This retrospective study was approved by our institutional review board (Hamamatsu University School of Medicine), and the requirement for written informed consent was waived. From July 2021 to April 2022, 33 patients underwent CTHA (a total of 38 procedures) for the treatment of hepatic lesions. We excluded six patients because of a lack of raw CTHA image data, and the remaining 27 patients (18 men, 9 women; age [mean ± standard deviation]: 75.7 ± 9.7 years; age range: 53–91 years) were enrolled in this study. For those patients who underwent TACE two or more times, the first session was used in this study. The target lesions for TACE were: HCC (n = 24), liver metastasis from rectal cancer (n = 1), and liver metastasis from pancreatic neuroendocrine neoplasm (n = 2). The indication for TACE was decided at a multidisciplinary meeting. The patients’ background data are shown in Table 1.

The CTHA and TACE procedures were performed using a hybrid angiography CT system. A 5-F preshaped loop catheter (GHC-A; Hanaco medical CO, LTD., Saitama, Japan) was inserted into the common hepatic artery. When a loop catheter was not suitable for the hepatic artery anatomy, a 4-F preshaped shepherd hook catheter (Medikit’s Angiographic Catheter; Medikit CO, LTD., Tokyo, Japan) was used. CTHA was obtained 6, 21.5, and 42 s after the initiation of the contrast media injection (dual injection of normal saline and 300 mgI/mL water-soluble iodine contrast media [Iopamiron 300; Bayer AG, Leverkusen, Germany] in a 1:1 ratio) for volume scans and 6, 23.5, and 41.5 s after for helical scans [20]. The flow rate was determined based on the DSA of the common hepatic artery. The duration of contrast media injection was 26 s for helical scan and 21 s for volume scan mode.

A 320-detector CT scanner (Aquilion ONE/NATURE Edition; Canon Medical Systems Corp., Tochigi, Japan) with automatic tube current modulation (auto exposure control; AEC) was used for CTHA. The tube current modulation program combined z-axis and angular modulation of the X-ray tube current adjusted for the patient’s body size and shape (monitored from a single positioning image) to account for all three dimensions. Using the scout scan and a preset noise index, the system modulates the tube current during CT scanning to achieve an acceptable image noise level. Volume and helical scans were used and selected according to the patient’s liver size.

The following CT imaging parameters were used: tube voltage, 120 kVp; tube current, 50–550 mA (AEC); noise index (SD value), 8.5; detector configuration, 320 detectors with 0.5-mm section thickness for volume scan and 80 detectors with 0.5-mm section thickness for helical scan; beam collimation, 160 mm for volume scan and 40 mm for helical scan; rotation time, 0.5 s; pitch factor, not applicable for volume scan and 0.813 for helical scan; scan field of view (FOV), M or L; display FOV, 32 cm; scan length, 160 mm for volume scan and 200 mm for helical scan; total exposure time, 0.5 s for volume scan and 4 s for helical scan.

All CTHA images were reconstructed using three different image-reconstruction methods: standard adaptive iterative dose reduction 3D (AIDR 3D Mild; Canon Medical Systems Corp.) (here defined as hybrid-IR), and mild- and strong-strength DLR (DLR-M and DLR-S, respectively; Advanced intelligent Clear-IQ Engine [AiCE] Body Sharp Mild and AiCE Body Sharp STR; Canon Medical Systems Corp.).

