Radiomics reproducibility in computed tomography through changes of ROI size, resolution, and hounsfield unit: A...

Area of Science:

• Quantitative radiology
• Medical imaging analysis
• Radiomics feature reproducibility

Background:

• Radiomics offers quantitative insights but is sensitive to scanning parameters.
• Investigating the impact of region of interest (ROI) sizes, Hounsfield Unit (HU), and computed tomography (CT) resolution on radiomics outcomes is essential.

Purpose of the Study:

• To evaluate the reproducibility of radiomics features under varying ROI sizes, HU values, and CT resolutions.
• To identify stable radiomics features and groups across different imaging conditions.

Main Methods:

• Utilized the GAMMEX 464 phantom for reproducibility assessment.
• Acquired data with consistent system setup, varying ROI sizes (1, 3, 5 mm) and HU.
• Employed LIFEx software for segmentation and feature extraction, focusing on textural features.
• Calculated coefficient of variation (COV) to quantify feature variability, with COV <5% indicating high stability.

Main Results:

• Out of 32 textural features, 16-18 showed high reproducibility (COV<5%) across different resolutions.
• Polyethylene, acrylic, and water exhibited stable textural features across parameter variations.
• Grey-level run length matrix (GLRLM) and grey-level zone length matrix (GLZLM) groups demonstrated high stability.

Conclusions:

• Standardization of pre-analysis steps and scanning parameters is vital for consistent radiomics feature values.
• Ensuring reproducibility enhances the reliability of radiomics applications in clinical and research environments.
• Further research is needed to build upon these findings and improve radiomics standardization.