ESTRO 2020 Abstract Book

S146 ESTRO 2020

past studies, we have reconstructed heart doses using a simple anatomy atlas-based whole heart model, i.e. stylized and without substructures. The primary objective of this study was to enhance the heart model, including substructures, such that it was more anatomically realistic in size, shape, and position using CT images and demonstrate that the refined heart model is representative across the typical age range of a pediatric cohort. Material and Methods Our atlas-based heart model was refined and expanded using whole heart and substructure contours from the National Cancer Institute (NCI) computational human phantom series. Within a commercial treatment planning system, we registered the 5-year-old NCI phantom with our in-house phantom. Using a Python script, whole heart and substructure contours were extracted, deformed, and converted to a grid of evenly spaced points in our in-house master phantom. We then scaled our phantom to ages 1, 5, 10, 15 years and adult. Scaling functions in our computational phantoms take into account non-uniform growth along all three dimensions. To evaluate if our new cardiac model was representative from infant to adolescent, we calculated Dice similarity coefficient (DSC) values comparing our atlas-based and new cardiac models scaled to aforementioned ages with NCI phantoms of same age and gender.

Conclusion The new cardiac model developed here is anatomically superior to the previous model and enables dose estimation for cardiac substructures for individuals in pediatric cohorts whose historic RT was not planned with CT-images. Our model can be used for late cardiac toxicity studies to evaluate the relationship between cardiac substructure doses and late cardiac disease. PH-0287 Transfer learning and Deep Neural Network for lung and heart dose prediction in breast treatments J. Perez-Alija 1 , P. Gallego 1 , M. Lizondo 1 , J. Nuria 1 , A. Latorre-Musoll 1 , I. Valverde-Pascual 1 , M. Barceló-Pagès 1 , N. Garcia-Apellaniz 1 , P. Carrasco de Fez 1 , P. Delgado- Tapia 1 , P. Simon Garcia 1 , M. Adria Mora 1 , A. Ruiz Martinez 1 , M. Ribas 1 , E. Ambroa 2 1 Hospital de la Santa Creu i Sant Pau, Medical Physics, Barcelona, Spain ; 2 Consorci Sanitari de Terrassa, Medical Physics Unit- Radiation Oncology Department, Terrassa, Spain Purpose or Objective Generating a convolutional neural network (CNN) model to predict lung and heart dose-volume histograms (DVH) in breast cancer patients with lymph nodes treated with 3D- CRT would help in the technique decision process. Usually, the work done in dose prediction using CNNs does not consider the plan quality of the training data. To ensure this quality, we propose a method for outliers detection within the dataset that can be used as a DVH predictor. Material and Methods We selected 195 patients with left breast cancer treated with 3D-CRT. We included patients with axillary and supraclavicular lymph nodes but excluded those with an internal mammary nodal (IMN) chain. For the model creation, we trained the CNN renormalizing all plans to 2 Gy/fraction, to take into account different prescribed doses. For our CNN model, we implemented a transfer learning approach using a pre-trained VGG-16 and replacing its three last layers with a fully connected neural network. Input data was the planning CT contour information. Output was a 2D lung and heart DVH for every slice. All slices were subsequently added up to account for the final whole OAR DVH. For the outliers detection, we partitioned our set in training, validation, and test (176, 10, and 10 patients,

Results We developed a new cardiac model with 14 substructures (including aorta, ventricles, atriums, arteries and valves) for our computational phantom. Importantly, the phantom (along with heart and substructures) can be scaled to any age at RT for individual dose reconstructions, which is essential for pediatric cohorts. The new cardiac model is anatomically similar to NCI cardiac model across all ages with an average DSC value of 0.80 ± 0.04. This is a significant improvement (Wilcoxon Signed-Rank Test, p < 0.01) over the atlas-based heart model with an average DSC value of 0.37 ± 0.05. The improvement is consistent for each age and both genders.