INSTITUTE OF PHYSICS PUBLISHING PHYSICS IN MEDICINE AND BIOLOGY Phys. Med. Biol. 46 (2001) 1651–1663 www.iop.org/Journals/pb PII: S0031-9155(01)21237-8 An SVM classifier to separate false signals from microcalcifications in digital mammograms Armando Bazzani 1 , Alessandro Bevilacqua 2 , Dante Bollini 1 , Rosa Brancaccio 1 , Renato Campanini 1 , Nico Lanconelli 1, 3 , Alessandro Riccardi 1 and Davide Romani 1 1 Department of Physics, University of Bologna, and INFN, Bologna, Italy 2 Department of Electronics, Computer Science and Systems, University of Bologna, and INFN, Bologna, Italy E-mail: nico.lanconelli@bo.infn.it Received 22 January 2001, in final form 22 March 2001 Abstract In this paper we investigate the feasibility of using an SVM (support vector machine) classifier in our automatic system for the detection of clustered microcalcifications in digital mammograms. SVM is a technique for pattern recognition which relies on the statistical learning theory. It minimizes a function of two terms: the number of misclassified vectors of the training set and a term regarding the generalization classifier capability. We compare the SVM classifier with an MLP (multi-layer perceptron) in the false-positive reduction phase of our detection scheme: a detected signal is considered either microcalcification or false signal, according to the value of a set of its features. The SVM classifier gets slightly better results than the MLP one (Az value of 0.963 against 0.958) in the presence of a high number of training data; the improvement becomes much more evident (Az value of 0.952 against 0.918) in training sets of reduced size. Finally, the setting of the SVM classifier is much easier than the MLP one. 1. Introduction Breast cancer is the most common form of cancer among women. The presence of microcalcifications in breast tissues is one of the main features considered by radiologists for its diagnosis. CAD (computer aided diagnosis) systems have been examined in order to assist doctors: the computer output is presented to radiologists as a second opinion and can improve the accuracy of the detection. Several techniques developed for the automated detection of microcalcifications can mainly be grouped into three different categories: multiresolution analyses (Yoshida et al 1994, Lado et al 1999), difference-image techniques (Chan et al 1987) and statistical methods (Karssemeijer 1993, Gurcan et al 1998, Poissonier et al 1998). 3 Address for correspondence: Department of Physics, Viale Berti-Pichat 6/2, 40127 Bologna, Italy. 0031-9155/01/061651+13$30.00 © 2001 IOP Publishing Ltd Printed in the UK 1651