Mammogram retrieval on similar mass lesions

Area of Science:

• Medical Imaging
• Computer Science
• Biomedical Engineering

Background:

• Vast digital mammogram archives exist, necessitating efficient retrieval systems for research and diagnosis.
• Content-based image retrieval (CBIR) is crucial for accessing and analyzing mammographic data.

Purpose of the Study:

• To develop a content-based mammogram retrieval system for identifying pathologically similar mass lesions.
• To extract and utilize shape and margin features for lesion characterization.
• To propose a similarity measure aligned with radiologists' observational methods.

Main Methods:

• Extraction of shape and margin features from mammographic lesions.
• Implementation of a hierarchical feature arrangement and weighting distance measure for similarity assessment.
• Evaluation using the DDSM dataset to assess feature effectiveness.

Main Results:

• Zernike moments effectively discriminate round-shape masses.
• Circumscribed margins are effectively discriminated among various margin types.
• The retrieval system achieves highest precision for round-shape and circumscribed-margin masses.

Conclusions:

• The proposed retrieval system accurately characterizes mammographic lesions based on shape and margin features.
• The system demonstrates high precision in retrieving specific types of mass lesions, aiding diagnostic processes.
• This approach enhances the utility of digital mammogram databases for clinical and research purposes.