Application of a New Deep Learning Image Reconstruction (DLIR) Algorithm in Pediatric Orbital horizontal level Ultra-Low Dose CT scan: A Pilot Study Based on Image Quality

Application of a New Deep Learning Image Reconstruction (DLIR) Algorithm in Pediatric Orbital horizontal level Ultra-Low Dose CT scan: A Pilot Study Based on Image Quality | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Application of a New Deep Learning Image Reconstruction (DLIR) Algorithm in Pediatric Orbital horizontal level Ultra-Low Dose CT scan: A Pilot Study Based on Image Quality Yang Li, Feng-Xian Wang, Xia Liu, Shan-shan Sun, Xiu-feng Song This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6243177/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Purpose: To evaluate the performance of a new deep learning image reconstruction (DLIR) algorithm in pediatric orbital ultra-low-dose CT by sinus CT for improving image quality and reducing radiation dose. Methods: A retrospective analysis of 50 children who underwent paranasal sinus or orbital CT examination at our hospital was performed. ultra-low-dose CT of the nasal sinus was used to simulate orbital scanning as the experimental group and compared with the control group of general-low dose orbital CT. Two groups scanning used 100 kVp tube voltage, automatic tube current (SmartmA, 60-250 mA), slice thickness 1.25 mm, interslice spacing 1.25 mm, and pitch 0.992:1 with rotation time 1.00 s. The experimental group configured a Noise Index of 22 and a pre-ASiR-V weight of 70%, while the control group set a Noise Index of 8 and a pre-ASiR-V weight of 30%. The radiation dose, image noise (SD), signal-to-noise ratio (SNR), contrast noise ratio (CNR), and subjective scores of both groups were compared. The CT dose index (CTDI) and dose-length product (DLP) values were documented, and the effective dose (ED) was subsequently calculated. The results indicated that the experimental group experienced reductions of 83.92% in DLP and 83.17% in ED compared to the control group.Compared between DL-H (the optimal set of DLIR) and ASiR-V 70% (the control group), the image noise was reduced by 41.6% (4.17±0.86 HU vs. 7.14±1.19 HU), and the CNR was improved by 39.3% (26.60±5.61 vs. 16.14±2.48). In the qualitative image assessment, DL-H obtained the highest score. Conclusion: Combined with the comparison results of quantitative and qualitative image assessment, DL-H under the ultra-low-dose scan was significantly better than ASiR-V 70% under the general-low dose scan, and ASiR-V 70% was better than DL-M and DL-L. We can infer that the use of DL-H may be able to save a significant amount of radiation dose while achieving the current image quality. Deep learning iterative reconstruction orbital CT children Figures Figure 1 Figure 2 Introduction As computed tomography (CT) technology rapidly evolves, the use of low-dose CT is becoming increasingly prevalent to minimize radiation exposure in patients. Reducing the CT radiation dose in children has been a hot spot of research in recent years [ 1 – 3 ]. CT is an excellent imaging modality for displaying bony structures, and it is intuitive and fast, and it is a significant examination method for the paranasal sinuses and orbits of children [ 4 – 5 ]. The paranasal sinuses are formed in air-bearing bony cavities, and the sinus wall is thin and has a natural contrast advantage. CT in children's paranasal sinuses is mostly used to evaluate paranasal sinus inflammatory lesions. The requirements for image noise and sharpness are relatively low, which provides us with a very low-dose scanning opportunity. In addition, the CT scan range of the paranasal sinuses will inevitably include the orbit, and the eyeball is extremely sensitive to radiation. However, orbital CT in children usually involves trauma diagnosis and requires relatively high image noise and sharpness. It is the relatively single disease spectrum of children's sinus and orbital CT scans but the same scan range that provides our research with unique conditions for the comparative evaluation of image quality under ultra-low-dose and orbital levels. Filtered back projection (FBP) stands as the industry standard for CT image reconstruction; however, this technology is constrained by considerable noise and artifacts [ 6 ]. Moreover, it fails to deliver exceptional diagnostic images when using lower radiation doses. In recent years, iterative reconstruction (IR) technology has advanced rapidly, prompting various CT manufacturers to incorporate sophisticated IR algorithms aimed at lowering radiation exposure. Adaptive statistical iterative reconstruction V (ASiR-V) is a cutting-edge technology that has garnered attention for its ability to produce high-quality diagnostic images while substantially reducing radiation doses [ 7 – 8 ]. Nevertheless, its application is significantly hindered due to the overly smooth and unnatural of the images. Deep learning image reconstruction (DLIR) represents a novel approach to image reconstruction. This method employs high-dose FBP datasets and deep neural network (DNN) models to further diminish noise and suppress artifacts [ 9 ]. The schematic diagram of the DLIR algorithm is depicted in Fig. 1. Although DLIR technology has only recently obtained FDA certification and has not been widely adopted, there are scant reports on its use in pediatric CT scans [ 10 – 12 ].There are no reports about the application of orbital CT scans in children. The objective of this study is to evaluate the effects of Deep Learning Iterative Reconstruction (DLIR) and Adaptive Statistical Iterative Reconstruction-V (ASiR-V) on the image quality of computed tomography (CT) scans at the orbital level in pediatric patients. Additionally, it seeks to explore the potential application value of DLIR technology in the context of ultra-low-dose CT imaging of the orbital region in children. Materials and methods 1.1 General information This retrospective analysis evaluated 50 pediatric patients undergoing paranasal sinus or orbital CT examinations at our institution. Based on clinical indications and scanning protocols, two cohorts were retrospectively reconstructed from archival data: 1) The ultra-low-dose group (n = 25, 12 males/13 females, age 2–14 years, mean 6.67 ± 2.83 years) comprised children with suspected sinusitis scanned under optimized low-radiation protocols; 2) The conventional-dose group (n = 25, 13 males/12 females, age 1–13 years, mean 7.03 ± 2.82 years) included trauma patients undergoing standard-dose CT scans. The study protocol, approved by Qingdao University Affiliated Women and Children’s hospital's Ethics Committee, parental/guardian informed consent was waived individual informed consent for anonymized image quality assessment. 1.2 Instruments and methods All examinations were performed using a 256-slice CT scanner (GE Revolution CT) with standardized imaging protocol: supine axial acquisition spanning from the maxillary alveolus to frontal sinus apex, fixed 100 kVp tube voltage, SmartmA modulation (60–250 mA), 0.992:1 pitch, and 1.25 mm isotropic reconstruction. Radiation protection included lead shielding for radiosensitive tissues, with oral/rectal chloral hydrate sedation administered as needed. Protocol differentiation focused on dose optimization parameters: the study cohort employed a high noise index (NI = 22) with 70% pre-ASiR-V weighting, while the control group utilized conventional settings (NI = 8, 30% pre-ASiR-V). Quantitative radiation metrics (CTDIvol and DLP) were systematically recorded for dosimetric analysis. 1.3 Image postprocessing and analysis The raw data of the experimental group were reconstructed with three levels of DLIR (high, medium, and low) and named groups DL-H, DL-M, and DL-L, respectively. The control group used post-ASIR-V reconstruction, in which the reconstruction weight set 70%, for the sake of distinction the control named the ASiR-V 70% group. The reconstructed images featured a slice thickness and interslice gap of 0.625 mm each. These images were subsequently transferred to the AW4.7 workstation for both objective and subjective evaluations. 1.4 Image quality evaluation and comparison 1.4.1 Objective evaluation Quantitative analysis utilized a standardized methodology wherein 10 mm² regions of interest (ROIs) were systematically placed across three consecutive axial slices at anatomically matched positions. Attenuation measurements were obtained for vitreous humor, retro-orbital adipose tissue, and pre-orbital air space. Objective noise quantification employed the standard deviation (SD) of voxel intensities within ROIs, with ambient air SD serving as baseline noise for computational metrics: signal-to-noise ratio (SNR = [Tissue CT Value]/[Air SD]) and contrast-to-noise ratio (CNR = [Vitreous CT - Fat CT]/[Air SD]). This computational framework ensured standardized parameter comparisons across reconstruction protocols. 