A study in Radiological Society of North America (RSNA) has provided valuable insights into the reproducibility of radiomics features extracted from T2-weighted MRI scans in patients with neuroblastoma. Conducted under the PRIMAGE project, this study underscores the importance of image processing steps in maintaining the reliability of radiomics features for personalized cancer diagnosis and prognosis. The research involved a retrospective analysis of 419 patients with neuroblastic tumors, diagnosed between 2002 and 2023, encompassing 746 T2/T2*-weighted MRI sequences taken at diagnosis and/or post-initial chemotherapy.
MRI images underwent several processing steps, including denoising, inhomogeneity bias field correction, normalization, and resampling. Tumors were automatically segmented, and 107 shape, first-order, and second-order radiomics features were extracted as the reference standard. Subsequent variations in image processing settings and volumetric masks were applied, and new radiomics features were extracted for comparison with the reference standard. Reproducibility was assessed using the concordance correlation coefficient (CCC), and intrasubject repeatability was measured using the coefficient of variation (CoV).
The study revealed that when normalization was omitted, only 5% of radiomics features demonstrated high reproducibility, highlighting normalization as a critical step for ensuring reliable radiomics features. Significant changes were observed in the normalization and resampling processes, emphasizing their impact on the reproducibility of radiomics features. In contrast, removing inhomogeneities had the least impact, with 83% of radiomics parameters remaining stable. Shape features remained stable after mask modifications, achieving a CCC greater than 0.90, indicating high reproducibility. Moreover, volumetric mask modifications were the most favorable for achieving high CCC values, with 70% of radiomics features showing stability. However, only 7% of second-order radiomics features demonstrated an excellent CoV of less than 0.10, indicating limited repeatability for these features.
The study concludes that modifications in the T2-weighted MRI preparation process for patients with neuroblastoma can significantly influence radiomics features, with normalization identified as the most influential factor for reproducibility. Inhomogeneities removal had the least impact. These findings are crucial for developing robust imaging biomarkers and improving personalized cancer diagnosis and prognosis. The PRIMAGE (PRedictive In-silico Multiscale Analytics to support cancer personalized diaGnosis and prognosis, Empowered by imaging biomarkers) project aims to advance cancer diagnostics and prognostics through integrating imaging biomarkers and multiscale analytics.