Radiomics feature reliability assessed by intraclass correlation coefficient: a systematic review

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Medicine

Background:

• Radiomics research is rapidly expanding, but concerns about its reliability are increasing.
• Intraclass correlation coefficient (ICC) is a key metric for assessing radiomics feature reliability.
• Understanding factors influencing radiomics reliability is essential for clinical translation.

Purpose of the Study:

• To systematically review the current status of radiomics reliability research.
• To analyze radiomics reliability based on the intraclass correlation coefficient (ICC).
• To identify common pitfalls in ICC usage and suggest improvements.

Main Methods:

• Systematic review following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines.
• Inclusion of 481 radiomics studies utilizing CT, PET, or MRI across diverse subjects and diseases.
• Analysis focused on ICC values reported for radiomics features related to various factors.

Main Results:

• Image segmentation showed high radiomics feature reliability.
• Image acquisition introduced more feature variability than segmentation, particularly in MRI.
• Image post-processing and feature quantification demonstrated varied reliability levels and potential to mitigate acquisition-induced variability.

Conclusions:

• Radiomics feature reliability is significantly influenced by image acquisition parameters, especially for MRI.
• While image segmentation appears reliable, overall radiomics reliability is impacted by multiple factors.
• High study heterogeneity and potential bias limit definitive conclusions; further research on radiomics reliability is warranted.