1.4.2 Subjective evaluation The five-point scoring system was defined as: one point (excellent) indicating artifact-free images with sharply defined intraorbital structures and precise anatomical delineation; two points (good) applied to images demonstrating intact anatomical details of ocular muscles and optic nerves despite mild edge softening; three points (medium) reserved for scans with notable artifacts yet retaining diagnostically sufficient structural visibility; four points (poor) assigned to images where severe artifacts compromised regional anatomical clarity; progressing to five points (non-diagnostic) when overwhelming artifacts precluded any clinical evaluation. This ordinal scale maintained strict progression from optimal visualization (1 point) down to complete diagnostic inadequacy (5 points). 1.4 Radiation dose The DLP was recorded to calculate the effective dose (ED) with the formula ED = DLP × K, where K is the conversion factor [ 13 ]. 1.5 Statistical analysis Objective data were subjected to Welch's ANOVA for analysis, while Dunnett's T3 test was employed to compare differences between two groups. Subjective scores underwent assessment through the Kruskal–Wallis test, with pairwise comparisons between two groups being conducted using the Mann–Whitney U test. The interrater reliability between the two physicians in scoring was verified by the kappa test. The radiation dose was compared between the experimental group and the control group using an independent samples t-test. Statistical analysis was conducted using SPSS 21 software, with statistical significance set at P < 0.05. Results The objective evaluation values and subjective score are listed in Table 1-2, and the radiation doses for the different groups are listed in Table 3. 2.1 Objective image assessment There were significant differences between the four groups of two groups in vitreous body noise, orbital fat noise, image noise, vitreous body SNR, orbital fat SNR and CNR and images (P<0.05), except vitreous body CT values and orbital fat CT values. In pairwise comparison, there were significant differences between DL-H (the optimal set of DLIR) and ASiR-V 70% (the control group) (Table 2); the image noise was reduced by 41.6% (4.17±0.86 HU vs. 7.14±1.19 HU), and the CNR was improved by 39.3% (26.60±5.61 vs. 16.14±2.48) (P < 0.05). 2.2 Subjective Image Assessment The evaluation results of the two physicians showed substantial interrater reliability (kappa=0.678). Scores for orbital images were significantly different between groups (P<0.05). The DL-H (1.30±0.51) obtained the highest score. The scores for images were ranked in descending order as follows: DL-H, DL-M, ASiR-V 70% and DL-L. 2.3 Radiation dose Assessment The CTDIvols of the experimental group and control group were 2.35mGy and 14.61mGy, respectively. The mean DLPs were 36.65±2.44mGy·cm and 227.92±27.58mGy·cm. The mean EDs were 0.17±0.03mSv and 1.01±0.18mSv. The differences were statistically significant (P<0.05). Compared with the control group, DLP and ED in the experimental group were reduced by 83.92% and 83.17%, respectively. Discussion Children are in a critical phase of growth and development, making them particularly vulnerable to the harmful effects of radiation, notably to the eye's lens [14-15]. Despite its effectiveness in visualizing bone structures and calcifications, CT scans remain extensively utilized. Consequently, significantly lowering the radiation exposure from CT scans, particularly for pediatric patients, is essential to maintain its clinical viability over time. Because the paranasal sinuses are formed in air-bearing bony cavities, the sinus wall is thin and has a natural contrast advantage. The requirements for image noise and sharpness are relatively low, which provides us with an ultra-low-dose scanning opportunity. However, orbital CT in children usually involves trauma diagnosis and requires relatively high image noise and sharpness. So that, we selected children who underwent sinus ultra-low dose CT for sinusitis in the experimental group. Control group included children who underwent general-low dose CT examinations for trauma diagnosis. Our study uses an ingenious experimental design. We used the characteristic that in pediatric patients, the paranasal sinus CT scan and orbital CT scan have approximately the same scan range, which provides our research with unique conditions for the comparative evaluation of image quality at the ultra-low-dose orbital level. At present, one of the cutting-edge techniques for low-dose CT scanning is the application of iterative reconstruction (IR). IR algorithms enhance the quality of images acquired under low signal conditions [16], thereby enabling the use of reduced radiation doses. Notable IR algorithms that have been proven effective include iterative model reconstruction (IMR, Philips), advanced modeled iterative reconstruction (ADMIRE, Siemens), and adaptive statistical iterative reconstruction (ASiR-V, GE Healthcare). The effectiveness of ASiR-V in improving image quality in a pediatric head phantom study has been reported [17]. Consequently, we selected this technique as the benchmark control for our research. Post-ASiR-V technology enhances image quality, and as the weight ratio progressively increases, image noise is diminished. Nonetheless, once the optimal weight threshold is surpassed, the clarity of the images begins to deteriorate, and the edges of the images gradually become less distinct and more distorted [18]. Our study employed appropriate weight ratios, specifically, post-ASiR-V weights of 70%. The experimental group and the control group both employed an identical 100 kV low tube voltage with automatic tube current modulation. We also applied a high helical pitch and fast rotation time to further reduce the radiation dose. We used different noise indexes and pre-ASiR-V weights to achieve ultra-low-dose and general-low dose scans. In our study, the experimental group experienced a significant reduction in radiation dose, with the dose-length product (DLP) decreasing to 36.65±2.44 mGy·cm and the effective dose (ED) to 0.17±0.03 mSv, compared to the control group's values of 227.92±27.58 mGy·cm and 1.01±0.18 mSv. This represents a reduction of 83.92% for DLP and 83.17% for ED, respectively. The thinner the image is, the more obvious the noise will be, so we chose to rebuild the thinnest image (0.625mm) to magnify the difference in image quality. Our study pioneers the evaluation of the DLIR algorithm's impact on image quality in pediatric orbital scans, utilizing ultra-low radiation doses. The DLIR algorithm (GE TrueFidelity), supplied by the manufacturer and employed in our research, is rooted in a deep convolutional neural network architecture. This deep learning algorithm is characterized by its reliance on high-quality, large-sample FBP datasets to train a deep neural network (DNN). Throughout the training phase, the DNN is engaged in analyzing the data, synthesizing and optimizing a reconstruction function via the learning process, and validating the inference engine against a diverse array of test datasets [19-21]. The DLIR algorithm we utilized offers reconstructions at high, medium, and low strengths, enabling the achievement of various levels of noise reduction. Our study showed that, even if the radiation dose dropped by over 80%, the high-level DLIR reconstruction yielded ultra-low-dose orbital images with significantly higher SNRs and CNRs and lower noise than the control group. The subjective overall image quality score for DLIR images also showed the same trend. The image quality of the control group was only between DL-M and DL-L. The control group with 70% post-ASiR-V demonstrated beam hardening artifact appearance in the images compared to DL-H (Figure 2). We believe this is related to the design of current IR algorithms. The IR algorithm adopts a step-by-step solution method, and through a preset reconstruction model, iteratively finds the optimal solution that matches the input data. Due to the simplification of the model complexity and the limitations of modelling, although image noise is reduced, wax-like and beam hardening image artifacts are generated. This is why our control group set post-ASiR-V to 70% instead of 100%. DLIR can avoid such wax-like and and beam hardening artifacts very well, and the reconstructed images are more real and sharp. The IR algorithm's image denoising ability is limited, so it is difficult to apply in tasks requiring good image quality. Currently, reported ultra-low dose examinations are mostly limited to organs with good natural contrast. CT scans of the orbital region require a thinner slice because the thin slice image enhances spatial resolution and provides more information. However, in conventional CT imaging (such as sinus CT scans), there is a need to balance the relationship between radiation dose and spatial resolution. Thinner