Breast pectoral muscle segmentation in mammograms using a modified holistically-nested edge detection network.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate segmentation of pectoral muscles in mammograms is crucial for breast cancer detection.
• Existing methods using hand-crafted models struggle with complex pectoral muscle boundary variations and unclear boundaries due to tissue overlap.

Purpose of the Study:

• To develop an automated method for pectoral muscle segmentation in mediolateral oblique mammograms.
• To overcome limitations of traditional segmentation techniques by employing a deep learning approach.

Main Methods:

• A Convolutional Neural Network (CNN) inspired by the Holistically-nested Edge Detection (HED) network was adapted for pectoral muscle segmentation.
• The framework utilizes multi-scale and multi-level learning to capture complex hierarchical features.
• Post-processing steps involving morphological property extraction refine the initial boundary estimation from the network's probability map.

Main Results:

• The proposed CNN method achieved high accuracy in pectoral muscle segmentation.
• Quantitative evaluation demonstrated Jaccard similarity of 94.8% ± 8.5% and Dice similarity of 97.5% ± 6.3% across multiple datasets.
• The results are comparable to state-of-the-art methods.

Conclusions:

• The developed deep learning framework effectively performs automatic pectoral muscle segmentation in mammograms.
• This approach offers a robust alternative to conventional methods, addressing challenges posed by complex anatomical variations.
• The method shows significant potential for improving the efficiency and accuracy of breast cancer screening workflows.