ESTRO 2025 - Abstract Book

S3729

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

[2] Reynaud, O., Time-dependent diffusion MRI in cancer: tissue modeling and applications. Frontiers in Physics, 2017. 5: p. 58. [3] Jokivuolle, M., et al., Assessing tumor microstructure with time-dependent diffusion imaging: Considerations and feasibility on clinical MRI and MRI-Linac. Med Phys, 2024.

779

Digital Poster Quantifying robustness of positron emission tomography radiomics features with a dedicated phantom Joel Poder 1,2,3 , Zhihe Tian 2 , Yu-Feng Wang 2 , Yu Sun 2 , Robba Rai 4 , Annette Haworth 2 1 Medical Physics, St George Hospital Cancer Care Centre, Kogarah, Australia. 2 Institute of Medical Physics, University of Sydney, Camperdown, Australia. 3 Centre for Medical Radiation Physics, University of Wollongong, Wollongong, Australia. 4 Liverpool Cancer Therapy Centre, Ingham Institute, Liverpool, Australia Purpose/Objective: The purpose of this study is to determine the robustness of positron emission tomograph (PET) radiomics features to changing image reconstruction parameters through the use of a phantom with dedicated radiomics inserts. Identification of radiomics features robust to changes in reconstruction parameters is crucial for the selection of features in multi-centre radiomics studies aiming to predict treatment response and outcome. Material/Methods: A set of 11 inserts were three-dimensional (3D) printed mimicking a range of shape and texture radiomics features. The inserts were hollow with a solid outer shell to allow for filling with 30kBq/ml of Ga-68. The radiomics inserts were imaged using a GE Discovery MI scanner for a 20 minute acquisition time. The Q.Clear reconstruction algorithm was used to reconstruct the raw images with varying parameters to quantify the robustness of radiomics features. On each reconstructed image, insert regions of interest were created in MIM (v7.1.3, MIM Software INC.) encompassing 40% of the maximum signal within the insert. Radiomics analysis was performed using PyRadiomics v3.1.0 and coefficient of variation (COV) calculated to assess the robustness of radiomics feature extraction. Results: Changes in image smoothing were found to have a larger impact on radiomics feature stability than varying image noise levels. Shape features were found to be the most robust to changes in image reconstruction parameters with 100% of features having a COV < 5%. This was followed by first order features (86.1% of features with COV < 5%) and texture features (78.8% of features with COV < 5%). Ninety one percent of all features have a COV < 10%. Conclusion: A novel phantom design was developed to quantify the robustness of PET radiomics feature extraction. The phantom was utilised to quantify radiomics feature robustness to varying reconstruction parameters on a single PET scanner. PET radiomics feature robustness appears to be promising, indicating their utility for studies aiming to predict treatment response and outcome. Future studies will aim to utilise the phantom to quantify PET radiomics feature extraction in a multi-centre setting.

Keywords: radiomics, phantom, robustness