Abstract
Purpose
To compare the image quality and diagnostic capability in detecting malignant liver tumors of low-dose CT (LDCT, 33% dose) with deep learning–based denoising (DLD) and standard-dose CT (SDCT, 100% dose) with model-based iterative reconstruction (MBIR).
Materials and Methods
In this prospective, multicenter, noninferiority study, individuals referred for liver CT scans were enrolled from three tertiary referral hospitals between February 2021 and August 2022. All liver CT scans were conducted using a dual-source scanner with the dose split into tubes A (67% dose) and B (33% dose). Blended images from tubes A and B were created using MBIR to produce SDCT images, whereas LDCT images used data from tube B and were reconstructed with DLD. The noise in liver images was measured and compared between imaging techniques. The diagnostic performance of each technique in detecting malignant liver tumors was evaluated by three independent radiologists using jackknife alternative free-response receiver operating characteristic analysis. Noninferiority of LDCT compared with SDCT was declared when the lower limit of the 95% CI for the difference in figure of merit (FOM) was greater than −0.10.
Results
A total of 296 participants (196 men, 100 women; mean age, 60.5 years ± 13.3 [SD]) were included. The mean noise level in the liver was significantly lower for LDCT (10.1) compared with SDCT (10.7) (P < .001). Diagnostic performance was assessed in 246 participants (108 malignant tumors in 90 participants). The reader-averaged FOM was 0.880 for SDCT and 0.875 for LDCT (P = .35). The difference fell within the noninferiority margin (difference, −0.005 [95% CI: −0.024, 0.012]).
Conclusion
Compared with SDCT with MBIR, LDCT using 33% of the standard radiation dose had reduced image noise and comparable diagnostic performance in detecting malignant liver tumors.
Keywords: CT, Abdomen/GI, Liver, Comparative Studies, Diagnosis, Reconstruction Algorithms
Clinical trial registration no. NCT05804799
© RSNA, 2024
Keywords: CT, Abdomen/GI, Liver, Comparative Studies, Diagnosis, Reconstruction Algorithms
Summary
Compared with standard-dose liver CT with model-based iterative reconstruction, low-dose liver CT with deep learning–based denoising reconstruction significantly reduced image noise and showed noninferior diagnostic performance in detecting malignant liver tumors.
Key Points
■ Low-dose CT (LDCT, 33% dose) with deep learning–based denoising (DLD) provided significantly lower noise level in the liver compared with standard-dose CT (SDCT, 100% dose) with model-based iterative reconstruction (MBIR) (mean, 10.1 vs 10.7; P < .001).
■ The reader-averaged figure of merit for detecting malignant liver tumors, evaluated using jackknife alternative free-response receiver operating characteristic analysis, showed that LDCT with DLD was noninferior to SDCT with MBIR (0.875 for LDCT vs 0.880 for SDCT [95% CI of the difference: −0.024, 0.012], P = .35).
Introduction
Contrast-enhanced multiphasic liver CT has been an essential imaging modality for detecting and characterizing liver tumors. In patients with cirrhosis, it is widely accepted as a noninvasive diagnostic tool for hepatocellular carcinoma (HCC) in major society guidelines (1). Liver CT is often used for repeated examination of patients with HCC or liver metastasis to monitor the response to treatment. However, over the past decades, the rapid increase in CT examinations has raised concerns about radiation exposure (2). To minimize the risk of radiation-induced cancers, CT scans should use the lowest possible radiation dose (3–6). Meanwhile, reducing radiation dose leads to increased image noise, which can degrade CT image quality. Therefore, balancing the need for reduced radiation exposure with the diagnostic effectiveness of CT scans is a substantial challenge.
In this regard, the model-based iterative reconstruction (MBIR) method has been introduced to reduce CT image noise. Previous studies reported that MBIR is effective in reducing image noise and improving image quality (7–10). More recently, there have been notable technical advances in artificial intelligence, and deep learning–based reconstruction has emerged as an effective method for noise reduction and image quality improvement (11–15). Unlike MBIR, which relies on manually designed prior functions for noise reduction and preservation of internal structures, deep learning methods can take advantage of more complex functions, allowing for more effective noise reduction even in low-dose CT (LDCT) examinations (12,13,16,17). A retrospective study by Park et al (18) found that LDCT of the liver, with only 33% of the standard radiation dose, plus deep learning–based denoising (DLD) reconstruction, could provide comparable diagnostic performance in detecting malignant liver tumors to standard-dose CT (SDCT) with MBIR. However, despite these promising results, Jensen et al (19) recently reported that LDCT with a 65% radiation dose reduction and DLD demonstrated an overall inferior characterization of focal liver lesions and lower reader confidence compared with SDCT with filtered back projection reconstruction, especially for small tumors less than 5 mm in size.
