Translate this page into:
Improving cone-beam computed tomography image quality for transarterial therapy of liver malignancies: Evaluation of a motion correction algorithm with and without automated bone removal
, Arun Kamireddy1, Dana Angelina Schaar2, Fereshteh Khorshidi3, Izzet Altun3, Heyuan Huang4, Christopher R. Bailey1, Robert P. Liddell1, Nariman Nezami5, Peter Fischer6, Alexander Preuhs6, Tina Ehtiati7, Clifford R. Weiss1,
*Corresponding author: Clifford R. Weiss, Russell H. Morgan Department of Radiology and Radiological Science, Johns Hopkins University School of Medicine, Baltimore, United States. cweiss@jhmi.edu
-
Received: ,
Accepted: ,
How to cite this article: Mustafa A, Khalil A, Kamireddy A, Schaar DA, Khorshidi F, Altun I, et al. Improving cone-beam computed tomography image quality for transarterial therapy of liver malignancies: Evaluation of a motion correction algorithm with and without automated bone removal. J Clin Imaging Sci. 2025;15:43. doi: 10.25259/JCIS_150_2025
Abstract
Objectives:
The objectives of the study are to compare the effect on image quality of a motion artifact correction algorithm (CAVAREC) alone versus in combination with an automated bone removal algorithm (ZIBOS) for motion-degraded intraprocedural liver cone-beam computed tomography (CBCT) images.
Material and Methods:
In this retrospective, two-center, Institutional Review Board (IRB)-approved study, 48 CBCTs from 41 patients were included. Inclusion criteria were (a) age ≥18; (b) liver CBCT with contrast injected at the main, right, or left hepatic artery; and (c) CBCT motion-degraded. Exclusion criteria were (a) no visible tumor and (b) CBCT not capturing the entire liver. The mean (standard deviation [SD]) age was 64 (7) years, 81% male. 65% had hepatocellular carcinoma and 35% metastatic cancer. 69% CBCTs were from transarterial radioembolization (TARE) mapping, 25% transarterial chemoembolization, and 6% TARE treatment. Mean (SD) maximum tumor diameter was 4.6 (2.7) cm. CBCT images were processed with CAVAREC (prototype, Siemens Healthineers AG, Forchheim, Germany) and CAVAREC + ZIBOS (work in progress, Comprehenso, Hannover, Germany). Using CoroEval, sharpness for two segmental arteries per case was obtained quantitatively. Three blinded interventional radiologists independently evaluated overall image quality on a 0–100 scale and large vessels, small vessels, vessel sharpness, tumor feeders, tumor blush, and streak artifacts on a −50 to +50 scale relative to uncorrected images. Results were analyzed with paired t-tests and Wilcoxon signed-rank tests, adjusting P-values with the Benjamini–Hochberg procedure.
Results:
On quantitative assessment, mean (SD) sharpness for uncorrected, CAVAREC, and CAVAREC + ZIBOS (C+Z) images was 0.281 (0.04), 0.287 (0.04), and 0.284 (0.04), respectively, P = 0.02 for uncorrected versus CAVAREC and P > 0.05 for other comparisons. On qualitative assessment, mean (SD) overall quality for uncorrected, CAVAREC, and C+Z images was 45 (14), 53 (16), and 53 (17), respectively, P < 0.001 for both uncorrected versus CAVAREC and uncorrected versus C+Z and P = 0.06 for CAVAREC versus C+Z. The mean preference for both CAVAREC and C+Z compared to uncorrected images for all parameters ranged from +4.3 to +9.5, P < 0.001, except tumor blush which was +1.6 for CAVAREC and +0.9 for C+Z, P > 0.05. CAVAREC versus C+Z was not significant for any parameter.
Conclusion:
Motion artifact correction of liver CBCT images using CAVAREC improves image quality significantly. According to observer studies, integrating the automated bone segmentation algorithm ZIBOS with CAVAREC does not degrade image quality.
