Automatic Classification of Malignant and Benign Microcalcifications on Mammograms: Texture Analysis using K-nearsetneighbours Classifier
N. Hamdi, Khalid Auhmani, Moha M’Rabet Hassani · International Journal of Mathematics & Statistics/International journal of mathematics and statistics · 2009
The goal of this paper is to present a computerized method for the discrimination between benign and malignant microcalcifications. We investigated the feasibility of using texture features extracted from mammograms to predict whether the presence of microcalcifications is associated with malignant or benign pathology. In our scheme, microcalcifications are first detected in regions of interest (ROIs) by using morphological filter and wavelet transform. Twenty two mammographic images with microcalcifications in the Mammographic Image Analysis Society database (MIAS) were used as case samples (9 benign and 13 malignant ROIs). Spatial grey level dependence (SGLD) matrices at one pixel distance in different directions were constructed from the background-corrected ROI. Twelve texture measures were extracted from each SGLD matrix. Then a KNN classifier with Euclidean distance is considered for the task of classifying the regions as malignant or benign MCs.The performance of our method on a set of abnormal images is then presented. As a result, the average accuracy was approximately 95.5% (sensitivity: 92.3%, specificity: 100%).