Feature Selection Using Neighborhood Component Analysis with Support Vector Machine for Classification of Breast Mammograms

摘要

Recognition of lumps in the breast region is well exploited through mam-mography. For radiologist, the identification of cancerous tissues is tedious and time-consuming, and many automated computer-aided techniques have been proposed to enhance the clinical diagnosis. This specific research work suggests the application of neighborhood component analysis (NCA) as a feature selection technique for breast mammograms classification. Tamura features (coarseness, contrast and directionality) which provide the characteristics of image surface and objects appearance in images and statistical features (mean, variance, skewness, kurtosis, energy and entropy) were extracted from the breast mammograms, and NCA was applied to identify the best features. Finally, support vector machine classification was performed to classify abnormal condition from normal. Simulation study using the local hospital datasets revealed an overall classification accuracy of 93% by making use of quadratic kernel with S VM classifier.