Multi-scale Radiomic Fingerprint: Quantifying Spatial Changes in Biology.

Area of Science:

• Radiomics
• Medical Imaging Analysis
• Quantitative Imaging

Background:

• Traditional radiomics uses single-voxel increments, potentially missing multi-scale texture information.
• Voxel-based scales limit interpretability and reproducibility in clinical applications.

Purpose of the Study:

• To propose and evaluate a multi-scale radiomic approach using millimeter-based units.
• To improve texture information capture, reproducibility, and clinical interpretability in radiomics.

Main Methods:

• Defined texture distances in millimeter-based units for multi-scale analysis.
• Examined radiomic feature variance across multiple spatial scales and diseases (venous malformations, gliomas, Alzheimer's, brain metastases, multiple sclerosis).
• Compared predictive model performance using millimeter-based vs. voxel-based radiomics and evaluated anisotropic datasets.

Main Results:

• Radiomic features at different millimeter scales were statistically different (p < 0.05) across five diseases.
• Multi-scale millimeter-based radiomics models consistently yielded higher mean F1 scores than voxel-based models.
• Significant differences in texture metrics between millimeter and voxel scales suggest capture of distinct biological information.

Conclusions:

• A multi-scale, millimeter-based radiomic approach captures more textural information than traditional voxel-based methods.
• This approach enhances clinical interpretability and may have broad implications for disease diagnosis, monitoring, and treatment evaluation.
• The findings support the use of millimeter-based scales for more robust and clinically relevant radiomic analysis.