S2498
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
performance differed by organ. Accordingly, interpretation requires a reliable, organ-matched RC, preferably estimated with long-term repeats to better reflect real-world variability. For multicenter analyses, harmonization can combine calibration on stable reference structures with post-processing corrections to limit site/system effects. Larger multicenter cohorts and ongoing sequence refinements will help
consolidate clinical standardization. Keywords: MR-Linac,Functional MRI repeatability,Rectal tumor
Digital Poster 5143
Visualization for Detecting Lung SBRT Failure Patterns Through CT-Biological Effective Dose Interaction – A Vision Transformer Approach Shuo Wang, Yu Lei, Chi Lin Radiation Oncology, University of Nebraska Medical Center, Omaha, USA Purpose/Objective: The purpose of this study was to leverage neural network visualization tools to detect high-attention areas indicative of treatment failure in lung SBRT by examining correlation matrices between CT and Biological Effective Dose (BED). Material/Methods: We retrospectively studied 179 lung cancer patients treated with SBRT between 2007 and 2022, among which 172 patients had quality 4D planning CT. Each clinical plan was recalculated on Free-Breathing (FB) CT or Average Intensity Projection CT (Average CT) to ensure dose was calculated on both CTs. We created the voxelated BED dose matrix for each plan based on their prescription and calculated four correlation matrices between CT and voxelated BED: Entropy, Spearman Rank, Jensen–Shannon Divergence (JSD) and Wasserstein Distance (WD)1. For each correlation matrix, we trained our data on a 3D Vision Transformer model2 to predict treatment failure and generated a high-attention region by applying a threshold at the 95th percentile of the values within the 3D attention maps generated by Grad-CAM.
Results: With a median follow-up time of 34.4 months, the FB/AIP dataset included 125/120 non-failure cases and 54/52 failure cases. All Vision Transformer models using correlation matrices achieved 100% accuracy in predicting treatment failure on the both datasets, with high-attention areas overlapping significantly more with a peritumoral region (2cm outside PTV) than PTV. The high-attention areas identified by models using WD, Spearman, JSD, and Entropy matrices overlapped with the peritumoral region at percentages of 45.4%, 18.6%, 41.4%, and 45.7% for the FB dataset, and 51.7%, 29.5%, 29.5%, and 39.0% for the average CT dataset. Overlap with the peritumoral region was significantly higher than with PTV, which showed percentages of 10.6%, 5.2%, 8.3%, and 13.2% for the FB dataset and 12.1%, 7.3%, 5.9%, and 9.0% for the AIP dataset.
Made with FlippingBook - Share PDF online