A Computer-Based Decision Support System for Breast Cancer Diagnosis
Abstract
In the second year of the project, we devoted efforts in developing effective feature extraction methods, constructing feature database, developing visual explanation tool for data mining and knowledge discovery, which is both statistically principled and visually effective. This method, as illustrated by the well-planned simulations and pilot applications in computer-aided diagnosis, can be very capable of revealing hidden structure within data. It is important to emphasize that the present algorithm is that the models are determined by the information theoretic criteria, and this criterion can not only select the most appropriate model structure but also allow a user-driven portfolio as a double check. In addition, since we perform model selection and parameter initialization firstly over the projection space, the computational complexity is greatly reduced in compared to the maximum likelihood estimation in full dimension. Other possible advantages include the determination of data projection by maximum the separation of clusters which in turn optimizes the other crucial operations such as model selection and parameter initialization, and data rendering algorithms which permit user or hypothesis driven nature of the data projection. Using the visual explanation tool, we are trying to discover the feature database structure for case, feature selection and classifier design.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2000
- Accession Number
- ADA389946
Entities
People
- Zuyi Wang
Organizations
- The Catholic University of America