Evaluation of an iterative motion-correction algorithm for hepatic cone-beam CT during transarterial interventions
Phillip S Engler1, Gerd Grözinger2, Sven S Walter1
1Department of Diagnostic and Interventional Radiology, Universitätsklinikum Tübingen, Hoppe-Seyler-Str. 3, 72076 Tübingen, Germany.
Purpose:
To evaluate an iterative motion-correction algorithm for periinterventional hepatic cone-beam computed tomography (CBCT) regarding its efficacy in reducing motion artifacts, improving vessel depiction, and enhancing diagnostic confidence during transarterial interventions.
Materials And Methods:
This retrospective single-centre study included 69 CBCT datasets from 69 patients undergoing TACE or SIRT between 2018 and 2021. One CBCT dataset per patient served as the unit of analysis. Each dataset was post-processed with a motion-correction algorithm at 100-iteration increments (It0 = Baseline; It100-It1000). Three interventional radiologists independently assessed motion artifacts (MA) and vessel depiction (VD) using a 5-point Likert scale (higher ratings representing better image quality). Newly generated artifacts (NGA) were assessed on a 0-5 scale ("0″ representing no evidence of NGA, higher ratings representing stronger artifacts). Target lesion (TL) and vessel tree (VT) visibility were defined as yes/no. Non-parametric tests were applied, including Friedman and Wilcoxon signed-rank tests for ordinal ratings and Cochran's Q and McNemar tests for binary visibility outcomes. Interrater reliability was determined using Fleiss' kappa.
Results:
Compared with baseline datasets, median ratings for motion artifacts and vessel depiction improved from 3 (IQR 1-2) to 4 (IQR 1) at iteration levels 200-300 (all p ≤ 0.001). Target lesion and vessel tree visibility increased from baseline values of 42-44% and 30-41% to peak rates of 65-75% and 54-64%, respectively, with significant differences across iteration levels for all readers (all p < 0.001). Beyond 500 iterations, diagnostic quality declined due to progressively increasing NGA. The proportion of non-diagnostic datasets increased continuously with higher iteration levels, ranging from 1.4% at It300 to 88.4% at It1000. Interrater agreement across parameters was moderate to substantial (κ = 0.40-0.70).
Conclusion:
Iterative motion correction significantly improves image quality in hepatic CBCT during transarterial interventions at moderate iteration levels (200-300). Excessive iteration introduces new artifacts, underscoring the importance of optimizing mid-range iteration settings to balance motion correction and artifact generation.
