Diagnostic assistant system for diabetic retinal complications based on convolution neural network

Abstract

The invention discloses a diagnostic assistant system for diabetic retinal complications based on convolution neural network, which includes: a training set and a test set preparation module, which are used to prepare the training set and the test set for type 2 diabetes mellitus complicated with retinopathy and non-type 2 diabetes mellitus complicated with retinopathy; a convolution neural network construction module is used to construct convolution neural network; a convolution neural network optimization module is used to optimize the constructed convolution neural network, a convolution neural network training module is to train the convolution neural network by using the training set, a classified output module, wherein the test set is input as a well-trained neural network, an outputvalue is type 2 diabetes mellitus complicated with retinopathy and non-type 2 diabetes mellitus complicated with retinopathy. The system provides the basis for the early diagnosis of diabetes mellitus complicated with retinopathy and optimizing the diagnosis process, in-depth study is combined with the electronic medical record information and a good result is achieved.

Description

technical field [0001] The invention relates to a diagnosis auxiliary system for diabetic retinal complications based on a convolutional neural network. Background technique [0002] Traditional Chinese medicine diagnoses diseases through "looking, smelling, asking, and knowing". Even with various accurate indicators, the traditional manual method still has many defects: 1) Doctors have to systematically judge and screen dozens or hundreds of indicators for each patient, which consumes a lot of money. Manpower, and there is a certain chance of misjudgment; 2) When there are more than one disease (ie, complications), complex pathogenic factors will make manual diagnosis less intuitive; 3) Individual differences will lead to incomplete disease and treatment effects The same, while individual differences are sometimes very subtle and will be ignored by manual diagnosis; [0003] In recent years, with the rise of cognitive computing technologies such as machine learning, big da...

AI Technical Summary

Problems solved by technology

[0002] Traditional Chinese medicine diagnoses diseases through "looking, smelling, asking, and knowing". Even with various accurate indicators, the traditional manual method still has many defects: 1) Doctors have to systematically judge and screen dozens or hundreds of indicators for each patient, which consumes a lot of money. Manpower, and there is a certain chance of misjudgment; 2) When there are more than one disease (ie, complications), complex pathogenic factors will make manual diagnosis less intuitive; 3) Individual differences will lead to incomplete disease and treatment effects The same, while individual differences are sometimes very subtle and will be ignored by manual diagnosis;

Classical models include LR, SVM, RF, GBDT, etc., which have the advantages of fast training, the ability to output feature importance, and good interpretability, but the disadvantages are that the model is relatively simple and the feature learning ability is general

In-depth models include DNN, CNN, RNN, etc., which have the advantages of high prediction accuracy, strong feature learning ability, and the characteristics of universal approximation, but the disadvantages are complex models, slow training speed, and high requirements for computing resources.

[0004] The present invention selects "diabetic retinal complications" as the entry point, which is one of the most common clinical microvascular complications of diabetic patients, but if not treated in time, it will lead to blindness of diabetic patients

Images