Classification of Breast Lesions using the Difference of Statistical Features

Abstract

This paperclassifies thebreast lesions using the difference of statistical features obtained from outside area of interest (OAOI) and inside area of interest (IAOI).The breast lesions are differentiated into two classes benign and malignant. Texture featuresare computed using statistical texture feature modelsincluding SFM, NGTDM, FOS, GLCM, GLRLM, and GLDS.The SVM classifier has been used to classify the lesions on the basis of the features extracted from (a) OAOI, (b) IAOI and (c) the difference of statistical features computed from OAOI and the corresponding IAOI. The texture features computed using SFM texture feature model from OAOIsyields the maximum accuracy of 75% withindividual class accuracy values of 79.2% for benign and 72.2 % for malignant where the same texture features when computed from IAOIs yield the maximum classification accuracy of 65% with individual classification accuracyvalues of 45.8% and 77.7% for benign and malignant lesions respectively. However, the overall accuracy of 85 % is achieved using the difference of the GLRLM texture featuresbetween the OAOIs and IAOIs with individual classification accuracy values of 70.4 % for benign and 94.4 % for malignant lesions. Thus it can be concluded that the difference of GLRLM texture features computed from OAOIs and the corresponding IAOIs contain significant information for differential diagnosis between benign and malignant focal breast lesions.