Two radiologists (INITIALS BLINDED, with 10 and 2 years, respectively, of post-training experience in interpreting body CT images and interventional radiology) measured the CT values of sub-segmental artery (A1–A8), sub-sub-segmental artery (A2s–A8s), tumor, liver parenchyma, and subcutaneous fat using 2–10-mm-diameter regions-of-interest (ROIs) on 0.5-mm-thick axial first-phase CTHA images using a 3D workstation (SYNAPSE VINCENT, FUJIFILM Co., Tokyo, Japan). ROIs for artery were placed on CTHA images, encompassing as much of the vascular lumen as possible while avoiding vascular walls, calcification, thrombus, and artifacts. The ROI size for vessel measurement was determined as the maximum size of the vessel without including vessel wall. ROIs for tumor were placed on the strongest enhancement area. ROIs for liver parenchyma were placed so as not to involve the major vessels, tumor, and artifact. When arterial branches were partly lacking because of prior treatment, they were managed as missing values. The sub-sub-segmental branch of A1 was not evaluated in this study because of its small diameter. In patients with multiple tumors, the largest and second largest tumor on which TACE was performed were evaluated. The SNR was calculated by dividing the CT values of artery by the standard deviation (SD) of the CT values of subcutaneous fat. The CNR of tumor was calculated by dividing the difference in CT values between lesion and liver parenchyma by the SD of the CT values of subcutaneous fat. Image noise was defined as the SD of subcutaneous fat.

Two radiologists (INITIALS BLINDED, with 10 and 2 years, respectively, of post-training experience in interpreting body CT images and interventional radiology), who were unaware of the reconstruction methods, reviewed the first-phase CTHA images using a 3D workstation (SYNAPSE VINCENT). CTHA images were initially presented with a preset soft-tissue window setting (320 HU width and 120 HU level), with the radiologists being allowed to modify the window setting at their own discretion.

The radiologists graded vascular depiction (sharpness and contrast), lesion contrast, and image quality (sharpness, granulation, artifact, and overall diagnostic availability) using a five-point rating scale: 5 for excellent, 4 for good, 3 for acceptable, 2 for suboptimal, and 1 for unacceptable. To minimize learning bias, each review for hybrid-IR, DLR-M, and DLR-S was performed with a time interval of at least 2 weeks.

A radiologist (INITIALS BLINDED, with 10 years of post-training experience in interpreting body CT images and interventional radiology) retrospectively reviewed the angiography and identified the tumor-feeder arteries. This radiologist also retrospectively applied automated feeder artery detection software (Embolization Plan; Canon Medical Systems Corp.) to the CTHA images and evaluated its detection rate for feeder arteries. The volume of interest (VOI) for the automated feeder artery detection software was manually drawn by the radiologist on the target lesion and peritumoral area. Partial mistracing of the route of a feeder artery by the software was counted as a failure.

Two radiologists (INITIALS BLINDED, with 2 years of post-training experience in interpreting body CT images and interventional radiology) who were unaware of the tumor-feeder artery independently reviewed the CTHA images and evaluated the number of tumor-feeder arteries and tumor-feeder artery visualization (contrast, continuity, and confidence level) using a 4-point rating scale, and then reviewed the evaluations in consensus. The reviewers scored feeder artery contrast according to a 4-point rating scale: 4 for excellent, 3 for good, 2 for fair, and 1 for bad. The reviewers scored feeder artery continuity using 4-point rating scale: 4 for no discontinuities, 3 for some discontinuity but less than 50%, 2 for around 50% discontinuity, and 1 for more than 50% subject to discontinuity. The reviewers also scored feeder artery confidence level using 4-point rating scale: 4 for 100% of confidence level, 3 for more than 50% but less than 100% of confidence level, 2 for around 50% of confidence level, and 1 for less than 50% of confidence level.

Statistical analyses were performed using SPSS for Windows (version 26.0; IBM Corp., Armonk, NY, USA). Quantitative values were compared between the three different reconstruction methods (hybrid-IR, DLR-M, and DLR-S) using repeated-measures analysis of variance and post hoc Tukey tests. Qualitative scores were compared using the Friedman test followed by pair-wise comparisons using the Wilcoxon signed-rank test. Cochran’s Q test was used for the analysis of feeder artery detection rate. P values less than 0.05 were considered statistically significant. The Bonferroni correction adopting a stricter P value of less than 0.017 was used for pair-wise comparisons.

Weighted kappa analysis was conducted to assess interobserver variability in the evaluation of image quality and feeder artery visualization. A kappa value of up to 0.20 was considered to indicate slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81 or greater, almost perfect agreement.