Breast Cancer Segmentation Methods: Current Status and Future Potentials.

Area of Science:

• Medical Imaging Analysis
• Computational Pathology
• Artificial Intelligence in Healthcare

Background:

• Early breast cancer detection significantly improves patient survival rates and reduces treatment costs.
• Mammography is a key tool for early breast cancer detection, with segmentation techniques crucial for tumor identification and analysis.
• Accurate segmentation aids in quantifying breast tissue volume for effective treatment planning.

Purpose of the Study:

• To categorize and review various segmentation methods for breast cancer detection in mammograms.
• To identify frequently used techniques, databases, and deep learning models in mammogram segmentation.
• To highlight the advantages of specific models, such as U-Net, in mammogram image analysis.

Main Methods:

• Categorization of segmentation methods into classical (region-, threshold-, edge-based), machine learning, and deep learning (supervised/unsupervised).
• Review of commonly used techniques within each category, including region growing and median filters for noise reduction.
• Analysis of frequently utilized databases, such as the MIAS database, and identification of popular deep learning architectures like U-Net.

Main Results:

• Region-based segmentation, particularly region growing, is prevalent in classical methods.
• Unsupervised machine learning methods are frequently employed in machine learning segmentation.
• The U-Net model is highly favored for mammogram segmentation due to its minimal need for annotated data and efficient training on GPUs.

Conclusions:

• Segmentation is vital for accurate tumor detection and quantification in mammograms.
• Deep learning models, especially U-Net, offer significant advantages for mammogram segmentation, streamlining analysis and potentially improving early detection rates.
• The study underscores the evolving landscape of mammogram analysis, emphasizing the role of advanced computational techniques.