Design And Implementation Of General System For Intelligent Visual Researches Based On Digital Pathological Images

Digital pathology images are the achievement of digital pathology technology.In recent years,with the continuous advancement of digital pathology technology,the quality level and amount of information of digital pathology images have increased significantly,and the potential value needs to be explored urgently.Meanwhile,deep learning technology has made a great splash in the field of computer vision.The digital pathology images,which contain rich information,fits well with deep learning,which is good at mining high-dimensional information.Intelligent visual researches based on digital pathological images have attracted increasing attention from researchers.The study of this type of researches has an important contribution to the development of computer vision,pathology,and auxiliary diagnosis,and also has far-reaching implications for the progress of the medical system.However,pathologists,who are the main researchers in this category,are constrained by insufficient programming skills,difficulties in applying hardware devices,and lack of tools to use data and medical knowledge.This problem is gradually becoming one of the main reasons that hinder the research process in this type of researches.In order to lower the threshold for pathologists to engage in such researches,this thesis designs and implements a general system for intelligent visual researches based on digital pathological images.The work of this thesis is narrated from 3 parts: demand analysis,system design and implementation,system deployment,applications and testing.First,based on the development status and characteristics of digital pathology images and deep learning,the functional and non-functional requirements of the system were analyzed based on the working experience of pathologists and deep learning researchers.Then,based on the conclusion of the demand analysis,the system was determined to use B/S architecture.Complete deep learning experiment function module including image annotation,dataset management,task management,model application was provided,and,to achieve the requirements of generality of intelligent vision tasks,FPN network applied to image classification task,Deep Lab-V3 network applied to image segmentation and Inception-V3 network applied to target detection were integrated.Next,Vue.js as the front-end framework,Django as the back-end framework,Postgre SQL as the core database,Redis as the cache,Rabbit MQ as the message queue,and Vuetify as the front-end UI library were used to develop the system.Finally,the system was deployed using Docker container technology,the system’s capability were introduced in detail through a specific case,and the system were fully tested to ensure the stable operation of the whole system.In summary,the system provides an easy-to-use and complete tool for applying digital pathology images and conducting deep learning experiments,which lowers the threshold of deep learning researches based on digital pathology images and promote the research process.This system had excellent performance in practical work and received good feedback from users.