Classification of mammographic lesion based in Completed Local Binary Pattern and using multiresolution representation

2nd International Conference on Mathematical Modeling in Physical Sciences 2013

物理学;数学

Duarte, Y.A.S.^1 ; Nascimento, M.Z.^2 ; Oliveira, D.L.L.^1

Center of Mathematics, Computing and Cognition, Federal University of ABC (UFABC), Santo-Andre, São Paulo, Brazil^1

Faculty of Computing (FACOM), Federal University of Uberlândia (UFU), Minas Gerais, Uberlândia, Brazil^2

关键词: Area under the ROC curve;    Digital database for screening mammographies;    Digitized mammograms;    Features extraction;    Local binary patterns;    Multi resolution representation;    Statistical techniques;    System effectiveness;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/490/1/012127/pdf
DOI  :  10.1088/1742-6596/490/1/012127

来源: IOP

【 摘 要 】

This paper presents a comparison of two methods for features extraction of mammograms based in completed local binary pattern (CLBP) and wavelet transform. In first part, CLBP was applied in digitized mammograms. In second part, we applied CLBP in the sub-bands obtained from the wavelet multi-resolution representation of the mammographies. In this study, we evaluated the CLBP in the image in the spatial domain and in the sub-bands obtained with wavelet transform. Then, the statistical technique of variance analysis (ANOVA) was used to reduce the number of features. Finally, the classifier Support Vector Machine (SVM) was applied in the samples. The proposed methods were tested on 720 mammographies which 240 was diagnosed as normal samples, 240 as benign lesion and 240 as malign lesion. The images were obtained randomly of the Digital Database for Screening Mammography (DDSM). The system effectiveness was evaluated using the area under the ROC curve (AUC). The experiments demonstrate that the textural feature extraction of the multi-resolution representation was more relevant with value of AUC=1.0. In our experiments, CLBP in the spatial domain resulted in value of AUC=0.89. The proposed method demonstrated promising results in the classification of different classes of mammographic lesions.