ESTRO 2020 Abstract book

S679 ESTRO 2020

radiomic features of rectum 3D dose distributions were used for predictive modelling of rectal toxicity. We compared models using clinical + DVH features and clinical + radiomics features of rectum 3D dose distributions. We have used 10 fold stratified cross-fold validation to evaluate the results with or without our workflow.

Low-dose rate brachytherapy (BT) utilizing permanent 125 iodine radioactive seeds is a highly effective treatment option for low and intermediate risk prostate cancer patients. In prostate brachytherapy, optimal implantation is not always achieved, due to edema or seed loss. One way to improve seed placement is the use of stranded seeds called intra-operatively built custom linked seeds (IBCLS) in opposition to loose seeds (LS) but there is no data comparing acute and late toxicities between these 2 techniques. The aim of this study was to compare the acute and late toxicities of both procedures in a matched paired population. Material and Methods Patients were considered for BT following the European guidelines. Patients with previous external radiotherapy to the pelvis or those who received androgen deprivation before the procedure were not considered. Between 2003 and 2018, 105 Patients in the “loose seeds” cohort were individually matched to 105 patients in the IBCLS cohort following the: age, prostate volume, preoperative IPSS score, clinical stage and Gleason score. Toxicities were assessed using the International Index of Erectile Function five items (IIEF5) and the International Prostate Symptom Score (IPSS) forms. Results Patients characteristics were comparable. There was no differences in the D90Gy (180.01Gy vs 179.78Gy, p=0.733), but the percentage of the prostate volume that receives 100% (V100%) and 150% (V150%) of the prescribed dose were higher in IBCLS (p<0.001). The organ at risks dose were lower for both rectum and urethra in IBCLS group (p<0.001). There was no difference in the IPSS at 1, 3 and 6 months between two groups, but at one year, the IPSS was lower in the IBCLS group (10.04 vs 7.77) (p=0.028). The sexual and gastrointestinal toxicity were similar in both groups. There are no grade 2 toxicities in the IBCLS group. Conclusion This study shows that prostatic brachytherapy by intra operatively built custom linked seeds is a safe technique, with a lower impact on IPSS score at 12 months compared to brachytherapy with loose seeds. PO-1199 Rectal toxicity prostate cancer treated with Brachytherapy: a radiomics-machine learning based NTCP G. Dissaux 1,2 , M. Ibrahim 2 , F. Lucia 1,2 , V. Bourbonne 1,2 , N. Boussion 1,2 , O. Pradier 1,2 , D. Visvikis 2 , A. Valeri 2,3 , J. Bert 2 , M. Hatt 2 , U. Schick 1 1 University Hospital, Radiation Oncology, Brest, France ; 2 Univ Brest, LaTIM INSERM UMR 1101, Brest, France ; 3 University Hospital, Urology, Brest, France Purpose or Objective Dose-volume histograms (DVH) do not account for the spatial relationships between voxel doses. We aimed to compare the prediction of rectal toxicities following prostate cancer (PC) Low Dose Rate Brachytherapy (LDR- BT) using the conventional approach based on DVH parameters with the use of dosimetric radiomic features using 4 ML machine learning algorithms and 3 deep learning (ML, DL) techniques as well majority voting techniques. Imbalanced class problem was also taken into account in our workflow. Material and Methods The records of 188 consecutives patients with more than 2 years follow up (including clinical data, DVH and rectal toxicities event) who underwent LRT-BT for localized PC were collected prospectively. 21 patients had ≥ grade 2 late rectal toxicity in our whole dataset. In our workflow we have used 4 ML algorithms such as Random Forest, Xtreme Gradient Boosting, LightGBM, CatBoost and 3 DL classifiers: Deep Neural Network, Deep Autoencoder+RF, Deep Variational Autoencoder+RF, as well as majority voting. Aparts from DVH parameters, also dosimetric

Results Best result was found using random forest for all the combinations we have tested using our workflow. For the late ≥ grade 2 toxicities, the mean area under the curve (AUC) for using clinical and DVH parameters was 0.57±0.20, using and clinical and radiomics features it increased to 0.65±0.17. Without following our workflow, the performance for the mean area under the curve (AUC) for clinical + DVH features and clinical + radiomics are 0.53±0.20 and 0.59±0.22, respectively.

Conclusion The predictive performance of Normal-Tissue Complications Probability (NTCP) models based on radiomics features is better than the commonly used DVH parameters. Dosimetric radiomics analysis using ML/DL techniques is a promising tool for NTCP modelling in brachytherapy. Besides, following the workflow we developed might increase the accuracy by overcoming class imbalance problem. PO-1200 Multimodality treatment for M1a prostate cancer: toxicity and early oncological results. C. Berghen 1 , S. Joniau 2 , G. Devos 2 , K. Poels 1 , K. Haustermans 1 , G. De Meerleer 1 1 University Hospital Gasthuisberg, Radiation Oncology, Leuven, Belgium ; 2 University Hospital Gasthuisberg, Urology, Leuven, Belgium Purpose or Objective In M1a prostate cancer (PC), disease has spread to non- regional lymph nodes and state-of-the-art treatment consist of systemic treatment with palliative androgen deprivation therapy (pADT) whether or not combined with second-line hormones or chemotherapy. However, using multimodality treatment (MMT) involving surgery, radiotherapy and ADT is gaining interest. We report on toxicity and early oncological results of MMT for M1a PC.