ESTRO 2022 - Abstract Book

S1426

Abstract book

ESTRO 2022

Conclusion We present the advantages and downsides of two DL algorithms for OAR segmentation in HN CT images. nnU-Net performed slightly better for every OAR, yet this superiority is only remarkable for the parotids. Training and inference times were meaningfully longer compared to DenseVNet. The aim of DL technologies should be to shrink the time spent in manual OAR delineation by providing fast but good enough segmentations still flexible to be modified by an expert without relying on too extensive computational resources. To optimize the potential of DL techniques in the radiotherapy field, the weight between accuracy and computational resources must be therefore carefully evaluated. It is likely more efficient to train and update periodically with new data a simpler network such as DenseVNet than to rely upon complex methods as nnU- Net.

PO-1633 Radiomics predicts the location of local recurrence after reirradiation for head and neck carcinoma

A. Beddok 1,2,3 , V. Calugaru 1,3 , L. Champion 4,2 , C. Nioche 2 , G. Crehange 1,3 , I. Buvat 2

1 Institut Curie, Radiation Oncology, Paris, France; 2 Institut Curie, Laboratory of Translational Imaging in Oncology (LITO). UMR (U1288). , Orsay, France; 3 Institut Curie, Proton Therapy Center, Orsay, France; 4 Institut Curie, Nuclear Medicine, Saint-Cloud, France Purpose or Objective Curative reirradiation (reRT) is a promising alternative for the treatment of local recurrence (LR) of head and neck cancer (HNC). However, up to 50% of patients may experience a second LR within two years after the end of reRT. The aim of our study was to evaluate whether radiomics from FDG PET imaging could predict the localization of this second LR. Materials and Methods Among 23 patients re-irradiated with curative intent from 30/08/2012 to 08/04/2019 for advanced HNC, 14 patients had a second LR. For each of them, the reirradiated GTV was segmented based on the hypermetabolism on PET defined as SUV greater than 5. Twenty-nine parameters including three standardized uptake values (SUVs), five first order statistics derived from the gray-level histogram and 31 texture indices were extracted from these GTV using the LIFEX software after spatial resampling: 2 x 2 x 2 mm, intensity discretization: fixed bin size of 0.157, SUV units from 0 - 20. The volume of the second LR, called recurrent tumor volume (Vrecur, red line on Figure 1 ), was identified on PET scans. The LR were categorized as “in-field”, “marginal”, or “outside” if 100%, < 50% or < 20% of Vrecur was within the 95% isodose of the first recurrence GTV (green line on Figure 1 ), respectively. Student t.tests were used to compare the parameters extracted from both “in- field” and “outside” groups. Principal component analysis (PCA) and Ascending Hierarchical Classification (AHC) were performed to separate both groups. Correlograms were calculated to identify an association between the selected parameters.