Preliminary results of breast cancer cell classifying based on gray-level co-occurrence matrix

Abstract

This study proposes and appraise a gray level cooccurrence matrix (GLCM) for extracting the feature of cell regions in microscopic image into four region types: positive cancer cell, negative cancer cell, lymphocyte and stromal cell. The classification task uses decision tree with cross validation. To give a high classification performance, the main focus of interest is feature extraction task. Twenty-two texture features of GLCM have used to analysis images at four directions and six scales of gray-level quantization. A set of these texture features is used in 2045 images for training and testing. The result shows that the classification accuracy obtained from decision tree is 95.21%. It is demonstrated that the proposed GLCM texture features and decision tree can classify the histological structures in microscopic image and can be applied to improve and to develop an accurate cell counting of computer-aided diagnosis system for breast cancer prognosis. Index Terms-Gray-level co-occurrence matrix, texture features, breast cancer, estrogen, immunohistochemistry, microscopic image

Automatic classification of histological images and histological knowledge modelling of the human cardivascular system = Clasificación automática de imágenes histológicas y modelado del conocimiento histológico del sistema cardiovascular humano

“…There are many works dedicated to object recognition for Content Based Information Retrieval (CBIR) on histological images. Some of them using techniques based on: (i) texture, many works have compared several feature types on healthy and pathological histological images demonstrating a clear improvement of the algorithms performance and classification process (Markkongkeaw et al, 2013;Peyret et al, 2015;Lai et al, 2011); (ii) shape-based features may capture the particular characteristics to recognise cells or some pathologies -tumours or cancer -, some results suggest that these shapes are not accurate diagnostic features but it can be used for specific aims (Kothari et al, 2013;Melnyk, 2015;Sharma et al, 2012); (iii) colour representation may be used for diagnosis diseases and as an sign to recognise cells or tissues. However, the colour of histological images may vary according to the stain, the preparation procedure 2.…”