal CT scans at Union Hospital, Tongji Medical College, Huazhong University of Science and Technology between September and December 2024 were retrospectively collected and divided into three groups: Groups A, B, and C, with 40 patients in each group. The clinical information of the patient is shown in Table 1. Inclusion criteria: all patients presented with abdominal discomfort and were referred for CT examination for further diagnostic evaluation. Exclusion criteria: cases with significant motion artifacts that hindered image evaluation.

Image acquisition and reconstruction

All patients underwent CT examinations using a Philips Incisive (Philips Healthcare) scanner. Prior to scanning, patients were instructed to remain still and follow voice prompts for breath-holding during the procedure. Scanning position: patients were placed in a supine position with head-first orientation on the scanning table, with both arms raised above the head. Scan range: the upper boundary included the superior margin of the diaphragm, while the lower boundary was determined based on the clinical indications specified in the examination request. Scanning parameters: the tube voltage was set to 120 kVp, and tube current was controlled using an automatic modulation technique. The X-ray tube rotation speed was 0.5 s per rotation, with a pitch of 0.8, and collimation of 64×0.625 mm. As determined by a pilot study, the DoseRight Index (DRI) was set as follows for the three groups: Group A: DRI =18, Group B: DRI =20, and Group C: DRI =22.

Post-processing and data analysis

All cases were reconstructed using four methods: FBP, iDose4, PI standard, and PI smooth. The PI algorithm is a deep learning-based reconstruction solution trained on a high-quality dataset of clinical CT images. Its architecture is designed to learn the mapping from low-dose raw data to high-quality images, effectively suppressing noise and artifacts while preserving structural details and texture. For all four reconstruction algorithms, the slice thickness was 1 mm with an inter-slice interval of 0.5 mm. Adaptive filtering was applied, with a window width of 210 Hounsfield Units (HU) and a window level of 35 HU. All reconstructed images were transferred to a Philips IntelliSpace Portal workstation, version 12.1, for quantitative analysis.

Image analysis

Objective image quality metrics included CT value, image noise [standard deviation (SD)], SNR, and contrast-to-noise ratio (CNR). The regions of interest (ROIs) were placed by an experienced radiologist with 5 years of experience in abdominal imaging to ensure consistency and clinical relevance. For each case, the ROIs for the same type of tissue were consistent in shape, size, and location, while avoiding obvious lesions, air, pancreatic duct, and calcifications. Measurements were taken at three adjacent axial slices for each case, and the average of the three values was used for analysis. Mean CT values and SDs were measured for the liver, pancreas, abdominal aorta. The SD of the erector spinae was used as the background noise of the image, and the corresponding SNR and CNR values of the target tissues were calculated. All values were reported as absolute values. The formulas were as follows:

SNRliver=SIliverSDliver,CNRliver=SIliver−SIpsoasmusclesSDpsoasmuscles[1]

SNRpancreas=SIpancreasSDpancreas,CNRpancreas=SIpancreas−SIpsoasmusclesSDpsoasmuscles[2]

SNRaorta=SIaortaSDaorta,CNRaorta=SIaorta−SIpsoasmusclesSDpsoasmuscles[3]

The dose saving percentage for the low-dose group (Group A or B) relative to the standard-dose group (Group C) was calculated using the following formula:

Dosesaving=(1−SSDElow-dosegroupSSDEgroupC)×100%[4]

SI represents the CT value of the ROI at the corresponding anatomical location. Intra-group comparisons were performed among the four reconstruction methods (FBP, iDose4, PI standard, and PI smooth) within each of Groups A, B, and C. Additionally, inter-group comparisons were conducted for the PI reconstruction images (PI standard and PI smooth) across Groups A, B, and C.

Subjective image quality was evaluated using a double-blind method by two radiologists with more than 5 years of experience in diagnostic imaging. A 5-point scoring system was used: a score of 5 indicated clear visualization of anatomical structures and fine details, no significant noise, and excellent lesion depiction; a score of 4 indicated relatively clear anatomical structures, mild image noise, and good lesion depiction; a score of 3 indicated that the visualization of anatomical structures and lesions met basic diagnostic requirements, with slightly increased noise; a score of 2 indicated unclear anatomical and lesion visualization, significant image noise, and insufficient detail for diagnosis; and a score of 1 indicated blurred anatomical structures and lesions, severe noise, and images unsuitable for diagnosis. The average score from the two radiologists was used for inter-group analysis. The inter-observer agreement between the two radiologists was quantitatively assessed using Cohen’s kappa (κ) statistic. To resolve any discrepancies in scoring, the following procedure was implemented: Initially, both radiologists evaluated all images independently. In cases of divergent scores, a consensus reading was conducted to discuss and reconcile the differences. If an immediate consensus could not be reached, a third senior radiologist with over 10 years of experience was consulted to make the final determination.

