S2077
Physics - Image acquisition and processing
ESTRO 2026
Results: Gamma-Area>0.3 was the most effective metric for identifying reconstruction differences, showing strong correlation with reader ratings (r=0.7–0.8) and providing spatial localization via heatmaps/Surface- Dice. Other metrics were less sensitive to spatial discrepancies. In cases where discrepancies arose between reader ratings and the chosen metric, further review revealed that some artifacts had not been detected by human observers.In the initial cohort, minor changes were observed in 75% of phase-based and 80% of amplitude-based reconstructions. Surface- Dice values ranged from 0.6-0.7 when organ-level changes occurred, exceeded 0.9 when anatomy was preserved. Substantial changed images were rated anatomically more accurate. These trends were confirmed in an independent cohort, where nearly 90% of amplitude-based and 80% of phase-based reconstructions showed only minor changes. Substantial differences were rare, occurring in <10% of cases. Examples in Figure 2.
anatomical correctness across a larger cohort is challenging because ZeeFree also affects organs-at- risk(OARs). This study aims to detect affected patients and assess whether anatomical accuracy is improved in clinically relevant cases. Material/Methods: This retrospective study analysed 4DCT scans from 80 patients with lung tumors affected by respiratory motion. All scans were acquired using on a SOMATOM-go.Open-Pro-scanner (Siemens- Healthineers) with breathing-adapted-binning (i4DCT)[2]. Two reconstruction methods were compared: the Standard reconstruction and a novel algorithm called ZeeFree, which is based on a DEMON- based joint 3D vector field approach designed to correct stack-misalignments and reduce motion- artifacts[1].Organs-at-risk were segmented using TotalSegmentator [3]. To assess anatomical and image quality differences, quantitative metrics were computed, including (Surface-)Dice coefficient and gamma analysis. The gamma evaluation integrated geometric and intensity-based differences, using a threshold of 50HU/2mm. This threshold was chosen to reflect clinically plausible soft-tissue contrast and spatial tolerances that align with typical inter-observer variability in radiotherapy contouring. To account for skewed distributions, thresholds were derived using the median-absolute-deviation based on a subset of 30 patients selected for strong breathing irregularities. Based on those thresholds, further 50 patients were evaluated and categorized.(see Figure 1)
Made with FlippingBook - Share PDF online