Texture feature standardization in digital mammography for improving generalizability across devices.

Area of Science:

• Medical Imaging
• Radiology
• Biophysics

Background:

• Mammographic texture analysis shows promise for breast cancer risk assessment.
• Variations between digital mammography (DM) systems can affect texture analysis results.
• Standardization is crucial for reliable cross-system comparisons.

Purpose of the Study:

• To compare texture features from two GE digital mammography systems (2000D and DS).
• To develop a methodological framework for identifying and standardizing system-induced effects in texture analysis.
• To identify texture features robust to inherent system differences for breast cancer risk assessment.

Main Methods:

• Compared GE Senographe 2000D and DS systems using a physical breast phantom (Rachel).
• Extracted 26 texture features (histogram, co-occurrence, run-length) from Cranio-Caudal (CC) view images.
• Applied z-score normalization and varied co-occurrence parameters for standardization.
• Used Kolmogorov-Smirnov (K-S) test to identify robust texture features (p>0.05).

Main Results:

• Identified specific texture features that are robust to differences between the GE 2000D and DS systems.
• Demonstrated that standardization steps can alleviate system-specific effects on texture features.
• The K-S test effectively distinguished between system-dependent and system-independent texture features.

Conclusions:

• A methodological framework for standardizing texture analysis across DM systems was established.
• The study provides a basis for selecting generalizable texture descriptors for breast cancer risk assessment.
• This approach enhances the reliability of texture analysis in multi-center or longitudinal studies.