ESTRO 2023 - Abstract Book

S127

Saturday 13 May

ESTRO 2023

Results The mean rectum shape was found by weighing the re-irradiation rectum shape by 50%, whereas the shape in the primary planCT and the PRS both were given 25% weight. Reason for reducing the weight of the older CT was that primary RT occurred in the years 2001-2010, thereby susceptible to alterations in preparation protocol. In Figure 2, the results for a patient (A) with small rectum for re-irradiation, a decrease of 30% as compared to the primary planCT and a patient (B) with larger rectum but no volumetric change compared to the primary planCT. Compared to the dose from the estimated mean+/-1SD, the rectal shape at planCT lead to an underestimation for patient A, whereas for patient B the dose to the estimated rectum shapes and the planCT was almost identical.

Conclusion We presented a novel method combining patient-specific and population rectal shapes for use in decision-making of re- irradiation for prostate cancer. The model adjust to the variation of the two patient-specific inputs, and should be evaluated for more patients.

PD-0166 Characterization of prostate cancer on bpMRI: comparison of zone and lesion-derived radiomic models

Abstract withdrawn

PD-0167 Predicted tumour probability maps improve deep learning outcome prediction in oropharyngeal cancer B. Ma 1 , A. De Biase 2,3 , J. Guo 2,3 , L. V. van Dijk 2 , J. A. Langendijk 2 , S. Both 2 , P. M.A. van Ooijen 2,2 , N. M. Sijtsema 2 1 University Medical Center Groningen, Department of Radiation Oncology, Groningen, The Netherlands; 2 University Medical Centre Groningen, Department of Radiation Oncology, Groningen, The Netherlands; 3 University Medical Centre Groningen, Data Science Centre in Health (DASH), Groningen, The Netherlands Purpose or Objective Recently, deep learning (DL) algorithms showed to be promising in predicting outcomes such as distant metastasis-free survival (DMFS) and overall survival (OS) using pre-treatment imaging in head and neck cancer. Gross Tumour Volume of the primary tumour (GTVp) segmentation used as additional channel of DL input improved model performance. However, the binary segmentation mask of the GTVp directs the focus of the network to the defined tumour region only and uniformly. DL models trained for tumour segmentation have also been used to generate predicted tumour probability maps (PM) where each pixel value corresponds to the degree of certainty of that pixel to be classified as tumour. PM are less affected by