ESTRO 2021 Abstract Book

S509

ESTRO 2021

Results On the whole dataset, age at irradiation was the only clinical variable significantly correlated with RP ( p =0.004). VBA highlighted (Figure) two large clusters (total volume 551 cc) significantly associated with RP in the lungs (346 cc) and in the heart (205 cc). The subregions mostly involved were the right lower and middle lung lobes, the left and right atrium with their walls and the pericardium. The mean BED in heart-S 0.05 for RP patients was 33.3 Gy, and for patients without RP was 25.2 Gy. The mean BED in lungs-S 0.05 for RP patients was 38.5 Gy, and for patients without RP was 25.2 Gy. PICA detected 63 dose clusters homogenously spread across the thorax. Connectograms showed that, while doses to main structures (cardiac chambers and lung lobes) were weakly correlated (Spearman coefficient Rs 2 <0.2), Rs 2 between adjacent lobe segments or chambers and related walls can reach values as high as 0.8.

Figure Coronal CT views fused with: a) voxel-wise mean of BED maps for the analyzed patients; b) difference of mean BED between patients with RP and without RP; c) voxel-wise standard deviation of BED for the analyzed patients; d) significance of the BED difference expressed as –log p Conclusion The spread of PICA clusters suggests a uniform power of possible VBA inferences on the dataset. This lays robust foundations to the obtained findings that, taking advantage from the pooled data analysis of different thoracic tumors and different RT modalities, confirmed the lower parts of the lungs and the heart as key players in the development of RP. Connectograms showed that the dataset can support a radiobiological differentiation among the heart and lung main substructures. OC-0642 A radiomics based prognostic model for patients with head and neck squamous cell carcinoma S. Keek 1 , F. Wesseling 2 , H. Woodruff 1,16 , J. van Timmeren 3 , I. Nauta 4 , T. Hoffmann 5 , S. Cavalieri 6 , G. Calareso 7 , S. Primakov 8 , R. Leijenaar 9 , L. Licitra 6,15 , M. Ravanelli 10 , K. Scheckenbach 11 , T. Poli 12 , D. Lanfranco 12 , M. Vergeer 13 , R. Leemans 4 , R. Brakenhoff 4 , F. Hoebers 2 , P. Lambin 14,16 1 Maastricht University, Precision Medicine, Maastricht, The Netherlands; 2 Maastricht University Medical Centre+, Department of Radiation Oncology (MAASTRO), Maastricht, The Netherlands; 3 University of Zürich, Department of Radiation Oncology, Zürich, Switzerland; 4 Vrije Universiteit Amsterdam, Otolaryngology/Head and Neck Surgery, Amsterdam, The Netherlands; 5 University of Ulm, Dept. of Otorhinolaryngology, Ulm, Germany; 6 Fondazione IRCCS Istituto Nazionale dei Tumori, Head and Neck Medical Oncology Unit, Milan, Italy; 7 Fondazione IRCCS Istituto Nazionale dei Tumori, Radiology Unit, Milan, Italy; 8 Maastricht Unversity, Precision Medicine, Maastricht, The Netherlands; 9 Oncoradiomics , -, Liège, Belgium; 10 University of Brescia, Department of Medicine and Surgery, Milan, Italy; 11 University Hospital Düsseldorf, Dept. of Otorhinolaryngology- Head and Neck Surgery, Düsseldorf, Germany; 12 University of Parma – University Hospital of Parma, Maxillofacial Surgery Unit, Department of Medicine and Surgery, Parma, Italy; 13 Vrije Universiteit Amsterdam, Department of Radiation Oncology, Amsterdam, The Netherlands; 14 Maastricht University, Department of Precision Medicine, Maastricht, The Netherlands; 15 University of Milan, Department of Oncology and Hemato-Oncology, Milan, Italy; 16 Maastricht University Medical Centre+, Department of Radiology and Nuclear Medicine, Maastricht, The Netherlands Purpose or Objective Patients with locoregionally advanced head-and-neck squamous cell carcinoma (HNSCC) have high relapse and mortality rates. RNA and DNA profiling have identified molecular subtypes of HNSCC with different prognosis. Some of these subtypes may include primary tumors with high heterogeneity which may react differently to treatment. Imaging-based decision support could improve outcome by optimizing personalized treatment, and support stratification in clinical trials. We hypothesized that a multifactorial prognostic model including computed tomography (CT) radiomics-based features for advanced HNSCC improves risk-stratification