Radiation dose parameters automatically generated by the CT scanner were recorded, including CT dose volume index (CTDI_vol), dose length product (DLP), and effective radiation dose.

Statistical analysis

Statistical analysis was performed using GraphPad Prism 10.1.2 software (GraphPad Software, San Diego, CA, USA). All continuous variables were expressed as mean ± SD. The Friedman test was employed for within-group comparisons to assess differences among the four reconstruction algorithms at each identical dose level. For comparisons of the same reconstruction algorithm across the different dose level groups (A, B, and C), the Kruskal-Wallis test was utilized. When a statistically significant difference was identified (P<0.05), post-hoc pairwise comparisons were performed using the Wilcoxon signed-rank test with a Bonferroni correction for multiple comparisons.

Results

Basic characteristics

Radiation dose distributions for Groups A, B, and C are presented in Table 1. The dose reduction percentages for Groups A and B compared to Group C were 40.14% and 24.95% for CTDI_vol, 34.23% and 18.02% for DLP, and 38.17% and 25.22% for size-specific dose estimate (SSDE).

The baseline clinical characteristics of the patients in the three groups are summarized in Table 2. There were no statistically significant differences in these characteristics among Groups A, B, and C (all P>0.05).

Objective image quality

For Group A, the PI smooth algorithm demonstrated significantly lower noise (SD) and higher SNR/CNR in the liver, pancreas, and abdominal aorta compared to PI standard, iDose4, and FBP (all P<0.05). The most pronounced improvements were observed relative to FBP in the liver and to iDose4/FBP in the pancreas (Figure 1). CT values showed statistically significant differences among reconstructions in the liver (P<0.001) and pancreas (P=0.04), but not in the abdominal aorta (P>0.05) (see Table 3).

Figure 1 Male, 40 years old, BMI =22.15 kg/m², DRI =18, IPMN of the pancreas duct. (A) FBP reconstruction exhibits high image noise and coarse noise texture, which obscure subtle structural details. (B) iDose4 reconstruction demonstrates moderate noise reduction, though residual noise remains visible around the duct. (C) PI standard reconstruction achieves a favorable balance between noise suppression and structural preservation. (D) PI smooth reconstruction provides the most effective noise reduction and the highest overall image clarity, optimally visualizing the pancreatic duct and surrounding parenchyma with superior signal-to-noise ratio. Arrows indicate the IPMN of the pancreatic duct. BMI, body mass index; DRI, DoseRight Index; FBP, filtered back projection; IPMN, intraductal papillary mucinous neoplasm; PI, precise imaging.

The within-group analysis for Group B revealed a qualitatively similar pattern of results to that of Group A, with consistent directional improvements observed for the PI smooth algorithm (see Figure 2 and Table 4).

Figure 2 Male, 26 years old, BMI =23.43 kg/m², DRI =20, calcified lesion in the right lobe of the liver. (A) FBP reconstruction shows significant noise and degraded contrast, impairing clear delineation of the calcified focus. (B) iDose4 reconstruction reduces noise moderately but retains granular texture adjacent to the lesion. (C) PI standard reconstruction improves noise suppression and edge definition of the calcification. (D) PI smooth reconstruction provides the clearest depiction of the calcified lesion, with minimal background noise and optimal visual clarity for confident detection. Arrows indicate the intrahepatic calcified lesion. BMI, body mass index; DRI, DoseRight Index; FBP, filtered back projection; PI, precise imaging.

In Group C, the trends for SD, SNR, and CNR aligned with those in Groups A and B. Regarding CT values, significant differences were identified in the liver and abdominal aorta (both P<0.001), whereas the difference for pancreas was not statistically significant (P=0.30) (see Table 5 and Figure 3).

