Design And Implementation Of Computer-aided Diagnosis System For Cell And Histopathology Based On Deep Learning

As one of the most difficult diseases for human beings,cancer is a serious threat to human life and health.Pathological examination can determine the stage and classification of cancer,and plays an important role in clinical diagnosis and treatment decision-making.Therefore,pathologists are also called "doctor’s doctor".However,the number of pathologists in China is seriously insufficient and unevenly distributed,and pathological diagnosis resources at the grassroots and remote areas are scarce.With the rapid development of information technology and artificial intelligence in recent years,artificial intelligence has achieved initial applications in the field of computer-aided diagnosis.In this context,in order to help pathologists perform pathological diagnosis efficiently and relieve the shortage of resources of pathologists,this article takes cervical exfoliated cell pathology and liver cancer histopathology as the research objects,and studies computer-aided diagnosis algorithms based on deep learning.It is applied to the remote pathological auxiliary diagnosis system,thereby improving the efficiency of doctors in diagnosing pathology and reducing the rate of misjudgment.The research content of this article is mainly divided into the following aspects:1)This thesis first conducts related research on the classification of cervical cells in cytopathology.An improved Res Net-M-T model based on Res Net is proposed,which improves the classification effect by optimizing residual blocks and connecting feature maps.On this basis,a cascade classification method is proposed.The classification task is divided into two stages.The first stage distinguishes negative cells and positive cells,and the second stage accurately distinguishes the types of positive cells.The experimental results show that the classification performance of cascade classification is better than single multi-classification.2)This thesis also conducts related research on the segmentation of lesions in liver cancer histopathology.Based on the Efficient Net segmentation network,an improved model,Dual-Efficient Net,is proposed according to the characteristics of different resolutions of digital pathology pictures,which introduces richer global and local information into the network.Perform image normalization according to the characteristics of liver pathological pictures,and design a new loss function for the uneven number of samples.This thesis uses the PAIP2019 public data set for experimental evaluation.The results show that the segmentation algorithm proposed in this thesis has relatively good results,and the Jaccard coefficient can reach 0.746.3)Finally,based on the algorithm of the above research,this paper proposes and implements a pathology-assisted diagnosis system solution according to the needs,including the realization of image input and preprocessing module,image training module,intelligent analysis module,and diagnosis report result display module.The system can upload and analyze pathological pictures,intelligently analyze and diagnose,and realize auxiliary diagnosis of pathology.After testing,the system can provide doctors with preliminary pathological diagnosis,which has practical application value.