slices usually mean higher noise levels and lower image quality. To maintain reasonable image noise at low radiation doses, 5 mm images are usually used for diagnosis at the cost of losing some disease-related information. Some studies have mentioned that the signal strength of children’s head CT images with a thickness of 0.625 mm is only one-eighth of that of 5 mm images [22]. Therefore, in this study, we skillfully simulated ultra-low-dose orbital scanning by combining pediatric sinus CT with 0.625 mm thickness. Both objective and subjective evaluation showed that the quality of DL-H images with 0.625 mm thickness was significantly better than that of ASiR-V 70%. From the fact that the current image noise and image quality of general-low dose ASIR-V 70% images are acceptable for current routine clinical use, we can infer that the use of DL-H may be able to save a significant amount of radiation dose (well beyond the current 83%). Our study has some limitations and can be improved upon. First, the sample size was small. Due to our experimental design, we could only use the raw data stored in the CT host for DLIR reconstruction. Second, we only compared image quality but did not evaluate the effectiveness of lesion detection. Nevertheless, orbital CT has been widely accepted in clinical practice, and the image quality of the DL-H group in our study was significantly better than that in our current routine practice, so we have reason to believe that DLIR technology will achieve better clinical application. Finally, we did not evaluate bone window images in this study, and our previous research on sinus CT in children showed almost no significant difference in the evaluation of bone window images at low radiation doses [23]. Therefore, we did not include skull evaluation in this study. Declarations Ethics approval and consent to participate This study was conducted in accordance with the principles of the Declaration of Helsinki. This retrospective study was approved by the Ethics Committees of Qingdao university affiliated Women and Children’s Hospital for using the data, and the informed consent was waved. The research protocol was in compliance with relevant Chinese laws. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Availability of data and materials The datasets used and/or analyzed during the current study available from the corresponding author on reasonable request. Funding No funding. Authors' contributions Yang Li: Conceptualization, Methodology, Writing-original draft. Feng-Xian Wang and Xia Liu: Data curation, Formal analysis. Shan-shan Sun: Investigation. Xiu-feng Song: Project administration, Writing-review & editing. Acknowledgments The authors would like to express their gratitude to Dr. Yun Sheng, Dr. Ming-Jun Wang, and Dr. Shuai Zhang for their assistance in generating DLIR images and for enhancing the understanding of the DLIR technique. The authors wish to acknowledge that portions of the data presented in this manuscript (or: A preliminary version of the abstract) has been previously published in [BMC Medical Imaging, https://doi.org/10.1186/s12880-022-00834-1]. The current work significantly expands upon the initial findings. All reused materials are appropriately cited within the text and comply with ethical guidelines for scholarly publication. References Gottumukkala RV, Kalra MK, Tabari A, Otrakji A, Gee MS. Advanced CT Techniques for Decreasing Radiation Dose, Reducing Sedation Requirements, and Optimizing Image Quality in Children. Radiographics. 2019;39(3):709-726. Nagayama Y, Oda S, Nakaura T, et al. Radiation Dose Reduction at Pediatric CT: Use of Low Tube Voltage and Iterative Reconstruction [published correction appears in Radiographics. 2019 May-Jun;39(3):912]. Radiographics. 2018;38(5):1421-1440. Schulz B, Beeres M, Bodelle B, Bauer R, Al-Butmeh F, Thalhammer A, Vogl TJ, Kerl JM. Performance of iterative image reconstruction in CT of the paranasal sinuses: a phantom study. AJNR Am J Neuroradiol. 2013 May;34(5):1072-6. Schulz B, Potente S, Zangos S, Friedrichs I, Bauer RW, Kerl M, Vogl TJ, Mack MM. Ultra-lowultra-low dose dual-source high-pitch computed tomography of the paranasal sinus: diagnostic sensitivity and radiation dose. Acta Radiol. 2012 May 1;53(4):435-40. Lam S, Bux S, Kumar G, Ng Kh, Hussain A. A comparison between low-dose and standard-dose non-contrastednoncontrasted multidetector CT scanning of the paranasal sinuses. Biomed Imaging Interv J. 2009 Jul;5(3):e13. Sagara Y, Hara AK, Pavlicek W, Silva AC, Paden RG, Wu Q. Abdominal CT: comparison of low-dose CT with adaptive statistical iterative reconstruction and routine-dose CT with filtered back projection in 53 patients. AJR Am J Roentgenol. 2010;195(3):713-719. De Marco P, Origgi D. New adaptive statistical iterative reconstruction ASiR-V: Assessment of noise performance in comparison to ASiR. J Appl Clin Med Phys. 2018;19(2):275-286. Tang H, Liu Z, Hu Z, et al. Clinical value of a new generation adaptive statistical iterative reconstruction (ASIR-V) in the diagnosis of pulmonary nodule in low-dose chest CT. Br J Radiol. 2019;92(1103):20180909. Mileto A, Guimaraes LS, McCollough CH, Fletcher JG, Yu L. State of the Art in Abdominal CT: The Limits of Iterative Reconstruction Algorithms. Radiology. 2019;293(3):491-503. Sun J, Li H, Gao J, Li J, Li M, Zhou Z, Peng Y. Performance evaluation of a deep learning image reconstruction (DLIR) algorithm in "double low" chest CTA in children: a feasibility study. Radiol Med. 2021 Sep;126(9):1181-1188. doi: 10.1007/s11547-021-01384-2. Epub 2021 Jun 16. PMID: 34132926. Sun J, Li H, Wang B, Li J, Li M, Zhou Z, Peng Y. Application of a deep learning image reconstruction (DLIR) algorithm in head CT imaging for children to improve image quality and lesion detection. BMC Med Imaging. 2021 Jul 8;21(1):108. doi: 10.1186/s12880-021-00637-w. PMID: 34238229; PMCID: PMC8268450. Sun J, Li H, Li J, Yu T, Li M, Zhou Z, Peng Y. Improving the image quality of pediatric chest CT angiography with low radiation dose and contrast volume using deep learning image reconstruction. Quant Imaging Med Surg. 2021 Jul;11(7):3051-3058. doi: 10.21037/qims-20-1158. PMID: 34249634; PMCID: PMC8250028. Mccollough C, Edyvean S, Gould B, et al. The Measurement, Reporting, and Management of Radiation Dose in CT. 2008. May MS, Brand M, Lell MM, Sedlmair M, Allmendinger T, Uder M, Wuest W. Radiation dose reduction in parasinus CT by spectral shaping. Neuroradiology. 2017 Feb;59(2):169-176. Sun J, Zhang Q, Duan X, Zhang C, Wang P, Jia C, Liu Y, Peng Y. Application of a full model-based iterative reconstruction (MBIR) in 80 kVp ultra-lowultra-low-dose paranasal sinus CT imaging of pediatric patients. Radiol Med. 2018 Feb;123(2):117-124. Fletcher JG, DeLone DR, Kotsenas AL, et al. Evaluation of lower-dose spiral head CT for detection of intracranial findings causing neurologic deficits. AJNR Am J Neuroradiol. 2019;40(11):1855–63. Kim HG, Lee HJ, Lee SK, et al. Head CT: Image quality improvement with ASIR-V using a reduced radiation dose protocol for children. Eur Radiol. 2017;27(9):3609-3617. Kwon H, Cho J, Oh J, et al. The adaptive statistical iterative reconstruction-V technique for radiation dose reduction in abdominal CT: comparison with the adaptive statistical iterative reconstruction technique [published correction appears in Br J Radiol. 2016;89(1058):20150463e]. Br J Radiol. 2015;88(1054):20150463. 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 Jan;22(1):131-138. Mileto A, Guimaraes LS, McCollough CH, Fletcher JG, Yu L. State of the Art in Abdominal CT: The Limits of Iterative Reconstruction Algorithms. Radiology. 2019;293(3):491-503. Jensen CT, Liu X, Tamm EP, et al. Image Quality Assessment of Abdominal CT by Use of New Deep Learning Image Reconstruction: Initial Experience. AJR Am J Roentgenol. 2020;215(1):50-57. Sun J, Li H, Wang B, et al. Application of a deep learning image reconstruction (DLIR) algorithm in head CT imaging for children to improve image quality and lesion detection. BMC Med Imaging. 2021;21(1):108. Published 2021 Jul 8. Li Y, Liu X, Zhuang XH, Wang MJ, Song XF. Assessment of low-dose paranasal sinus CT imaging using a new deep learning image reconstruction technique in children compared to adaptive statistical iterative reconstruction V (ASiR-V). BMC Med Imaging. 2022;22(1):106. Published 2022 Jun 3. Tables Table. 1 Evaluation of images quality with different groups（ x ( _ ) ± s ） Group Objective evaluation Subjective evaluation vitreous body CT value vitreous body noise vitreous body SNR orbital fat CT value orbital fat noise orbital fat SNR image noise CNR Image subjective score of Orbital level DL-H 8.05±3.46 9.86±1.41 2.04±1.00 -98.53±7.14 12.62±2.24 24.56±5.04 4.17±0.86 26.60±5.61 1.30±0.51 DL-M 8.29±3.24 12.25±1.86 1.46±0.69 -100.11±7.09 15.05±2.06 17.33±3.30 5.98±1.18 18.79±3.65 1.88±0.56 DL-L 8.97±3.53 14.67±2.56 1.26±0.54 -96.07±15.91 15.74±3.31 13.60±3.97 7.43±1.35 14.86±4.25 2.68±0.65 ASiR-V 70% 9.37±2.74 8.14±1.18 1.18±0.40 -103.27±7.11 10.88±2.75 10.08±2.48 7.14±1.19 16.14±2.48 2.18±0.66 F 0.952 59.120 5.587 2.495 16.951 64.355 51.534 27.978 86.929 (χ² value) P ＞0.05 ＜0.05 ＜0.05 ＞0.05 ＜0.05 ＜0.05 ＜0.05 ＜0.05 ＜0.05 Table. 