Keywords
Bone segmentation
Cone-beam computed tomography
Liver malignancies
Motion artifact correction
Transarterial therapy
INTRODUCTION
C-arm cone-beam computed tomography (CBCT) allows three-dimensional imaging during minimally invasive surgical procedures by continuous rotation of the X-ray source and detector around the patient.[1] In interventional oncology, intraprocedural CBCT can enable treatment planning, device navigation, and assessment of therapy in real-time in the interventional radiology suite.[2] However, since acquisition of CBCT can take 5–20 s,[1] any motion due to breathing, cardiac cycle, or patient movement can introduce motion artifacts[2,3] and degrade the image.
Algorithms to correct CBCT motion artifacts after image acquisition have been developed and tested, generally providing a modest improvement in image quality.[4-10] CAVAREC (prototype, Siemens Healthineers AG, Forchheim, Germany) is an iterative motion-compensated reconstruction algorithm for CBCT data. In each iteration, it estimates the motion of each projection image relative to a sparse reference image. The sparse reference image is created from the motion-corrupted initial reconstruction by windowing to focus on the high intensities.[11] The reference image determines what the motion compensation focuses on. For the interventional oncology use case, the focus should mainly be on the contrast-enhanced vessels, but the intensity range of the vessels overlaps with the intensity of bones at 1000 HU and above.
High-contrast objects within the images, such as bones, which are not relevant to the procedure, can interfere with correction by the algorithms of desired high-contrast areas such as arteries injected with contrast media.[5] Therefore, bone removal before running the motion correction algorithm could theoretically improve algorithm performance. One study has investigated the effect of combining bone removal by manual volume punching with the motion correction algorithm CAVAREC and found little additional benefit of prior bone removal.[5] However, the use of automated bone removal algorithms, which could be more effective and efficient, has not been reported previously.
By removing bones from the initial reconstruction, windowing will mostly leave vessels and thus focus the motion compensation on the vessels. ZIBOS (Comprehenso, Hannover, Germany) is a proprietary deep-learning-based bone segmentation algorithm. CAVAREC has not previously been studied with automated bone removal.
The purpose of this study is to compare the motion correction of intraprocedural liver CBCTs with CAVAREC alone versus CAVAREC combined with ZIBOS, using a quantitative approach by measuring vessel sharpness as well as qualitative reader studies.
MATERIAL AND METHODS
Subjects
This was a retrospective, IRB-approved study. Consecutive clinical CBCT images were obtained from two academic medical centers. Inclusion criteria were (a) age ≥18; (b) CBCT with contrast injection acquired during liver-directed arterial therapies; (c) non-selective proximal contrast injection with catheter tip at the main, right, or left hepatic artery; and (d) presence of motion artifacts degrading the CBCT image. Exclusion criteria were (a) absence of a visible tumor; (b) CBCT not capturing the entire liver; and (c) negligible or no motion artifact. Electronic medical records were reviewed to obtain baseline patient demographics and clinical data.
CBCT acquisition
Center 1 protocol
A dual-phase CBCT protocol was employed for transarterial chemoembolization (TACE) and transarterial radioembolization (TARE) procedures, with each phase consisting of a 5-s X-ray delay followed by an 8-s rotational acquisition covering 200° at 0.5° angular increments (397 projections). Intra-arterial contrast was injected at rates of 0.5–2.0 mL/s, adjusted according to target vessel size and flow dynamics, with total injection volumes ranging from 4 mL to 10 mL per phase.
Patients were instructed to exhale and maintain breath-hold at end-expiration. During acquisition arms were elevated when feasible to minimize streak artifacts. Imaging was performed on two angiographic systems, Siemens Artis Zeego system (VD11 software version; Siemens Healthineers AG, Forchheim, Germany). Both systems were equipped with a 30 × 40 cm2 flat-panel detector, operating at 90 kVp and 0.36 μGy/frame detector dose. Images were reconstructed using syngo Application software (Siemens Healthineers AG) with the DynaCT Dual Arterial reconstruction algorithm, using full field-of-view, a 512 × 512 matrix, 0.5 mm slice thickness, and HU kernel.
