The long and winding road of radiomics: learnings from two meta-analyses of the radiomics quality score

The high-throughput extraction of radiomics features from medical images for predictive modelling holds great promise to improve the clinical management of patients. Previous meta-analyses into the radiomics quality score (RQS) applied in the literature have shown that after more than a decade of investigation, issues with workflow standardisation, model reproducibility, validation, and data accessibility persist and impede the clinical translation of radiomics-based models. These systematic findings have informed a timely review of the best practices and pitfalls to avoid within radiomics and predictive modelling, with a focus on realistic radiomics modelling in the context of limited sample sizes. Each section covers a radiomics topic that encompasses one or more RQS criteria and is broken into subsections as follows: (1) a discussion of the background and recommendations on the respective topic, (2) key findings from our meta-analyses and discovered pitfalls, and (3) a succinct list of actionable items that reflect best practice. New and emerging quality appraisal tools and the future direction of radiomics are also discussed.