Mammogram retrieval through machine learning within BI-RADS standards

Area of Science:

• Medical Imaging
• Computer Science
• Machine Learning

Background:

• Physicians often compare mammograms to recall similar cases for diagnosis.
• Content-based image retrieval (CBIR) systems can assist in this process by searching for visually similar images.

Purpose of the Study:

• To develop and evaluate a content-based mammogram retrieval system.
• To improve the accuracy of retrieving similar mammograms based on pathological characteristics.

Main Methods:

• Interpreting mammographic lesions using Breast Imaging Reporting and Data System (BI-RADS) standards.
• Employing a hierarchical similarity measurement scheme with a distance weighting function.
• Utilizing a machine learning approach with support vector machines and user relevance feedback.

Main Results:

• The proposed machine learning approach with a Radial Basis Function (RBF) kernel achieved the best performance.
• The system demonstrated improved retrieval performance for mammograms with similar mass and calcification lesions.

Conclusions:

• The developed content-based mammogram retrieval system effectively aids physicians in finding similar cases.
• Machine learning, particularly with RBF kernel and relevance feedback, enhances the accuracy of mammogram similarity retrieval.