Center 2 protocol
CBCT acquisitions were performed during TACE and TARE procedures using a single-phase arterial protocol. Each scan consisted of a 6-s X-ray delay followed by an 8-s rotational acquisition, resulting in a total injection duration of 14 s. Intra-arterial contrast (50% diluted Visipaque® 320 mg I/mL) was injected at a rate equal to half the vessel diameter in millimeters (e.g., a 4-mm vessel received 2 mL/s), and total contrast volume was calculated as contrast volume (mL) = (Vessel Diameter/2) × 14 s. Alternatively, a fixed protocol was sometimes used with an injection rate of 2 mL/s for a total volume of 24 mL, with a 6-s X-ray delay.
Patients were instructed to exhale and hold their breath at end-expiration approximately 2–3 s before image acquisition. Arms were elevated when feasible to reduce streak artifacts. Image acquisition was performed using a Siemens Artis Zee floor-mounted system (VC21C software version; Siemens Healthineers AG, Forchheim, Germany) equipped with a 30 × 40 cm2 flat-panel detector. Scans were acquired over a 200° rotational arc using a 0.5° angular increment (397 projections), with acquisition parameters of 90 kVp and 0.36 μGy/frame detector dose.
Images were reconstructed using syngo Application software with the DynaCT Body Nat Fill reconstruction algorithm, using full field-of-view, 512 × 512 matrix, HU/Normal kernel, and 0.5 mm slice thickness.
Motion correction
CBCT images were obtained from PACS and reconstructed on a dedicated workstation using syngo DynaCT (Siemens Healthcare, Forchheim, Germany). The 3D reconstructions were processed iteratively by CAVAREC to produce motion-corrected images. 250, 350, and 450 iterations were obtained. The process was repeated using CAVAREC + ZIBOS to produce motion-corrected images with automated bone segmentation. This resulted in seven sets of images per case: Uncorrected, 250, 350, 450 (CAVAREC alone), 250, 350, and 450 (CAVAREC + ZIBOS).
Vessel segmentation for quantitative assessment
One set of corrected images per case was imported into CoroEval.[12] Two hepatic segmental or subsegmental arteries were segmented per case by a radiologist (AK). Sharpness metrics for each vessel were exported, as well as point coordinates of the arteries. This process was repeated for each of the remaining five sets of corrected images, using the previously exported point coordinates to locate the arteries at the same points.
The uncorrected images were also imported into CoroEval, and the diameters of the arteries were exported.
Qualitative assessment
A questionnaire [Appendix 1] to assess clinically relevant features of the images was developed by adapting questionnaires from previous studies.[4-6] The questionnaire was administered using a secure, web-based application, research electronic data capture (REDCap) hosted at Johns Hopkins University.[13] Three blinded interventional radiologists (CRW, RPL, and NN) with 3, 18, and 2 years of experience participated in reading images for the qualitative assessment.
Three sets of images were displayed side-by-side using the RadiAnt DICOM Viewer (Medixant, Poznan, Poland) for each case: uncorrected on the middle monitor and CAVAREC and CAVAREC + ZIBOS on the left and right monitors, randomized by case. The 250 iteration was chosen for CAVAREC and CAVAREC + ZIBOS images due to higher vessel sharpness. The images were anonymized using the OsiriX DICOM Viewer (Pixmeo SARL, Bernex, Switzerland). The monitors on which CAVAREC and CAVAREC + ZIBOS images were displayed were randomized by case, and the readers were blinded to the sequence of images. Readers were allowed to adjust windowing and zoom according to their preference, and these settings were automatically synced across the three images.