2 Comparison of images quality with DL-H and ASiR-V 70%（ x ( _ ) ± s ） Group Objective evaluation Subjective evaluation vitreous body CT value vitreous body noise vitreous body SNR orbital fat CT value orbital fat noise orbital fat SNR image noise CNR Image subjective score of Orbital level DL-H 8.05±3.46 9.86±1.41 2.04±1.00 -98.53±7.14 12.62±2.24 24.56±5.04 4.17±0.86 26.60±5.61 1.30±0.51 ASiR-V 70% 9.37±2.74 8.14±1.18 1.18±0.40 -103.27±7.11 10.88±2.75 10.08±2.48 7.14±1.19 16.14±2.48 2.18±0.66 P ＞0.05 ＜0.05 ＜0.05 ＜0.05 ＞0.05 ＜0.05 ＜0.05 ＜0.05 ＜0.05 Table. 3 Comparison of radiation dose（ x ( _ ) ± s ） Group DLP（mGy*cm） ED（mSv） A 36.65±2.44 0.17±0.03 B 227.92±27.58 1.01±0.18 t -34.56 -23.00 P P＜0.05 P＜0.05 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6243177","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":452124467,"identity":"2ab56d0a-6b27-46ad-8bb1-2d3720268693","order_by":0,"name":"Yang Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAuElEQVRIiWNgGAWjYBACPmYwZSPHT7QWNoiWNGPJBqK1QKjDiRsOEK2Fncfwc8GvtMTNx5M3MPyo2EaMw3iMpWf22RhvO/OsgLHnzG2itBhI8/akyW67kWPAzNhGnBbj37w9hxk3zyBBi5k0z4/DihskiNfCVmbN25BmLAH0y0Gi/MLPf3jzbZ4/wKhsT9744EcFEVoYGDgMGBjbQIwEgwPEqAcC9gcMDH8gWojUMQpGwSgYBSMNAABR3DfAxf7D/gAAAABJRU5ErkJggg==","orcid":"","institution":"Qingdao University Affiliated Women and Children’s Hospital","correspondingAuthor":true,"prefix":"","firstName":"Yang","middleName":"","lastName":"Li","suffix":""},{"id":452124468,"identity":"7d9b641a-9539-47e6-8f42-61523567c1de","order_by":1,"name":"Feng-Xian Wang","email":"","orcid":"","institution":"Qingdao University Affiliated Qingdao Women and Children’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Feng-Xian","middleName":"","lastName":"Wang","suffix":""},{"id":452124469,"identity":"18a89c76-5e2e-4651-bffd-42be179c7728","order_by":2,"name":"Xia Liu","email":"","orcid":"","institution":"The Affiliated Hospital of Qingdao University","correspondingAuthor":false,"prefix":"","firstName":"Xia","middleName":"","lastName":"Liu","suffix":""},{"id":452124470,"identity":"2d750d74-fa59-494f-8b41-11d0639da52e","order_by":3,"name":"Shan-shan Sun","email":"","orcid":"","institution":"Qingdao University Affiliated Women and Children’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Shan-shan","middleName":"","lastName":"Sun","suffix":""},{"id":452124471,"identity":"eddd81c5-c91e-44c8-9606-5630ac6c3ce6","order_by":4,"name":"Xiu-feng Song","email":"","orcid":"","institution":"Qingdao University Affiliated Women and Children’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Xiu-feng","middleName":"","lastName":"Song","suffix":""}],"badges":[],"createdAt":"2025-03-17 09:38:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6243177/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6243177/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82299158,"identity":"3611fdcf-4968-4334-b599-553df74a0dd6","added_by":"auto","created_at":"2025-05-08 20:31:35","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":77856,"visible":true,"origin":"","legend":"\u003cp\u003eSee image above for figure legend.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6243177/v1/1adb5c4ef660258111ab4ea2.png"},{"id":82300033,"identity":"1945c887-8316-4764-bd7c-0dc90d0f1e76","added_by":"auto","created_at":"2025-05-08 20:39:35","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":297584,"visible":true,"origin":"","legend":"\u003cp\u003eAxial CT images at the orbital level in different groups. (A) DL-H; (B) DL-M; (C) DL-L; (D) ASiR-V 70%. The experimental groups (A, B, C) represent ultra-low dose CT reconstructions of a male subject, while the control group (D) represents general-low dose CT reconstructions of a male subject of the same age. The subjective image quality scores, primarily assessed based on image sharpness and the presence of beam hardening artifacts, were ranked in descending order as follows: DL-H, DL-M, ASiR-V 70%, and DL-L.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6243177/v1/5147f081fcf1318bd7ff9bbd.png"},{"id":109078545,"identity":"165dde1f-5247-43eb-83a2-52d951cb2520","added_by":"auto","created_at":"2026-05-12 11:15:01","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":576185,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6243177/v1/e661499c-ca68-4d6c-8379-d3028a770b57.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Application of a New Deep Learning Image Reconstruction (DLIR) Algorithm in Pediatric Orbital horizontal level Ultra-Low Dose CT scan: A Pilot Study Based on Image Quality","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAs computed tomography (CT) technology rapidly evolves, the use of low-dose CT is becoming increasingly prevalent to minimize radiation exposure in patients. Reducing the CT radiation dose in children has been a hot spot of research in recent years [\u003cspan additionalcitationids=\"CR2\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. CT is an excellent imaging modality for displaying bony structures, and it is intuitive and fast, and it is a significant examination method for the paranasal sinuses and orbits of children [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. The paranasal sinuses are formed in air-bearing bony cavities, and the sinus wall is thin and has a natural contrast advantage. CT in children's paranasal sinuses is mostly used to evaluate paranasal sinus inflammatory lesions. The requirements for image noise and sharpness are relatively low, which provides us with a very low-dose scanning opportunity. In addition, the CT scan range of the paranasal sinuses will inevitably include the orbit, and the eyeball is extremely sensitive to radiation. However, orbital CT in children usually involves trauma diagnosis and requires relatively high image noise and sharpness. It is the relatively single disease spectrum of children's sinus and orbital CT scans but the same scan range that provides our research with unique conditions for the comparative evaluation of image quality under ultra-low-dose and orbital levels.\u003c/p\u003e \u003cp\u003eFiltered back projection (FBP) stands as the industry standard for CT image reconstruction; however, this technology is constrained by considerable noise and artifacts [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Moreover, it fails to deliver exceptional diagnostic images when using lower radiation doses. In recent years, iterative reconstruction (IR) technology has advanced rapidly, prompting various CT manufacturers to incorporate sophisticated IR algorithms aimed at lowering radiation exposure. Adaptive statistical iterative reconstruction V (ASiR-V) is a cutting-edge technology that has garnered attention for its ability to produce high-quality diagnostic images while substantially reducing radiation doses [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Nevertheless, its application is significantly hindered due to the overly smooth and unnatural of the images. Deep learning image reconstruction (DLIR) represents a novel approach to image reconstruction. This method employs high-dose FBP datasets and deep neural network (DNN) models to further diminish noise and suppress artifacts [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]. The schematic diagram of the DLIR algorithm is depicted in Fig.\u0026nbsp;1. Although DLIR technology has only recently obtained FDA certification and has not been widely adopted, there are scant reports on its use in pediatric CT scans [\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].There are no reports about the application of orbital CT scans in children. The objective of this study is to evaluate the effects of Deep Learning Iterative Reconstruction (DLIR) and Adaptive Statistical Iterative Reconstruction-V (ASiR-V) on the image quality of computed tomography (CT) scans at the orbital level in pediatric patients. Additionally, it seeks to explore the potential application value of DLIR technology in the context of ultra-low-dose CT imaging of the orbital region in children.\u003c/p\u003e"},{"header":"Materials and methods","content":"\u003cp\u003e1.1 General information\u003c/p\u003e\n\u003cp\u003eThis retrospective analysis evaluated 50 pediatric patients undergoing paranasal sinus or orbital CT examinations at our institution. Based on clinical indications and scanning protocols, two cohorts were retrospectively reconstructed from archival data: 1) The ultra-low-dose group (n\u0026thinsp;=\u0026thinsp;25, 12 males/13 females, age 2\u0026ndash;14 years, mean 6.67\u0026thinsp;\u0026plusmn;\u0026thinsp;2.83 years) comprised children with suspected sinusitis scanned under optimized low-radiation protocols; 2) The conventional-dose group (n\u0026thinsp;=\u0026thinsp;25, 13 males/12 females, age 1\u0026ndash;13 years, mean 7.03\u0026thinsp;\u0026plusmn;\u0026thinsp;2.82 years) included trauma patients undergoing standard-dose CT scans. The study protocol, approved by Qingdao University Affiliated Women and Children\u0026rsquo;s hospital\u0026apos;s Ethics Committee, parental/guardian informed consent was waived individual informed consent for anonymized image quality assessment.