This prospective single-center randomized study was approved by the Institutional Review Board of Seoul National University Hospital, and written informed consent was obtained from all participants (NCT04024514). Financial support was provided by Riyeon Pharmaceuticals (Seoul, South Korea); however, the authors retained full control over the data and information submitted for publication.

Between December 2019 and December 2020, we enrolled participants who met the following eligibility criteria: (a) older than 20 years; (b) at high risk of developing HCC (chronic hepatitis B, C, or liver cirrhosis except for congestive hepatopathy) and scheduled for liver CT for diagnostic or follow-up purposes; (c) decreased renal function (estimated glomerular filtration rate [eGFR] < 60 mL/min/1.73 m2) and not on dialysis; and (d) provided informed consent. Exclusion criteria were as follows: (a) younger than 20 years; (b) not at high risk for developing HCC; (c) on dialysis; (d) no venous access in the forearm; (e) anticipated beam hardening artifacts due to a prosthesis; and (f) any relative or absolute contraindication of CECT except renal dysfunction. The eGFR was calculated according to the CKD-EPI 2009 formula [20]: eGFR = 141 × min (serum creatinine [Cr]/κ, 1)α × max (serum Cr/κ, 1)−1.209 × 0.993age × 1.018 [if female] × 1.159 [if Black], where κ is 0.7 for female and 0.9 for male patients, and α is − 0.329 for female and − 0.411 for male patients.

Participants were randomly assigned to either the standard liver CT protocol group (the standard group) or the renal protection protocol liver CT group (the RPP group). A computer-generated permuted block randomization process, managed by our medical research collaboration center, was utilized. The block sizes were 4 and 6 for a 1:1 allocation. Two stratification factors were employed by research coordinators: (a) BMI (< 25 kg/m2 vs. ≥ 25 kg/m2) and (b) eGFR (< 45 mL/min/1.73 m2 vs. ≥ 45 mL/min/1.73 m2). Participants, investigators, and outcome assessors were all blinded to the participant allocation.

All participants were asked to fast for at least 6 h prior to the CECT examination. Intravenous hydration was administered before CT scan, and oral hydration was encouraged after examination in accordance with our institutional protocol to prevent CM-induced nephrotoxicity. All CT scans were conducted using a dual-layer CT scanner (IQon; Philips Healthcare, Amsterdam, The Netherlands) with the following settings: 120 kVp, a gantry rotation time of 0.33 s, 64 × 0.625 mm collimation, and a slice thickness of 3 mm with 2-mm reconstruction intervals. The precontrast, portal venous, and delayed phases were acquired before, 70 s after, and 180 s after CM administration, respectively. The arterial phase was obtained using the bolus tracking technique, initiated 17 s after reaching a trigger threshold of 150 Hounsfield units at the abdominal aorta. The scan parameters remained consistent across all four phases.

In our study, we used 525 mgI/kg of iodinated CM for the standard liver CT protocol, while 300 mgI/kg was used for the RPP liver CT. In the standard group, iodinated CM (Ioversol 350 mgI/mL, Optiray350; Riyeon Pharmaceuticals) was administered via a power injector (Stellant®; Medrad, Pittsburgh, PA, USA) for 35 s, followed by a 30-mL saline flush. In the RPP group, iodinated CM with a lower iodine concentration (Ioversol 320 mgI/mL, Optiray320; Riyeon Pharmaceuticals) was used in the same manner, with the minimum contrast injection rate set at 2 mL/sec. In both groups, the maximum volume of administered CM was limited to one vial (130 mL) according to our institutional safety protocol.

In the standard group, CT images were reconstructed using a hybrid iterative reconstruction (iDose level 4). In the RPP group, arterial, portal, and delayed phase images were reconstructed using a monoenergetic image (50 keV) and a DL-based iodine-boosting method, in addition to the automatically generated iDose images of precontrast, arterial, portal, and delayed phases.

DL-based iodine enhancement — DL-based reconstruction was carried out using commercially available software (ClariACE; ClariPi, Seoul, South Korea). This method employs a two-stage U-net architecture, in which image denoising and contrast augmentation occur sequentially. Initially, the DL-based denoising algorithm extracts a noise component image from the input image (iDose in this study), which is then subtracted from the input image [21, 22]. Next, the augmentation process begins, during which the iodine component image is extracted using the DL-based iodine enhancement algorithm. Ultimately, the extracted iodine component image is added to the denoised image, thereby improving the signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) of the augmented CECT image [19].