The readers were instructed to use the uncorrected image on the middle monitor as the reference image and complete the questionnaire for each case on REDCap. Each reader completed the image review independently and using a randomized sequence of cases. Overall image quality was assessed for all images for all cases independently on an absolute scale from 0 (poor) to 100 (excellent). Preference for one or more of the three images was recorded. For six parameters (larger vessels, smaller vessels, sharpness, tumor feeders, tumor blush, and streak artifacts), preference for each corrected image compared to the reference uncorrected image was recorded on a relative scale from −50 (strongly prefer uncorrected image) to +50 (strongly prefer corrected image).
Statistical analysis
Statistical analysis was completed in R version 4.2.2 and RStudio version 2023.06.0. A P-value smaller than 0.05 was considered statistically significant.
Quantitative assessment
Mean sharpness and diameter were calculated for each artery. Then mean sharpness of all arteries for each set of images was calculated. Repeated measures analysis of variance (ANOVA) was performed to evaluate for differences in vessel sharpness across the 7 sets of images. Paired t-tests were performed for all possible pairwise comparisons of vessel sharpness, and the P-values were adjusted for multiple comparisons using the Benjamini–Hochberg procedure.
Subset analysis of sharpness by breaking down the data by diameter was performed using the same statistical tests to assess differences in algorithm performance by diameter.
Three subset analyses were performed, using the 25th, 50th, and 75th quartiles of vessel diameter as the cutoff for small versus large vessels.
Qualitative assessment
Means and standard deviations of overall image quality were calculated for uncorrected, CAVAREC, and CAVAREC + ZIBOS images, as well as percentage improvements for the two corrected images compared to the uncorrected images.
Repeated measures ANOVA was performed on overall image quality for the three groups. Paired t-tests were performed for all three pairwise comparisons.
For image preference, percentages for each category were calculated. A Chi-squared test was performed to evaluate differences in proportions for each response. The Marascuillo procedure was employed to perform all pairwise comparisons between responses, with a calculated difference exceeding the critical range considered significant.[14]
To compare clinical imaging features between CAVAREC and CAVAREC + ZIBOS, means and 95% confidence intervals were plotted. Wilcoxon signed-rank tests were used to compare the two groups across each parameter. The P-values were adjusted for multiple comparisons using the Benjamini–Hochberg procedure.
Subset analysis of the qualitative assessment was performed by stratifying the data by the institution from which images were obtained to evaluate potential differences in algorithm performance related to CBCT acquisition protocols.
In addition, subset analysis of overall quality was performed by stratifying the data into low and high uncorrected image quality groups, based on the median quality score, to evaluate potential differences in algorithm performance by baseline image quality.
RESULTS
Subjects
Based on the inclusion criteria, a total of 48 cases were included in the study. Patient characteristics are shown in Table 1. The mean age (standard deviation [SD]) at procedure was 64 (7) years, and 9/48 (19%) patients were female. The mean (SD) body mass index was 26.5 (5.3) kg/m2. The most common tumor was hepatocellular carcinoma, 31/48 (65%), followed by metastatic colorectal, 9/48 (19%); metastatic pancreatic, 3/48 (6%); intrahepatic cholangiocarcinoma, 3/48 (6%); and other metastases, 2/48 (4%). Liver cirrhosis was present in 21/48 (44%) patients, most commonly due to hepatitis C virus (HCV) in 12 patients (25%), followed by alcohol/HCV and nonalcoholic steatohepatitis in 3 patients (6%) each; and hepatitis B virus (HBV) and HBV/HCV in 1 (2%) patient each. The most common procedure was TARE mapping, 33/48 (69%); followed by TACE, 12/48 (25%); and TARE, 3/48 (6%). The mean (SD) largest diameter of the tumor targeted by the procedure was 4.6 (2.7) cm.
