ESTRO 2024 - Abstract Book

S366

Beachytherapy - Physics

ESTRO 2024

Over 13 Years. Clin Oncol (R Coll Radiol). 2023 Apr;35(4):237-244. doi: 10.1016/j.clon.2022.12.008. Epub 2022 Dec 31.

2 Lobo S, Gaze MN, Slater O, Hoskin P, Sands G, Sullivan T, Cho A, Eminowicz G, Smeulders N. Bladder function after conservative surgery and high-dose rate brachytherapy for bladder-prostate rhabdomyosarcoma. Pediatr Blood Cancer. 2022 Aug;69(8):e29574. doi: 10.1002/pbc.29574. Epub 2022 Jan 19. 3 Tomas Palmqvist, Anne Dybdahl Wanderas, Anne Beate Langeland Marthinsen, Marit Sundset, Ingrid Langdal, Signe Danielsen, Iuliana Toma-Dasu. Dosimetric evaluation of manual and inverse optimized treatment planning for high dose rate brachytherapy of cervical cancer. Acta Oncologica. 2014; 53: 1012-1018. doi/pdf/10.3109/0284186X.2014.928829.

1783

Digital Poster

Dwell locations and dwell time prediction of cervical cancer brachytherapy using 3D U-net

tao Liu 1 , rui Luo 2 , shijing Wen 1 , siqi Wang 1 , bin bi 1 , xianliang Wang 1

1 Sichuan Cancer Hospital and Institute, Department of radiation oncology, Chengdu, China. 2 Mianyang Central Hospital, Nuclear medicine department, Chengdu, China

Purpose/Objective:

The purpose of this study is to predict the dose for cervical cancer brachytherapy using a three-dimensional (3D) U-net model. This involves building radiation source-dwell labels and incorporating the concept of forward dose calculation.

Material/Methods:

A dataset comprising 200 radiotherapy plans with CT/MR fletcher applicator was collected from April 2021 to February 2022 at the Sichuan Cancer Hospital. Deep learning techniques were employed to predict the dwell labels of the radioactive sources. The predicted dwell labels were then used to calculate the dwell locations and corresponding time weights using K-means clustering. The predicted results of Case1 and Case2 were compared with the clinical plan, including source dwell locations and dwell time weights , as well as the cross-sectional, sagittal, and coronal dose distributions. The conformity index (CI) and homogeneity index (HI) were calculated for Case1 and Case2, and the D1cc and D2cc of the bladder, rectum, small intestine, and colon were quantitatively evaluated.

Results:

The model demonstrates good accuracy in predicting the dwell locations of the radioactive source, with an average difference of 1.01 mm. There are no remarkable changes in the dwell time of the radioactive source. The predicted dose distributions in the cross-section, sagittal plane, and coronal plane are consistent with the clinical results, and are able to accommodate HRCTV contour changes. The average CI and HI indices are 0.7 and 0.465, respectively. The predicted D90 values for HRCTV closely matched the prescribed dose (6Gy), with values of 6.04Gy