Figure 3 Male, 29 years old, BMI =23.89 kg/m², DRI =22, low-density hepatic nodule. (A) FBP reconstruction displays pronounced noise and suboptimal contrast, limiting clear visualization of the nodule margins. (B) iDose4 reconstruction offers partial noise reduction, though the nodule boundary remains partially obscured by residual noise. (C) PI standard reconstruction achieves balanced noise control and improved delineation of the nodule’s periphery. (D) PI smooth reconstruction provides the most homogeneous background and superior contrast resolution, optimally defining the low-density nodule with enhanced diagnostic confidence. Arrows indicate the location of the low-density hepatic nodule. BMI, body mass index; DRI, DoseRight Index; FBP, filtered back projection; PI, precise imaging.

The PI smooth of all three groups (A, B, and C) exhibited good objective image quality. The results of the comparison of objective image quality among the three groups of PI smooth were as follows (see Table 6): CT values of the liver, pancreas, and abdominal aorta showed no statistically significant differences among the three dose groups (all P>0.05).

In contrast, image noise characteristics exhibited significant dose-dependent variations. SD values differed significantly across groups for all evaluated organs (liver, pancreas, and abdominal aorta; all P<0.05), demonstrating a general trend of increasing noise with decreasing radiation dose. However, pairwise comparisons revealed that Group B (moderate dose) showed no significant difference in liver SD compared to Group A (P=0.60), nor in pancreatic SD compared to Group C (P=0.18).

SNR analysis showed statistically significant differences between Group A (low dose) and Group C (standard dose). Notably, Group B demonstrated comparable SNR to Group C in both the liver (P=0.95) and pancreas (P=0.54).

For CNR, no statistically significant differences were observed among the three dose groups for any of the evaluated organs (all P>0.05).

Subjective image quality

In all three groups (A–C), subjective scores declined across the reconstruction methods in the order of PI smooth, PI standard, iDose4, and FBP (Figure 4A-4C). A1 and B1 had significantly higher scores than A3/A4 and B3/B4 (all P<0.001). In contrast, there were no significant differences between the scores of A1 and A2 (P=0.92), or between B1 and B2 (P=0.65). In Group C, C1 was significantly higher than C2 (P=0.03) and markedly higher than C3/C4 (P<0.001). The subjective image quality scores for the PI smooth mode in Groups A, B, and C were 4.65±0.53, 4.72±0.45, and 4.87±0.33, respectively. The difference between Groups A and C reached statistical significance (P=0.03), while no significant differences were found between Groups A and B or between Groups B and C (Figure 4D). Throughout this analysis, the inter-observer agreement between the two radiologists was substantial (Cohen’s κ =0.77).

Figure 4 Subjective scores of images for each reconstruction algorithm in Groups A, B, and C. (A) Group A. A1, PI smooth; A2, PI standard; A3, iDose4; A4, FBP. (B) Group B. B1, PI smooth; B2, PI standard; B3, iDose4; B4, FBP. (C) Group C. C1, PI smooth; C2, PI standard; C3, iDose4; C4, FBP. (D) Subjective image quality comparison of PI smooth Modes. A1, Group A; B1, Group B; C1, Group C. *, P<0.05; ****, P<0.0001; ns, not significant. FBP, filtered back projection; PI, precise imaging.

Discussion

This study systematically evaluated the image quality and subjective scores of abdominal organs using four reconstruction algorithms: PI smooth, PI standard, iDose4, and FBP, under three different DRI levels (Groups A, B, and C). The main findings are as follows: (I) intra-group comparisons (Groups A, B, and C): although statistically significant differences in CT values were observed for some organs, the absolute mean differences were minimal and considered clinically negligible. This indicates that tissue density measurements remained consistent across reconstruction algorithms for clinical purposes. However, in terms of SD, SNR, and CNR, the PI smooth mode consistently yielded the lowest SD and highest SNR and CNR values for all organs (all P<0.05), with significantly better performance than the other three algorithms. Subjective scores for PI smooth images were also higher than those of the other reconstruction techniques. A progressive decline in subjective image quality scores was observed across the four reconstruction methods in Group A (A₁–A₄). The difference between A₁ and A₂ was not statistically significant (P=0.92); however, A₁ showed significantly higher scores compared to A₃ and A₄ (both P<0.001). A similar downward trend in subjective ratings was also observed in Groups B and C. (II) Inter-group comparisons: for CNR, no statistically significant differences were observed among the three groups for any of the evaluated organs (liver: P=0.70; pancreas: P=0.77; abdominal aorta: P=0.051; all P>0.05). Regarding SNR, statistically significant differences were found across all three organs (liver: P=0.02; pancreas: P=0.009; abdominal aorta: P<0.001).