Given the controversial findings from prior studies, it is imperative to conduct a prospective multicenter study with a robust design to determine the equivalence of LDCT with DLD in terms of image quality and diagnostic performance in detecting malignant liver tumors compared with SDCT. We hypothesized that LDCT with DLD can provide image quality and diagnostic performance in detecting malignant liver tumors comparable to those achieved with SDCT with MBIR. The aim of this prospective, multicenter, noninferiority trial was to compare the image quality and diagnostic accuracy between LDCT with DLD and SDCT with MBIR, using a predefined noninferiority margin.
Materials and Methods
Study Design and Participants
This study was a prospective multicenter noninferiority trial involving two university-affiliated tertiary referral hospitals in South Korea (Seoul National University Hospital and Korea University Guro Hospital) and one in Germany (Tübingen University Hospital). This study was approved by the institutional review board of each participating center, and all participating individuals provided written informed consent. The study (ClinicalTrials.gov study identifier no. NCT05804799) was conducted in accordance with the principles of the Helsinki Declaration. Appendix S1 contains the study protocol, including the calculation of the required sample size. The primary variable assessed was the noise level in liver images, while the secondary variable was the diagnostic performance for detecting malignant liver tumors. Between February 2021 and August 2022, all three participating centers prospectively enrolled individuals who were suspected of having focal liver lesions. The inclusion criteria for this study were the following: (a) age between 20 and 85 years old and (b) patients referred to the radiology department for contrast-enhanced liver CT due to suspicion of focal liver lesions. The exclusion criteria for this study were as follows: (a) patients with an estimated glomerular filtration rate of less than 60 mL/min/1.73 m2 and (b) patients with a previous history of severe adverse reactions to iodinated contrast media.
Acquisition and Reconstruction of Contrast-enhanced Liver CT Images
All participants underwent a contrast-enhanced liver CT scan using a 192-channel, third-generation dual-source CT scanner (SOMATOM Force; Siemens Healthineers). The scanner's dual-source mode divided the standard radiation dose into two: tube A (reference: 250 mAs, 67% of dose) and tube B (reference: 125 mAs, 33% of dose). Blended images from tubes A and B were used to create the standard dose for liver CT scans. The detailed liver CT protocol is available in Appendix S1. The dose-length product was recorded for each liver CT scan and converted to an effective dose using an organ-weighted factor (k = 0.0151 mSv ∙ m Gy−1 ∙ cm−1) (18).
After the acquisition of the CT images, SDCT images were created using the MBIR method (Advanced Modeled Iterative Reconstruction, or ADMIRE, level 3; Siemens Healthineers). LDCT images were reconstructed using data from only tube B and a commercially available, vendor-agnostic deep learning–based reconstruction technique (ClariCT.AI; ClariPI). As a result, LDCT used 33% of the radiation dose of SDCT. The detailed method of DLD-based image reconstruction is provided in Appendix S1. In brief, the ClariCT.AI method is a DLD method based on a U-Net convolutional neural network. It was trained using a CT image with noise addition as the input and an original CT image without noise addition as the output.
Evaluation of Focal Liver Lesions and Lesion Confirmation
Among all enrolled participants, focal liver lesions were evaluated for those meeting the following criteria: (a) confirmed focal liver lesions through either histopathology or clinical diagnosis and (b) five or fewer focal liver lesions. In this study, histopathologic examination and clinical diagnosis were used to diagnose malignant liver tumors. Whenever surgical resection or liver biopsy was performed for focal liver lesions, the final diagnosis was confirmed through histopathology. The imaging diagnosis for HCC was established based on the guidelines set forth by the Korean Liver Cancer Association–National Cancer Center (20,21). For participants with recognized primary malignancies exhibiting liver metastasis, the imaging diagnosis was based on either observation of tumor growth at follow-up imaging or lesions showing typical imaging features of metastasis with at least two different imaging modalities (22–24). For all other malignant liver tumors excluding HCC and metastases, histopathology served as the definitive diagnostic method.
