ESTRO 2025 - Abstract Book

S4363

RTT - Treatment planning, OAR and target definitions

ESTRO 2025

Dong, Haoyu, et al. "Segment anything model 2: an application to 2d and 3d medical images.", arXiv:2408.00756 (2024) Zhu, Jiayuan, et al. "Medical sam 2: Segment medical images as video via segment anything model 2.", arXiv:2408.00874 (2024) Smith, Gillian Adair, et al. "Interobserver variation of clinical oncologists compared to therapeutic radiographers (RTT) prostate contours on T2 weighted MRI." Technical Innovations & Patient Support in Radiation Oncology 25 (2023): 100200 Huang, Yuhao, et al. "Segment anything model for medical images?." Medical Image Analysis 92 (2024)

2736

Digital Poster Validation of Deep Learning-Based Dose Prediction for Prostate SBRT: A Dosimetric Comparison Study

Dániel Gugyerás, Ádám Miovecz, Ferenc Lakosi Oncoradiology, SVMKMOK, Kaposvár, Hungary

Purpose/Objective: Prostate stereotactic body radiation therapy (SBRT) delivery demands highly conformal dose distributions to achieve optimal tumor coverage while ensuring critical organ-at-risk (OAR) sparing. The steep dose gradients and proximity of critical structures make the planning process challenging and time-consuming. This study evaluates the performance of an artificial intelligence-based dose prediction model for prostate SBRT planning, with particular focus on its potential to guide consistent planning objectives through automated dose prediction. Material/Methods: Twenty prostate cancer patients treated with SBRT (36.25Gy in 5 fractions) using dual-arc VMAT on a TrueBeam system (6FFF) were analyzed retrospectively. For each case, dose distributions were generated using the prostate model of Dose+ 1.0 (MVision AI, Helsinki, Finland), a commercial AI-based dose prediction software, and compared with the clinical treatment plans. The software requires as input a CT series and structure set with target names including the prescribed dose, with typical processing times below two minutes for dose generation. Dosimetric evaluation included mean doses and dose-volume histograms (DVHs) for all relevant OARs including bladder, rectum, femoral heads, penile bulb, and urethra PRV. The predicted dose distributions were analyzed for their potential to serve as patient-specific planning objectives. Results: The AI model demonstrated comparable performance to clinical plans across all evaluated OARs. Mean dose differences (AI vs. clinical) showed strong agreement for critical structures: bladder (10.7Gy vs. 9.9Gy), rectum (9.2Gy vs. 9.1Gy), femoral heads (left: 5.3Gy vs. 5.9Gy; right: 5.3Gy vs. 5.9Gy), and urethra PRV (39.0Gy vs. 37.8Gy). DVH analysis revealed consistent dose-volume relationships between predicted and clinical plans, particularly in high dose regions critical for SBRT planning. The predicted dose distributions provided achievable patient-specific planning objectives, potentially facilitating standardization of the planning process.