Sparse Non-negative Matrix Factorization (SNMF) based color unmixing for breast histopathological image analysis

作者：Xu, Jun^*; Xiang, Lei; Wang, Guanhao; Ganesan, Shridar; Feldmand, Michael; Shih, Natalie N. C.; Gilmore, Hannah; Madabhushi, Anant

来源：Computerized Medical Imaging and Graphics, 2015, 46: 20-29.

DOI：10.1016/j.compmedimag.2015.04.002

• Color unmixing
• Breast histology
• Immunohistochemistry image
• Hematoxylin and Eosin image

Color deconvolution has emerged as a popular method for color unmixing as a pre-processing step for image analysis of digital pathology images. One deficiency of this approach is that the stain matrix is pre-defined which requires specific knowledge of the data. This paper presents an unsupervised Sparse Non-negative Matrix Factorization (SNMF) based approach for color unmixing. We evaluate this approach for color unmixing of breast pathology images. Compared to Non-negative Matrix Factorization (NMF), the sparseness constraint imposed on coefficient matrix aims to use more meaningful representation of color components for separating stained colors. In this work SNMF is leveraged for decomposing pure stained color in both Immunohistochemistry (IHC) and Hematoxylin and Eosin (H&E) images. SNMF is compared with Principle Component Analysis (PCA), Independent Component Analysis (ICA), Color Deconvolution (CD), and Non-negative Matrix Factorization (NMF) based approaches. SNMF demonstrated improved performance in decomposing brown diaminobenzidine (DAB) component from 36 IHC images as well as accurately segmenting about 1400 nuclei and 500 lymphocytes from H & E images.

• 出版日期2015-12
• 单位rutgers; 南京信息工程大学