Image Analysis
Primary variable: quantitative analysis of image noise.— One radiologist (D.H.L., with 17 years of experience in liver imaging) measured the noise level indicated by the SD of CT attenuation values at several sites, including the liver parenchyma, spleen, paraspinal muscles, anterior abdominal fat, main portal vein, and inferior vena cava. To measure the noise level, circular or ovoid-shaped regions of interest were placed on both the SDCT and LDCT image sets. For each site, two regions of interest were placed in homogeneous areas, and the average value of SDs of the two regions of interest was recorded for further analysis. In addition, 60 participants were randomly selected to calculate the structural similarity index measure between SDCT with MBIR and LDCT with DLD. The process of structural similarity index measure calculation is provided in Appendix S1.
Secondary variable: diagnostic performance in detection of malignant liver tumors.— Three board-certified abdominal radiologists (Bo Yun Hur, MD, with 15 years of experience, Eun Sun Lee, MD, with 16 years of experience, and Seungchul Han, MD, with 6 years of experience in liver imaging) independently evaluated both SDCT with MBIR and LDCT with DLD images. The readers were aware that participants might have malignant liver tumors, and clinical history, including information regarding the presence of risk factors for HCC or history of extrahepatic primary malignancy, was provided. However, the readers were blinded to detailed information regarding the final diagnosis of liver lesions. To minimize recall bias, at least a 2-week interval was held between two image interpretation sessions, and the two image sets were presented in random order.
Each reader was asked to identify and record any potentially malignant focal liver lesions. When potentially malignant liver tumors were found, the readers recorded the size and location of each lesion and assigned a probability of malignancy using a five-point confidence scale: 1 = definitely benign, 2 = probably benign, 3 = indeterminate, 4 = probably malignant, and 5 = definitely malignant. Lesions with scores of 4 or 5 were classified as malignant liver tumors. See Appendix S1 for detailed information regarding the assessment of malignant liver tumors.
Statistical Analysis
All continuous variables are presented as means ± SDs, and categorical variables are presented as percentages. The primary outcome of this study was the noise level in the liver. To compare the noise level between SDCT and LDCT at each site, we used the paired t test. We also performed jackknife alternative free-response receiver operating characteristic (JAFROC) analysis to analyze and compare the diagnostic performance in detecting malignant liver tumors between SDCT with MBIR and LDCT with DLD (JAFROC version 4.1; http://www.devchakraborty.com/) (25). The figure of merit (FOM), defined as the probability of rating lesions (including unmarked lesions) higher than nonlesion marks on control images (18), was calculated for each image set. To compare the FOM between the two image sets, the F statistic test was used. Noninferiority of LDCT compared with SDCT was determined when the lower limit for the 95% CI of the difference in FOM between the two methods in detecting malignant liver tumors was greater than −0.10 (26–28). The sensitivities of SDCT and LDCT in detecting malignant liver tumors on a per-lesion basis were calculated and compared using the generalized estimating equation method to handle correlated observations. The sensitivities of SDCT and LDCT for detecting malignant liver tumors were recalculated based on lesion size (ie, equal to or smaller than 1 cm and larger than 1 cm), type of malignant liver tumors (ie, HCC and non-HCC malignancies, including metastasis and cholangiocarcinoma), and vascularity of liver tumors (ie, hypervascular and hypovascular) and were compared using the generalized estimating equation method. In addition, per-patient sensitivity and specificity of SDCT and LDCT in detecting malignant liver tumors were also assessed and compared using the McNemar test. Interreader agreement in assessing focal liver lesions was evaluated using Fleiss κ statistics: poor (<0.20), fair (0.20–0.39), moderate (0.40–0.59), good (0.60–0.79), and excellent (0.80–1.00) agreements. Statistical analyses were performed using SPSS software (version 27.0.0; IBM) and MedCalc program (version 20; MedCalc). A P value less than .05 was considered statistically significant.
Results
Participant Characteristics
Our study initially evaluated 300 individuals who underwent a contrast-enhanced liver CT scan in the radiology department. Among them, four individuals were excluded from this study; two refused to participate, and two withdrew informed consent after the acquisition of the scan. Noise level was measured at both SDCT and LDCT in all 296 enrolled participants (male to female ratio = 196:100; mean age, 60.5 years ± 13.3). Table 1 summarizes the baseline characteristics of study participants, and Figure 1 presents the participant enrollment process.
Table 1:
Participant Characteristics
Figure 1:
Participant enrollment process.
