Is a study on radiomics reproducibility reproducible? Let’s see, but an open door anyway

Excerpt

Artificial intelligence (AI), regarded as the key technology of the fourth industrial revolution, shows great potential in predicting diagnosis and prognosis [1]. Radiomics, one of the AI study areas, is the method to quantify tumor phenotypes by extracting high-throughput quantitative image biomarkers. These effective image biomarkers, captured from the region of interest (ROI), are associated with underlying gene-expression patterns [2]. In many studies, radiomics showed broad prospects and was applied to such as diagnosis, prognosis, risk assessment, treatment selection, biopsy, or resect. Although the performance and realistic demand of AI were amazing, many challenges remain in achieving and applying AI products. For example, sometimes AI models applied in clinical settings may be unreliable and impractical, especially in different development environments. Hence, reproducibility [3] and interpretability [4] issues are urgent questions to be solved. …