Microwave imaging has the potential to be a highly sensitive modality for breast cancer detection. The rationale for using microwaves to detect breast tumors is the dielectric-properties contrast that exists between malignant and normal breast tissue at microwave frequencies. This malignant-to-normal dielectric contrast, reported to be 2:1 or higher, implies that even small malignant tumors exhibit relatively large radar cross sections that can be exploited for non-invasive detection, localization and characterization of tumors in the breast.; We investigate three ultrawideband (UWB) radar techniques for detecting and localizing tumors in numerical breast phantoms. The detectors operate on array backscatter data in the 1-11 GHz band and are based on a penalized least-squares beamformer design, a generalized likelihood ratio test (GLRT), and a penalized expectation-maximization (EM) algorithm. We demonstrate the capability of detecting small ( 1 cm) tumors through several examples and investigate robustness to mismatch between the actual and assumed scattering scenarios.; One of the foremost challenges faced by radar-based detection techniques is the clutter that arises due to scattering within normal heterogeneous breast tissue. We address this issue by first conducting a statistical characterization of clutter computed from 3-D MRI-derived breast phantoms. Then we mitigate the clutter by implementing a whitening transformation based on a parametric model for the clutter covariance matrix. Incorporating this whitening transformation into our detector designs, we show a significant reduction in falsely-detected clutter.; Finally, we consider the problem of classifying salient tumor characteristics from the radar data. The shape and size of a tumor are compelling classification features that may be indicative of the nature of a tumor. Using a basis expansion approach and a simple linear classifier design, we demonstrate shape and size classification on backscatter obtained from a variety of 3-D numerical tumor models. This feasibility study indicates that both the tumor size and shape have the potential to be reliably classified directly from backscatter data. In this manner, tumor classification may be seamlessly integrated with UWB radar detection without requiring any special hardware or additional data collection.
展开▼
