| Diabetes is one of the serious threats to the health of the world.The latest data released by the 9th edition of the IDF Global Diabetes Overview shows that 463 million adults currently suffer from diabetes.For diabetic patients,other complications that may be caused after illness are the main pressure on medical expenses,and they are also the biggest factor in the death of patients.If it is possible to predict the complications of diabetic patients in advance,it will provide doctors and patients with great convenience and will greatly reduce medical expenses.However,the current medical testing of diabetes complications is still at the stage of special medical testing after patients show symptoms.With the development of the artificial intelligence industry,how to use machine learning to predict the complications of diabetic patients has gradually attracted attention.On the other hand,because of the privacy of medical data,how to protect patients’ medical data from being leaked is also a very important issue.This paper uses an in-depth method,based on the case data of diabetic patients,to study how to accurately predict the probability of a patient suffering from several selected complications.At the same time,while obtaining high accuracy,the method of federated learning is used to protect the privacy of patients,and the problem of centralized computing power distribution can also be solved through federated learning.Specifically,three deep learning method models of recurrent neural network(RNN)long short-term memory(LSTM)and RNN recurrent unit(GRU)were constructed at the same time,and the same data was used for training and comparison to obtain the most accurate model parameters.After that,the federated learning FATE framework is applied to build a federated learning platform,and the prediction of the model is placed on the mobile device to protect the privacy of users.Through a large number of tests on the trained model found that the RNN GRU model works best,followed by the RNN LSTM model.The prediction accuracy of the RNN GRU model is between 73%(myocardial infarction)and 83%(chronic ischemic heart disease).Federal learning combined with artificial intelligence to predict diabetes complications can predict medical data to achieve early prediction and early prevention,while protecting the privacy of data,improving reference for expert consultation and treatment,benefiting the people,and contributing to the implementation of smart medical care. |
