Click here to

Session: Multi-Disciplinary General ePoster Viewing [Return to Session]

A Deep Learning Method Approach for Machine Modeling of Elekta Linear Accelerator From Limited Beam Data

S Ahn1*, C Kim2,M Han2,S Han2,Y Lee2,H Kim2,C Hong2,J Kim2,J Kim2**, (1) Department of Radiation Oncology, Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul, 06351, South Korea, (2) Department of Radiation Oncology, Yonsei Cancer Center, Yonsei University College of Medicine, 50-1 Yonsei-ro, Seodaemun-gu, Seoul 03722, South Korea

Presentations

PO-GePV-M-33 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: The purpose of this study is to investigate whether accurate modeling is possible by using the limited beam data with the deep learning-based linac beam modeling.

Methods: The physical characteristics of linac beam data depend on the equipment, and we investigated whether profiles and PDDs could be predicted with only a reference field for different field sizes. Eleckta linac beam limited data sets (n=3) were obtained from one institution, and the profile and PDD were predicted and evaluated. The framework of linac beam modeling deep neural network (LBMnet) that predicts PDDs and profiles is configured to predict the field size from 1×1 cm2 to 40×40cm2 fields when a 10×10 cm2 field is input data. The predicted PDDs and profiles were quantitatively compared with the results measured by ion-chamber, and a linac beam model was created based on the beam data generated from RTP, and dose calculation results and gamma analysis were performed for ten different clinical cases.

Results: The percentage mean absolute error of the predicted profiles and PDDs showed the largest difference in the build-up region and the penumbra region, and the maximum difference was ±3% except for the two regions. 2D-gamma passing rate of clinical case dose calculation was 90% (criteria of 0.5%/0.5 mm) and 99% (criteria of 1.0%/1.0 mm).

Conclusion: The accuracy of linac beam data prediction based on deep learning was shown through the results of clinical case, and this method has the potential to help increase accuracy and simplify the procedure while reducing time in the linac QA and commissioning process.

Keywords

Quality Assurance, Modeling, Linear Accelerator

Taxonomy

IM- Dataset Analysis/Biomathematics: Machine learning

Contact Email

Share: