Radiomics, the extraction of quantitative features from medical images, has shown great potential in improving precision diagnosis, prognosis, and treatment planning. However, the reproducibility of radiomics features remains a major challenge due to the variability introduced by differences in imaging devices, acquisition protocols, and image reconstruction methods. This study introduces the first open-access cone-beam computed tomography (CBCT) phantom dataset specifically designed to test reproducibility in on-board imaging systems used in C-arm linear accelerators for radiotherapy. Using a widely recognized Catphan phantom, CBCT images were acquired from multiple devices across different imaging parameters, including variations in mAs, slice thickness, and reconstruction filters. The dataset includes 120 CBCT volumes with corresponding region of interest (ROI) segmentations and radiomics features enabling comprehensive testing of radiomics feature stability across intra- and inter-vendor comparisons. By providing this open-access dataset, the study aims to facilitate the standardization of CBCT radiomics research, improve feature reproducibility, and support the development of robust radiomics models for clinical applications.
Hatamikia et al. (2026) studied this question.