Radiation Dose for Contrast-enhanced Liver CT
The mean dose-length product of SDCT was 801.0 mGy ∙ cm ± 247.3, and the mean estimated effective radiation dose was 12.1 mSv ± 4.0. As a result, the estimated median effective radiation dose of LDCT was 4.0 mSv, which is 33% of the SDCT dose.
Primary Variable: Quantitative Analysis of Image Noise
Table 2 and Figure S1 summarize the noise levels measured in images at each site at both SDCT with MBIR and LDCT with DLD. The mean noise level in the liver at LDCT with DLD was 10.1, which was significantly lower than the mean noise level of 10.7 at SDCT with MBIR (P < .001). The 95% CI of the difference in noise level in the liver between LDCT with DLD and SDCT with MBIR was −0.79 to −0.37. LDCT with DLD also exhibited significantly lower noise levels than SDCT with MBIR in the spleen (10.4 vs 10.9, P < .001) and main portal vein (11.8 vs 12.1, P = .03). However, there was no evidence of a difference in noise level observed in the anterior abdominal fat, paraspinal muscle, and inferior vena cava between LDCT with DLD and SDCT with MBIR. The mean structural similarity index measure was 0.89 ± 0.02.
Table 2:
Noise Level at Each Site at SDCT with MBIR and LDCT with DLD Reconstruction
Secondary Variable: Diagnostic Performance in Detecting Malignant Liver Tumors
Among the 296 participants enrolled in this study, 50 were excluded from the analysis of malignant liver tumors due to the following reasons: (a) presence of more than five lesions in the liver (n = 37) and (b) absence of a final diagnosis (ie, focal liver lesions were not confirmed by either histopathologic examination or clinical diagnosis) (n = 13). Consequently, the diagnostic performance of SDCT with MBIR and LDCT with DLD in detecting malignant liver tumors was evaluated in 246 participants. Among those evaluated, 108 malignant liver tumors were found in 90 participants. Information regarding the malignant liver tumors, such as size, number, diagnosis method, and type of tumors, are summarized in Table 1.
The reader-averaged FOM determined by JAFROC analysis revealed noninferior results between SDCT with MBIR and LDCT with DLD (Table 3). The FOM was 0.880 for SDCT with MBIR and 0.875 for LDLT with DLD. There was no evidence of a difference in FOM between the two image sets (−0.005 [95% CI: −0.024, 0.012], P = .35).
Table 3:
Diagnostic Performance in Detecting Malignant Liver Tumors Determined by JAFROC Analysis
Table 4 summarizes the sensitivity of both SDCT with MBIR and LDCT with DLD in detecting malignant liver tumors. There was no evidence of a difference in sensitivity between the two techniques for all three readers (P > .05). Furthermore, there was no evidence of a difference in sensitivity between the image sets for lesion size (size equal to or smaller than 1 cm and size greater than 1 cm, both P > .05) or detection of HCCs or non-HCC malignancies, including metastasis and cholangiocarcinoma (P > .05) (Figs 2, 3). In fact, SDCT and LDCT showed similar sensitivity for the detection of both hypervascular and hypovascular malignant liver tumors. In terms of per-patient sensitivity and specificity, no significant disparities were identified between SDCT and LDCT in the detection of malignant liver tumors across all three readers (as detailed in Table S2).
Table 4:
Sensitivity for Detecting Malignant Liver Tumors
Figure 2:
Images in a 52-year-old male participant with chronic hepatitis B viral infection. An arterial phase image shows a 3.3-cm hypervascular hepatocellular carcinoma (HCC) in the right lobe of the liver. This mass shows washout on the portal venous phase image. Surgical resection was performed, and the diagnosis of HCC was made with histopathologic examination. The images shown are (A) standard-dose arterial phase image created with model-based iterative reconstruction (MBIR), (B) standard-dose portal venous phase image created with MBIR, (C) low-dose arterial phase image reconstructed by deep learning-based denoising (DLD), and (D) low-dose portal venous phase image reconstructed by DLD.
Figure 3:
Images in a 48-year-old male participant who underwent anterior resection for sigmoid colon cancer. Contrast-enhanced CT images show a 9-mm peripherally enhancing low-attenuation nodular lesion in segment IV of the liver (arrow). Surgical resection was performed, and the diagnosis of metastasis from colon cancer was made with histopathologic examination. (A) Standard-dose arterial phase image created with model-based iterative reconstruction (MBIR), (B) standard-dose portal venous phase image created with MBIR, (C) low-dose arterial phase image reconstructed by deep learning–based denoising (DLD), and (D) low-dose portal venous phase image reconstructed by DLD.