As for subjective image quality scores, significant differences were observed among the three groups using PI smooth reconstruction (Group A: 4.65±0.53; Group B: 4.72±0.45; Group C: 4.87±0.33). Specifically, the difference between Groups A and C reached statistical significance (P=0.03), while no significant differences were found between Groups A and B or between Groups B and C.

This study demonstrates that the PI smooth algorithm significantly reduces image noise and enhances SNR and CNR across different DRI levels, with overall image quality superior to PI standard, iDose4, and conventional FBP. Intra-group comparisons in Groups A, B, and C consistently demonstrated that PI smooth outperformed the other reconstruction algorithms. Additionally, for the CNR, no statistically significant differences were observed among the three groups for any of the evaluated organs (all P>0.05). It is noteworthy, however, that the highest CNR values for both the pancreas and the abdominal aorta were recorded in Group B. Subjective image quality scores showed no statistically significant differences between Group B and the other two groups. We suggest that low-dose abdominal CT scans (Group B, DRI =20) combined with PI smooth reconstruction can effectively reduce radiation exposure while maintaining diagnostic image quality.

The PI algorithm employs an artificial neural network architecture and reconstructs images through deep learning based on large volumes of CT imaging data. Depending on the reconstruction strength, the PI algorithm offers five modes: sharper, sharp, standard, smooth, and smoother. Based on preliminary experiments and previous literature (21), this study selected the standard and smooth PI reconstruction modes, with conventional FBP and the iDose4 iterative reconstruction algorithm used as controls. An intra-group comparison was performed for the four reconstruction algorithms in Group C. Regarding SNR and CNR, the values for the liver, pancreas, and abdominal aorta increased progressively from the iDose4 iterative reconstruction algorithm to the PI standard mode and then to the PI smooth mode. This indicates that, in terms of objective image quality metrics, the PI algorithm provides superior image quality in routine-dose abdominal CT scans compared to both FBP and iDose4 algorithms, with the smooth mode demonstrating the best performance. These findings are consistent with those reported by Yang et al. (22) and Sun et al. (23) in their studies on abdominal organ CT imaging, as well as similar results reported by Heinrich et al. (24) in aortic CT Angiography. The SNR and CNR values for the organs included in this study showed the same trend in Groups A and B, further suggesting that the PI algorithm outperforms FBP and iDose4 reconstruction under different radiation dose levels in abdominal CT scans. Regarding SD, under the same DRI setting, SD values for the liver, pancreas, and abdominal aorta decreased sequentially from FBP to iDose4, PI standard, and PI smooth, indicating progressively enhanced noise suppression capabilities across the four reconstruction algorithms. This trend is in line with previous research findings (25,26).

The results of this study demonstrated that under identical dose index conditions, there were no significant differences in CT values of the same organ across FBP, iDose4, and PI reconstructions. The meta-analysis conducted by multiple papers on abdominal deep learning reconstruction algorithms by van Stiphout et al. (27) also supports that there, although statistically significant differences in CT values were observed among some reconstruction algorithms, the absolute mean differences were minimal and considered clinically negligible. Similar conclusions have also been reported in studies on brain CT by Lei et al. (28) and Alagic et al. (29), indicating consistency in CT value reconstruction across these algorithms.

Despite the use of different dose indices in Groups A, B, and C, there were no statistically significant differences in the CT values of the liver, pancreas, and abdominal aorta among the three groups (P>0.05). This suggests that within a certain dose range, reducing radiation exposure does not significantly affect the CT attenuation values of major anatomical structures.

However, a different trend was observed in terms of image noise (SD values). Significant differences in SD values of the liver, pancreas, and abdominal aorta were found between Groups A, B, and C (P<0.05), This indicates that as radiation dose decreases, image noise correspondingly increases. Similar findings were reported by Yang et al. (30) who demonstrated that image noise exhibits a strong power law relationship with radiation dose across all reconstruction algorithms. However, no significant difference was observed between Groups B and A in the liver (P=0.60), nor between Groups B and C in the pancreas (P=0.18). This suggests that the dose setting for Group B represents a favorable trade-off between image noise and radiation dose in these specific contexts.

