Design and Clinical Efficacy of a Computer-Aided Detection Tool for Masses in Mammograms
Abstract
Our hypothesis is that a highly sensitive and highly specific CAD scheme, incorporating unique preprocessing techniques and advanced Decision Theory methods, can detect masses and improve the performance of mammographers. To test this hypothesis, we propose to construct a CAD system from two key components: 1) a highly sensitive mass detector, and 2) statistical models designed to reduce false-positives. We feel that it is essential to develop a tool that can identify a high percentage of masses, both spiculated and nonspiculated. It is important for computerized tools to detect as many masses as possible, but not to detect too many regions that are not actual masses. Thus, our program will first concentrate on finding many suspicious regions. Once suspicious regions are identified within the mammogram, we will explore several classification techniques to determine whether the regions are actually masses or some other structure in the breast. The techniques we plan to explore, for both detecting masses and classifying them, include standard, well-known techniques as well as new and novel approaches.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2005
- Accession Number
- ADA441280
Entities
People
- Joseph Y. Lo
- Swatee Singh
Organizations
- Duke University Hospital