Regarding the interreader agreement in assessing focal liver lesions, the κ value was 0.726–0.785 for SDCT with MBIR and 0.709–0.774 for LDCT with DLD, respectively, indicating good agreement. The detailed κ values are provided in Table S3.
Discussion
In this prospective multicenter noninferiority trial, LDCT using 33% of the standard radiation dose with DLD reconstruction provided a significantly lower noise level in the liver parenchyma compared with SDCT with MBIR (P < .001). Moreover, the diagnostic performance of LDCT with DLD in detecting malignant liver tumors was not inferior to that of SDCT with MBIR, as the difference between the two techniques did not cross the predefined margin for noninferiority (the difference in FOM: −0.005 [95% CI: −0.024, 0.012]). Based on these results, we cautiously conclude that LDCT with DLD provides comparable diagnostic capability to detect malignant liver tumors as SDCT with MBIR while offering a lower noise level in the liver. Our results are in line with those of a previous study by Park et al (18).
Prior to the development of the DLD technique, MBIR was commonly used to improve CT image quality by reducing image noise (7–10). MBIR was also applied to low-dose abdominal CT scans, and Choi et al (29) reported that LDCT using 67% of the radiation dose with MBIR could provide comparable image quality and diagnostic performance for the detection of focal lesions to that of SDCT. However, our study results contrasted with these previous findings. In the study by Choi et al, LDCT with 33% of the radiation dose (equivalent to the radiation dose of LDCT used in this study) with MBIR resulted in reduced sensitivity in detecting focal lesions compared with SDCT, likely due to blurred lesion conspicuity with artifacts resulting from a markedly reduced radiation dose. The authors concluded that LDCT with MBIR using 33% of the SDCT dose may not be appropriate for evaluating focal liver lesions, including metastases (29).
In contrast, our study found that LDCT, using 33% of the standard dose, with DLD provided noninferior diagnostic performance in detecting malignant liver tumors compared with SDCT with MBIR. This discrepancy between the results of our study and the previous study may be due to the fact that more complex functions can be implemented in DLD algorithms, enabling more efficient and effective reduction of image noise while maintaining the noise spectrum and image texture compared with MBIR, which utilizes manually designed prior functions. Moreover, DLD preserves image quality and diagnostic capability to evaluate focal liver lesions, outperforming the MBIR method. Thus, the use of DLD algorithms can allow for the accurate detection of small, low-contrast liver lesions at decreased radiation doses. Lee et al (30) reported that DLD provided significantly better image quality and diagnostic acceptability for the detection of liver metastases than MBIR. Several previous studies have also reported that DLD could reduce noise level and improve image quality compared with the filtered back projection method and MBIR (14,17,30–34).
However, a recent study by Jensen et al (19) reported that LDCT with a 65% radiation dose reduction and DLD reconstruction may be less effective than SDCT with filtered back projection reconstruction in detecting liver metastases, especially for metastatic liver tumors that are smaller than 5 mm in size. In our study, LDCT using 33% of the radiation dose with DLD provided comparable diagnostic performance in detecting malignant liver tumors to SDCT with MBIR, even for tumors that were 1 cm or smaller. This discrepancy between our study and the previous study by Jensen et al may be attributed to several factors. First, it is worth noting that the median tumor size in our study was 2.6 cm, which appears to be larger than the size of 0.7 cm in the aforementioned study (19). Second, our study included individuals with suspected primary malignant liver cancers and metastasis, while the previous study only included patients with suspected liver metastases. Last, the DLD software program used was different between the two studies. In our study, we used a vendor-agnostic DLD program, while the previous study by Jensen et al used a vendor-specific DLD algorithm. Furthermore, a recent study (35) showed that the vendor-agnostic DLD program used in our study demonstrated better overall image quality with fewer artifacts due to the preservation of the noise spectrum compared with the vendor-specific DLD technique used in the previous study by Jensen et al.
Our study had several limitations. First, because of the study design, where radiation dose was split into two simultaneously operating x-ray tubes, we used only one specific CT scanner from one vendor. Therefore, our study results need to be validated in future studies, including various types of CT scanners from various vendors. However, we used vendor-agnostic DLD programs, and several previous studies using the same DLD programs but different CT scanners from other vendors have also reported promising results in noise reduction with improved image quality (30,35–37). Second, 10% of malignant liver tumors were clinically diagnosed, which could have affected the readers’ performance.