\u003c/p\u003e\n\u003cp\u003e1.2 Instruments and methods\u003c/p\u003e\n\u003cp\u003eAll examinations were performed using a 256-slice CT scanner (GE Revolution CT) with standardized imaging protocol: supine axial acquisition spanning from the maxillary alveolus to frontal sinus apex, fixed 100 kVp tube voltage, SmartmA modulation (60\u0026ndash;250 mA), 0.992:1 pitch, and 1.25 mm isotropic reconstruction. Radiation protection included lead shielding for radiosensitive tissues, with oral/rectal chloral hydrate sedation administered as needed. Protocol differentiation focused on dose optimization parameters: the study cohort employed a high noise index (NI\u0026thinsp;=\u0026thinsp;22) with 70% pre-ASiR-V weighting, while the control group utilized conventional settings (NI\u0026thinsp;=\u0026thinsp;8, 30% pre-ASiR-V). Quantitative radiation metrics (CTDIvol and DLP) were systematically recorded for dosimetric analysis.\u003c/p\u003e\n\u003cp\u003e1.3 Image postprocessing and analysis\u003c/p\u003e\n\u003cp\u003eThe raw data of the experimental group were reconstructed with three levels of DLIR (high, medium, and low) and named groups DL-H, DL-M, and DL-L, respectively. The control group used post-ASIR-V reconstruction, in which the reconstruction weight set 70%, for the sake of distinction the control named the ASiR-V 70% group. The reconstructed images featured a slice thickness and interslice gap of 0.625 mm each. These images were subsequently transferred to the AW4.7 workstation for both objective and subjective evaluations.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e1.4 Image quality evaluation and comparison\u003c/p\u003e\u003cspan\u003e\n \u003cp\u003e1.4.1 Objective evaluation\u003c/p\u003e\n\u003c/span\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eQuantitative analysis utilized a standardized methodology wherein 10 mm\u0026sup2; regions of interest (ROIs) were systematically placed across three consecutive axial slices at anatomically matched positions. Attenuation measurements were obtained for vitreous humor, retro-orbital adipose tissue, and pre-orbital air space. Objective noise quantification employed the standard deviation (SD) of voxel intensities within ROIs, with ambient air SD serving as baseline noise for computational metrics: signal-to-noise ratio (SNR = [Tissue CT Value]/[Air SD]) and contrast-to-noise ratio (CNR = [Vitreous CT - Fat CT]/[Air SD]). This computational framework ensured standardized parameter comparisons across reconstruction protocols.\u003c/p\u003e\n\u003cp\u003e1.4.2 Subjective evaluation\u003c/p\u003e\n\u003cp\u003eThe five-point scoring system was defined as: one point (excellent) indicating artifact-free images with sharply defined intraorbital structures and precise anatomical delineation; two points (good) applied to images demonstrating intact anatomical details of ocular muscles and optic nerves despite mild edge softening; three points (medium) reserved for scans with notable artifacts yet retaining diagnostically sufficient structural visibility; four points (poor) assigned to images where severe artifacts compromised regional anatomical clarity; progressing to five points (non-diagnostic) when overwhelming artifacts precluded any clinical evaluation. This ordinal scale maintained strict progression from optimal visualization (1 point) down to complete diagnostic inadequacy (5 points).\u003c/p\u003e\n\u003cp\u003e1.4 Radiation dose\u003c/p\u003e\n\u003cp\u003eThe DLP was recorded to calculate the effective dose (ED) with the formula ED\u0026thinsp;=\u0026thinsp;DLP \u0026times; K, where K is the conversion factor [\u003cspan class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e\n\u003cp\u003e1.5 Statistical analysis\u003c/p\u003e\n\u003cp\u003eObjective data were subjected to Welch\u0026apos;s ANOVA for analysis, while Dunnett\u0026apos;s T3 test was employed to compare differences between two groups. Subjective scores underwent assessment through the Kruskal\u0026ndash;Wallis test, with pairwise comparisons between two groups being conducted using the Mann\u0026ndash;Whitney U test. The interrater reliability between the two physicians in scoring was verified by the kappa test. The radiation dose was compared between the experimental group and the control group using an independent samples t-test. Statistical analysis was conducted using SPSS 21 software, with statistical significance set at P\u0026thinsp;\u0026lt;\u0026thinsp;0.05.\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThe objective evaluation values and subjective score are listed in Table 1-2, and the radiation doses for the different groups are listed in Table 3.\u003c/p\u003e\n\u003cp\u003e2.1 Objective image assessment\u003c/p\u003e\n\u003cp\u003eThere were significant differences between the four groups of two groups in vitreous body noise, orbital fat noise, image noise, vitreous body SNR, orbital fat SNR and CNR and images (P\u0026lt;0.05), except vitreous body CT values and orbital fat CT values. In pairwise comparison, there were significant differences between DL-H (the optimal set of DLIR) and ASiR-V 70% (the control group) (Table 2); the image noise was reduced by 41.6% (4.17±0.86 HU vs. 7.14±1.19 HU), and the CNR was improved by 39.3% (26.60±5.61 vs. 16.14±2.48) (P \u0026lt; 0.05).\u003c/p\u003e\n\u003cp\u003e2.2 Subjective Image Assessment\u003c/p\u003e\n\u003cp\u003eThe evaluation results of the two physicians showed substantial interrater reliability (kappa=0.678). Scores for orbital images were significantly different between groups (P\u0026lt;0.05). The DL-H (1.30±0.51) obtained the highest score. The scores for images were ranked in descending order as follows: DL-H, DL-M, ASiR-V 70% and DL-L.\u003c/p\u003e\n\u003cp\u003e2.3 Radiation dose Assessment\u003c/p\u003e\n\u003cp\u003eThe CTDIvols of the experimental group and control group were 2.35mGy and 14.61mGy, respectively. The mean DLPs were 36.65±2.44mGy·cm and 227.92±27.58mGy·cm. The mean EDs were 0.17±0.03mSv and 1.01±0.18mSv. The differences were statistically significant (P\u0026lt;0.05). \u0026nbsp;Compared with the control group, DLP and ED in the experimental group were reduced by 83.92% and 83.17%, respectively.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eChildren are in a critical phase of growth and development, making them particularly vulnerable to the harmful effects of radiation, notably to the eye's lens [14-15]. Despite its effectiveness in visualizing bone structures and calcifications, CT scans remain extensively utilized. Consequently, significantly lowering the radiation exposure from CT scans, particularly for pediatric patients, is essential to maintain its clinical viability over time. Because the paranasal sinuses are formed in air-bearing bony cavities, the sinus wall is thin and has a natural contrast advantage. The requirements for image noise and sharpness are relatively low, which provides us with an ultra-low-dose scanning opportunity. However, orbital CT in children usually involves trauma diagnosis and requires relatively high image noise and sharpness. So that, we selected children who underwent sinus ultra-low dose CT for sinusitis in the experimental group. Control group included children who underwent general-low dose CT examinations for trauma diagnosis. Our study uses an ingenious experimental design. \u0026nbsp;We used the characteristic that in pediatric patients, the paranasal sinus CT scan and orbital CT scan have approximately the same scan range, which provides our research with unique conditions for the comparative evaluation of image quality at the ultra-low-dose orbital level.