Intra-scan inter-tissue variability can help harmonize radiomics features in CT

Area of Science:

• Medical Imaging
• Radiomics
• Quantitative CT

Background:

• Radiomics analysis in computed tomography (CT) is sensitive to variations in acquisition protocols.
• Understanding the repeatability and variability of radiomics features is crucial for reliable quantitative imaging.

Purpose of the Study:

• To assess the repeatability and relative intra-scan variability of radiomics features across different CT acquisition protocols.
• To identify radiomics features suitable for improving tissue characterization and inter-site harmonization.

Main Methods:

• Utilized CT scans from a radiomics phantom and unenhanced abdominal patient series.
• Extracted 9 radiomics features, defining "tandems" as measurements of specific tissues by specific features.
• Assessed variability sources (repetition, protocol, material, patient) and intra-scan correlation.

Main Results:

• Radiomics feature repeatability is material-dependent; 56% of tandems were highly repeatable.
• Histogram-derived radiomics showed lower repeatability; Gray Level Zone Length Matrix (GLZLM) and Gray Level Co-occurrence Matrix (GLCM) features showed high correlation.
• Nearly 60% of relative radiomics measurements exceeded a 0.90 correlation coefficient, enabling paired measurements to improve material/tissue differentiation.

Conclusions:

• Identified specific radiomics features with linearly correlated intra-scan measurements between tissues.
• Relative intra-scan measurements offer a method to improve quantitative imaging reliability on CT.
• Linear correlation property enhances tissue characterization and facilitates inter-site harmonization.