| Parameter | Value |
|---|---|
| Age (years), mean±SD | 64±7 |
| Sex, female: Male, n(%) | 9:39 (19%: 81) |
| BMI (kg/m2) | 26.5 (5.3) |
| Tumor, n(%) | |
| HCC | 31 (65) |
| Metastatic colorectal | 9 (19) |
| Metastatic pancreatic adenocarcinoma | 3 (6) |
| ICC | 3 (6) |
| Other metastases | 2 (4) |
| Liver cirrhosis, n(%) | 21 (44) |
| Liver cirrhosis etiology, n(%) | |
| HCV | 12 (25) |
| Alcohol/HCV | 3 (6) |
| NASH | 3 (6) |
| HBV | 1 (2) |
| HBV/HCV | 1 (2) |
| Procedure, n(%) | |
| TARE mapping | 33 (69) |
| TACE | 12 (25) |
| TARE | 3 (6) |
| Largest diameter of target tumor (cm) | 4.6 (2.8) |
SD: Standard deviation, BMI: Body mass index, HCC: Hepatocellular carcinoma, ICC: Intrahepatic cholangiocarcinoma, HCV: Hepatitis C virus, HBV: Hepatitis B virus, NASH: Non-alcoholic steatohepatitis, TARE: Transarterial radioembolization, TACE: Transarterial chemoembolization
Quantitative assessment
The mean sharpness of hepatic segmental/subsegmental arteries is presented in Table 2. Uncorrected images had the lowest mean sharpness, 0.281 (SD 0.04), and the 250 iteration of CAVAREC alone had the highest mean sharpness, 0.287 (SD 0.04). Figure 1 displays a boxplot of sharpness for uncorrected, CAVAREC (iteration 250), and CAVAREC + ZIBOS (iteration 250) images. Illustrative axial, sagittal, and coronal reconstructions of uncorrected, CAVAREC, and CAVAREC + ZIBOS images are shown in Figure 2.
- Boxplot of sharpness of hepatic segmental/subsegmental arteries for 250 iterations.
- A 55-year-old man with colorectal liver metastases undergoing Y-90 transarterial radioembolization. Improvement in image sharpness with motion correction in intraprocedural liver CBCT during Y-90 transarterial radioembolization is shown. Axial (left), sagittal (middle), and coronal (right) reformatted cone-beam CT images are shown for the original uncorrected reconstruction (top row), CAVAREC motion-corrected reconstruction (middle row), and CAVAREC + ZIBOS reconstruction (bottom row). Red arrows in the axial views indicate hepatic segmental arteries, which appear sharper in the CAVAREC and CAVAREC + ZIBOS images compared to the original images. Blue arrows in the axial views highlight the catheter in the aorta, which demonstrates reduced motion blur and streak artifact in both corrected images. Yellow arrows in the coronal views show improved visualization of a hepatic artery branch with CAVAREC, with further improvement in the CAVAREC + ZIBOS image.
| Image set | Iterations | Mean sharpness (SD) |
|---|---|---|
| Uncorrected | 0.281 (0.04) | |
| CAVAREC | 250 | 0.287 (0.04) |
| 350 | 0.285 (0.04) | |
| 450 | 0.284 (0.04) | |
| CAVAREC+ZIBOS | 250 | 0.284 (0.04) |
| 350 | 0.284 (0.04) | |
| 450 | 0.284 (0.04) |
SD: Standard deviation
Repeated measures ANOVA across the seven groups for vessel sharpness was statistically significant, P = 0.03. The results of paired t-tests for all pairwise comparisons are shown in Figure 3. The only statistically significant comparison was CAVAREC alone (250 iterations) versus uncorrected, mean 0.287 versus 0.281, P = 0.02.
- Paired t-test P-values for all pairwise comparisons for mean sharpness. Statistical significance (P < 0.05) is highlighted in yellow.
The mean vessel diameter was 2.42 mm. The 25th quartile, median, and 75th quartile diameters were 2.01, 2.38, and 2.69 mm, respectively. Three subset analyses were performed using each of the three quartiles as cutoffs for small versus large vessels. The only statistically significant t-test was the CAVAREC alone (250 iterations) versus uncorrected comparison, mean 0.270 versus 0.264, P = 0.03, for large vessels with a cutoff for large vessels of 2.01 mm, the first quartile.