In terms of SNR, statistically significant differences were observed between Groups A and C. However, no significant differences were found between Groups B and C in both the liver (P_20–22=0.95) and pancreas (P_20–22=0.54), suggesting that that when the radiation dose is increased to a moderate level (Group B), the image quality may already meet clinical diagnostic standards, and further increasing the dose (Group C) yields limited additional benefit. However, the significantly lower SNR in the low-dose group (Group A) compared to the standard-dose group implies that excessively low doses may compromise image signal quality. A study by Kawashima et al. (31) highlighted that deep learning image reconstruction (DLIR) possesses greater dose optimization potential than conventional iterative reconstruction algorithms, based on system performance function and dose reduction potential analyses.

Regarding CNR, no statistically significant differences were observed among the three groups (P>0.05). This aligns with findings by Lyu et al. (32), who demonstrated that deep learning reconstruction techniques could maintain lesion detectability comparable to full-dose scans even with a 34% radiation dose reduction, confirming that iterative reconstruction algorithms can preserve tissue contrast characteristics under moderate dose reduction. However, in contrast to our findings, a study by Bie et al. (33) on renal CT reported significant differences in SD, SNR, and CNR across different dose levels with DLIR images. The discrepancy may stem from methodological differences: while Bie et al.’s study varied the radiation dose by adjusting kilovoltage (kV), our study altered the dose by modifying the DRI, which primarily reflects changes in tube current.

The results of this study indicate that although Groups A, B, and C demonstrated significant differences in radiation dose parameters—CTDI_vol values of 6.74, 8.45, and 11.26 mGy; DLP values of 303.1, 377.8, and 460.9 mGy·cm; and SSDE values of 9.07, 10.97, and 14.67 mGy, respectively—there were only small differences in subjective image quality scores among the three groups, with a statistically significant but clinically minor difference between Groups A and C (P=0.03), and no significant differences between Groups A and B or between Groups B and C. This suggests that moderate radiation dose reduction is feasible without compromising diagnostic image quality. Specifically, compared to Group C (the highest dose group), Group A (the lowest dose group) achieved approximately a 40% reduction in CTDI_vol, a 34% reduction in DLP, and a 38% reduction in SSDE. Importantly, these reductions did not negatively impact the subjective assessment of image quality. This finding is consistent with the study by Chu et al. (34), which also reported that in abdominal CT examinations, a dose reduction of 30–40% can still yield diagnostically acceptable image quality.

This study had several limitations. First, this study is a retrospective, single-center analysis with a relatively small sample size, which may limit the generalizability of the findings to broader patient populations or other clinical settings. Second, the study lacked pathological analysis, formal disease classification, and quantitative assessment of lesion characteristics. Future work will integrate these elements to better evaluate the diagnostic utility of the algorithms.

Conclusions

In conclusion, this study demonstrates that the Philips PI smooth reconstruction algorithm consistently provides superior objective and subjective image quality compared to conventional techniques across all tested radiation dose levels in non-contrast abdominal CT. Reducing the radiation dose (DRI =20) combined with the PI smooth reconstruction algorithm can lower patient radiation exposure while maintaining diagnostic image quality, providing a feasible approach for low-dose abdominal CT examinations.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://amj.amegroups.com/article/view/10.21037/amj-25-57/rc

Data Sharing Statement: Available at https://amj.amegroups.com/article/view/10.21037/amj-25-57/dss

Peer Review File: Available at https://amj.amegroups.com/article/view/10.21037/amj-25-57/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://amj.amegroups.com/article/view/10.21037/amj-25-57/coif). S.G. is employed by Philips Healthcare, the company provided no financial or material support for this work. 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. Ethical approval was not required for this retrospective study by the Huazhong University of Science and Technology Institutional Review Board/Research Ethics Committee. This determination was made because the study involved the analysis of existing, fully anonymised imaging and clinical data collected as part of routine clinical care. All patient identifiers were permanently removed prior to analysis, rendering the data non-identifiable. Individual consent for this retrospective analysis was waived.

