MRI-based radiomics in breast cancer: feature robustness with respect to inter-observer segmentation variability

Area of Science:

• Medical Imaging
• Radiology
• Quantitative Imaging

Background:

• Radiomics extracts quantitative features from medical images for tissue characterization.
• MRI-based radiomics shows promise in breast cancer diagnosis and therapy assessment.
• Feature quantification and robustness are key challenges in radiomics.

Purpose of the Study:

• To determine the robustness of radiomics features extracted by two software packages.
• To assess feature robustness concerning variability in manual breast tumor segmentation on MRI.

Main Methods:

• 129 breast tumors were manually segmented in 3D on T1-weighted MRI by four observers.
• Robustness was assessed using the intraclass correlation coefficient (ICC > 0.9).
• Inter-observer variability was evaluated using the volumetric Dice Similarity Coefficient (DSC).

Main Results:

• Mean DSC was 0.81, indicating good segmentation overlap across observers.
• 41.6% of RadiomiX features and 32.8% of Pyradiomics features were robust.
• Robust features were independent of inter-observer manual segmentation variability.

Conclusions:

• A significant proportion of radiomics features are robust to inter-observer segmentation variability.
• This finding supports the potential of MRI-based radiomics for reliable breast cancer assessment.
• Standardization of segmentation may further enhance radiomics feature reliability.