Qualitative assessment
Overall image quality of uncorrected, CAVAREC alone, and CAVAREC + ZIBOS images is presented in Table 3. A boxplot is presented in Figure 4. The mean image quality improved by 18% from 45 on uncorrected images to 53 for both CAVAREC and CAVAREC + ZIBOS. Repeated measures ANOVA to compare the three groups was significant, P < 0.001. Paired t-tests for both uncorrected and CAVAREC and uncorrected versus CAVAREC + ZIBOS were significant, P < 0.001. CAVAREC versus CAVAREC + ZIBOS was not significant, P = 0.06.
| Institution | Uncorrected (mean±SD) | CAVAREC | CAVAREC+ZIBOS | ||
|---|---|---|---|---|---|
| Quality (mean±SD) | Improvement (%)* | Quality (mean±SD) | Improvement (%)* | ||
| Center 1 | 49±14 | 57±14 | 16 | 56±13 | 14 |
| Center 2 | 44±14 | 52±17 | 18 | 52±18 | 18 |
| Both combined | 45±14 | 53±16 | 18 | 53±17 | 18 |
- Overall image quality on qualitative assessment boxplot.
Readers’ responses for overall image preference are shown in Figure 5. CAVAREC and CAVAREC + ZIBOS images were preferred in 31% and 32% of the responses, respectively. The uncorrected images had a 4% preference. Both CAVAREC and CAVAREC + ZIBOS were preferred equally 17%, and there was no preference for any of the three images 17%. A chi-square test to assess differences in the proportions of responses in each category was significant, P < 0.001. The results of the Marascuillo procedure to perform all pairwise comparisons for proportions of responses are shown in Table 4 and revealed significant results for CAVAREC vs uncorrected as well as CAVAREC + ZIBOS vs uncorrected. The remaining comparisons were not statistically significant.
| Group 1 | Group 2 | Difference | Critical range | Significant |
|---|---|---|---|---|
| C | C+CZ | 0.14 | 0.2 | No |
| C | CZ | 0.01 | 0.19 | No |
| C | None | 0.14 | 0.2 | No |
| C | Uncorrected | 0.26 | 0.21 | Yes |
| C+CZ | CZ | 0.15 | 0.2 | No |
| C+CZ | None | 0 | 0.21 | No |
| C+CZ | Uncorrected | 0.12 | 0.22 | No |
| CZ | None | 0.15 | 0.2 | No |
| CZ | Uncorrected | 0.28 | 0.21 | Yes |
| None | Uncorrected | 0.12 | 0.22 | No |
Difference exceeding the critical range indicates statistical significance. C: Prefer CAVAREC, CZ: Prefer CAVAREC+ZIBOS, C+CZ: Prefer both CAVAREC and CAVAREC+ZIBOS equally, None: Prefer none of the images over the other, Uncorrected: Prefer the uncorrected/original images
- Image preference of readers.
Readers’ assessment of six clinically relevant imaging features is displayed in Figure 6. On a relative scale from −50 to +50, the mean preference for all parameters for both CAVAREC and CAVAREC + ZIBOS images ranged from +4.3 to +9.5, except tumor blush which was +1.6 for CAVAREC and +0.9 for CAVAREC + ZIBOS. Wilcoxon signed-rank test results are shown in Table 5 and were statistically significant with adjusted P < 0.001 for both CAVAREC and CAVAREC + ZIBOS compared to uncorrected images for all features except tumor blush, which was not statistically significant. CAVAREC compared to CAVAREC + ZIBOS was not statistically significant for any parameter.
