Click here to

Session: Applications of AI in Radiotherapy Planning and Adaptation [Return to Session]

AI-Dose Predictor for Secondary Dose Calculation for a Novel Compensator-Based IMRT System for Low- and Middle-Income Countries

K Oh1,2*, M Gronberg2, S Gay2, A Olanrewaju2, B Sengupta1, T Netherton2, C Cardenas3, L Court2, E Ford1, (1) University of Washington, Seattle, WA, (2) UT MD Anderson Cancer Center, Houston, TX, (3) The University of Alabama at Birmingham, Birmingham, AL

Presentations

SU-J-BRC-1 (Sunday, 7/10/2022) 4:00 PM - 5:00 PM [Eastern Time (GMT-4)]

Ballroom C

Purpose: We are developing a cost-effective radiation delivery system for low- and middle-income countries consisting of a cobalt-60 source coupled to a novel refillable-compensator system, and have commissioned this system in Eclipse. Here we describe the treatment planning evaluation as well as the development of an AI-based dose predictor for dose verification.

Methods: 1)Treatment planning: Cobalt-60 compensator-based IMRT plans were generated for comparison to the clinically-used VMAT plans for 46 head-and-neck cancer patients, and dosimetric parameters were evaluated. 2)Dose verification tool: A deep learning(DL)-based 3D dense dilated U-Net was used to predict dose distributions of static fields for verification of simple treatment fields. For training, validation, and testing, 600 datasets were split into 480-60-60 sets. The DL network was then extended to predict dose for patient-specific dose verification. Individual field doses from 46 9-field compensator-based IMRT plans for head-and-neck cancer were separated into each field, yielding 405 datasets. These were divided into 333-36-36 sets for training, validation, and testing. Inputs were patient CT scans, binary masks for the beam shape in the patient, and fluence maps.

Results: 1)The compensator-based IMRT plans provided equivalent plan quality compared to linac-based VMAT plans, with average dose differences of <1Gy for PTVs and OARs. 2)DL-based predictions of static fields of PDDs and profiles provided excellent agreement with beam data, with an average percent deviation of 0.41±0.38% and 0.28±0.54% across the different field sizes, phantom sizes, and SSDs investigated. DL dose prediction models for patient plans demonstrated a mean absolute error 0.42±0.30Gy on the test sets vs. clinical plans.

Conclusion: The novel cobalt-60 compensator-based IMRT system allows for high-quality plans which might be beneficial to resource-limited regions throughout the world. We have developed a DL-based dose verification tool that accurately predicts doses for this system and may have more general applications for quality assurance.

Funding Support, Disclosures, and Conflict of Interest: Funded by National Cancer Institute

Keywords

Co-60, Compensators, Dose

Taxonomy

TH- External Beam- Photons: Development (new technology and techniques)

Contact Email

Share: