ESTRO 2024 - Abstract Book

S3087

Physics - Autosegmentation

ESTRO 2024

1 Baroudi H, Brock KK, Cao W, Chen X, Chung C, Court LE, El Basha MD, Farhat M, Gay S, Gronberg MP, Gupta AC, Hernandez S, Huang K, Jaffray DA, Lim R, Marquez B, Nealon K, Netherton TJ, Nguyen CM, Reber B, Rhee DJ, Salazar RM, Shanker MD, Sjogreen C, Woodland M, Yang J, Yu C, Zhao Y. Automated Contouring and Planning in Radiation Therapy: What Is “Clinically Acceptable”? Diagnostics (Basel). 2023 Feb 10;13(4):667. doi: 10.3390/diagnostics13040667. PMID: 36832155; PMCID: PMC9955359 2 Sherer MV, Lin D, Elguindi S, Duke S, Tan LT, Cacicedo J, Dahele M, Gillespie EF. Metrics to evaluate the performance of auto-segmentation for radiation treatment planning: A critical review. Radiother Oncol. 2021 Jul;160:185-191. doi: 10.1016/j.radonc.2021.05.003. Epub 2021 May 11. PMID: 33984348; PMCID: PMC9444281.

2071

Digital Poster

Deep Learning based CTV segmentation on a large cohort of patients: the case of breast Radiotherapy

Maria Giulia Ubeira Gabellini 1 , Gabriele Palazzo 1 , Luciano Rivetti 2 , Martina Mori 1 , Zan Klanecek 2 , Andrei Fodor 3 , Nadia Gisella Di Muzio 3,4 , Robert Jeraj 2,5 , Antonella del Vecchio 1 , Claudio Fiorino 1 1 IRCCS San Raffaele Scientific Institute, Medical Physics, Milan, Italy. 2 University of Ljubljana, Faculty of Mathematics and Physics, Ljubljana, Slovenia. 3 IRCCS San Raffaele Scientific Institute, Radiotherapy, Milan, Italy. 4 Vita-Salute San Raffaele University, Medicine and Surgery, Milan, Italy. 5 University of Wisconsin, Department of Medical Physics, Madison, USA

Purpose/Objective:

In order to optimize the radiotherapy treatment and reduce toxicities, organs-at-risk (OARs) and clinical target volume (CTV) must be segmented; this is a highly time-consuming task, typically done manually and prone to large variability [1]. Nowadays, the interest toward automatic tools supporting OARs/CTV segmentation is rapidly increasing, also if just few studies dealt with CTV breast segmentation using large cohorts [2]. Atlas-based methods were first considered [3], but Deep Learning (DL) based methods (e.g., ResNet, Unet, nnUnet) are preferred to train/test reliable models [4-7], particularly for soft tissue-based regions. This may help in optimizing and reducing the clinical workload in the radiotherapy workflow, also if particular attention must be given to reach high segmentation precision considering the direct influence on treatment planning.

The aim of this work is to create for the first time an advanced DL 3D model able to automatically segment right and left CTV breast together thanks to the availability of a large patients' cohort treated all in a single institute.

Material/Methods:

The dataset (3D CT images) used is of 611 patients, who underwent breast-conserving surgery followed by radiotherapy (RT), of which 20% was used for cross-validation. Patients collected are respectively 332 for right-sided breast cancer (BC) and 279 for left-sided BC. Another 75 patients (with both right/left BC) were used for internal test. The images were acquired through different scans and thus have variable 3D size. CTV delineation was performed by different clinicians in the period 2017-2021.