Using multiscale texture and density features for near-term breast cancer risk analysis

Area of Science:

• Radiology
• Medical Imaging Analysis
• Biomedical Engineering

Background:

• Mammography screening aims to detect breast cancer early.
• Current screening paradigms may not be optimal for all individuals.
• Personalized screening strategies could enhance efficacy.

Purpose of the Study:

• To investigate the potential of quantitative multiscale texture and density feature analysis of digital mammograms.
• To predict near-term breast cancer risk for personalized screening.
• To improve the efficacy of screening mammography.

Main Methods:

• Utilized a dataset of 340 digital mammograms (141 positive, 199 negative/benign).
• Extracted five subregions at different scales from mammograms based on intensity value distributions.
• Developed and calculated density and texture features, using sequential forward floating selection and a support vector machine (SVM) with tenfold validation for risk prediction.
• Assessed performance using the area under the receiver operating characteristic curve (AUC).

Main Results:

• An optimal feature set of 12 features was selected from an initial 765 computed features.
• The SVM classifier achieved an AUC of 0.729 ± 0.021.
• Reported positive predictive value of 0.657 and negative predictive value of 0.755.

Conclusions:

• Quantitative multiscale mammographic image feature analysis shows a moderately high association with actual near-term breast cancer risk.
• This approach holds promise for developing a personalized breast cancer screening paradigm.
• The findings support the use of advanced image analysis for improved breast cancer risk prediction.