Optimization of volumetric breast density estimation in digital mammograms

Area of Science:

• Radiology
• Medical Imaging
• Biophysics

Background:

• Mammography is crucial for breast cancer screening.
• Accurate breast density estimation is vital for risk assessment.
• Current physics-based methods may underestimate density in dense breasts due to lack of fatty tissue reference pixels.

Purpose of the Study:

• To investigate alternative approaches for obtaining tissue reference values in mammography.
• To improve the accuracy of breast density estimations, particularly in dense breasts.
• To develop a hybrid method combining multiple reference values for robust density assessment.

Main Methods:

• Evaluated three novel reference values (F1, F2, F3) derived from pixel value distributions and dense tissue regions.
• F1 and F2 utilized percentiles from breast interior pixel values.
• F3 used dense tissue regions and fat layer thickness measurements.
• A hybrid approach combined density estimations from F1-F3.
• Compared results with MRI data from 119 women (202 mammogram pairs).

Main Results:

• Individual reference values (F1-F3) showed varying correlations with MRI-based density estimates.
• The hybrid method, combining density estimations from all three references, yielded the highest correlation ([Formula: see text]).
• The hybrid approach demonstrated improved volumetric density estimates, particularly for dense breasts.

Conclusions:

• A hybrid method combining multiple reference values enhances breast density estimation accuracy.
• This approach is particularly beneficial for dense breasts, overcoming limitations of traditional methods.
• The developed algorithms offer a more reliable tool for volumetric breast density assessment irrespective of breast composition.