Repeatability and Reproducibility of Radiomic Features: A Systematic Review

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Predictive models for clinical decision-making increasingly use quantitative imaging biomarkers, known as radiomic features.
• Concerns regarding the reproducibility of these radiomic features can impede the broad generalizability and clinical adoption of radiomics-assisted models.

Purpose of the Study:

• To qualitatively synthesize findings from 41 studies investigating the repeatability and reproducibility of radiomic features.
• To identify factors influencing the stability of radiomic features and highlight areas for improved reporting.

Main Methods:

• A systematic review of the published peer-reviewed literature was conducted.
• Studies were identified through a PubMed search using specific filters and keywords related to cancer, radiomics, reproducibility, and repeatability.
• Information on cancer type, feature class, reporting quality, and statistical metrics was extracted from 41 full-text articles.

Main Results:

• First-order radiomic features demonstrated higher reproducibility compared to shape metrics and textural features.
• Entropy was consistently identified as a stable first-order feature.
• Feature reproducibility is sensitive to image acquisition, reconstruction, preprocessing details, and software used, with no consensus on stable shape or texture features.

Conclusions:

• Current investigations into radiomic feature repeatability and reproducibility are primarily focused on a limited range of cancer types.
• Improved reporting quality is needed, particularly concerning feature extraction software, image preprocessing, and the criteria for defining feature stability.