Reproducibility and standardization in radiomics: Are we there yet?

Abstract

Radiomics aims for ways to personalize patient’s treatments based on the extraction of texture features from routine clinical images. However, the implementation of Radiomics-based models in routine patient care poses a significant number of challenges, since there are multi-step processes in the implementation of the model. These include image selection (e.g CT, MRI, PET, or a combination of these), region of interest (ROI) segmentation, texture-feature extraction, model development, and model-validation. Each of these processes needs careful evaluation to ensure reproducibility and accuracy of the final Radiomics-based model before it can be introduced into the clinical practice. Image selection, ROI segmentation, and texture feature extraction are the most critical component in Radiomics modeling, because they define the reliability of the model. In this short review, after presenting a brief description of a Radiomics-based model representative workflow, we will concentrate in particular on the status of various technical aspects of imaging modality, segmentation selection, and texture feature extraction, including their shortcomings and possible solutions with emphasis in radiation therapy.