Mammographic Mass Classification Of Benign And Malignant Based On BoW And Sparse Representation

Breast cancer is one of the most common malignant tumor diseases among women,which greatly threatens women’s health. A patient will not miss the best time fortreatment only with early detection and early diagnosis. As a common symptom ofbreast cancer, mammographic mass is one of the most important evidence for doctor’sdiagnosis. The differences on texture, shape and margin between benign and malignantmasses also make it a very important evidence for distinguishing benign with malignant.Therefore, it is undoubtedly significant to develop an artificial intelligence system forautomatically classifying benign and malignant masses.This paper is based on the study of Bag of Words and its application in imageprocessing. With deeply study on latent semantic topic model, LDA and sparserepresentation, this paper proposes two novel feature extraction methods, SPM-LDAand SPM-SC. SPM-LDA combines spatial pyramid matching with LDA while SPM-SCcombines spatial pyramid matching with sparse coding.For classification, support vector machine (SVM) is employed to learn the modelsof mass classes. The experimental results show that, SPM-LDA and SPM-SC all havegood performance on the classification of benign and malignant masses. To solve theproblem of the high feature dimension, this paper proposes a feature dimensionreduction method based on ReliefF algorithm. Experimental results show that ReliefFsuccessfully promotes the classification accuracy and meanwhile reduces the timeconsuming. The experimental results illustrate the effectiveness of the proposedalgorithms.