Design And Implementation Of Pulmonary Fibrosis Analysis System Based On Convolution Neural Network

In recent years,Convolution Neural Network has been widely studied and applied in many fields,such as image recognition,natural language processing,object detection and so on.In the field of image recognition,the combination of Convolution Neural Network and digital pathology makes the automatic analysis of medical images be possible.Applying Convolution Neural Network to the analysis tasks of pulmonary fibrosis pathological images to diagnose the degree of pulmonary fibrosis lesions can reduce the work intensity of pathologists,reduce misdiagnosis rate,and eliminate the diagnostic differences caused by subjective factors.At the same time,the computer-aided diagnosis system can fill the talent gap of pathologists and greatly improve the work efficiency of pathologists.At present,there are few researches on Convolution Neural Network in the diagnosis of pulmonary fibrosis diseases.In the medical image analysis task of other diseases,convolutional neural network model is usually used to classify or segment pathological images or CT images to distinguish the background,blood vessels,normal tissues,diseased tissues and other regions,mainly focusing on regional classification,while the related research on quantitative analysis of diseased tissue region is less.In order to solve the above problems,this paper designs and implements a pulmonary fibrosis analysis system based on convolution neural network,which uses the method of supervised learning to train the convolution neural network model,combined with Ashcroft scoring method,to quantitatively analyze the focus area in the pathological image of pulmonary fibrosis and complete the diagnosis of pulmonary fibrosis.The specific research contents are as follows:1.In order to solve the problems of data imbalance,low brightness and inconsistent staining in the data set,the pathological images and their corresponding XML tag files are segmented to expand the pictures in the data set.The HSV algorithm is used to adjust the brightness of the images and the staining standardization algorithm is used to standardize the data set.2.When analyzing the pathological images of pulmonary fibrosis,the processing flow of pathological images is optimized.First of all,two classifications are carried out,and the binary classification model based on Res Net50 network is trained to distinguish lung tissue from blood vessels,large bubbles and other non-pulmonary tissues in pathological images.According to the lung tissue regions identified by the two-classification model,multi-classification processing was carried out,combined with Ashcroft scoring method,the multi-classification model based on Res Net50＿IBN＿b was trained to classify the pulmonary fibrosis lesions in detail,and the quantitative analysis of pulmonary fibrosis lesions was realized.3.For evaluating the performance of two-classification model and multi-classification model,we add the process of predicting pathological images and visualization of prediction results.The two-classification model and multi-classification model are used to predict the pathological image,and the visual prediction results are compared with the diagnosis results of doctors,which verifies the accuracy of the model more intuitively and makes the analysis results of the model more convincing.4.Based on the above research results,this paper designs and implements a pulmonary fibrosis diagnosis system,which can automatically analyze the focus area in the pathological image of pulmonary fibrosis and complete the diagnosis of pulmonary fibrosis disease.The research results of this paper can assist doctors to complete the diagnosis of pulmonary fibrosis diseases,and have important application prospects in the diagnosis and treatment of pulmonary fibrosis diseases.