- Mean preference of readers for CAVAREC (C) or CAVAREC + ZIBOS (C + Z) compared to uncorrected images across clinically relevant imaging features.
| Feature | Group 1 | Group 2 | P-value |
|---|---|---|---|
| Larger vessels | CAVAREC | Uncorrected | <0.001 |
| CAVAREC+ZIBOS | Uncorrected | <0.001 | |
| CAVAREC | CAVAREC+ZIBOS | 0.92 | |
| Smaller vessels | CAVAREC | Uncorrected | <0.001 |
| CAVAREC+ZIBOS | Uncorrected | <0.001 | |
| CAVAREC | CAVAREC+ZIBOS | 1 | |
| Sharpness | CAVAREC | Uncorrected | <0.001 |
| CAVAREC+ZIBOS | Uncorrected | <0.001 | |
| CAVAREC | CAVAREC+ZIBOS | 0.92 | |
| Tumor feeders | CAVAREC | Uncorrected | <0.001 |
| CAVAREC+ZIBOS | Uncorrected | <0.001 | |
| CAVAREC | CAVAREC+ZIBOS | 0.97 | |
| Tumor blush | CAVAREC | Uncorrected | 0.06 |
| CAVAREC+ZIBOS | Uncorrected | 0.41 | |
| CAVAREC | CAVAREC+ZIBOS | 0.9 | |
| Streak artifacts | CAVAREC | Uncorrected | <0.001 |
| CAVAREC+ZIBOS | Uncorrected | <0.001 | |
| CAVAREC | CAVAREC+ZIBOS | 0.92 |
P< 0.05 is considered significant.
Subset analyses did not show significant differences in performance of either CAVAREC or CAVAREC + ZIBOS when stratified by institution or baseline uncorrected image quality. The results of these analyses are provided in Supplements 1 and 2, respectively.
DISCUSSION
In this study, motion compensation with CAVAREC improved overall quality of motion-degraded intra-procedural liver CBCT images by 18%, and the addition of automated bone segmentation (ZIBOS) to CAVAREC did not provide additional improvement. Quantitative analysis of vessel sharpness demonstrated a statistically significant improvement with CAVAREC (mean, 0.287 corrected vs. 0.281 uncorrected, P = 0.02) but not with CAVAREC + ZIBOS (mean, 0.284 corrected vs. 0.281 uncorrected, P = 0.3). Finally, reader studies showed small but statistically significant improvement in multiple clinically relevant imaging characteristics with CAVAREC, with ZIBOS providing no additional benefit.
Performance of automated bone segmentation as in this study has not been reported previously. However, manual bone segmentation has been tested previously, with similar results to this study: Vessel sharpness improved with motion correction, but bone segmentation provided little additional benefit.[5] The use of motion correction algorithms alone without bone segmentation for liver CBCTs has been reported previously, including the same CAVAREC algorithm as in this study[4] as well as a different algorithm, Motion Freeze (GE Healthcare),[6,7] with results largely concordant with the motion correction alone portion of the present study.
CAVAREC has also been tested for motion correction in other settings. Correction of pulmonary arteries CBCTs led to a similar small but statistically significant improvement across clinically relevant imaging features.[9] In addition, CAVAREC improved image sharpness in phantom and clinical lung CBCTs obtained during diagnostic bronchoscopy, which unlike the previous applications does not involve contrast injection.[8] Motion correction has also been investigated in brain CBCTs obtained during acute stroke imaging, where it increased the proportion of images adequate for clinical decision making from 76% to 93%.[10] Thus, multiple motion correction algorithms in multiple settings demonstrate similar performance on motion-degraded CBCT images, although no study has previously reported on the concurrent use of an automated bone segmentation algorithm.
This study had several limitations. It was retrospective. The strength of significance observed in most statistical tests suggests that results would be similar with a larger sample size; however, the sample size in this study was 48 images. Only three readers participated in evaluating the images for qualitative assessment, and although they were blinded to CAVAREC versus CAVAREC + ZIBOS, they were aware which image was the reference image and, by implication, the uncorrected image. Images from two centers were included in the study, which provided more generalizability than a single-center study but less than a large, prospective multicenter study. Finally, the algorithm was not tested on images without motion artifacts.
CONCLUSION
Motion correction of liver CBCT images degraded by motion artifacts using CAVAREC provides modest improvement in image quality, and combining ZIBOS, an automated bone segmentation algorithm, with CAVAREC does not appear to degrade image quality.
