ESTRO 2024 - Abstract Book

S4796

Physics - Quality assurance and auditing

ESTRO 2024

Analysis of Complexity Indices and Radiomic Features for Predicting Delivery Accuracy in VMAT

Sung Jin Kim 1 , Sang Hoon Jung 1 , Youngyih Han 1,2

1 Samsung Medical Center, Department of Radiation Oncology, Seoul, Korea, Republic of. 2 Sungkyunkwan University, Department of Health Sciences and Technology, SAIHST, Seoul, Korea, Republic of

Purpose/Objective:

To predict delivery accuracy, we calculated complexity indices (CIs) based on the mechanical variations during irradiation and extracted radiomic features from fluence maps. We analyzed the correlation of these quantified indices and features with the results of pre-treatment quality assurance (PTQA), as well as their effect on delivery accuracy.

Material/Methods:

A total of 59 volumetric modulated arc therapy (VMAT) plans were enrolled. Twenty-eight of these plans were created using Eclipse Ver. 15.6 (Varian Medical System, USA), and the remaining 31 were created with Pinnacle Ver. 9.10 (Philips Healthcare, USA). Thirty plans used TrueBeams with HD120 multi-leaf collimator (MLC), and 29 plans used TrueBeams with M120 MLC (Varian Medical System, USA). Portal dosimetry was employed for pre-treatment QA. We calculated CIs and radiomic features, including 2D shape, first order, gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRLM), gray level size zone matrix (GLSZM), Neighboring gray tone difference matrix (NGTDM), and gray level dependence matrix (GLDM). We analyzed the statistical differences in CIs and radiomic features based on the classification of treatment planning systems (TPSs) and MLC using the Mann-Whitney test. We also identified correlations between these factors and the results of VMAT QA using Spearman’s test.

Results:

Significant differences were found in leaf travel (LT) and LT mean for MLC, as well as in dose-rate variation (DR variation ) for the TPSs. Among the 120 features analyzed, 58 showed significant differences based on TPS classification. We found significant correlations between the CIs and QA results, particularly with factors like LT and LT mean , as shown in Fig.1. Regarding radiomic features, we observed statistically significant correlations in 11 features. Specifically, 21, 10, and 11 features demonstrated significant correlations for Eclipse, Pinnacle, and M120, respectively. We also found a significant correlation with 38 features among the plans using HD120, as shown in Fig. 2.