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. Brenner DS, Fong TC. Approach to Abdominal Imaging in 2022. Emerg Med Clin North Am 2021;39:745-67. [Crossref] [PubMed]
2. Buchberger B, Scholl K, Krabbe L, et al. Radiation exposure by medical X-ray applications. Ger Med Sci 2022;20:Doc06. [Crossref] [PubMed]
3. Brenner DJ, Hall EJ. Computed tomography--an increasing source of radiation exposure. N Engl J Med 2007;357:2277-84. [Crossref] [PubMed]
4. Miyoshi K, Tanabe M, Ihara K, et al. Dual-Source Contrast-Enhanced Multiphasic CT of the Liver Using Low Voltage (70 kVp): Feasibility of a Reduced Radiation Dose and a 50% of Contrast Dose. Tomography 2023;9:1568-76. [Crossref] [PubMed]
5. McCollough CH, Bartley AC, Carter RE, et al. Low-dose CT for the detection and classification of metastatic liver lesions: Results of the 2016 Low Dose CT Grand Challenge. Med Phys 2017;44:e339-52. [Crossref] [PubMed]
6. Rampado O, Depaoli A, Marchisio F, et al. Effects of different levels of CT iterative reconstruction on low-contrast detectability and radiation dose in patients of different sizes: an anthropomorphic phantom study. Radiol Med 2021;126:55-62. [Crossref] [PubMed]
7. Patino M, Fuentes JM, Singh S, et al. Iterative Reconstruction Techniques in Abdominopelvic CT: Technical Concepts and Clinical Implementation. AJR Am J Roentgenol 2015;205:W19-31. [Crossref] [PubMed]
8. Shi H, Luo S. A novel scheme to design the filter for CT reconstruction using FBP algorithm. Biomed Eng Online 2013;12:50. [Crossref] [PubMed]
9. Hong L, Lin L, Chen J, et al. CT Image Features of the FBP Reconstruction Algorithm in the Evaluation of Fasting Blood Sugar Level of Diabetic Pulmonary Tuberculosis Patients and Early Diet Nursing. Comput Math Methods Med 2021;2021:1101930. Retracted Publication. [Crossref] [PubMed]
10. Lim WH, Choi YH, Park JE, et al. Application of Vendor-Neutral Iterative Reconstruction Technique to Pediatric Abdominal Computed Tomography. Korean J Radiol 2019;20:1358-67. [Crossref] [PubMed]
11. Arapakis I, Efstathopoulos E, Tsitsia V, et al. Using “iDose4” iterative reconstruction algorithm in adults’ chest-abdomen-pelvis CT examinations: effect on image quality in relation to patient radiation exposure. Br J Radiol 2014;87:20130613. [Crossref] [PubMed]
12. Ploussi A, Alexopoulou E, Economopoulos N, et al. Patient radiation exposure and image quality evaluation with the use of iDose4 iterative reconstruction algorithm in chest-abdomen-pelvis CT examinations. Radiat Prot Dosimetry 2014;158:399-405. [Crossref] [PubMed]
13. Khawaja RD, Singh S, Gilman M, et al. Computed tomography (CT) of the chest at less than 1 mSv: an ongoing prospective clinical trial of chest CT at submillisievert radiation doses with iterative model image reconstruction and iDose4 technique. J Comput Assist Tomogr 2014;38:613-9. [Crossref] [PubMed]
14. Kordolaimi SD, Argentos S, Mademli M, et al. Effect of iDose4 iterative reconstruction algorithm on image quality and radiation exposure in prospective and retrospective electrocardiographically gated coronary computed tomographic angiography. J Comput Assist Tomogr 2014;38:956-62. [Crossref] [PubMed]
15. Caruso D, De Santis D, Del Gaudio A, et al. Low-dose liver CT: image quality and diagnostic accuracy of deep learning image reconstruction algorithm. Eur Radiol 2024;34:2384-93. [Crossref] [PubMed]
16. Noda Y, Kaga T, Kawai N, et al. Low-dose whole-body CT using deep learning image reconstruction: image quality and lesion detection. Br J Radiol 2021;94:20201329. [Crossref] [PubMed]
17. Cao J, Mroueh N, Mercaldo N, et al. Detectability of Hypoattenuating Liver Lesions with Deep Learning CT Reconstruction: A Phantom and Patient Study. Radiology 2024;313:e232749. [Crossref] [PubMed]