Ethical approval:
The research/study was approved by the Institutional Review Board at Johns Hopkins University School of Medicine, number IRB00316420, dated 10/17/2023.
Declaration of patient consent:
Patient’s consent is not required as the patients identity is not disclosed or compromised.
Conflicts of interest:
CRW Grant support (within last 12 months): Siemens Healthineers AG, Medtronic, Boston Scientific, Guerbet, NIH (NHLBI), DoD CRW Consultant: Palvella Therapeutics, Cook Medical, Guerbet, Medtronic, Boston Scientific, Siemens Healthineers AG. PF, AP, and TE are employed by Siemens Healthineers AG.
Use of artificial intelligence (AI)-assisted technology for manuscript preparation:
The authors confirm that there was no use of artificial intelligence (AI)-assisted technology for assisting in the writing or editing of the manuscript and no images were manipulated using AI.
Financial support and sponsorship: This work was funded by grant support from Siemens Healthineers AG (CRW).
References
- Technology Assessment Committee of the Society of Interventional Radiology. C-arm cone-beam CT: General principles and technical considerations for use in interventional radiology. J Vasc Interv Radiol. 2008;19:814-20.
- [CrossRef] [PubMed] [Google Scholar]
- C-arm cone-beam computed tomography in interventional oncology: Technical aspects and clinical applications. Radiol Med. 2014;119:521-32.
- [CrossRef] [PubMed] [Google Scholar]
- Artefacts in CBCT: A review. Dentomaxillofac Radiol. 2011;40:265-73.
- [CrossRef] [PubMed] [Google Scholar]
- A motion artifact correction algorithm for cone-beam CT in patients with hepatic malignancies treated with transarterial chemoembolization. J Vasc Interv Radiol. 2022;33:1367-74.e2.
- [CrossRef] [PubMed] [Google Scholar]
- Evaluation of a motion correction algorithm for C-arm computed tomography acquired during transarterial chemoembolization. Cardiovasc Intervent Radiol. 2021;44:610-8.
- [CrossRef] [PubMed] [Google Scholar]
- Clinical impact of a new cone beam CT angiography respiratory motion artifact reduction algorithm during hepatic intra-arterial interventions. Eur Radiol. 2020;30:163-74.
- [CrossRef] [PubMed] [Google Scholar]
- Retrospective use of breathing motion compensation technology (MCT) enhances vessel detection software performance. Cardiovasc Intervent Radiol. 2021;44:619-24.
- [CrossRef] [PubMed] [Google Scholar]
- Feasibility of a prototype image reconstruction algorithm for motion correction in interventional cone-beam CT scans. Acad Radiol. 2024;31:2434-43.
- [CrossRef] [PubMed] [Google Scholar]
- Motion reduction for C-arm computed tomography of the pulmonary arteries: Image quality of a motion correction algorithm in patients with chronic thromboembolic hypertension during balloon pulmonary angioplasty. Rofo. 2021;193:1074-80.
- [CrossRef] [PubMed] [Google Scholar]
- Motion artifact correction for cone beam CT stroke imaging: A prospective series. J Neurointerv Surg. 2023;15:e223-8.
- [CrossRef] [PubMed] [Google Scholar]
- Interventional 4D motion estimation and reconstruction of cardiac vasculature without motion periodicity assumption. Med Image Anal. 2010;14:687-94.
- [CrossRef] [PubMed] [Google Scholar]
- CoroEval: A multi-platform, multi-modality tool for the evaluation of 3D coronary vessel reconstructions. Phys Med Biol. 2014;59:5163-74.
- [CrossRef] [PubMed] [Google Scholar]
- The REDCap consortium: Building an international community of software platform partners. J Biomed Inform. 2019;95:103208.
- [CrossRef] [PubMed] [Google Scholar]
- 7.4.7.4. Comparing multiple proportions: The marascuillo procedure. 2024 Available from: https://www.itl.nist.gov/div898/handbook/prc/section4/prc474.htm [Last accessed on 2025 Jun 24]
- [CrossRef] [Google Scholar]