Histological Image Analysis For Quantifying Breast Tumor

Breast tumor grading is usually adopted in the diagnosis and prognosis of breast cancer:the doctor observes different pathological markers, through the microscope, in the tissue sections to score a pathological grade. However, the way of manual analysis is time consuming and has the subjectivity of the doctors, and the differences between doctors may lead to inappropriate treatment. Therefore, the study of computer aided diagnosis system is meaningful for doctors and patients: provide an accurate and quantitative supplementary analysis to speed up the diagnosis process. The Nottingham Grading System is highly correlated with the morphological and topological features of the nuclei in breast tumor, so detection and segmentation cells from histopathology images is the foundation of automatic analysis of histopathology image. However, cell detection and segmentation is a challenging task:the appearance of the cell are highly heterogeneous and the tissue structure in histopathology image is complex; in addition, the overlap between cells may lead to wrong segmentation. Aiming at this problem, this paper proposes an convolution neural network based active contour adaptive ellipse fitting method.The method uses a convolution neural network model combined with a sliding window to detect the cell, then initialization active contour model based on the results of cell detection; the problem of segmentation of overlapping cells is solved by using an adaptive ellipse fitting approach. In order to verify the performance of the proposed method, the proposed method was tested on three data sets. Experimental results show that:the detection accuracy of the proposed method on three data sets is 73.33%,83.91% and 76.88%; and the segmentation accuracy on the dataset 1 and dataset 2 is 85.03% and 90.33% respectively. This shows that the performance of proposed method is obviously better than other methods. In the study of automatic pathological grading, this paper presents a multi-feature description based automatic grading method for breast tumor. The method uses a convolution neural network model to detect the epithelial cells and lymphocytes in histopathology images. After color deconvolution, adaptive threshold, morphological operations, watershed algorithm and ellipse fitting were used to get the boundaries of the cells. Then the shape features and texture features of the cells and the spatial structure features of the cell distribution were extracted. After reducing the dimension of features, a support vector machine classifier was used to discriminate histopathology images with three grades. Experimental results show that:the proposed method could distinguish histopathology images of low, intermediate and high grades with classification accuracy of 90.20%. Moreover, the proposed algorithm was able to distinguish high, intermediate and low grade with accuracy of 92.87%,2.88% and 93.61%, respectively. This shows that performance of the proposed method is far better than the other methods.