Research On Key Technologies In Nuclear Detection And Segmentation For Multi-Disease Pathological Images

As a discipline of basic medical,the study of pathological starts from the level of human organs,tissues or cells,and goes into the analysis on the pathogenesis,development process and changes of bodies.Pathological image analysis is one of the important branches,and using deep learning technology aided diagnosis has become a current trends.Therefore,focused on cell level analysis of pathological images,this paper conducts an intensive research on cell detection and segmentation technology and provides a solution for the general analysis of different diseases.The research content of this paper mainly includes the following three aspects:For image feature extraction,in order to adapt to the diversity of cell morphology and scale,a triplet-path multi-scale feature extraction and fusion algorithm is proposed to improve the representation ability of image features under multi-task.For cell detection,an improved location-aware truncated loss function is introduced into the classification training,through the analysis of various factors such as model structure design and pathological data characteristics,which greatly improved the detection accuracy under the condition of limited data amount.For cell segmentation,considering the different effects of common segmentation functions on foreground and background,a new adaptive background-enhanced strategy is adopted for segmentation training,which effectively improves the precision of cell segmentation.These three kinds of technologies are integrated into the same framework to form a general end-to-end model effectively.Through comparison experiments on various datasets,the effectiveness and generalization of this model have been fully verified.The task of cell detection and segmentation under multiple diseases has been successfully realized,which provides a simple and efficient solution for the intelligent analysis of pathological images...