| Human beings have been fighting against cancer for thousands of years.In recent years,the emergence of immunotherapy has provided a powerful weapon for human beings to overcome cancer.Among them,PD-1/PD-L1 immunotherapy is the most active one,but it is only effective for specific populations.PD-L1 detection can screen out the target population by analyzing the IHC images,but the artificial analysis method at the present stage is slow and has low accuracy.In this study,the traditional image segmentation algorithm and the depth learning segmentation model are used to extract two target regions respectively,and a set of automatic process is proposed to complete the automatic interpretation of the image,which optimizes the analysis process of the detection method and will promote the further development of cancer immunotherapy.The main research goal of this subject is to extract invasive cancer and PD-L1 positive cell area from PD-L1 staining image and calculate the proportion of the two.Centering on this goal,we discuss the segmentation method of glands and positive staining in PD-L1 staining image and the quantitative analysis of the degree of positive staining.The automatic PD-L1 immunohistochemical image analysis system developed in this research can effectively assist the pathologist to complete the clinical diagnosis.The research object of this subject is the micro-image of breast cancer tissue.The main work of this paper is divided into the following aspects:(1)A method for segmenting positive products in breast cancer IHC images.Distinguishing between normal glands and invasive carcinoma in gland segmentation results requires the use of myoepithelial cell segmentation results in P63 images to obtain PD-L1 positive cell segmentation results after obtaining the location of invasive carcinoma.In this study,we proposed the corresponding segmentation schemes for the immunohistochemical images with two different staining patterns.After appropriate pretreatment and enhancement,we used the segmentation algorithm based on region growth for the P63 images with relatively concentrated positive staining and fixed morphology.For the SP142 images with disordered staining information,the optimal fuzzy c-means algorithm which is mainly based on the staining information and supplemented by the spatial information is used.And finally obtaining segmentation results of two positive staining targets.(2)A method of gland segmentation in breast cancer IHC images.Breast cancer exists as glands in its IHC images.In this study,we adopted the scheme of segmenting the glands first and then using P63 images to assist in determining whether it belongs to invasive cancer.In this study,convolutional neural network is used for gland segmentation.the performance of U-Net,PSPNet,Seg Net and Deeplab v3+ existing segmentation models in the immunohistochemical image of this study is compared and analyzed.the effects of different cutting sizes on the segmentation are compared,and the appropriate pretreatment and staining standardization method is proposed.the UNet optimization model is designed and validated,and the glands in the image are finally segmented.(3)A PD-L1 image system is developed,which can assist in the acquisition of invasive cancer,segmentation and quantitative analysis of positive staining targets,and gives the reference diagnosis results.The system can analyze the staining degree while segmenting the target region,and can manually adjust the staining intensity threshold to segment the positive staining target more flexibly,which not only utilizes image processing technology to assist clinical diagnosis,but also lays a foundation for the application of image technology in this field.The successful development of automatic immunohistochemical image analysis system in this subject will realize the automatic interpretation of PD-L1 stained images,and promote the continuous development of cancer immunotherapy-related detection technology in China for the benefit of more cancer patients. |
PDF Full Text Request