Research On Grading Of HE-stained Histopathological Image Of Breast Cancer

Breast cancer is a common female cancer,and it ranks first in the incidence of female cancers.Image processing technology is also becoming more and more mature,which makes pathological diagnosis based on computer appear more and more frequently in the diagnosis of various diseases,and researchers have also done many research about that.At present,the diagnosis of breast cancer based on hematoxylin-eosin stained histopathological images is regarded as the gold standard of the diagnosis of breast cancer.The surgeon cut out a piece of tissue from the patient's lesion,then a series of operations such as hematoxylin-eosin staining is done with it,and finally tissue slice is made.Then,the pathologist places the slice under a microscope and watch it,and give a diagnostic result.In this way,the diagnostic result of patient seriously depends on the subjective judgment of pathologist,so the requirements is very highly for doctors,and the efficiency is less efficient.So for hematoxylin-eosin stained histopathological images of breast cancer,there is a need for a computer-aided automatic diagnosis method that can reduce the burden of the pathologist,what's more by means of image processing technology,the diagnosis of the patient will become objective and efficient.To achieve that,this article has do some research of hematoxylin-eosin stained histopathological images.At first,a segmentation model based on pixel classification is proposed to segment the cancerous region of hematoxylin-eosin stained breast cancer histopathological images.This model is designed according to common machine learning task.In this model,a preprocess step is done with the images firstly,the extract the color features and texture features of pixels,and then classify the pixels into cancerous or non-cancerous areas by trained classifiers.The final step is postprocess,which aims to improve the accuracy of segmentation.The experiments verify the good performance of the model.Secondly,this paper proposes a method based on two-times-watershed to identify the nucleus region in the image,which is a preparing step for later experiment,and the experimental results have a good visual effect.Finally,the result of cancerous region segmentation and nucleus recognition are combined,the pixel level feature and the object level feature of the image are extracted for classification,then classify the image,the result of the experiment are regarded as histological grades of the image.Experiments show that it is feasible to grade histopathological images.