In conclusion, the use of DLD reconstruction allowed for a 67% reduction in liver CT radiation dose while providing significantly lower noise level in the liver compared with SDCT with MBIR. Furthermore, the diagnostic performance of LDCT with DLD in detecting malignant liver tumors was found to be comparable to that of SDCT, with the difference between techniques not crossing the predefined noninferiority margin, thus demonstrating the clinical feasibility of using LDCT with DLD.
Supported by the Korea Medical Device Development Fund grant funded by the Korean government (the Ministry of Science and ICT; the Ministry of Trade, Industry and Energy; the Ministry of Health & Welfare; the Ministry of Food and Drug Safety; fund no. RS-2020-KD000226, 1711174549). The study design, participant enrollment, examinations, data analyses, and interpretations were independently conducted by the authors.
Data sharing: Data generated or analyzed during the study are available from the corresponding author by request.
Disclosures of conflicts of interest: D.H.L. Grant from the Korea Medical Device Development Fund funded by the Korean government (the Ministry of Science and ICT; the Ministry of Trade, Industry and Energy; the Ministry of Health & Welfare; the Ministry of Food and Drug Safety, fund no. RS2020-KD000226, 1711174549), the study design, participant enrollment, examinations, data analyses, and interpretations were independently conducted by the authors; research grants from Canon Medical Systems. J.M.L. Grant from Korea Medical Device Development Fund funded by the Korean government (the Ministry of Science and ICT; the Ministry of Trade, Industry and Energy; the Ministry of Health & Welfare; the Ministry of Food and Drug Safety, fund no. RS2020-KD000226, 1711174549); grants/contracts from Bayer Healthcare, Canon Health, Central Medical Service, GE HealthCare, Guerbet, Samsung Medison, Dongkook Lifescience, StarMed, Bracco, RF Medical, Clarify, and Medical IP; payment or honoraria from Samsung Medison, Philips, GE HealthCare, Bayer, Guerbet, and Clarify; associate editor for Radiology. C.H.L. No relevant relationships. S.A. No relevant relationships. A.O. No relevant relationships.
Abbreviations:
- DLD
- deep learning–based denoising
- FOM
- figure of merit
- HCC
- hepatocellular carcinoma
- JAFROC
- jackknife alternative free-response receiver operating characteristic
- LDCT
- low-dose CT
- MBIR
- model-based iterative reconstruction
- SDCT
- standard-dose CT
References
- 1. Kim TH , Kim SY , Tang A , Lee JM . Comparison of international guidelines for noninvasive diagnosis of hepatocellular carcinoma: 2018 update . Clin Mol Hepatol 2019. ; 25 ( 3 ): 245 – 263 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Brenner DJ . Should we be concerned about the rapid increase in CT usage? Rev Environ Health 2010. ; 25 ( 1 ): 63 – 68 . [DOI] [PubMed] [Google Scholar]
- 3. Brenner DJ , Hall EJ . Computed tomography--an increasing source of radiation exposure . N Engl J Med 2007. ; 357 ( 22 ): 2277 – 2284 . [DOI] [PubMed] [Google Scholar]
- 4. Mahesh M . Medical radiation exposure with focus on CT . Rev Environ Health 2010. ; 25 ( 1 ): 69 – 74 . [DOI] [PubMed] [Google Scholar]
- 5. Pearce MS , Salotti JA , Little MP , et al . Radiation exposure from CT scans in childhood and subsequent risk of leukaemia and brain tumours: a retrospective cohort study . Lancet 2012. ; 380 ( 9840 ): 499 – 505 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Mathews JD , Forsythe AV , Brady Z , et al . Cancer risk in 680,000 people exposed to computed tomography scans in childhood or adolescence: data linkage study of 11 million Australians . BMJ 2013. ; 346 : f2360 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Brady SL , Yee BS , Kaufman RA . Characterization of adaptive statistical iterative reconstruction algorithm for dose reduction in CT: A pediatric oncology perspective . Med Phys 2012. ; 39 ( 9 ): 5520 – 5531 . [DOI] [PubMed] [Google Scholar]