\u003c/p\u003e\n\u003cp\u003eAt present, one of the cutting-edge techniques for low-dose CT scanning is the application of iterative reconstruction (IR). IR algorithms enhance the quality of images acquired under low signal conditions [16], thereby enabling the use of reduced radiation doses. Notable IR algorithms that have been proven effective include iterative model reconstruction (IMR, Philips), advanced modeled iterative reconstruction (ADMIRE, Siemens), and adaptive statistical iterative reconstruction (ASiR-V, GE Healthcare). The effectiveness of ASiR-V in improving image quality in a pediatric head phantom study has been reported [17]. Consequently, we selected this technique as the benchmark control for our research. Post-ASiR-V technology enhances image quality, and as the weight ratio progressively increases, image noise is diminished. Nonetheless, once the optimal weight threshold is surpassed, the clarity of the images begins to deteriorate, and the edges of the images gradually become less distinct and more distorted [18]. Our study employed appropriate weight ratios, specifically, post-ASiR-V weights of 70%. The experimental group and the control group both employed an identical 100 kV low tube voltage with automatic tube current modulation. We also applied a high helical pitch and fast rotation time to further reduce the radiation dose. We used different noise indexes and pre-ASiR-V weights to achieve ultra-low-dose and general-low dose scans. In our study, the experimental group experienced a significant reduction in radiation dose, with the dose-length product (DLP) decreasing to 36.65±2.44 mGy·cm and the effective dose (ED) to 0.17±0.03 mSv, compared to the control group's values of 227.92±27.58 mGy·cm and 1.01±0.18 mSv. This represents a reduction of 83.92% for DLP and 83.17% for ED, respectively. The thinner the image is, the more obvious the noise will be, so we chose to rebuild the thinnest image (0.625mm) to magnify the difference in image quality.\u003c/p\u003e\n\u003cp\u003eOur study pioneers the evaluation of the DLIR algorithm's impact on image quality in pediatric orbital scans, utilizing ultra-low radiation doses. The DLIR algorithm (GE TrueFidelity), supplied by the manufacturer and employed in our research, is rooted in a deep convolutional neural network architecture. This deep learning algorithm is characterized by its reliance on high-quality, large-sample FBP datasets to train a deep neural network (DNN). Throughout the training phase, the DNN is engaged in analyzing the data, synthesizing and optimizing a reconstruction function via the learning process, and validating the inference engine against a diverse array of test datasets [19-21]. The DLIR algorithm we utilized offers reconstructions at high, medium, and low strengths, enabling the achievement of various levels of noise reduction. Our study showed that, even if the radiation dose dropped by over 80%, the high-level DLIR reconstruction yielded ultra-low-dose orbital images with significantly higher SNRs and CNRs and lower noise than the control group. The subjective overall image quality score for DLIR images also showed the same trend. The image quality of the control group was only between DL-M and DL-L. The control group with 70% post-ASiR-V demonstrated beam hardening artifact appearance in the images compared to DL-H (Figure 2). We believe this is related to the design of current IR algorithms. The IR algorithm adopts a step-by-step solution method, and through a preset reconstruction model, iteratively finds the optimal solution that matches the input data. Due to the simplification of the model complexity and the limitations of modelling, although image noise is reduced, wax-like and beam hardening image artifacts are generated. This is why our control group set post-ASiR-V to 70% instead of 100%. DLIR can avoid such wax-like and and beam hardening artifacts very well, and the reconstructed images are more real and sharp.\u003c/p\u003e\n\u003cp\u003eThe IR algorithm's image denoising ability is limited, so it is difficult to apply in tasks requiring good image quality. Currently, reported ultra-low dose examinations are mostly limited to organs with good natural contrast. CT scans of the orbital region require a thinner slice because the thin slice image enhances spatial resolution and provides more information. However, in conventional CT imaging (such as sinus CT scans), there is a need to balance the relationship between radiation dose and spatial resolution. Thinner slices usually mean higher noise levels and lower image quality. To maintain reasonable image noise at low radiation doses, 5 mm images are usually used for diagnosis at the cost of losing some disease-related information. Some studies have mentioned that the signal strength of children’s head CT images with a thickness of 0.625 mm is only one-eighth of that of 5 mm images [22]. Therefore, in this study, we skillfully simulated ultra-low-dose orbital scanning by combining pediatric sinus CT with 0.625 mm thickness. Both objective and subjective evaluation showed that the quality of DL-H images with 0.625 mm thickness was significantly better than that of ASiR-V 70%. From the fact that the current image noise and image quality of general-low dose ASIR-V 70% images are acceptable for current routine clinical use, we can infer that the use of DL-H may be able to save a significant amount of radiation dose (well beyond the current 83%).\u003c/p\u003e\n\u003cp\u003eOur study has some limitations and can be improved upon. First, the sample size was small. Due to our experimental design, we could only use the raw data stored in the CT host for DLIR reconstruction. Second, we only compared image quality but did not evaluate the effectiveness of lesion detection. Nevertheless, orbital CT has been widely accepted in clinical practice, and the image quality of the DL-H group in our study was significantly better than that in our current routine practice, so we have reason to believe that DLIR technology will achieve better clinical application. Finally, we did not evaluate bone window images in this study, and our previous research on sinus CT in children showed almost no significant difference in the evaluation of bone window images at low radiation doses [23]. Therefore, we did not include skull evaluation in this study.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eEthics approval and consent to participate\u003c/p\u003e\n\u003cp\u003eThis study was conducted in accordance with the principles of the Declaration of Helsinki. This retrospective study was approved by the Ethics Committees of Qingdao university affiliated Women and Children’s Hospital for using the data, and the informed consent was waved. The research protocol was in compliance with relevant Chinese laws.\u003c/p\u003e\n\u003cp\u003eConsent for publication\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003eCompeting interests\u003c/p\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; The authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003eAvailability of data and materials\u003c/p\u003e\n\u003cp\u003eThe datasets used and/or analyzed during the current study available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003eFunding\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eNo funding.\u003c/p\u003e\n\u003cp\u003eAuthors' contributions\u003c/p\u003e\n\u003cp\u003eYang Li: Conceptualization, Methodology, Writing-original draft.\u0026nbsp;Feng-Xian Wang and\u0026nbsp;Xia Liu: Data curation, Formal analysis. Shan-shan Sun: Investigation. Xiu-feng Song: Project administration, Writing-review \u0026amp; editing.\u003c/p\u003e\n\u003cp\u003eAcknowledgments\u003c/p\u003e\n\u003cp\u003eThe authors would like to express their gratitude to Dr. Yun Sheng, Dr. Ming-Jun Wang, and Dr. Shuai Zhang for their assistance in generating DLIR images and for enhancing the understanding of the DLIR technique.\u003c/p\u003e\n\u003cp\u003eThe authors wish to acknowledge that portions of the data presented in this manuscript (or: A preliminary version of the abstract) has been previously published in [BMC Medical Imaging, https://doi.org/10.1186/s12880-022-00834-1]. The current work significantly expands upon the initial findings. All reused materials are appropriately cited within the text and comply with ethical guidelines for scholarly publication.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eGottumukkala RV, Kalra MK, Tabari A, Otrakji A, Gee MS. Advanced CT Techniques for Decreasing Radiation Dose, Reducing Sedation Requirements, and Optimizing Image Quality in Children. Radiographics. 2019;39(3):709-726.\u003c/li\u003e\n \u003cli\u003eNagayama Y, Oda S, Nakaura T, et al. Radiation Dose Reduction at Pediatric CT: Use of Low Tube Voltage and Iterative Reconstruction [published correction appears in Radiographics. 2019 May-Jun;39(3):912]. Radiographics. 2018;38(5):1421-1440.\u003c/li\u003e\n \u003cli\u003eSchulz B, Beeres M, Bodelle B, Bauer R, Al-Butmeh F, Thalhammer A, Vogl TJ, Kerl JM. Performance of iterative image reconstruction in CT of the paranasal sinuses: a phantom study. AJNR Am J Neuroradiol. 2013 May;34(5):1072-6.\u003c/li\u003e\n \u003cli\u003eSchulz B, Potente S, Zangos S, Friedrichs I, Bauer RW, Kerl M, Vogl TJ, Mack MM. Ultra-lowultra-low dose dual-source high-pitch computed tomography of the paranasal sinus: diagnostic sensitivity and radiation dose. Acta Radiol. 2012 May 1;53(4):435-40.\u003c/li\u003e\n \u003cli\u003eLam S, Bux S, Kumar G, Ng Kh, Hussain A. A comparison between low-dose and standard-dose non-contrastednoncontrasted multidetector CT scanning of the paranasal sinuses. Biomed Imaging Interv J. 2009 Jul;5(3):e13.