The Diagnosis Method Of Hirschsprung’s Disease Based On Deep Learning And Its Software Deployment

Hirschsprung’s disease is one of the most common developmental malformations of the digestive tract in children,and the children are mainly manifested as intractable constipation or digestive tract obstruction.The Ach E stained image of transanorectal mucosal biopsy can be used for the auxiliary diagnosis of Hirschsprung’s disease,but the manual analysis of Ach E stained image is a lot of work.This paper takes Ach E stained image data set as the research object,based on deep learning and transfer learning to study the segmentation and classification algorithm of Ach E stained image,designs and develops an intelligent diagnosis system,and realizes the data collection and computer assistance of Hirschsprung’s disease diagnosis.The specific work of this paper is as follows:(1)Proposed the image segmentation model Res-DSC-UNet based on U-Net combined with residual structure and depth separable convolution,and proposed the image segmentation model Efficient based on U-Net and using the pre-trained model as the encoder skeleton-UNet and Mobile-UNet,train these four segmentation models on the HD-Ach E data set to obtain the accuracy evaluation indicators such as PA,Io U,Dice Coefficient of the segmentation model,and efficiency evaluations such as training time,calculation efficiency,model volume,etc.index.The experimental results show that although the accuracy index of Efficient-UNet is the best,the improvement is not obvious compared with other models.Mobile-UNet has the best efficiency evaluation index,and its accuracy is similar to other models.Considering the accuracy and efficiency,Mobile-UNet is selected as the segmentation model for the intelligent diagnosis system.(2)An image classification model HD-Ach E-Net based on deep learning is proposed.According to whether the HD-Ach E data set is segmented and preprocessed and the form of segmentation preprocessing,three data sets of HD-Ach E-Raw,HD-Ach E-Auto and HD-Ach EManual are obtained.Train HD-Ach E-Net on these three data sets,and perform migration learning and model fine-tuning on the pre-training model,and compare the classification accuracy and related efficiency evaluation indicators of the classification model obtained by HD-Ach E-Net and migration learning.Experiments show that pre-processing the data set segmentation in advance can effectively improve the classification accuracy.The classification accuracy of the classification model obtained by transfer learning is better than HD-Ach E-Net,indicating that transfer learning has more advantages in small-scale data sets.Since the model obtained by migration learning based on Mobile Net V2 has more advantages in classification accuracy,calculation efficiency,model volume and other indicators,this model is selected as the classification model selected by the intelligent diagnosis system.(3)Designed and implemented an intelligent diagnosis system,which consists of three parts: database,server and client.The database is responsible for the storage of HD-Ach E data sets and related model files;the server provides a network interface for interacting with the database,regularly retrains and releases the model,and provides a background management system for visual management of the database;the client provides model version updates,Intelligent segmentation diagnosis,manual segmentation diagnosis and data collection and uploading functions.