S2453
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
multicenter randomized controlled trial. J Clin Oncol. 2025 Aug 10;43(23):2583-2594. [2] van den Bosch S, Cox MC, Doornaert PAH, et al. Dose de-escalation of elective neck irradiation in head and neck cancer: A secondary analysis of acute toxicity findings from the randomized controlled UPGRADE-RT trial. Radiother Oncol. 2025 Oct 5;213:111196. Keywords: dose surface map, tube feeding
air in the upper aero-digestive tract was systematically delineated including the area between soft palate and the distal mobile tong, the oropharyngeal- and laryngopharyngeal air cavity, and the hypopharynx (figures 1A and 1B). Areas with adjacent mucosa in case of collapsed mucosal folds were manually extended. Three landmarks were placed to subdivide the pharyngeal mucosa into anatomical subregions (figure 1B). Dose surface maps (DSM) were acquired using equidistant sampling (figure 1C). DSMs were geographically matched by primary tumor laterality and by using the 3 landmarks. DSMs were accumulated for patients with and without tube feeding (acute dysphagia grade ≥ 3). Absolute dose difference between accumulated DSMs was calculated (figure 2B) and subregions with statistically significant dose differences (p ≤ 0.01) were identified by pixel-wise permutation tests (figure 2C). A subgroup analysis based on tumor site was performed (oropharynx versus larynx/hypopharynx). Results: DSMs could be successfully created for all 295 evaluable patients. In patients requiring tube feeding during treatment, the most pronounced subregion of pharyngeal mucosa with statistically significant higher dose compared to those without tube feeding was identified at the level of epiglottis, caudal part of the base of the tongue and at the contralateral laryngopharyngeal wall (figure 2C). Similar results were found for the subgroup of larynx/hypopharynx tumors (n=178), while no subregions with significant dose differences could be identified for the subgroup of oropharynx tumors (n=117). Conclusion: The use of DSMs of the pharyngeal mucosa in head and neck cancer patients is feasible. High dose to the subregions of the pharyngeal mucosa near the epiglottis, the caudal part of the base of the tongue and at the contralateral laryngopharyngeal wall are associated with tube feeding (acute dysphagia grade ≥ 3). References: [1] van den Bosch S, Doornaert PAH, Hoebers FJP, et al. Clinical benefit and safety of reduced elective dose
Digital Poster Highlight 2911
Radiomics and 4DCT phase analysis for predicting post-SBRT liver outcomes: a retrospective cohort study Tiago Engenheiro 1,2 , Vitali Moiseenko 3 , Casey Bojechko 3 , Austin Hopper 3 , Jona Hattangadi-Gluth 3 , Sangwoo Kim 3 , Michael Sherer 3 , Renata Georgia Raidou 1 1 Institute of Visual Computing & Human-Centered Technology, TU Wien, Vienna, Austria. 2 Informatics, Instituto Superior Tecnico, Lisbon, Portugal. 3 Department of Radiation Medicine and Applied Sciences, UC San Diego, La Jolla, USA Purpose/Objective: To explore personalized evaluation of post-SBRT liver outcomes through radiomics and machine learning, we (i) assess model performance for predicting liver toxicity, and identify key radiomic predictors; (ii) evaluate the impact of optimal 4DCT phase selection on prediction accuracy; and (iii) explore whether visualizations support identifying phase stability [1]. Material/Methods: Data from 63 patients with primary (32 patients, hepatocellular carcinoma or cholangiocarcinoma) or metastatic liver cancer (31 patients) treated with gated SBRT (25-60 Gy in 3-5 fractions) at UC San Diego were retrospectively analyzed. 4DCT data for all breathing phases (0-90%), CT averaged over phases selected for gating, and delineated structures were included. Gastrointestinal toxicity, lab values, and Child-Pugh (CP) scores were extracted pre- and post-SBRT. Radiomic features [2] were extracted from healthy liver tissue and standardized. Feature stability was assessed across phases using the intra-class correlation coefficient (ICC ≥ 0.75), and redundant features (PCC>0.9) were excluded. Feature selection employed logistic regression with LASSO regularization within a stratified 5-fold cross-validation [3]. Logistic regression and random forest models [4] to predict post-SBRT liver toxicity were evaluated using ROC-AUC metrics. SHAP values [5] quantified feature importance. Radial plots investigated stable vs. average phase deviations and associations with clinical outcomes.
in definitive radiotherapy for head and neck squamous cell carcinoma: the UPGRADE-RT
Made with FlippingBook - Share PDF online