ESTRO 2023 - Abstract Book

S1061

Digital Posters

ESTRO 2023

Conclusion PCPC could be recognized as a clinical endpoint to evaluate the complications and predict survival of CCRT in the NSCLC. Hypoalbuminemia might strongly predict PCPC in locally advanced NSCLC.

PO-1327 Radiomics and Dosiomics Analysis to predict Pneumonitis after Stereotactic Radioimmunotherapy

K.M. Kraus 1,3,4 , M. Oresheko 2 , D. Bernhardt 2,4 , S.E. Combs 2,4,5 , J.C. Peeken 2,5,4 , M. Oreshko 6

1 Technical Universty of Munich (TUM), Radiation Oncology, Munich, Germany; 2 Technical University of Munich (TUM), Radiation Oncology, Munich, Germany; 3 Helmholtz Zentrum München (HMGU) GmbH, German Research Center for Environmental Health, Institute of Radiation Medicine (IRM), Neuherberg, Germany; 4 German Cancer Consortium (DKTK), Partner Site Munich and German Cancer Research Center (DKFZ), Munich, Germany; 5 Helmholtz Zentrum München (HMGU) GmbH, German Research Center for Environmental Health, Institute of Radiation Medicine (IRM), Munich, Germany; 6 LMU Munich, Medical Faculty, University hospital, Munich, Germany Purpose or Objective The impact of immune checkpoint inhibition (ICI) on the occurrence of post-therapy pneumonitis (PTP) is unclear. We aim to find predictors for the occurrence of pneumonitis after lung stereotactic body radiation therapy using radiomics and dosiomics analysis of 3 D dose distributions for lung cancer and to evaluate the potential impact of ICI on the risk for PTP after combined radioimmunotherapy. Materials and Methods 110 cases of primary lung cancer or pulmonary metastases that received SBRT between 2010 and 2021 were retrospectively collected. Eleven patients received radioimmunotherapy with ICI. In total, 24 suffered from PTP. Clinical data including the occurrence of pneumonitis > grade 0 according to the Common Terminology Criteria for Adverse Events version 5.0 and the use of ICI within an interval of 50 days around SBRT were extracted from patient files. Planning CTs, 3D dose distributions, lung and treatment volume segmentations were selected from the radiotherapy treatment planning system. 104 radiomics and dosiomics features were extracted from the planning CT and 3D dose distributions from each volume of interest (total/ipsilateral lung – GTV, PTV+2cm) using the open-source library Pyradiomics in python, respectively. Feature reduction was performed using 1000-fold bootstrapping using pearson intercorrelation coefficient and the Boruta algorithm. We studied three single predictive models (radiomics, dosiomics, clinical data) and one combined model (dosiomics + radiomics). Different machine learning models including random forest (rf), logistic elastic net regression (glmnet), support vector machine (svmRadial) and logitBoost were trained and tested using a 5-fold nested cross validation approach and Synthetic Minority Oversampling Technique resampling in R [1]. Model performance was analyzed using the area under the receiver operating characteristic curve (AUC) on the test sets of the outer folds. Results The best radiomics and dosiomics models (rf, each) predicted PTP better than random (AUC = 0.73 (95 % confidence interval 0.72 – 0.74) and AUC = 0.70 (0.68 – 0.71), respectively). The combination of dosiomics and radiomics features showed the highest predictive performance using random forest (rf) classifier (AUC = 0.79, 0.78 – 0.80). The clinical model did not predict PTP better than random. The AUC values for all models and classifiers applied are depicted in table 1. There was no influence of ICI therapy with an AUC ranging from 0.45 to 0.5 depending on the classifier used.