S2156
Physics - Inter-fraction motion management and daily adaptive radiotherapy
ESTRO 2026
Material/Methods: A retrospective study of twenty pelvic cancer patients compared clinical CBCT (cCBCT) images against the novel Iris CBCT reconstruction. The Iris reconstruction integrates two key components: (1) a convolutional neural network for scatter estimation trained using Monte Carlo simulations, providing inline scatter correction during acquisition, and (2) a polyquant model-based iterative reconstruction that directly reconstructs virtual monoenergetic images at 64 keV using a polychromatic forward model with total variation regularization1,2,3. Evaluation assessed: 1) DL auto-segmentation (DLAS)4 accuracy for six organs- at-risk (Bladder, Left/Right Femoral Heads, Prostate, Rectum, Sacrum) against MD drawn contours using Dice Similarity Coefficient (DSC) and Mean Distance to Agreement (MDA); 2) Hounsfield Unit (HU) consistency with planning CT; and 3) dosimetric agreement via 3D gamma analysis.To target gas motion artifacts, a five- patient subset with pronounced streaking underwent additional processing with the IMAC algorithm. This neural network, trained on synthetic paired data (artifact-laden/clean images), performs slice-wise artifact suppression while preserving unaffected anatomy. IMAC's efficacy was quantified by comparing DLAS performance (DSC/MDA) of IMAC-processed volumes (Iris-IMAC) to Iris-CBCT. Results: Iris CBCT significantly improved image quality (Figure1) and quantitative metrics. DLAS performed robustly, showing excellent agreement with reference contours (mean DSC=0.9, mean MDA=1.6 mm). Iris CBCT HU values showed markedly improved consistency with the planning CT with mean HU difference of 42 compared to 297 for cCBCT. Dosimetrically, Iris CBCT outperformed cCBCT, with a mean gamma pass rate (1%/2mm) of 97.3±1.7% compared to 85.6±17.3%. The implementation of additional artifact reduction (Iris- IMAC) led to a visible decrease in streaking and shading caused by gas artifacts (Figure 2). DLAS improvement was quantified by a reduction in contour variability, evidenced by an overall lower standard deviation in DSC and MDA for the Iris-IMAC.
direction: ML (1.97 ± 1.54 mm vs 2.63 ± 2.06 mm, p = 0.03); CC (3.13 ± 2.34 mm vs 4.03 ± 2.67 mm, p = 0.001); AP (3.49 ± 2.26 mm vs 2.82 ± 2.29 mm, p = 0.001). Rotational variations were not statistically significant (p = 0.14). The mean 3D vector shift, calculated as the root-mean-square of translational displacements, was significantly smaller with SGRT (1.91 ± 0.36 mm; 95% CI 1.78–2.04 mm) than with mask-based setup (2.26 ± 0.48 mm; 95% CI 2.09–2.43 mm; p < 0.05). Corresponding PTV margins were reduced for SGRT: ML (4.9 mm vs 6.8 mm), CC (8.1 mm vs 9.9 mm), and AP (7.7 mm vs 8.7 mm). Mean setup duration was significantly shorter with SGRT (2.51 ± 0.81 min vs 3.55 ± 0.99 min, p = 0.007). Conclusion: Maskless SGRT achieved improved inter-fraction setup accuracy, smaller PTV margins, and faster setup times compared with conventional mask-based laser alignment. SGRT offers a precise, patient-friendly, and time-efficient approach for pelvic cancer radiotherapy. References: 1. Sotiropoulou V, Kouloulias V, Kouvaris J et al. Comparison between the SGRT and the conventional setup method for patients undergoing VMAT for pelvic malignancies. Appl Radiat Isot. 2025 Mar; 217:111659.2. Rudat V, Nour A, Alaradi A et al. Set up accuracy and margins for surface-guided radiotherapy (SGRT) of head, thorax, abdomen, and pelvic target volumes. Sci Rep. 2023 Oct 9;13(1):17018.3. Van Herk M, Remeijer P, Rasch C et al. The probability of correct target dosage: dose-population histograms for deriving treatment margins in radiotherapy. Int J Radiat Oncol Biol Phys. 2000 Jul 1;47(4):1121-35. Keywords: Surface Guided RT, PTV margin, Setup Accuracy Evaluation of A Deep-Learning and Polyquant CBCT Reconstruction with Gas Motion Artifact Reduction for Robust Pelvic Adaptive Radiotherapy Eenas A Omari 1 , Andrew Keeler 1 , Thomas Joyce 2 , Jonathan H Mason 2 , Monica E. Shukla 1 , Eric Paulson 1 1 Radiation Oncology, Medical College of Wiscosnin, Milwaukee, USA. 2 Elekta, AB, Stockhom, Sweden Purpose/Objective: CBCT image quality from c-arm linacs is often insufficient for accurate dose calculation and segmentation in adaptive radiotherapy, primarily due to scatter, beam hardening, and artifacts. This study evaluates a novel reconstruction algorithm which integrates deep-learning (DL) scatter correction with polyquant model-based iterative reconstruction (MBIR), with an additional investigation of a mage- space Motion Artifact Correction (IMAC) algorithm. Digital Poster Highlight 3374
Made with FlippingBook - Share PDF online