Support vector machines for the classification of early-stage breast cancer based on radar target signatures

Abstract

Microwave Imaging (MI) has been widely investigated as a method to detect early stage breast cancer based on the dielectric contrast between normal and cancerous breast tissue at microwave frequencies. Furthermore, classification methods have been developed to di®erentiate between malignant and benign tumours. To successfully classify tumours using Ultra Wideband (UWB) radar, other features have to be examined other than simply the dielectric contrast between benign and malignant tumours, as contrast alone has been shown to be insuficient. In this context, previous studies have investigated the use of the Radar Target Signature (RTS) of tumours to give valuable information about the size, shape and surface texture. In this study, a novel classification method is examined, using Principal Component Analysis (PCA) to extract the most important tumour features from the RTS. Support Vector Machines (SVM) are then applied to the principal components as a method of classifying these tumours. Finally, several di®erent classification architectures are compared. In this study, the performance of classifiers is tested using a database of 352 tumour models, comprising four di®erent sizes and shapes, using the cross validation method.