- 8. Gramer BM , Muenzel D , Leber V , et al . Impact of iterative reconstruction on CNR and SNR in dynamic myocardial perfusion imaging in an animal model . Eur Radiol 2012. ; 22 ( 12 ): 2654 – 2661 . [DOI] [PubMed] [Google Scholar]
- 9. Smith EA , Dillman JR , Goodsitt MM , Christodoulou EG , Keshavarzi N , Strouse PJ . Model-based iterative reconstruction: effect on patient radiation dose and image quality in pediatric body CT . Radiology 2014. ; 270 ( 2 ): 526 – 534 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Kataria B , Althén JN , Smedby Ö , Persson A , Sökjer H , Sandborg M . Assessment of image quality in abdominal CT: potential dose reduction with model-based iterative reconstruction . Eur Radiol 2018. ; 28 ( 6 ): 2464 – 2473 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Chen H , Zhang Y , Kalra MK , et al . Low-Dose CT With a Residual Encoder-Decoder Convolutional Neural Network . IEEE Trans Med Imaging 2017. ; 36 ( 12 ): 2524 – 2535 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Akagi M , Nakamura Y , Higaki T , et al . Deep learning reconstruction improves image quality of abdominal ultra-high-resolution CT . Eur Radiol 2019. ; 29 ( 11 ): 6163 – 6171 . [Published correction appears in Eur Radiol 2019;29(8):4526-4527.] [DOI] [PubMed] [Google Scholar]
- 13. Greffier J , Hamard A , Pereira F , et al . Image quality and dose reduction opportunity of deep learning image reconstruction algorithm for CT: a phantom study . Eur Radiol 2020. ; 30 ( 7 ): 3951 – 3959 . [DOI] [PubMed] [Google Scholar]
- 14. 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 . [DOI] [PubMed] [Google Scholar]
- 15. Singh R , Digumarthy SR , Muse VV , et al . Image Quality and Lesion Detection on Deep Learning Reconstruction and Iterative Reconstruction of Submillisievert Chest and Abdominal CT . AJR Am J Roentgenol 2020. ; 214 ( 3 ): 566 – 573 . [DOI] [PubMed] [Google Scholar]
- 16. Kang E , Min J , Ye JC . A deep convolutional neural network using directional wavelets for low-dose X-ray CT reconstruction . Med Phys 2017. ; 44 ( 10 ): e360 – e375 . [DOI] [PubMed] [Google Scholar]
- 17. Brady SL , Trout AT , Somasundaram E , Anton CG , Li Y , Dillman JR . Improving Image Quality and Reducing Radiation Dose for Pediatric CT by Using Deep Learning Reconstruction . Radiology 2021. ; 298 ( 1 ): 180 – 188 . [DOI] [PubMed] [Google Scholar]
- 18. Park S , Yoon JH , Joo I , et al . Image quality in liver CT: low-dose deep learning vs standard-dose model-based iterative reconstructions . Eur Radiol 2022. ; 32 ( 5 ): 2865 – 2874 . [DOI] [PubMed] [Google Scholar]
- 19. Jensen CT , Gupta S , Saleh MM , et al . Reduced-Dose Deep Learning Reconstruction for Abdominal CT of Liver Metastases . Radiology 2022. ; 303 ( 1 ): 90 – 98 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Korean Liver Cancer Association (KLCA) and National Cancer Center (NCC) Korea . 2022 KLCA-NCC Korea practice guidelines for the management of hepatocellular carcinoma . Clin Mol Hepatol 2022. ; 28 ( 4 ): 583 – 705 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Korean Liver Cancer Association (KLCA) and National Cancer Center (NCC) Korea . 2022 KLCA-NCC Korea Practice Guidelines for the Management of Hepatocellular Carcinoma . Korean J Radiol 2022. ; 23 ( 12 ): 1126 – 1240 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Danet IM , Semelka RC , Leonardou P , et al . Spectrum of MRI appearances of untreated metastases of the liver . AJR Am J Roentgenol 2003. ; 181 ( 3 ): 809 – 817 . [DOI] [PubMed] [Google Scholar]
- 23. Ba-Ssalamah A , Uffmann M , Saini S , Bastati N , Herold C , Schima W . Clinical value of MRI liver-specific contrast agents: a tailored examination for a confident non-invasive diagnosis of focal liver lesions . Eur Radiol 2009. ; 19 ( 2 ): 342 – 357 . [DOI] [PubMed] [Google Scholar]
