ESTRO 2023 - Abstract Book

S27

Saturday 13 May

ESTRO 2023

future work to determine their ability to identify patients at risk of future disease progression during their treatment planning in clinical practice.

MO-0059 Prediction of pathological response to chemo-radiotherapy in rectal cancer using federated learning I. Bermejo 1 , P. Mateus 1 , M. Savino 2 , B. Osong 1 , Y. Willems 1 , N.D. Capocchiano 2 , M. Berbée 1 , M.A. Gambacorta 3 , A. Damiani 2 , V. Valentini 2 , A. Dekker 1 1 Maastricht University, Department of Radiation Oncology (Maastro), GROW School for Oncology and Reproduction, Maastricht, The Netherlands; 2 Università Cattolica S. Cuore, Fondazione Policlinico Universitario A. Gemelli IRCCS, Rome, Italy; 3 Fondazione Policlinico Universitario A. Gemelli IRCCS, Università Cattolica S. Cuore, Rome, Italy Purpose or Objective After (chemo)radiotherapy, patients with rectal cancer who achieve pathological complete response (pCR) have a significantly better prognosis and can avoid surgery, thus retaining a higher quality of life. The likelihood of pCR increases with the dose of radiation, but so do side effects. Therefore, it is of particular interest to predict which patients will achieve a near complete response, so these patients might be considered for treatment intensification. In this study, we have trained a model to predict near-complete pathological response using data from different clinics. Materials and Methods We used data extracted from two European radiation oncology clinics to develop a prediction model for near complete response (defined as a residual tumour diameter < 2cm and pT1-2N0M0). The variables included in the model based on availability and expert knowledge were: T, N and M stages, mesorectal fascia involvement, WHO status, tumour volume, distance to the anal junction, treatment, age, and gender. We imputed missing values using multiple imputation with chained equations for the training dataset. We trained a Bayesian network (BN) applying the nonparametric bootstrap using the hill climbing algorithm and the Bayesian Dirichlet equivalent score without sharing data, using federated learning. We asked three radiation oncologists to propose the structure of the BNs and then compared the performance of the BN in terms of area under the ROC curve (AUC) when the structure was determined by experts, when it was learnt from data, and when expert-provided structure was fine-tuned based on data. We split the data 70/30 for training and testing. Results In total, we included data from 1325 patients, 77 of which achieved near complete response. The training and testing AUCs for the BN whose structure was learnt from data were 66% and 58% percent respectively. The BNs whose structure was determined by experts, achieved average AUCs of 76% and 54% for training and testing respectively. Finally, the BNs with expert provided structures fine-tuned with data achieved average AUCs of 66% and 66% for training and testing respectively. See Table 1 for more detailed results. Conclusion Expert-provided structures fine-tuned with data resulted in higher testing AUCs. Near complete response is a difficult outcome to predict, but our model could be used as an additional tool when determining a patient’s treatment plan as long as the uncertainty of the predictions is taken into account. Increasing the sample used for training would likely lead to performance improvements. Fortunately, the federated learning infrastructure we have set up allows for a straightforward addition of new clinics while circumventing issues with data sharing. MO-0060 From voxel-level to patient-level NTCP: an enhanced EUD concept to incorporate tissue heterogeneity E. Bahn 1,2,3,4 , J. Bauer 2,3,4 , S. Harrabi 2,3,4,5 , K. Herfarth 2,3,4,5 , J. Debus 2,3,4,5,1 , M. Alber 2,3,4 1 German Cancer Research Center (DKFZ), Clinical Cooperation Unit Radiation Oncology, Heidelberg, Germany; 2 Heidelberg University Hospital, Department of Radiation Oncology, Heidelberg, Germany; 3 Heidelberg Institute of Radiation Oncology, (HIRO), Heidelberg, Germany; 4 National Center for Tumor diseases, (NCT), Heidelberg, Germany; 5 Heidelberg Ion-Beam Therapy Center, (HIT), Heidelberg, Germany Purpose or Objective Voxel-level NTCP modelling approaches of image-based endpoints have recently gained momentum due to their ability to resolve localized dose response from e.g., variable proton RBE or heterogeneous sensitivity of organs and substructures. A central challenge is hereby the transition from voxel-level to patient-level NTCP to allow clinical decision making. This requires the use of sophisticated dose-volume concepts, the choice of which may profoundly impact model performance. Here, we present a theoretical concept for the voxel to patient level transition of NTCP models for the broad class of serial complications. This permits to account for heterogeneous radiosensitivities of organ substructures even with a single dose- volume measure and thus to harness the synergy of combining both model levels. Materials and Methods The enhanced EUD concept is based on the equivalent uniform dose. We derived a mechanistic motivation for the use of EUD for serial complications and broadened its scope by, briefly, introducing virtual doses to account for effects of tissue heterogeneity on dose response. This gives a functional form: NTCP = 1 - exp ⁡ (-exp ⁡ ( σ ⋅ eEUD - k)) with the enhanced EUD eEUD = ( ∑ d_c^k )^(1/k), whereby doses are virtually adjusted by suborgan-specific parameters θ : d_c = d + θ / σ . We demonstrate the use of this concept at a clinical dataset of 110 low-grade glioma patients treated with proton beam