Enhancing the Clinical Utility of Radiomics: Addressing the Challenges of Repeatability and Reproducibility in CT and...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Radiomics leverages imaging phenotypes and machine learning for cancer diagnosis and prognosis.
• Repeatability and reproducibility of radiomic features are significant barriers to clinical translation.
• Existing literature often lacks comprehensive discussion on factors influencing radiomic feature reliability.

Purpose of the Study:

• To review and synthesize factors affecting the repeatability and reproducibility of radiomic features.
• To examine the impact of these variations on radiomic model performance.
• To provide recommendations for improving the reliability of radiomic studies.

Main Methods:

• Literature synthesis focusing on CT/MR-based radiomic features.
• Analysis of sources of variation, including random effects and biases (scanner, inter-reader).
• Evaluation of the number of reproducible features and availability of repeatability indices.

Main Results:

• Identified random effects and biases as key sources of variation impacting generalizability.
• Highlighted scanner variability and inter-reader differences as significant biases.
• Found a paucity of detailed reporting on variation sources and reliability metrics in current studies.

Conclusions:

• Emphasized the critical role of random effects in feature selection and bias harmonization.
• Recommended detailed reporting of variation sources, calculation parameters, and reliability metrics.
• Stressed the need for standardized approaches to ensure successful clinical translation of radiomic models.