18. Cao L, Liu X, Qu T, et al. Improving spatial resolution and diagnostic confidence with thinner slice and deep learning image reconstruction in contrast-enhanced abdominal CT. Eur Radiol 2023;33:1603-11. [Crossref] [PubMed]
19. Kaga T, Noda Y, Mori T, et al. Unenhanced abdominal low-dose CT reconstructed with deep learning-based image reconstruction: image quality and anatomical structure depiction. Jpn J Radiol 2022;40:703-11. [Crossref] [PubMed]
20. Chandran M O, Pendem S. Comparison of image quality between Deep learning image reconstruction and Iterative reconstruction technique for CT Brain- a pilot study. F1000Res 2024;13:691. [Crossref] [PubMed]
21. Greffier J, Durand Q, Frandon J, et al. Improved image quality and dose reduction in abdominal CT with deep-learning reconstruction algorithm: a phantom study. Eur Radiol 2023;33:699-710. [Crossref] [PubMed]
22. Yang C, Wang W, Cui D, et al. Deep learning image reconstruction algorithms in low-dose radiation abdominal computed tomography: assessment of image quality and lesion diagnostic confidence. Quant Imaging Med Surg 2023;13:3161-73. [Crossref] [PubMed]
23. Sun Y, Sun DZ, Han CL. An Evaluation Analysis for Computed Tomography Image Quality of Primary Liver Cancer Lesions Based on Deep Learning Image Reconstruction. Curr Med Imaging 2024;20:e15734056261849. [Crossref] [PubMed]
24. Heinrich A, Streckenbach F, Beller E, et al. Deep Learning-Based Image Reconstruction for CT Angiography of the Aorta. Diagnostics (Basel) 2021;11:2037. [Crossref] [PubMed]
25. Watanabe S, Sakaguchi K, Kitaguchi S, et al. Pulmonary nodule volumetric accuracy of a deep learning-based reconstruction algorithm in low-dose computed tomography: A phantom study. Phys Med 2022;104:1-9. [Crossref] [PubMed]
26. Delabie A, Bouzerar R, Pichois R, et al. Diagnostic performance and image quality of deep learning image reconstruction (DLIR) on unenhanced low-dose abdominal CT for urolithiasis. Acta Radiol 2022;63:1283-92. [Crossref] [PubMed]
27. van Stiphout JA, Driessen J, Koetzier LR, et al. The effect of deep learning reconstruction on abdominal CT densitometry and image quality: a systematic review and meta-analysis. Eur Radiol 2022;32:2921-9. [Crossref] [PubMed]
28. Lei L, Zhou Y, Guo X, et al. The value of a deep learning image reconstruction algorithm in whole-brain computed tomography perfusion in patients with acute ischemic stroke. Quant Imaging Med Surg 2023;13:8173-89. [Crossref] [PubMed]
29. Alagic Z, Diaz Cardenas J, Halldorsson K, et al. Deep learning versus iterative image reconstruction algorithm for head CT in trauma. Emerg Radiol 2022;29:339-52. [Crossref] [PubMed]
30. Yang K, Cao J, Pisuchpen N, et al. CT image quality evaluation in the age of deep learning: trade-off between functionality and fidelity. Eur Radiol 2023;33:2439-49. [Crossref] [PubMed]
31. Kawashima H, Ichikawa K, Takata T, et al. Performance of clinically available deep learning image reconstruction in computed tomography: a phantom study. J Med Imaging (Bellingham) 2020;7:063503. [Crossref] [PubMed]
32. Lyu P, Li Z, Chen Y, et al. Deep learning reconstruction CT for liver metastases: low-dose dual-energy vs standard-dose single-energy. Eur Radiol 2024;34:28-38. [Crossref] [PubMed]
33. Bie Y, Yang S, Li X, et al. Impact of deep learning-based image reconstruction on image quality and lesion visibility in renal computed tomography at different doses. Quant Imaging Med Surg 2023;13:2197-207. [Crossref] [PubMed]
34. Chu B, Gan L, Shen Y, et al. A Deep Learning Image Reconstruction Algorithm for Improving Image Quality and Hepatic Lesion Detectability in Abdominal Dual-Energy Computed Tomography: Preliminary Results. J Digit Imaging 2023;36:2347-55. [Crossref] [PubMed]