Description:

Medical imaging is often used to support clinical decision-making, but only through visual inspection or simple measures. Additional relevant information, e.g., disease phenotypes, may be present in medical images but remains unassessed. Radiomics characterizes regions of interest in medical images using quantitative measures, e.g., morphology, image intensities, and texture. Although radiomics features (RF) are increasingly extracted via standardized radiomics software packages for more reproducible research, different feature-generation hyperparameters, fusion techniques, and segmentation methods may still lead to variable RFs. Employing robust RFs to process variations is another critical step toward a reproducible study. In particular, the interoperability of radiomics software is hindered by the lack of consensus concerning the exact calculation of image biomarkers. As a result, considerable variations in biomarker values have been reported, even if computed from the same image. Hence, we aim to improve the reproducibility of radiomics features by standardizing the computational process of extracting image features using convolutional image filters based on the Image Biomarker Standardisation Initiative (IBSI) guideline