Research On The Application Value Of Deep Learning Technology In Establishing Medical Image Standardization Platform

Background: During the course of the patient’s medical treatment,the phenomenon of repeated medical imaging was common,and most of the examinations are unnecessary repeated examinations.This phenomenon not only aggravated the economic burden on patients,but also affected the health of patients and the doctor-patient relationship.The reasons for the unnecessary repeated medical imaging are many,including the unreasonable allocation of medical resources and the incomplete relevant rules and regulations.At the same time,with more and more deep learning technologies have been applied in the medical field,such as convolutional neural networks which have been recognized as routine image recognition algorithms in the medical field.And the study of unsupervised learning algorithms using in medical field continues.With regard to the problem of unnecessary repeated medical imaging,combined with the popular deep learning technology,this study attempts to provide an medical image standardization algorithm research for constructing a medical imaging standardization platform.Object: This study used CT images as the research object,which was mainly obtained by searching CT images shared by mainstream public online databases(UCI,TCIA,etc.).The size of the sample size is obtained by the sample size calculation formula.Finally,1,000 CT images were screened for inclusion in the study.At the same time,the CT images included in the study were divided into training sets and test sets according to the ratio of 7:3.Method: In this study,a convolutional neural network algorithm with supervised learning is used in the CT image classification stage,and the image standardization stage is based on the improved algorithm based on the generative network architecture in unsupervised learning.In image classification stage,we compared three popular convolutional neural networks: CifarNet,AlexNet and GoogLeNet and made an algorithm selection.In image standardization stage,we compared three popular image transformation and improve algorithm: CycleGAN,WGAN and WGAN-GP.Result: Afte the comparision of the three convolutional neural networks and combined the existing hardware conditions and highest image classification performence,we found that GoogLeNet was the best classifaction algorithm.So,finally,in the CT image classification stage we used GoogLeNet-CNN algorithm for training.Among the 700 images in the group,231 were low “noise”-high-resolution CT images,and 469 high “noise” contents-low-resolution CT images.By comparing three kinds of image transformation and quality improvement algorithms,we found that WGAN was the most suitable unsupervised learning algotithm for image standardization under existing hardware conditions.So,in the standardization stage of CT images,WGAN algorithm was used for training.After training,the sharpness of CT images was significantly improved.In addition to using subjective judgment,based on quantitative analysis,the overall peak signal-to-noise ratio(PSNR)of CT images after training was from the initial 19.790.Raised to 23.017,structural similarity(SSIM)is 0.775.Conclution: This study demonstrated the ability of GoogLeNet-CNN to classify medical imaging data.It also proved that it is the best convolutional neural network algorithm currently developed,with the highest efficiency and highest recognition resolution.Moreover,the application of supervised deep learning in the medical field remains to be explored.In addition to image recognition,the future can also be used as an aid to other medical activities.In addition,the "CT image standardization algorithm" attempted in this study is to provide the algorithm and theoretical basis for the Medical Imaging Standardization Platform to be set up later.Establishing the Medical Image Standardization Platform can not only improve the construction of China’s regional health information exchange platform,integrate and share information on regional medical imaging,but also integrate the sharing of medical imaging diagnostic expert resources,and improve telemedicine imaging services.