ESTRO 2021 Abstract Book

S1549

ESTRO 2021

Materials and Methods Inclusion criteria were: biopsy proved non-mucinous LARC, 3.0T MRI, availability of clinical outcome and long-course CRT. MRI scans, with T2-weighted and DWI sequences, of 72 LARC patients were included. Two readers independently delineated each tumor. Radiomic features were extracted from both the “tumor core” (TC) and the “tumor border” (TB). Partial least square (PLS) regression was used as the multivariate, machine learning, algorithm of choice and leave-one-out nested cross- validation was used to optimize hyperparameters of the PLS. Nine MRI-based features were defined “clinical features”: tumor location (high, middle, low rectum), whole tumor volume, cranio-caudal extension, distance from the internal anal sphincter, mesorectal fascia infiltration, extramural vascular invasion, extramural depth invasion, T and N-stage. The extraction of the radiomic features from TB and TC T2w images was performed using PyRadiomics. The classification performances were assessed through Receiver Operating Characteristic (ROC) analyses. The Statistical Analysis was performed in Matlab. Results Nine MRI-based features were used in the machine learning model and 1405 radiomic features extracted were used for analysis. Combining the 9 clinical features with the 1405 radiomic features, the MRI-Based “clinical-radiomic” machine learning model properly predicted treatment response (AUC=0.793, p=5.6*10 -5 ). The only important feature with negative weight was tumor location, obtaining a better response for low tumors. When using the 9 clinical and 790 radiomic features extracted only from the TC, an AUC = 0.689 was obtained; using the 9 clinical and 626 radiomic features extracted only from the TB, the AUC was 0.541. A highly synergistic effects was obtained combining TB and TC features, replicating the results previously found with an AUC=0.793.T Conclusion Tumor prediction improved combining MRI-based clinical features and radiomic features, the latter extracted from both TC and TB. Prospective validation studies in randomized clinical trials are warranted to better define radiomics role in the development of rectal cancer precision medicine. PO-1820 Intra-mandible radio-sensitivity for osteoradionecrosis: effect of local dose and teeth extractions N. Sijtsema 1,2 , G. Verduijn 1 , Y. van Norden 1 , H. Mast 3 , A. van der Lugt 2 , M. Hoogeman 1,4 , S. Petit 1 1 Erasmus MC Cancer Institute, Department of Radiotherapy, Rotterdam, The Netherlands; 2 Erasmus MC, Department of Radiology and Nuclear Medicine, Rotterdam, The Netherlands; 3 Erasmus MC, Department of Oral and Maxillofacial Surgery, Rotterdam, The Netherlands; 4 HollandPTC, Department of Medical Physics and Informatics, Delft, The Netherlands Purpose or Objective Osteoradionecrosis (ORN) of the mandible is a severe late effect of radiotherapy for head and neck tumors. It is most often observed in premolar and molar regions, which suggest not all mandibular regions are equally prone for development of ORN. Therefore, we investigated the interaction between the ORN location, the dose to that location and the location of teeth extractions in a group of 324 oropharyngeal squamous cell carcinoma patients treated with (chemo)radiotherapy. Materials and Methods For patients with ORN after radiotherapy, the ORN volume was delineated on the planning CT. All dose distributions were converted to 2 Gy equivalent dose based on α/β=3 Gy, accumulated, and deformed non-rigidly to a reference patient using ADMIRE 3.7.7 (Elekta AB, Stockholm, Sweden). The reference patient contained delineations of 16 mandible subregions corresponding to the 16 dental elements. Multi-variable logistic regressions were performed per mandible subregion to relate the presence of ORN in a subregion to its mean dose, whether the dental element in the subregion was extracted and whether its neighbor elements (i.e. other dental elements on the same side of the mandible) were extracted. Likelihood ratio tests were performed to compare models, with p≤0.05 considered statistically significant. In the first regression all dental elements from all patients were combined. Next, separate regression analyses were performed per subregion (left and right combined). Results Patients could be registered accurately to the reference patient. The 95 percentile of the distance between the mandible delineations was on average 3.50 mm (range 1.51-8.12 mm). 27 Patients developed ORN (8%). In the combined analysis the mean dose to the mandible subregion, pre-radiotherapy teeth extraction in the mandible subregion, and pre-radiotherapy teeth extractions of neighboring dental elements were all associated with an increased risk of ORN (Figure 1). The odds ratios were 1.073 per unit increase in Gy (95% confidence interval (CI): (1.059,1.088)), 1.79 (95% CI: (1.023,3.11)), and 2.24 (95% CI: (1.39,3.54)), respectively. In the analysis per mandible subregion (Table 1) the dose was associated with an increased risk of ORN in the second premolar, first molar, second molar, and third molar. Extractions were associated with an increased risk of ORN only in the second premolar (p-value = 0.03) and in the second molar (p-value = 0.004). Due to limited statistical power, extraction of neighboring dental elements was not taken into account in the regressions per subregion.