(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: Daily-CBCT guidance has become standard of care for many sites to assist in patient positioning, adaptive radiotherapy, and radiomics-based modelling. One of the main challenges of radiomics is stability and reproducibility of radiomics-based texture features (TF). Studies involving radiomics must pre-select features prior to their use in clinical-based models. In this study, a novel 3D-printed phantom was used to evaluate the robustness of TF from CBCT images.
Methods: A 3D-printed phantom with five custom-printed rods of various materials, infill shapes and densities was used to acquire clinical CBCT images on two Varian TrueBeam Linacs. The TF reproducibility was evaluated across and within two Linacs from three consecutive CBCT-scans each from head/neck and thorax protocols. Regions of interest (ROI) of ~6.2cm³ from each rod were used to extract TF using the PyRadiomics software, three rods were inserted in this study. One-way ANOVA model of the intra-class correlation coefficient (ICC(3,1)) was used to evaluate TF reproducibility, with features showing ICC values above 0.9 considered robust if their Bonferroni-corrected p-value was below 0.05.
Results: The 3D-printed phantom (cylinder: d=22cm, h=25.5cm) comprises slots for interchangeable rods (d=1cm, h=10-20cm) to evaluate different types of materials. A total of 45 and 86 features were respectively reproducible for intra-Linac reproducibility with the thorax and head protocols; 43 of these features were shared between the two. For inter-Linac reliability, 31 and 54 features had ICC(3,1)>0.9 for thorax and head respectively; 23 were reproducible across Linacs for both protocols. No features exhibited ICC(3, 1)>0.9 for inter-protocol stability.
Conclusion: We have developed a 3D-phantom for consistent evaluation of texture feature stability/reproducibility. Our study found a substantial number of features available for robust radiomic modelling from CBCT imaging. However, some features showed variations across Linacs and CBCT scan protocol.
Keywords
Radiomics, CBCT Texture Features
Taxonomy
IM/TH- Image Analysis (Single Modality or Multi-Modality): Imaging biomarkers and radiomics
Contact Email
