ESTRO 2024 - Abstract Book

S4938

Physics - Quality assurance and auditing

ESTRO 2024

2. M. Folkerts, E. Abel, S. Busold, J. Perez, V. Krishnamurthi and L. CC, “A Framework for Defining FLASH Dose Rate for Pencil Beam Scanning,” Med Phys, vol. 47, no. 2, pp. 6396-6404, 2020.

3091

Mini-Oral

Predicting outcome of IROC’s thoracic moving dosimetry audit with random forest modeling.

Hunter Mehrens 1 , Andrea Molineu 1 , Nickolas Pajot 1 , Paola Alvarez 1 , Paige Taylor 1 , Laurence Court 1 , Rebecca Howell 1 , David Jaffray 1 , Christine Peterson 2 , Julianne Pollard-Larkin 1 , Stephen Kry 1 1 MD Anderson Cancer Center, Radiation Physics, Houston, USA. 2 MD Anderson Cancer Center, Biostatistics, Houston, USA

Purpose/Objective:

Radiotherapy facilities participating in national clinical trials in the United States must successfully irradiate credentialing phantoms from the Imaging and Radiation Oncology Core (IROC).This work analyzes what factors led to unacceptable plan delivery in the IROC moving thoracic phantom.

Material/Methods:

IROC’s thoracic phantom contains an oblong spheroid primary target made of solid water encased by cork. The accuracy of the delivered dose in the primary target are measured with thermoluminescent dosimeters (TLDs) and GAFchromic film. The phantom is paired with an acrylic motion table that provides sinusoidal motion in the superior inferior direction. To pass the phantom audit, measured point dose must agree with the treatment planning system (TPS) calculation within -8% and +5%, and ≥ 80% of pixels must pass a 7%/5mm gamma analysis on each of film plane with the average needing ≥ 85%. Thoracic phantom data collected from 2012-2020 comprising 1,643 irradiations was reviewed. Our analysis considered three subsets of data: (1) three phantom performance metrics, (2) eight treatment parameters and (3) six complexity metrics. We first evaluated these parameters using univariate analysis (independent t-test and Pearson chi-squared test with Tukey post-hoc analysis) to identify parameters associated with irradiation failure. Random forest modeling was used to predict phantom performance across all treatment parameters. In addition, 16 random forest models were run based on individual treatment parameters. Missing data was imputed using 100 trees and failures were upsampled to ensure a balanced data set between pass and fail. Data set was split into a training (70%) and a testing set (30%). The implementation of these random forest models were run ten times to assess fluctuations within the data and results. Random forest modeling was performed using R (4.0.2) via the ranger and missRanger packages.

Results: