ESTRO 2021 Abstract Book

S614

ESTRO 2021

post-prostatectomy radiotherapy (SRT), comprehensive explaining models are still lacking. The objective of this study was to fit long-term bRFS of a multi-center cohort of patients (pts), including the impact of Pelvic Nodal area Irradiation (PNI) using a TCP-based formula. Materials and Methods We previously showed that bRFS may be modelled by the Poisson-like expression: (1 - B x PSA) x (1- exp(- α eff D)) CxPSA ,being PSA the pre-SRT PSA, α eff the radiosensitivity, C the number of tumor cells at PSA=1 ng/mL, D the delivered dose and (1-BxPSA) the loss of bRFS due to tumor cells outside the irradiated volume at SRT. This formula was adapted to incorporate PNI by replacing this latter term with (1-B x λ x PSA), being λ the recovery of bRFS in case of PNI (equal to 1 without PNI). A cohort (n=725) of pts treated in 7 Institutes with SRT at a median dose of 72Gy (IQR:70-72.8Gy), with/without PNI (according to radiation oncologist’s preference), with 5-year minimum follow-up and pre-SRT PSA<2 ng/mL was randomly split in 2 groups (training, TR: n=483, validation, VA: n=242). According to multi-variable logistic regression (MVR), the most informative clinical predictor was pT stage: then, pts were further stratified in “high” and “low” risk sub- groups, composed of 315 and 156 pT2 and pT3 pts respectively. Then, bRFS data of TR were fitted by least square minimization and the best-fit parameters of the model derived, according to pT stratification. The resulting predicted values of long-term bRFS were compared against true data by calibration plots; model’s performances were then tested in the VA group. Dose-effect curves of bRFS for pT2/pT3 pts, according to different pre-SRT PSA values, were generated. Results The median follow-up was 8.5 years (IQR: 6.5-11.5); median pre-RT PSA was 0.43ng/mL (0.24-0.8); 270 and 277 patients received PNI and adjuvant hormonal therapy; of note, this latter factor was not associated to bRFS at MVR. Fits were successful: the resulting model was well calibrated (slope 0.88, R 2 : 0.76) and performances were confirmed in the V set (slope 0.88, R 2 : 0.79). Best fit values of the parameters were α eff =0.27Gy -1 , C=10 7 , B=0.37/0.56, λ =0.85/0.84 (for pT2/pT3). In Figure, the predicted long-term bRFS vs D for few pre-SRT PSA values, with/without PNI are shown: for sufficiently high doses, the bRFS plateau value is well predicted by PSA and pT stage. The impact of PNI, as shown in Figure, is up to 5-8 % for pT3 pts with PSA=0.5-1ng/mL.

Conclusion A radiobiologically consistent formula well fits long-term bRFS of a large, multi-centric group of pts treated with SRT with/without PNI. It can be easily used to individually predict bRFS, potentially assisting clinicians during therapeutic choice. The impact of PNI was predicted to be consistent with recent findings of the RTOG0534 SPORRT trial. PD-0783 Automated sarcopenia assessment in the neck and survival analysis in head and neck cancer patients H. Warr 1 , O. Murray 1 , D. McSweeney 2 , A. McWilliam 2 , A. Green 2 1 The University of Manchester, Physics and Astronomy, Manchester, United Kingdom; 2 The University of Manchester, Division Of Cancer Sciences, Radiotherapy Related Research, Manchester, United Kingdom Purpose or Objective Sarcopenia can be used to quantify the frailty of a patient, and can be assessed by extracting skeletal muscle characteristics. Models have been developed to measure skeletal muscle at the third lumbar vertebra level (L3); however, head and neck cancer scans often do not extend into the abdomen. The purpose of this investigation was to develop a model that automatically extracts muscle characteristics in the cervical spine (C3), and use these to assess head and neck (H&N) cancer patients for sarcopenia, before analysing the impact of depleted skeletal muscle on survival. Materials and Methods Skeletal muscle at C3 was manually delineated for 35 patients using pre-treatment CT scans from an open database of Head and Neck Squamous Cell Carcinoma (HNSCC) patients. Each patient was segmented by two observers to allow an estimation of interobserver variability. Convolutional neural networks (CNNs) are often used for the automatic segmentation of medical images. Transfer learning is a common technique used when there is limited training data available. A pretrained CNN (FCN-ResNet-50) was fine-tuned to segment the sternocleidomastoid and paravertebral muscles (training=25, testing=5, validation=5), shown in Figure 1.