Our hospital’s Institutional Review Board (Ethics committee Hannover Medical School; Nr. 8316_BO_K_2019) approved this retrospective study. All patients gave written consent. The indication for TACE was obtained by an inter-disciplinary tumor board. All TACE procedures from 1/2015 to 5/2021 (n = 644) were retrospectively reviewed. Based on previously published work determining influencing factors on image quality in CACTs as well as the IQ-improving effect of the 3D motion compensating algorithm, we purposefully selected datasets with poor image quality as main inclusion requirement (see Fig. 1), represented by substantially limited visualization of both central as well as peripheral hepatic arteries, including distinct blurriness of vessel margins and/ or severe cardiorespiratory motion artifacts [22, 23]. The study population thus comprised of 27 CACTs in 26 patients (18 m, 8 f; mean age: 69.7 years ± 10.7 SD)– one patient having received a CACT of both the left and the right liver artery due to variant hepatic anatomy—with patient characteristics shown in Table 1.

Flow chart of patient selection in the present study

Ultrasound-guided assessment of the femoral artery was performed after application of local anesthesia. Mesenterico-portography, sequential hepatic angiography and a CACT were acquired to assess vascular anatomy, to analyze tumor-feeding arteries and to plan an adequate, supraselective catheter position for treatment. The arms were elevated to avoid streaking artifacts. Patients were instructed to hold their breath for 10 s, which—unlike most patients in every-day practice—the majority of patients included in this study was not able to perform sufficiently, causing breathing motion artifacts (n = 16) of different proportions. CACT was obtained using the manufacturer’s preset (Artis Q: 6 s DR DynaCT, acquisition time of 6 s, fixed-tube-detector distance of 0.9 m, total acquisition angle of 200°, projection increment of 0.5°, 396 projectional images, 1 k-matrix, zoom factor 0, FOV 48 cm, system-detector dose per image of 0.36 lGy; Artis pheno: 5 s DR DynaCT, acquisition time of 5 s, fixed-tube-detector distance of 0.9 m, total acquisition angle of 200°, projection increment of 0.5°, 396 projectional images, 1 k-matrix, zoom factor 0, FOV 52 cm, system-detector dose per image of 0.36 lGy). Contrast medium was injected in accordance with our standard protocol for diagnostic catheter (n = 15, flow rate 5 ml/sec) or microcatheter injections (n = 12, flow rate 2.5 ml/sec). Doxorubicin-loaded drug-eluting beads of 30–60 um size (HepaSphere®, Merit Medical Europe, Maastricht, Netherlands) were injected into the tumor-feeding artery via the microcatheter (Merit Maestro™, Tenor™ 0.014 guidewire, Merit Medical Systems, Utah, USA), with stasis in the tumor-feeding arteries delineating the endpoint of the intervention.

All TACE procedures were performed by board-certified interventional radiologists at our institution on either a monoplane, ceiling-mounted or monoplane, robotic-arm-mounted angiographic system (Artis Q®, ARTIS pheno®, Siemens Healthcare, Forchheim, Germany) as available during clinical routine. Image acquisition commenced simultaneously to contrast injection, with the C-arm mounted X-ray source and detector rotating around the patient on a circular trajectory. The 3D reconstruction prototype software, developed and modified by the manufacturer (Siemens Healthcare, Forchheim, Germany), was installed on a dedicated workstation (syngo X Workplace®, Software Version VD20C, Siemens Healthcare, Forchheim, Germany) and applied to the original raw datasets. These were retrospectively modified by the algorithm’s utilization of iterative motion estimation and compensation of a 4D deformable motion, as previously published elsewhere [19,20,21,22,23]. For motion correction of selective CACT images of the hepatic arteries, we used 500 iterations in total and an iteration update/display every 100 iterations. The step size was 2 mm and motion resolution 25 mm. Manual volume punching of stationary high-contrast objects such as bones or extraneous materials was performed by a blinded radiologist (SKM), as these could lead to potential falsifications of motion correction in the liver. The aforementioned post-processing, enabled quantitative and qualitative comparisons between two datasets: the original CACT (CACTOrg) and the CACT after motion correction and bone removal (CACTMC_no bone).

Two radiologists of 10 and 3 years of experience (JBH, LSB: reader 1,2), blinded to the nature of the data set (CACTOrg vs. CACTMC_no bone), performed side-by-side grading of randomly assigned images before and after the application of the motion compensating algorithm, with concern to the following categories: 1) overall image quality (grade 1–3: good, moderate, poor), 2) vessel visualization and sharpness (i.e., clear visualization of all hepatic arteries, including subsegmental and subcapsular branches, grade 1–5), presence of artifacts (preponderantly induced by breathing or cardiac motion), and their image preference for TACE intervention (for details see Table 2).

To further validate the algorithm’s effect and to reproduce a realistic clinical setting, both the original as well as the post-processed datasets were additionally offered to two blinded radiologists of 11 and 4 years of experience (CvF, CLAD: reader 3,4), to document their preferred dataset for a transarterial chemoembolization of the liver, their recommendation for additional imaging (e.g., repeat CACT), and to compare the datasets by using a five-point Likert scale.

To quantitatively compare the hepatic artery delineation on both cone-beam CT datasets, the furthest peripheral position with preserved demarcation of the contrast-enhanced vessel lumen within the hepatic artery branch of each segment to the liver capsule was determined in a thin, axial multiplanar reformation (1 mm in both modalities; JBH and LSB in consensus). Additional measurements of intraarterial enhancement in the common, right or left hepatic artery as well as in areas of maximum liver parenchyma enhancement were performed (JBH and LSB in consensus), creating a ratio of the contrast in the liver arteries to the contrast in the liver parenchyma. For intravascular contrast quantification, a circular region of interest, fitted to at least two thirds of the vessel diameter and located 2 cm distally of the catheter tip in a thin-sliced, coronal reformat. For image assessment, all readers were able to use thin-sliced multiplanar reformats (slice thickness B 0.49 mm) in axial, coronal, sagittal or oblique orientation and maximum intensity projections (MIP) on a 3D PACS workstation (Visage 7.1, Visage Imaging, Berlin, Germany).

Descriptive statistical analyses of the patient’s demographic and angiographic data are presented as mean values ± standard deviation (sd). Performance of a Shapiro–Wilk test showed a normal distribution of values (p > 0.05). A two-sided t-test was performed to assess the differences between CACTOrg and CACTMC_no bone by measuring the distance of delineated subcapsular vessels in each liver segment to the capsule. For the comparison of image quality with regards to vessel visualization, presence and amount of movement artifacts as well as overall image quality, a paired Wilcoxon-test, an interrater and an intermodality agreement were calculated, using the two-way intra-class correlation coefficient with absolute agreement (ICC 2.1). This was also used to calculate interobserver agreement of the aforementioned two radiologists. Values ≥ 0.75 represented excellent agreement, those from 0.6–0.74 good, from 0.4–0.59 fair, and < 0.4 poor agreement [24].

Statistical analysis was conducted with R (version 3.6.1, http://www.r-project.org with package ‘‘IRR’’ version 0.84.1).