Classifying breast cancer regions in microscopic image using texture features

This study proposes and evaluates the application of two classifiers: decision tree (DT) and neural network (NN) to discriminate three region types: cancer (CC), lymphocyte (LC), and stromal (SC) in the breast cancer cell images. The feature extraction from area based texture information of BCCI is studied to compare results from the segmented cells. A combination between texture features based on energy information and fractal dimension (FD) is used as the feature extraction. One-hundred forty-one images of CC, 23 images of SC and 21 images of LC are analyzed. Results of the p values obtained from analysis-of-variance indicate that the difference between mean of three regions from FD feature and entropy feature are statistically less significant than other features. The classification accuracies obtained from NN and DT is 94.05% and 95.14%, respectively. The results show the feasibility of the proposed method in classifying the histological structures in breast cancer cell image and can be used for improving the cell counting accuracy in the computer-aided system.