Casamitjana A, Sala-Llonch R, Tudela R, Andrés A, Orío S, Casas J, et al.. Deep Learning CT segmentation for dosimetry in postoperative endometrial carcinoma treatment. In XLI Congreso Anual de la Sociedad Española de Ingeniería Biomédica. Cartagena: Ediciones UPCT. Universidad Politécnica de Cartagena; 2023.

Abstract

Prophylactic vaginal brachytherapy (VBT) is a common treatment after tumor resection in endometrial cancer that requires individual delineation of the Clinical Target Volume (CTV). The main aim of this study was to assess the viability of automatic Deep Learning (DL) algorithms for segmenting the CTV from pelvic Computed Tomography images. We collected a dataset of 220 CT images, labeled manually. We implemented and trained V-Net and UNETR networks and we assessed the performance in a cross-validated setting, by dividing them into training, validation, and test sets. We assessed the quality of the segmentations using the Dice Coefficient (DICE) and the 95% Hausdorff Distance (HD) and using dose-volume histogram parameters. We also evaluated data augmentation (DA) strategies. Both algorithms gave HD values between 8.2-8.7 and DICE of 0.78-0.79. There were no statistical differences in the volume-dose parameters between automatic and manual labels. DA slightly improved the performance of the algorithms. We proved the
applicability of DL for CTV segmentation in postoperative endometrial carcinoma, using two different model networks, UNETR and V-Net, and with DA.