Validation of A Method to Compensate Multicenter Effects Affecting CT Radiomics

Area of Science:

• Medical imaging analysis
• Radiomics and quantitative imaging

Background:

• Radiomics extracts quantitative features from medical images, surpassing visual assessment accuracy.
• CT scanner parameters (e.g., reconstruction kernel, section thickness) introduce variability in radiomic features, masking true biological textures.
• Standardizing radiomic features across different CT protocols is crucial for reliable analysis.

Purpose of the Study:

• To evaluate a compensation method for correcting radiomic feature variations arising from diverse CT imaging protocols.
• To determine if this method can eliminate protocol-specific effects on radiomic features.

Main Methods:

• Retrospective analysis of phantom data (10 texture patterns) and patient data (74 patients) from two cohorts scanned with different CT protocols.
• Development of a compensation approach to transform radiomic features into a common, protocol-independent space.
• Statistical assessment using Friedman tests and Principal Component Analysis (PCA) to evaluate feature distribution and texture pattern separation before and after compensation.

Main Results:

• Phantom data showed significant differences in feature distributions between protocols (P < .05) that were eliminated after compensation (P > .05).
• PCA confirmed that texture patterns were no longer confounded by imaging protocols post-compensation.
• Patient data demonstrated a substantial reduction in protocol effects, with significant P-values decreasing from 100% and 98% to 30% and 15% post-compensation.

Conclusions:

• The developed image compensation method effectively realigns radiomic feature distributions across different CT imaging protocols.
• This standardization facilitates robust and reproducible multicenter radiomic studies.
• The findings support the broader adoption of radiomics in clinical research by addressing scanner variability.