- 24. Lee DH , Lee JM , Hur BY , et al . Colorectal Cancer Liver Metastases: Diagnostic Performance and Prognostic Value of PET/MR Imaging . Radiology 2016. ; 280 ( 3 ): 782 – 792 . [DOI] [PubMed] [Google Scholar]
- 25. Chakraborty DP . Analysis of location specific observer performance data: validated extensions of the jackknife free-response (JAFROC) method . Acad Radiol 2006. ; 13 ( 10 ): 1187 – 1193 . [DOI] [PubMed] [Google Scholar]
- 26. Fletcher JG , Yu L , Li Z , et al . Observer Performance in the Detection and Classification of Malignant Hepatic Nodules and Masses with CT Image-Space Denoising and Iterative Reconstruction . Radiology 2015. ; 276 ( 2 ): 465 – 478 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Fletcher JG , Fidler JL , Venkatesh SK , et al . Observer Performance with Varying Radiation Dose and Reconstruction Methods for Detection of Hepatic Metastases . Radiology 2018. ; 289 ( 2 ): 455 – 464 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Nam JG , Lee JM , Lee SM , et al . High Acceleration Three-Dimensional T1-Weighted Dual Echo Dixon Hepatobiliary Phase Imaging Using Compressed Sensing-Sensitivity Encoding: Comparison of Image Quality and Solid Lesion Detectability with the Standard T1-Weighted Sequence . Korean J Radiol 2019. ; 20 ( 3 ): 438 – 448 . [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Choi SJ , Park SH , Shim YS , et al . Comparison of image quality and focal lesion detection in abdominopelvic CT: Potential dose reduction using advanced modelled iterative reconstruction . Clin Imaging 2020. ; 62 : 41 – 48 . [DOI] [PubMed] [Google Scholar]
- 30. Lee T , Lee JM , Yoon JH , et al . Deep learning-based image reconstruction of 40-keV virtual monoenergetic images of dual-energy CT for the assessment of hypoenhancing hepatic metastasis . Eur Radiol 2022. ; 32 ( 9 ): 6407 – 6417 . [DOI] [PubMed] [Google Scholar]
- 31. Seo JY , Joo I , Yoon JH , et al . Deep learning-based reconstruction of virtual monoenergetic images of kVp-switching dual energy CT for evaluation of hypervascular liver lesions: Comparison with standard reconstruction technique . Eur J Radiol 2022. ; 154 : 110390 . [DOI] [PubMed] [Google Scholar]
- 32. Szczykutowicz TP , Nett B , Cherkezyan L , et al . Protocol Optimization Considerations for Implementing Deep Learning CT Reconstruction . AJR Am J Roentgenol 2021. ; 216 ( 6 ): 1668 – 1677 . [DOI] [PubMed] [Google Scholar]
- 33. Parakh A , Cao J , Pierce TT , Blake MA , Savage CA , Kambadakone AR . Sinogram-based deep learning image reconstruction technique in abdominal CT: image quality considerations . Eur Radiol 2021. ; 31 ( 11 ): 8342 – 8353 . [DOI] [PubMed] [Google Scholar]
- 34. Lyu P , Neely B , Solomon J , et al . Effect of deep learning image reconstruction in the prediction of resectability of pancreatic cancer: Diagnostic performance and reader confidence . Eur J Radiol 2021. ; 141 : 109825 . [DOI] [PubMed] [Google Scholar]
- 35. Nam JG , Ahn C , Choi H , et al . Image quality of ultralow-dose chest CT using deep learning techniques: potential superiority of vendor-agnostic post-processing over vendor-specific techniques . Eur Radiol 2021. ; 31 ( 7 ): 5139 – 5147 . [Published correction appears in Eur Radiol 2021;31(8):6410.] [DOI] [PubMed] [Google Scholar]
- 36. Kolb M , Storz C , Kim JH , et al . Effect of a novel denoising technique on image quality and diagnostic accuracy in low-dose CT in patients with suspected appendicitis . Eur J Radiol 2019. ; 116 : 198 – 204 . [DOI] [PubMed] [Google Scholar]
- 37. Yeoh H , Hong SH , Ahn C , et al . Deep Learning Algorithm for Simultaneous Noise Reduction and Edge Sharpening in Low-Dose CT Images: A Pilot Study Using Lumbar Spine CT . Korean J Radiol 2021. ; 22 ( 11 ): 1850 – 1857 . [DOI] [PMC free article] [PubMed] [Google Scholar]