Breast density classification to reduce false positives in CADe systems.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Biomedical Engineering

Background:

• Breast parenchymal density is a significant breast cancer risk factor.
• Dense breast tissue complicates mammogram interpretation and lesion detection.
• Automated breast density classification can assist in early breast cancer detection and analysis.

Purpose of the Study:

• To develop and evaluate a novel weighted voting tree classification scheme for breast density classification.
• To compare the proposed method with existing classification techniques.
• To integrate the classification scheme into a computer-aided detection (CADe) system to improve lesion detection.

Main Methods:

• A novel hierarchical classification procedure combining classifiers with linear discriminant analysis (LDA).
• Utilized 298 texture features, with statistical analysis for feature selection based on tissue type influence.
• Incorporated the classification scheme into a CADe system and tested on 1459 mammograms (322 SFM, 1137 FFDM).

Main Results:

• Achieved 99.75% classification accuracy on the mini-MIAS screen-film mammogram (SFM) dataset.
• Demonstrated 91.58% agreement on the full-field digital mammogram (FFDM) dataset.
• Integration into the CADe system showed improved lesion detection rates and enhanced lesion detectability.

Conclusions:

• The proposed weighted voting tree classification scheme is effective for automated breast density classification.
• Prior breast tissue classification significantly improves lesion detection performance in CADe systems.
• The developed tools aid in distinguishing local attenuation without local tissue density constraints, enhancing diagnostic accuracy.