[Paper Review] Segmentation of Breast Regions in Mammogram Based on Density: A Review

Nafiza Saidin, Harsa Amylia Mat Sakim+2 more|arXiv (Cornell University)|Sep 25, 2012

Digital Radiography and Breast Imaging55 references21 citations

TL;DR

This review examines segmentation techniques for breast regions in mammograms based on tissue density, focusing on fibroglandular tissue detection and anatomical region segmentation. It evaluates methods for classifying and segmenting dense breast tissue—critical for early breast cancer detection—while addressing challenges in performance evaluation due to ground truth limitations.

ABSTRACT

The focus of this paper is to review approaches for segmentation of breast regions in mammograms according to breast density. Studies based on density have been undertaken because of the relationship between breast cancer and density. Breast cancer usually occurs in the fibroglandular area of breast tissue, which appears bright on mammograms and is described as breast density. Most of the studies are focused on the classification methods for glandular tissue detection. Others highlighted on the segmentation methods for fibroglandular tissue, while few researchers performed segmentation of the breast anatomical regions based on density. There have also been works on the segmentation of other specific parts of breast regions such as either detection of nipple position, skin-air interface or pectoral muscles. The problems on the evaluation performance of the segmentation results in relation to ground truth are also discussed in this paper.

Motivation & Objective

• To analyze and synthesize existing approaches for segmenting breast regions in mammograms based on tissue density.
• To identify gaps in segmentation techniques focused on fibroglandular tissue and anatomical structures like the nipple, skin-air interface, and pectoral muscle.
• To evaluate the performance of segmentation methods in relation to ground truth data, highlighting inconsistencies and limitations.
• To provide a comprehensive overview of density-based segmentation techniques for use in computer-aided diagnosis systems.
• To guide future research by identifying underexplored areas and methodological challenges in segmentation accuracy and validation.

Proposed method

• Systematic review of peer-reviewed literature on breast density-based segmentation in mammograms from 2000 to 2012.
• Categorization of methods into three main groups: classification of glandular tissue, segmentation of fibroglandular tissue, and segmentation of specific anatomical regions.
• Analysis of image processing techniques such as thresholding, edge detection, region growing, and active contours used in segmentation.
• Evaluation of performance metrics used in studies, including sensitivity, specificity, and Dice similarity coefficient, with emphasis on ground truth reliability.
• Comparison of algorithmic approaches across studies, focusing on preprocessing, feature extraction, and post-processing steps.
• Identification of common limitations in dataset availability, annotation quality, and lack of standardized evaluation protocols.

Experimental results

Research questions

• RQ1 What are the dominant techniques used for segmenting fibroglandular tissue in mammograms based on breast density?
• RQ2 How do different segmentation methods perform in identifying anatomical structures such as the nipple, skin-air interface, and pectoral muscle?
• RQ3 What are the main challenges in evaluating segmentation accuracy due to inconsistencies in ground truth data?
• RQ4 How does breast density influence the choice and effectiveness of segmentation algorithms?
• RQ5 What gaps exist in current research regarding standardized evaluation and reproducibility of segmentation results?

Key findings

• Most studies focus on classifying or segmenting fibroglandular tissue rather than full breast anatomical regions based on density.
• Segmentation of specific structures like the nipple and pectoral muscle is less explored, despite their clinical relevance.
• Performance evaluation is often hampered by poor or inconsistent ground truth annotations, reducing reliability of reported metrics.
• Thresholding and edge-based methods remain widely used, but their accuracy is limited in dense breast tissues with low contrast.
• Active contour and region-growing methods show improved results but require careful parameter tuning and initialization.
• There is a lack of standardized datasets and evaluation protocols, limiting reproducibility and cross-study comparison.