Platform Of Medical Imaging Research And Auxiliary Diagnosis Based On Machine Learning

The intelligent medical field formed by the combination of artificial intelligence technology and traditional medical industry has gradually matured.However,how to get a high-precision and stable model in intelligent medical field and how to make medical researchers more portable to use artificial intelligence technology for basic research are still urgent problems to be solved.In this paper,an intelligent medical image research platform is developed to solve the above problems,which is named RIAS(Radiology intelligent analysis software).At the same time,this paper also discusses the algorithms of radiomics,deep learning classification and segmentation network of RIAS.This paper focuses on the image preprocessing of radiomics,the way of plotting the region of interest,feature extraction,feature selection,feature visualization,model building and model evaluation.In the discussion of image preprocessing,two methods of image resampling are put forward.The composition of image features is analyzed.The general rules of feature engineering are summarized through the elaboration of the principle of feature selection,and the process of obtaining the optimal parameters of the model by the optimal search algorithm is discussed.In addition to image features,other features such as clinical information are analyzed.In deep learning,the development history of convolutional neural network is discussed The structure of image classification network and image segmentation network is analyzed,and the common points,advantages and disadvantages between two networks are discussed.At last,this paper expounds how to use convolutional neural network effectively in medical image,and analyzes the network theory,network application and other aspects.We take the practical research of colorectal cancer liver metastasis and liver segmentation as an example to discuss the application of radiomics,transfer learning network and segmentation network of deep learning in RIAS.In the radiomics part,we study according to the process of image preprocessing,feature extraction,feature engineering and model evaluation.Finally,by introducing multi feature fusion algorithm and comparing the results of multiple models,we get an efficient logistic regression model.Under the ROC(receiver operating characteristic)curve in the test set,the area of the model is 0.899,the sensitivity is 0.78,the specificity is 0.91,the positive predictive value is 0.88,and the negative predictive value is 0.83.When using transfer learning network of RIAS,through the application of ROI automatic positioning and clipping technology,image augmentation,image wavelet transformation and other methods,the research on different training rounds,different clipping image size and other variables,finally get the model of 0.75 classification accuracy in the test set.Finally,the liver is segmented by RIAS segmentation network of deep learning,and the segmentation accuracy is 0.95 in test set by training network At the same time,this paper further discusses the advantages and disadvantages of RIAS platform in algorithm,engineering and other aspects.This paper introduces the development version,one key operation by GUI(Graphical User Interface),modular programming and the application of docker to encapsulate the cloud environment Through the introduction of RIAS platform in machine learning and deep learning of algorithm and application,it can be found that RIAS can make researchers easily process data,images,convenient training,call high-quality models through the visual operation interface.