Signal Detection Theory-Based Information Processing for the Detection of Breast Cancer at Microwave Frequencies
Abstract
This research addressed directly the decision-theoretic task of detection and localization of breast tumor, using microwave diffraction measurements. Microwave energy has the advantages that at low power levels there are no radiation dangers, no contrast agents, and the examinations are comfortable. Although there is considerable scattering of a microwave signal in tissue, the presence, location, and nature of tumors is "coded" in the combination of amplitude and phases in the signals received at multiple sensors. Bayesian detection theory was used in this research to improve the probability of correct detection and localization. This improvement in performance is possible because conventional imaging techniques, by themselves, usually emphasize resolution and contrast, and leave the incorporation of uncertainties and decisions primarily to algorithms or human observers that post process the reconstructed image. This approach augments conventional medical image processing and provides additional processing of he scattered microwave field to aid the radiologist in dealing with uncertainties that are an inherent part of the decisions. Using the receiver operating characteristic (ROC) and other performance measures, and simulation, bounds on the performance attainable for various uncertainties in malignant tissue properties (permittivity), sizes, location, and signal-to-noise ratios were obtained.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 2003
- Accession Number
- ADA419213
Entities
People
- Loren W. Nolte
Organizations
- Duke University