\u003c/li\u003e\n \u003cli\u003eSagara Y, Hara AK, Pavlicek W, Silva AC, Paden RG, Wu Q. Abdominal CT: comparison of low-dose CT with adaptive statistical iterative reconstruction and routine-dose CT with filtered back projection in 53 patients. AJR Am J Roentgenol. 2010;195(3):713-719.\u003c/li\u003e\n \u003cli\u003eDe Marco P, Origgi D. New adaptive statistical iterative reconstruction ASiR-V: Assessment of noise performance in comparison to ASiR. J Appl Clin Med Phys. 2018;19(2):275-286.\u003c/li\u003e\n \u003cli\u003eTang H, Liu Z, Hu Z, et al. Clinical value of a new generation adaptive statistical iterative reconstruction (ASIR-V) in the diagnosis of pulmonary nodule in low-dose chest CT. Br J Radiol. 2019;92(1103):20180909.\u003c/li\u003e\n \u003cli\u003eMileto A, Guimaraes LS, McCollough CH, Fletcher JG, Yu L. State of the Art in Abdominal CT: The Limits of Iterative Reconstruction Algorithms. Radiology. 2019;293(3):491-503.\u003c/li\u003e\n \u003cli\u003eSun J, Li H, Gao J, Li J, Li M, Zhou Z, Peng Y. Performance evaluation of a deep learning image reconstruction (DLIR) algorithm in \u0026quot;double low\u0026quot; chest CTA in children: a feasibility study. Radiol Med. 2021 Sep;126(9):1181-1188. doi: 10.1007/s11547-021-01384-2. Epub 2021 Jun 16. PMID: 34132926.\u003c/li\u003e\n \u003cli\u003eSun J, Li H, Wang B, Li J, Li M, Zhou Z, Peng Y. Application of a deep learning image reconstruction (DLIR) algorithm in head CT imaging for children to improve image quality and lesion detection. BMC Med Imaging. 2021 Jul 8;21(1):108. doi: 10.1186/s12880-021-00637-w. PMID: 34238229; PMCID: PMC8268450.\u003c/li\u003e\n \u003cli\u003eSun J, Li H, Li J, Yu T, Li M, Zhou Z, Peng Y. Improving the image quality of pediatric chest CT angiography with low radiation dose and contrast volume using deep learning image reconstruction. Quant Imaging Med Surg. 2021 Jul;11(7):3051-3058. doi: 10.21037/qims-20-1158. PMID: 34249634; PMCID: PMC8250028.\u003c/li\u003e\n \u003cli\u003eMccollough C, Edyvean S, Gould B, et al. The Measurement, Reporting, and Management of Radiation Dose in CT. 2008.\u003c/li\u003e\n \u003cli\u003eMay MS, Brand M, Lell MM, Sedlmair M, Allmendinger T, Uder M, Wuest W. Radiation dose reduction in parasinus CT by spectral shaping. Neuroradiology. 2017 Feb;59(2):169-176.\u003c/li\u003e\n \u003cli\u003eSun J, Zhang Q, Duan X, Zhang C, Wang P, Jia C, Liu Y, Peng Y. Application of a full model-based iterative reconstruction (MBIR) in 80 kVp ultra-lowultra-low-dose paranasal sinus CT imaging of pediatric patients. Radiol Med. 2018 Feb;123(2):117-124.\u003c/li\u003e\n \u003cli\u003eFletcher JG, DeLone DR, Kotsenas AL, et al. Evaluation of lower-dose spiral head CT for detection of intracranial findings causing neurologic deficits. AJNR Am J Neuroradiol. 2019;40(11):1855\u0026ndash;63.\u003c/li\u003e\n \u003cli\u003eKim HG, Lee HJ, Lee SK, et al. Head CT: Image quality improvement with ASIR-V using a reduced radiation dose protocol for children. Eur Radiol. 2017;27(9):3609-3617.\u003c/li\u003e\n \u003cli\u003eKwon H, Cho J, Oh J, et al. The adaptive statistical iterative reconstruction-V technique for radiation dose reduction in abdominal CT: comparison with the adaptive statistical iterative reconstruction technique [published correction appears in Br J Radiol. 2016;89(1058):20150463e]. Br J Radiol. 2015;88(1054):20150463.\u003c/li\u003e\n \u003cli\u003eKim 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 Jan;22(1):131-138.\u003c/li\u003e\n \u003cli\u003eMileto A, Guimaraes LS, McCollough CH, Fletcher JG, Yu L. State of the Art in Abdominal CT: The Limits of Iterative Reconstruction Algorithms. Radiology. 2019;293(3):491-503.\u003c/li\u003e\n \u003cli\u003eJensen CT, Liu X, Tamm EP, et al. Image Quality Assessment of Abdominal CT by Use of New Deep Learning Image Reconstruction: Initial Experience. AJR Am J Roentgenol. 2020;215(1):50-57.\u003c/li\u003e\n \u003cli\u003eSun J, Li H, Wang B, et al. Application of a deep learning image reconstruction (DLIR) algorithm in head CT imaging for children to improve image quality and lesion detection. BMC Med Imaging. 2021;21(1):108. Published 2021 Jul 8.\u003c/li\u003e\n \u003cli\u003eLi Y, Liu X, Zhuang XH, Wang MJ, Song XF. Assessment of low-dose paranasal sinus CT imaging using a new deep learning image reconstruction technique in children compared to adaptive statistical iterative reconstruction V (ASiR-V). BMC Med Imaging. 2022;22(1):106. Published 2022 Jun 3.\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"624\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"10\" style=\"width: 624px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable. 1\u003c/strong\u003e Evaluation of images quality with different groups（\u003cem\u003e\u003cruby\u003ex\u003crp\u003e(\u003c/rp\u003e\n \u003crt\u003e_\u003c/rt\u003e\n \u003crp\u003e)\u003c/rp\u003e \u0026nbsp;\u0026nbsp;\n \u003c/ruby\u003e\u003c/em\u003e\u0026plusmn;\u003cem\u003es\u003c/em\u003e）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 59px;\"\u003e\n \u003cp\u003eGroup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"8\" style=\"width: 489px;\"\u003e\n \u003cp\u003eObjective evaluation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003eSubjective evaluation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003evitreous body CT value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003evitreous body noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003evitreous body SNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eorbital fat CT value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eorbital fat noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eorbital fat SNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eimage noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eCNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003eImage subjective score of Orbital level\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eDL-H\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e8.05\u0026plusmn;3.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e9.86\u0026plusmn;1.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e2.04\u0026plusmn;1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-98.53\u0026plusmn;7.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e12.62\u0026plusmn;2.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e24.56\u0026plusmn;5.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e4.17\u0026plusmn;0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e26.60\u0026plusmn;5.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e1.30\u0026plusmn;0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eDL-M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e8.29\u0026plusmn;3.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e12.25\u0026plusmn;1.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e1.46\u0026plusmn;0.69\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-100.11\u0026plusmn;7.09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e15.05\u0026plusmn;2.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e17.33\u0026plusmn;3.30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e5.98\u0026plusmn;1.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e18.79\u0026plusmn;3.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e1.88\u0026plusmn;0.56\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eDL-L\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e8.97\u0026plusmn;3.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e14.67\u0026plusmn;2.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e1.26\u0026plusmn;0.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-96.07\u0026plusmn;15.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e15.74\u0026plusmn;3.31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e13.60\u0026plusmn;3.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e7.43\u0026plusmn;1.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e14.86\u0026plusmn;4.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e2.68\u0026plusmn;0.65\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eASiR-V 70%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e9.37\u0026plusmn;2.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e8.14\u0026plusmn;1.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e1.18\u0026plusmn;0.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-103.27\u0026plusmn;7.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e10.88\u0026plusmn;2.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e10.08\u0026plusmn;2.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e7.14\u0026plusmn;1.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e16.14\u0026plusmn;2.