Disease Early Warning And Diagnosis Based On Deep Learning

With the increasing development of artificial intelligence in recent years,the research on clinical diagnosis and treatment decision-making assisted by deep learning has always been a hot topic in the medical field.It is mainly used in medical big data,such as medical image recognition,electronic medical record analysis,genomics based precision medicine research and clinical decision support system.The disease prediction model based on clinical manifestation is an important research content of clinical decision support system,this article is based on the depth of the level of attention mechanism neural network characteristic analysis of clinical data sets,the clinical cases as samples,the real world of complex data from high dimensional space into a low dimensional vector,and then by adopting the method of statistical analysis,to set up the neural network modeling disease forecast conclusion,and attempts to analyze the model of "interpretability",the main research work is as follows:(1)In order to carry out the real world of clinical research,the data resources of the existing information system of hospital are integrated through data integration technology,including the classification and analysis of data structure of hospital clinical information system,to set up the index data and public data model,a lightweight data warehouse for specific diseases(acute kidney injury)has been formed.A multi-scale convolutional neural network early warning model based on mixed time series was proposed.Taking acute kidney injury as an example,an empirical study of clinical disease early warning was carried out,which changed the cross-sectional research method of disease in traditional medical statistics in terms of technical methods.This method combines the irregular clinical data of risk factors associated with disease and the time series information of different patients as index data to improve the accuracy of disease prediction.(2)The research aims at the shortcomings of the proposed model based on(1),such as the redundancy of bedside data,the loss of key indicators and the different frequency of data sampling,the hierarchical attention network model is innovatively proposed to be used for critical early warning of time series based on physiological indicators,so that the model has the adaptability to complex data structure.Compared with the previous model,the prediction accuracy of the experiment is improved,and it is more practical.At the same time,the research through adding of the interpretable layer,combines the intermediate term generated during the training process with the gradual progression of the actual patient’s vital signs,marked out which indicators were more important for the treatment process of patients,and conducted an empirical study on the interpretability of the model.Experiments proved that the model had good reliability compared with other deep learning models.(3)The research adopts real world clinical data to build a research platform and uses the currently popular data mining components.The proposed method is real and effective.The main contents include: data integration of the existing system of the hospital with SQL Sever;The code was written by Tensor Flow/Py Torch and other third-party deep learning development kits in Python,which laid a foundation for the practical application of medical artificial intelligence in hospital information system.