Authors: Dayakshini Sathish; Surekha Kamath; Keerthana Prasad; Rajagopal Kadavigere
Addresses: ICE Department, Manipal Institute of Technology, Manipal University, Manipal 576104, Karnataka, India ' ICE Department, Manipal Institute of Technology, Manipal University, Manipal 576104, Karnataka, India ' School of Information Sciences Manipal, Manipal University, Manipal 576104, Karnataka, India ' Kasturba Medical College Manipal, Manipal University, Manipal 576104, Karnataka, India
Abstract: Breast cancer is the leading cancer in women worldwide. Early detection can reduce the mortality rate of breast cancer. Breast thermography is a non-invasive and simple imaging technique used for early detection of breast cancer. Feature extraction and selection of appropriate features play a major role in computer-aided detection of breast cancer using breast thermograms. In this article, texture features are extracted from automatically segmented breast thermograms by computing neighbourhood grey tone difference matrix (NGTDM) and run length matrix (RLM). Significance of these features in differentiating the abnormal breast from the normal breast is found by statistical test. NGTDM extracted coarseness, busyness, complexity, strength and RLM extracted long run emphasis and run percentage are found to be significant by statistical test. Extracted features are computationally less expensive and attained an average accuracy of 80%, sensitivity of 94% and specificity of 71.4% using back propagation neural network classifier.
Keywords: asymmetry analysis; breast cancer; breast thermography; neighbourhood grey tone difference matrix; statistical test.
International Journal of Bioinformatics Research and Applications, 2018 Vol.14 No.1/2, pp.104 - 118
Received: 20 Dec 2016
Accepted: 02 Sep 2017
Published online: 09 Jan 2018 *