Nipple Localization in Automated Whole Breast Ultrasound Coronal Scans Using Ensemble Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• The nipple is a critical landmark in breast lesion diagnosis.
• Existing computer-aided detection (CADe) systems primarily focus on mammograms, with limited research on breast ultrasound (BUS) coronal views.
• Accurate nipple localization is essential for reliable analysis of BUS images.

Purpose of the Study:

• To develop and evaluate a novel computer-aided detection (CADe) system for locating the Nipple Shadow Area (NSA) in coronal plane breast ultrasound (BUS) images.
• To integrate shape and texture features with machine learning algorithms for improved NSA detection.
• To address the gap in CADe systems for nipple detection in BUS coronal views.

Main Methods:

• Utilized Hu Moments and Gray-level Co-occurrence Matrix (GLCM) for shape and texture feature extraction via an iterative sliding window.
• Employed an Artificial Neural Network (ANN) to identify probable Nipple Shadow Areas (NSA).
• Computed contour features (fractal dimension, edge distance, contour area) and used a Support Vector Machine (SVM) for accurate NSA identification.

Main Results:

• The proposed CADe system achieved 91.99% accuracy on a custom dataset of 64 BUS coronal images from 13 patients.
• Demonstrated high performance metrics: 97.55% specificity, 82.46% sensitivity, and an 88% F-score.
• Successfully identified the Nipple Shadow Area (NSA) in breast ultrasound coronal views.

Conclusions:

• The developed CADe system shows significant potential for accurate nipple detection in breast ultrasound coronal images.
• The integration of Hu Moments, GLCM, ANN, and SVM provides a robust approach for NSA localization.
• This system can serve as a valuable tool to aid radiologists in breast lesion diagnosis from BUS images.