A statistical methodology for mammographic density detection

Area of Science:

• Medical Imaging
• Biostatistics
• Radiology

Background:

• Mammographic density is a recognized breast cancer risk factor.
• Accurate quantification of breast tissue composition is crucial for risk assessment.
• Existing methods may lack automated and precise tissue discrimination capabilities.

Purpose of the Study:

• To develop and validate a statistical methodology for automated discrimination of fat and fibroglandular tissue in digitized mammograms.
• To establish a foundation for incorporating tissue density into breast cancer risk analysis.
• To enable quantitative analysis of breast tissue composition directly from image data.

Main Methods:

• A chi-square probability analysis based statistical methodology is employed.
• The method utilizes a reversible linear filtering operation analogous to deconvolution.
• A relaxation method estimates a global reference variance, with local variances compared using chi-square analysis to label tissue as fat or nonfat.

Main Results:

• The methodology successfully discriminates between radiolucent (fat) and dense (fibroglandular) tissues.
• Preliminary results demonstrate encouraging accuracy in region-by-region tissue labeling.
• The approach identifies two distinct random events in the input field with differing variances, corresponding to fat and other tissues.

Conclusions:

• The developed statistical method offers automated and reliable discrimination of breast tissue types in mammograms.
• This technique provides a basis for integrating mammographic density into breast cancer risk assessment models.
• Future applications include quantitative risk analysis based on fat/dense tissue percentages and amounts.