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e2.18\u0026plusmn;0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eF\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e0.952\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e59.120\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e5.587\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e2.495\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e16.951\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e64.355\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e51.534\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e27.978\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e86.929 (\u0026chi;\u0026sup2;\u0026nbsp;value)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＞0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＞0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"624\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"10\" style=\"width: 624px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable. 2\u003c/strong\u003e Comparison of images quality with DL-H and ASiR-V 70%（\u003cem\u003e\u003cruby\u003ex\u003crp\u003e(\u003c/rp\u003e\n \u003crt\u003e_\u003c/rt\u003e\n \u003crp\u003e)\u003c/rp\u003e \u0026nbsp;\u0026nbsp;\n \u003c/ruby\u003e\u003c/em\u003e\u0026plusmn;\u003cem\u003es\u003c/em\u003e）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 59px;\"\u003e\n \u003cp\u003eGroup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"8\" style=\"width: 489px;\"\u003e\n \u003cp\u003eObjective evaluation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003eSubjective evaluation\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003evitreous body CT value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003evitreous body noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003evitreous body SNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eorbital fat CT value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eorbital fat noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eorbital fat SNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eimage noise\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003eCNR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003eImage subjective score of Orbital level\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eDL-H\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e8.05\u0026plusmn;3.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e9.86\u0026plusmn;1.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e2.04\u0026plusmn;1.00\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-98.53\u0026plusmn;7.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e12.62\u0026plusmn;2.24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e24.56\u0026plusmn;5.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e4.17\u0026plusmn;0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e26.60\u0026plusmn;5.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e1.30\u0026plusmn;0.51\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eASiR-V 70%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e9.37\u0026plusmn;2.74\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e8.14\u0026plusmn;1.18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e1.18\u0026plusmn;0.40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e-103.27\u0026plusmn;7.11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e10.88\u0026plusmn;2.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e10.08\u0026plusmn;2.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e7.14\u0026plusmn;1.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e16.14\u0026plusmn;2.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 77px;\"\u003e\n \u003cp\u003e2.18\u0026plusmn;0.66\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 59px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＞0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＞0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 61px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp; \u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"385\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"3\" style=\"width: 385px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTable. 3\u0026nbsp;\u003c/strong\u003eComparison of radiation dose（\u003cem\u003e\u003cruby\u003ex\u003crp\u003e(\u003c/rp\u003e\n \u003crt\u003e_\u003c/rt\u003e\n \u003crp\u003e)\u003c/rp\u003e \u0026nbsp;\u0026nbsp;\n \u003c/ruby\u003e\u003c/em\u003e\u0026plusmn;\u003cem\u003es\u003c/em\u003e）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 56px;\"\u003e\n \u003cp\u003eGroup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 196px;\"\u003e\n \u003cp\u003eDLP（mGy*cm）\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 134px;\"\u003e\n \u003cp\u003eED（mSv）\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 56px;\"\u003e\n \u003cp\u003eA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 196px;\"\u003e\n \u003cp\u003e36.65\u0026plusmn;2.44\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 134px;\"\u003e\n \u003cp\u003e0.17\u0026plusmn;0.03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 56px;\"\u003e\n \u003cp\u003eB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 196px;\"\u003e\n \u003cp\u003e227.92\u0026plusmn;27.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 134px;\"\u003e\n \u003cp\u003e1.01\u0026plusmn;0.18\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 56px;\"\u003e\n \u003cp\u003et\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 196px;\"\u003e\n \u003cp\u003e-34.56\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 134px;\"\u003e\n \u003cp\u003e-23.00\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 56px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 196px;\"\u003e\n \u003cp\u003eP＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 134px;\"\u003e\n \u003cp\u003eP＜0.05\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Deep learning, iterative reconstruction, orbital, CT, children","lastPublishedDoi":"10.21203/rs.3.rs-6243177/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6243177/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003ePurpose:\u003c/strong\u003e To evaluate the performance of a new deep learning image reconstruction (DLIR) algorithm in pediatric orbital ultra-low-dose CT by sinus CT for improving image quality and reducing radiation dose.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eA retrospective analysis of 50 children who underwent paranasal sinus or orbital CT examination at our hospital was performed. ultra-low-dose CT of the nasal sinus was used to simulate orbital scanning as the experimental group and compared with the control group of general-low dose orbital CT. Two groups scanning used 100 kVp tube voltage, automatic tube current (SmartmA, 60-250 mA), slice thickness 1.25 mm, interslice spacing 1.25 mm, and pitch 0.992:1 with rotation time 1.00 s. The experimental group configured a Noise Index of 22 and a pre-ASiR-V weight of 70%, while the control group set a Noise Index of 8 and a pre-ASiR-V weight of 30%. The radiation dose, image noise (SD), signal-to-noise ratio (SNR), contrast noise ratio (CNR), and subjective scores of both groups were compared. The CT dose index (CTDI) and dose-length product (DLP) values were documented, and the effective dose (ED) was subsequently calculated. The results indicated that the experimental group experienced reductions of 83.92% in DLP and 83.17% in ED compared to the control group.Compared between DL-H (the optimal set of DLIR) and ASiR-V 70% (the control group), the image noise was reduced by 41.6% (4.17±0.86 HU vs. 7.14±1.19 HU), and the CNR was improved by 39.3% (26.60±5.61 vs. 16.14±2.48). In the qualitative image assessment, DL-H obtained the highest score.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003e Combined with the comparison results of quantitative and qualitative image assessment, DL-H under the ultra-low-dose scan was significantly better than ASiR-V 70% under the general-low dose scan, and ASiR-V 70% was better than DL-M and DL-L. We can infer that the use of DL-H may be able to save a significant amount of radiation dose while achieving the current image quality.\u003c/p\u003e","manuscriptTitle":"Application of a New Deep Learning Image Reconstruction (DLIR) Algorithm in Pediatric Orbital horizontal level Ultra-Low Dose CT scan: A Pilot Study Based on Image Quality","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-08 20:31:30","doi":"10.21203/rs.3.rs-6243177/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"2c908ee8-c17d-4524-b45f-edf9bff9cac5","owner":[],"postedDate":"May 8th, 2025","published":true,"recentEditorialEvents":[{"type":"decision","content":"Rejected","date":"2026-05-12T09:33:03+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-30T04:58:51+00:00","index":97,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-12T10:54:19+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-08 20:31:30","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6243177","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6243177","identity":"rs-6243177","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}