Classification of Histological Images Based on the Stationary Wavelet Transform

3rd International Conference on Mathematical Modeling in Physical Sciences

物理学;数学

Nascimento, M.Z.^1 ; Neves, L.^2 ; Duarte, S.C.^3 ; Duarte, Y.A.S.^3 ; Batista, V Ramos^3

Faculty of Computer Science (FACOM), Federal University of Uberlândia (UFU), MG, Uberlândia, Brazil^1

Institute of Biosciences, Literature and Exact Sciences, Department of Computer Science and Statistics, São Paulo State University (UNESP), São José do Rio Preto, SP, Brazil^2

Mathematics, Computer Science and Cognition Centre, Federal University of ABC (UFABC), Santo André, SP, Brazil^3

关键词: Automatic method;    Classification system;    Histological images;    Lymphoma cells;    Non-Hodgkin lymphoma;    RBF kernels;    Stationary wavelet transforms;    Support vector machine algorithm;   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/574/1/012133/pdf
DOI  :  10.1088/1742-6596/574/1/012133

来源: IOP

【 摘 要 】

Non-Hodgkin lymphomas are of many distinct types, and different classification systems make it difficult to diagnose them correctly. Many of these systems classify lymphomas only based on what they look like under a microscope. In 2008 the World Health Organisation (WHO) introduced the most recent system, which also considers the chromosome features of the lymphoma cells and the presence of certain proteins on their surface. The WHO system is the one that we apply in this work. Herewith we present an automatic method to classify histological images of three types of non-Hodgkin lymphoma. Our method is based on the Stationary Wavelet Transform (SWT), and it consists of three steps: 1) extracting sub-bands from the histological image through SWT, 2) applying Analysis of Variance (ANOVA) to clean noise and select the most relevant information, 3) classifying it by the Support Vector Machine (SVM) algorithm. The kernel types Linear, RBF and Polynomial were evaluated with our method applied to 210 images of lymphoma from the National Institute on Aging. We concluded that the following combination led to the most relevant results: detail sub-band, ANOVA and SVM with Linear and RBF kernels.