Qualitative analysis — Four fellowship-trained body radiologists (J. H. Y., E. S. L., J. Y. P., and S. M. L., with 12, 12, 8, and 8 years of experience after fellowship, respectively) independently reviewed the images, ensuring that the window width and level were always adjustable. Image noise, image contrast, image texture, and overall image quality were scored on a 5-point scale for both arterial and portal venous phases, with higher scores indicating better image quality (Table E1). The location and size of non-cystic focal liver lesions (FLLs), excluding lipiodol uptake, were also documented. Lesion conspicuity during the arterial and portal venous phases was scored on a 5-point scale as follows: 1, not visible (automatically assigned to missed lesions); 2, barely delineated; 3, visible with a blurry margin; 4, visible lesion with a relatively sharp margin and acceptable contrast; and 5, clear contrast and a sharp margin [13].

Quantitative analysis — One fellowship-trained body radiologist (J.H.Y.) drew three circular regions of interest (ROIs) on consecutive slices at the level of the celiac trunk in the aorta during the arterial phase and at the main portal vein during the portal venous phase. Additionally, three ROIs were drawn in the liver parenchyma at the hilar level, avoiding vessels and FLLs. The average Hounsfield unit value was used as a representative value for the aorta, main portal vein, and liver parenchyma. Finally, circular ROIs were drawn in the subcutaneous fat layer of the anterior abdominal wall on three consecutive slices during both the arterial and portal venous phases. The standard deviation of the ROI values was considered to represent the image noise for each phase.

Two fellowship-trained radiologists (J. M. L. and J. H. K., with 20 and 4 years of experience, respectively, after fellowship) who did not participate in the review session reviewed the images to establish the reference standard for FLL detection. This reference standard was based on the most recent follow-up imaging, including a standard dose of liver CT or liver MRI taken within 3 months for patients with LR-3, -4, -5, or -M observations. For both groups, interval cancers identified during follow-up were considered true FLLs that had been missed on CECT. For those with no FLLs or only benign FLLs, remote cross-sectional images were used instead of 3-month follow-up images. A comprehensive description of the reference standard can be found in the supplementary material 1. In brief, HCC was diagnosed using histology or typical imaging characteristics on CECT, CE-MRI, CE-ultrasound, or tumor staining on angiography for transarterial chemoembolization. Benign FLLs were confirmed based on the stability of size on follow-up imaging and the imaging features.

The primary endpoint is to prove non-inferiority of image quality in RPP group compared with standard group. Secondary outcome is to compare the lesion conspicuity and detection rates between the two groups. Lastly, we collected information of participant-reported anuria, follow-up creatinine level and eGFR within a month after CT when available.

The sample size was determined based on a retrospective study examining improved image quality in monoenergetic images with reduced CM (4.3 ± 0.6 in monoenergetic images vs. 3.6 ± 0.3 in standard images) [14]. We assumed a one-sided significance level of 0.05, a target power of 0.95, and an allocation ratio of 1 for each group, along with a non-inferiority margin of − 0.2, as informed by a previous study of low-dose CT angiography [23]. As a result, the minimum number of participants required for each group was 22, and the final sample size was set at 52, considering an 18% dropout rate.

The independent-samples t-test was conducted to assess the differences between the two groups. To evaluate differences based on reconstruction methods within the same group, we employed either the paired t-test or the Friedman test, as appropriate. The intraclass correlation coefficient (ICC) was calculated to determine the inter-observer agreement. Due to the presence of multiple lesions in a patient, lesion conspicuity was analyzed using the generalized estimating equation method. A normal distribution and the identity link function were applied, and lesion conspicuity was reported with a 95% confidence interval (CI).

To compare the detection rate of FLLs between groups, we estimated the reader-averaged figure of merit (FOM) using the random reader, random lesion method in a weighted free response receiver operating characteristic analysis. This considered multiple lesions and lesion locations within the same participant. FLLs were considered to have been detected by each reviewer if the lesion conspicuity score was 2 or higher. Lesion conspicuity and detection were evaluated for all participants and in subgroups stratified by FLL size (less than 20 mm vs. 20 mm or greater).

All statistical analyses were conducted using commercially available software packages (SAS version 9.4, SAS Institute Inc., Cary, NC, USA; SPSS version 27, IBM Corp., Armonk, NY, USA; MedCalc version 20.216, MedCalc Software, Ostend, Belgium) and R version 4.1.1 (R Foundation for Statistical Computing, Vienna, Austria). A p-value of less than 0.05 was considered to indicate a statistically significant difference.