ESTRO 2020 Abstract Book

S107 ESTRO 2020

procedures (24.2% versus 21.2%). H-RT was associated with small increases in adjusted mean EPIC-26 sexual (3.3 points; 95% CI 2.1-4.5, p<0.001) and hormonal function scores (3.2 points; CI 1.8-4.6, p<0.001). These differences failed to meet established thresholds for a clinically meaningful change. There were no statistically significant differences in urinary or bowel function and quality of life. Conclusion This is the first national cohort study comparing functional outcomes after H-RT and C-RT reported by patients. These ‘real-world’ results further support the use of H-RT as the standard for radiotherapy in men with non-metastatic PCa. OC-0213 External beam radiotherapy alone or with high-dose-rate brachytherapy boost for prostate cancer P. Hoskin 1 , A. Rojas 2 , G. Lowe 2 , L. Bryant 2 , G. Pandey 2 , P. Ostler 2 1 Mount Vernon Hospital and University of Manchester, Cancer Centre, Northwood Middlesex, United Kingdom ; 2 Mount Vernon Hospital, Cancer Centre, Northwood, United Kingdom Purpose or Objective Purpose: A randomised phase-III trial compared external beam radiotherapy (EBRT) alone with EBRT combined with high-dose-rate brachytherapy boost (HDR-BTb) in localised prostate adenocarcinoma. Previous analysis after a median follow up of 85 months demonstrated an improved biochemical relapse free survival (bRFS) in the arm receiving the brachytherapy boost. This data has now been updated with a median follow up of 131 months Material and Methods Methods: From December 1997 to August 2005, 218 patients were assigned either to EBRT alone (108) delivering 55 Gy in 20 fractions over 4 weeks or EBRT (35·75 Gy in 13 fractions; n = 110) followed by a temporary high-dose-rate implant delivering 2 x 8·5 Gy over 24h. A balanced one-to-one randomisation was used with stratification according to tumour stage, PSA (Prostate Specific Antigen), and Gleason score. Exclusion criteria were evidence of metastases, PSA > 50µg/l, co-existing malignancy, co-existing medical condition that precluded general anæsthesia. The primary endpoint was Relapse- Free Survival (RFS) defined by either a PSA rise ≥2.0µg/l above nadir or clinical progression. Actuarial survival rates were calculated using the Kaplan-Meier method and Cox’s Proportional Hazard model to calculate HRs (Hazard ratios), which for bRFS represents the risk of biochemical, clinical relapse or death. Secondary endpoints were overall survival (OS), urinary and bowel toxicity measured using an adapted version of the Dische Morbidity Scales. Treatment and assessments were not masked. Results Results: 106 patients received EBRT alone and 110 received EBRT + HDR-BTb; of these 165 patients were given androgen deprivation treatment (ADT). The median age was 70yrs (range 47-80yrs) and 9 (4%) were low risk, 91 (42%) were intermediate risk and 116 (54%) were high risk. After a median follow-up time of 131 months, a 30% increase in the risk of recurrence with EBRT alone ( p = 0·001) resulting in a 17% improvement in RFS at 10 years with EBRT + HDR-BTb was seen. The median time to relapse was 137 months compared to 82 months for EBRT alone (log rank p = 0·01). In multivariate analysis treatment arm, risk category and no androgen deprivation therapy (ADT) were significant covariates for risk of relapse (see Table ). In subgroup analysis the effect of EBRT + HDR-BTb was greater in patients not receiving ADT compared to those who did receive ADT. Differences in overall survival were not significant (Table 1). Both treatments were equitoxic for severe late urinary and bowel morbidity.

Conclusion Conclusion: After a median follow up of 131 months there remains a significant improvement in RFS after EBRT + HDR-BTb compared to EBRT alone and no increase in the incidence of severe urinary and rectal morbidity at five and seven years after treatment.

Proffered Papers: Proffered papers 11: Dose calculation for advanced techniques

OC-0214 Comparison of deep convolutional neural networks to denoise Monte Carlo proton dose distributions U. Javaid 1 , K. Souris 1 , S. Huang 2 , J.A. Lee 1 1 UCLouvain, IREC/MIRO, Brussels, Belgium ; 2 Memorial Sloan Kettering Cancer center, Department of Medical Physics, New York, USA Purpose or Objective Monte Carlo (MC) simulations can be used to model dose distribution in radiation oncology. However, MC simulations rely on repeated but limited random sampling that leaves noise in the resulting dose distributions, limiting their usability for accurate clinical decisions. Simulations using a huge number of particles can partly address this issue but these are computationally expensive. Hence, a compromise between computation time and residual noise must be found in MC doses. Previous work on denoising MC dose distributions is based on image filtering approaches and recently using deep learning. In this work, we do a comparative study between two convolutional neural network (CNN) architectures to denoise whole-volume noisy MC dose distributions. Material and Methods We evaluate two convolutional neural networks (CNNs), i.e., dilated UNet and simpleNet (network with 3 convolutional layers). Dilated UNet has 3 downsampling layers with skip connections, whereas neither downsampling nor skip connections are used in simpleNet. Mean-squared error (MSE) measures pixel intensity differences between the denoised and noisefree reference images by averaging all squared differences. Low MSE indicates good similarity between the two images under observation therefore, we use it as a loss function to train the networks. We train our models on proton therapy MC dose distributions for different tumor sites ( brain, head & neck, liver, lung, prostate ) acquired from 35 patients. In