ESTRO 2023 - Abstract Book

S409

Sunday 14 May 2023

ESTRO 2023

Materials and Methods A total of 96 treatment plans with free-hand interstitial needles of HDR brachytherapy for cervical cancer were selected in this work. The dataset was randomly divided into the training set (81 cases) and the validation set (15 cases). We use a transformer as the encoder to effectively capture remote dependencies and global context information from the embedded input patches, and the encoded features maps were subsequently combined with the CNN-based decoder through skip connection at multi-level (Figure.1). The performances of our proposed model were compared with 3D attention-gated Unet. The differences of dose metrics between predicted dose and ground truth were assessed to demonstrate the effectiveness of the deep-learning model, including D90, D95 for HRCTV and D2cc for bladder, rectum, and bowel. Results The quantitative dose differences between the prediction and the actual dose calculated by different dose metrics were shown in Table 1, which demonstrated that the 3D transformer-based deep-learning model outperformed the 3D attention gated Unet. The average differences of D90 and D95 for HRCTV were both less than 0.1 Gy for the 3D transformer-based deep learning model, and the mean D2ccof OARs were both less than 0.6 Gy. The experimental results showed that the dose distribution predicted by our proposed methods is consistent with the clinical plans. Conclusion A novel 3D transformer-based deep learning method for dose prediction in HDR interstitial brachytherapy was proposed and evaluated in terms of dosimetric accuracy. The approach has great potential to generate accurate dose distribution and further determine the quality and outcome of HDR brachytherapy treatment plans for cervical cancer.

PD-0501 Superficial HDR brachytherapy - dosimetric verification of dose distribution with lead shielding G. Biel ę da 1,2 , G. Zwierzchowski 1,2 , A. Szymbor 3 , M. Boehlke 4 1 Greater Poland Cancer Centre, Medical Physics, Pozna ń , Poland; 2 Poznan University of Medical Sciences, Electroradiology, Pozna ń , Poland; 3 MCO Affidea, Medical Physics, Pozna ń , Poland; 4 West Pomeranian Oncology Center, Medical Physics, Szczecin, Poland Purpose or Objective Surface brachytherapy, usually characterized by a high dose gradient, allows the dose to be precisely deposited in the irradiated area while protecting critical organs. When the lesion is located in the nasal or ocular region, the organ of vision must be protected. The aim of this study was to verify the dose distributions near critical organs in the head and neck region during brachytherapy procedure using lead shielding of the eye using an anthropomorphic phantom and an applicator manufactured from PLA (poli-lactic acid) using FDM (fused deposition modeling) 3D printing technology. Materials and Methods An anthropomorphic head phantom with preserved air spaces and imitation cranial bones filled with plaster of a density comparable to real bones (600-700 HU) was prepared (fig1).