Model and Expansion Based Methods of Detection of Small Masses in Radiographs of Dense Breasts
Abstract
This report describes progress made in during the first year of study. Our goal is to detect masses m dense mammograms having a diameter less than 1 Cm. The "idea" of this project is to detect subtle masses by tuning the central frequency and width of a basis function generating overcomplete expansions. By modeling the shape of a mass through this flexibility we hope to detect small and subtle masses in dense breasts and improve the chances of early detection in screening mammography. In the first part of our investigation, we first evaluated existing tools to compute overcomplete expansions of multiscale signals. We compared in one dimension the CwT and the DWT for a proof of concept concerning any advantage of pursuing refinement of scale. We processed phantom masses, and iD intensity profiles of real masses mammograms to evaluate feasibility. In order to identify the best scale, we evaluated the use of maxima singularities and a correlated model using three masses of different size. Our study answered the question of weather of not dyadic scales were sufficient to detect masses in a dense mammograms. We clearly showed that reasonable approximations of mass shapes could be obtained through overcomplete expansions that computed voices between the traditional dyadic scales.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2000
- Accession Number
- ADA385385
Entities
People
- Andrew F. Laine
Organizations
- Columbia University