Fine Grained Classification of Mammographic Lesions using Pixel N-grams
Pradnya Vaibhav Kulkarni · 2019
Breast cancer is the most common type of cancerworldwide. Early diagnosis of breast cancer can result inbetter treatment options increasing the survival chances of apatient. Automated or computer aided detection of breastcancer is applied in order to improve the accuracy andturnover time. However, the accuracy of automated detectionsystems can still be improved. Most of the efforts in thecomputer aided detection systems classify the images intocancerous and non-cancerous categories. The aim of this paperis to classify the mammographic lesions into three categoriesnamely circumscribed, speculation and normal. The novelPixel N-gram features have been used for classification ofthese lesions. Pixel N-grams are originated from character Ngramconcept of text categorization. Classificationperformance is noted in order to analyse the effect ofincreasing N and effect of using different classifiers (MLP,SVM and KNN). It was observed that the classificationperformance increases with increase in N and then startsdecreasing again. Moreover, classification performanceachieved using MLP classifier was better than the performanceusing SVM or KNN classifiers.