Automatic Recognition Of Immunohistochemistry Images

Breast cancer is one of the most common malignant tumors for women,the incidence rate increases year by year.Now,it is a serious threat for women's health.Immunohistochemistry(IHC)is a kind of widely used technology for Pathological examination which helps a lot for the confirmed diagnosis for breast cancer,later treatment plan and Prognosis.Currently,most of the pathological examination results are checked by pathologists using microscope.This method is very complicated and will be affected by human factors.Therefore,automatic identification for IHC images play significant roles for social meanings and application values.This paper focus on designing a set of automatically procedures to identify the positive targets zone in the breast cancer IHC images,which may automatically complete the preprocessing of the images,segmentation of positive target areas,notification and separation of overlapping cells.This technology will help doctors to fulfill the clinically pathological diagnosis in high efficiency.Preprocessing Part:Most of the original IHC images existed some problems as overall image is dark and unclear because that it's very complicated to make the pathological slide and color the images,and maybe because of different equipment are used to collect the images.This article chose the piece wise-linear level transformation method to do a series of comparative experiments and finally confirmed the suitable parameters to achieve the following goals:enhance the brightness and contrast,keep the cell edge complete at the same time removes the unclear factors by bilateral filter method.The positive zone of regional segmentation:since the background of the IHC images are complicated and colorful,this article brought into the color decenvolution technology to realize the color separation of all kinds of stains in IHC images and finally got the single stained grey image.During the procedure of segmentation,aiming to solve the problem which is caused by traditional Fuzzy C-means Algorithm that the traditional way will ignore the spatial information and the result is not stable,I defined the membership degree matrix which is restricted by neighboring membership constraint,brought in the spatial information into membership degree matrix and finally get a new way which is called restricted neighboring membership fuzzy c-means algorithm.At the same time,the traditional Fuzzy C-means Algorithm usually met the problem that local optimal solution which is cause by random clustering center,in this article,I improved the way to take the clustering center on the basis of histogram.It's been demonstrated that this way will strengthen the robustness and decrease the number of iterations of clusteringSeparations Part of Overlapping Cells:Overlapping cells are often present in IHC images.In this paper,I used the strategy of searching the image concave points to pairing and segregating them.On the basis of Harris Corner Detection Method,I combined the method of US AN(Univalve Segment Assimilating Nucleus)Regional Concept together.Overlapping cells are divided into three types as series connection,parallel connection and ring connection by the relationship of image concave points and cell numbers.Overlapping cells were separated and isolated by using different matching strategies then.