A Constraint Satisfaction Neural Network Approach for Data Mining Classification and Association Rules in Breast Cancer Databases
Abstract
We propose to explore an innovative, data mining (DM) process for application in breast cancer (BC) databases. The DM process is the Constraint Satisfaction Neural Network (CSNN). Contrary to feed-toward networks and statistical models, the CSNN has a non- hierarchical architecture that allows it to be used either as a prediction/classification tool or as an analysis tool for mining association rules in databases. This is a feasibility study to investigate to what degree the CSNN can deliver the above promises for the mammographic diagnosis of breast lesion malignancy. The main objectives of the study are the following: (1) to develop a CSNN for mining a database of patients suspected with BC who underwent breast biopsy; (2) to evaluate the CSNN as a diagnostic tool; (3) to evaluate the CSNN as a patient prototype analysis tool to discover prevalent trends and associations among the variables; (4) to assess the network's robustness with missing data. Initially, the CSNN is intended as a computer-assisted diagnostic tool to help physicians optimize the decision to refer a probably benign breast lesion to short-term follow-up instead of biopsy. Ultimately, the CSNN can be applied as a support tool individualizing the decision process in BC patient management.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2003
- Accession Number
- ADA414891
Entities
People
- Georgia D. Tourassi
Organizations
- Duke University Hospital