ESTRO 2023 - Abstract Book

S1360

Digital Posters

ESTRO 2023

PO-1658 Specificity and limits of Ethos template-based intelligent dose optimization engine

B. Roberfroid 1 , A. Barragán-Montero 1 , E. Sterpin 1,2 , J.A. Lee 1 , X. Geets 1,3

1 UCLouvain, Center of Molecular Imaging, Radiotherapy and Oncology (MIRO), Brussels, Belgium; 2 KULeuven, Department of Oncology, Laboratory of Experimental Radiotherapy, Leuven, Belgium; 3 Cliniques Universitaires Saint-Luc, Department of Radiation Oncology, Brussels, Belgium Purpose or Objective Ethos Varian linear accelerator proposes some innovative features such as automatic dose planning to enable online adaptive radiation therapy. Ethos automatic planning is based on an optimization of generic clinical goals from a user defined template. This approach allows to deliver satisfying plans for the average patient population but might encounter difficulties to generate the best patient-specific plans. Concurrently to Ethos, another automatic planning approach has been developed which uses deep learning of convolutional neural networks to directly predict the supposed-optimal three dimensional dose distribution for a given patient complemented with a subsequent step of inverse planning called “dose mimicking”. Assuming that dose prediction models are meant to learn anatomy-specific doses, their specificity might actually be higher than a template-based approach. The goal of this study is to investigate how well the dose prediction with deep learning approach and the Ethos template based approach perform on challenging prostate plans to depict limitations and ways of improvement for both of them. Materials and Methods A U-NET dose prediction model was trained on a retrospective database of 35 prostate manual plans. 10 patients that are not part of the previous database and whose anatomy is challenging for the Ethos automatic planning were selected. Treatment plans for these 10 patients were generated with the Ethos automatic planning (EG_init) and with our dose prediction approach (DP). Ethos goals template was then slightly updated based on results of manual planning ground truth. The 10 patients were therefore re-planned with this more patient-specific template (EG_upd). Each approach was eventually compared based on clinical deliverability, clinical dose metrics, and mean dose to organs at risk. Results All DP plans were deemed clinically deliverable, whereas 1 EG_init plan and 3 EG_upd plans were not. V35Gy and V20Gy to anal canal were statistically lower for DP plans than for EG_init/upd plans (see table below). V30Gy to rectum was statistically lower for DP plans than for EG_init/upd plans. V40.8Gy to bladder was statistically lower for DP plans than for EG_init/upd plans. V42.5Gy to penile bulb was statistically lower for DP plans than for EG_init/upd plans. V60Gy to rectum was statistically higher for EG_upd than for DP, explaining the higher number of not clinically deliverable plans. After template update, mean doses to anal canal and bladder were still statistically lower for DP plans than for EG_upd plans (see figure below).