S2421
Physics - Radiomics, functional and biological imaging, and outcome prediction
ESTRO 2026
treated with moderately hypofractionated radiotherapy. Material/Methods:
The study included 1172 consecutive breast cancer patients treated at IRCCS San Raffaele Hospital with three-dimensional conformal radiotherapy (3DCRT) to the whole breast (40Gy in 15 fractions) before 2017[1]. Eighteen patients developed clinically and radiologically confirmed moderate/severe pulmonary events within three years. For each case, DVHs and planning CT–based densitometric parameters were extracted for both lungs, focusing on the ipsilateral one. Additional clinical factors and quantitative cardiac calcification scores (Agatston score, calcification volume, maximum HU value) were also available. Univariate and multivariate logistic regression analyses were performed to identify predictors of toxicity. Before multivariate modeling, collinearity and correlations were assessed using Spearman’s rank correlation and the Variance Inflation Factor. Results: At univariate logistic regression, several densitometry ipsilateral lung features (HU median, mean 90% percentiles), V37Gy of the ipsilateral lung, lung volume and CAC scores were associated with an increased risk. The strongest ones in terms of OR and p-value were the lung volume (AUC=0.68,P=0.0084) and the CAC scores (for instance MaxHU: AUC=0.64 P =0.004). Based on the Youden index, the optimal cut-offs were identified as 1745 cc for lung volume, 232 HU for Max HU, and 7.63 for the Agatston score. When dichotomized, patients with lung volume>1745cc exhibited a markedly increased risk of pulmonary toxicity (P=0.0003,OR=6.21,AUC=0.69), while Max HU>232 was associated with a moderate but significant risk (P=0.045,OR=2.64,AUC=0.61). Similarly, Agatston score>7.63 predicted pulmonary events with P=0.0005,OR=6.54,AUC=0.70.In bivariate logistic models selecting the best combination of two predictors, lung volume and one CAC score resulted in the best performing models: the model combining Vol_Lung>1745 and MaxHU continuous (P <0.0001,OR=5.97+1.0032) showed good performances (Hosmer–Lemeshow P = 0.83, AUC=0.71), as shown in Fig 1. Calibration plot demonstrated excellent agreement between predicted and observed probabilities (R2=0.974), supporting its robustness. An alternative model (Vol_Lung continuous+MaxHU >232, Fig 2) showed similar performance (P=0.0059,OR=1.0015+2.42, HL P=0.72, AUC=0.69,R2=0.99). Fig 1
Conclusion: Several previously known and reported GBM prognostic factors were evaluated. The newly introduced CE parameter, quantifying viable tumor within the GTV prior to RT, demonstrated superior prognostic power for survival compared with established clinical variables. The addition of MGMT and other clinical factors only marginally improved model performance, highlighting the potential for the proposed imaging-derived CE as an independent prognostic biomarker in GBM. Keywords: Glioblastoma, radiotherapy, radiology CT-based Predictors of Radiation-Induced Pulmonary Toxicity in Breast Cancer Patients Marco Fois 1 , Alfonso Belardo 1 , Andrei Fodor 2 , Lucia Perna 1 , Laura Giannini 2 , Paola Mangili 1 , Gabriele Palazzo 1 , Marcella Pasetti 2 , Miriam Torrisi 2 , Roberta Tummineri 2 , Antonella Del Vecchio 1 , Nadia Gisella Di Muzio 2,3 , Tiziana Rancati 4 1 Medical Physics, IRCCS San Raffaele Scientific Institute, Milan, Italy. 2 Radiation Oncology, IRCCS San Raffaele Scientific Institute, Milan, Italy. 3 Faculty of Medicine and Surgery, Vita-Salute San Raffaele University, Milan, Italy. 4 Data Science Unit, Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, Italy Purpose/Objective: The primary objective of this study was to identify dosimetric, clinical and planning CT-based densitometric predictors of radiation-induced pulmonary events in a cohort of breast cancer patients Digital Poster 1103
Made with FlippingBook - Share PDF online