Breast tissue segmentation from x-ray radiographs.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Accurate segmentation of mammograms is crucial for breast cancer detection.
• Existing methods may lack robustness or accuracy in distinguishing key regions.

Purpose of the Study:

• To develop a robust and accurate method for segmenting mammograms into three distinct regions: breast tissue, pectoral muscle, and background.
• To evaluate the proposed method's performance against manual expert segmentation.

Main Methods:

• A two-layer neural committee machine was employed for segmentation.
• The first layer utilized local features (intensity, histograms, LBP, HOG) and support vector machines (SVMs).
• The second layer used a gating network with SVMs to combine first-layer outputs and a prior map.

Main Results:

• The method successfully segmented breast tissue in all tested mammograms without failure.
• Experimental results on 495 mammograms demonstrated accuracy comparable to manual expert segmentation.
• The approach showed robustness in distinguishing breast tissue, pectoral muscle, and background.

Conclusions:

• The proposed neural committee machine offers a robust and accurate solution for mammogram segmentation.
• This method has the potential to assist radiologists by providing precise segmentation results.
• The findings suggest this technique can challenge the accuracy of manual segmentation in clinical practice.