PeMNet for Pectoral Muscle Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Mammography is crucial for early breast cancer detection.
• Computer-Aided Diagnosis (CAD) systems accelerate mammography analysis.
• Accurate breast pectoral muscle segmentation is vital for specific CAD tasks.

Purpose of the Study:

• To propose a novel deep learning framework, PeMNet, for efficient breast pectoral muscle segmentation in mammography.
• To introduce and integrate the Global Channel Attention Module (GCAM) to enhance segmentation performance.
• To improve the accuracy and efficiency of pectoral muscle segmentation for CAD systems.

Main Methods:

• Developed PeMNet, a deep learning framework incorporating the Global Channel Attention Module (GCAM).
• GCAM extracts and refines channel attention maps (CAMs) iteratively for improved feature representation.
• Implemented elementwise multiplication of refined CAMs with feature maps to enhance segmentation.
• Validated the framework on a merged INbreast and OPTIMAM dataset.

Main Results:

• PeMNet achieved superior performance compared to state-of-the-art methods.
• Key metrics included IoU of 97.46%, global pixel accuracy of 99.48%, Dice similarity coefficient of 96.30%, and Jaccard index of 93.33%.
• The GCAM module effectively improved segmentation accuracy with minimal parameter overhead.

Conclusions:

• PeMNet offers a highly accurate and efficient solution for breast pectoral muscle segmentation in mammography.
• The integration of GCAM significantly enhances the capabilities of deep learning models for medical image analysis.
• This framework has the potential to advance CAD systems for earlier and more reliable breast cancer diagnosis.