Pectoral muscle identification in mammograms.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Image Processing

Background:

• Pectoral muscle removal is crucial for accurate mammogram analysis in computer-aided detection (CAD) of breast cancer.
• Its presence can negatively impact cancer detection outcomes.
• Existing methods often struggle with precise pectoral muscle segmentation.

Purpose of the Study:

• To propose an efficient and automatic method for pectoral muscle identification in mediolateral oblique (MLO) view mammograms.
• To improve the preprocessing stage of breast cancer detection systems.

Main Methods:

• Utilizing watershed transformation to identify the pectoral muscle edge.
• Developing a merging algorithm to correct oversegmentation caused by multiple catchment basins.
• Validating the approach on 84 mammograms from the mammographic image analysis database.

Main Results:

• Achieved a mean false positive rate of 0.85% and a mean false negative rate of 4.88% compared to ground truth.
• Demonstrated superior performance over state-of-the-art methods, particularly in reducing the mean false negative rate.
• Showcased low Hausdorff distances (mean ± std dev: 3.85 ± 1.07 mm) when compared to manual segmentations.

Conclusions:

• The proposed watershed transformation-based method offers an efficient and accurate solution for pectoral muscle segmentation in mammograms.
• This automated approach enhances the reliability of computer-aided detection systems for breast cancer.
• The method's improved accuracy, especially in reducing false negatives, holds significant potential for clinical application.