Diabetes Complications Discovery Method Based On Similarity Constrained LDA Model

The use of intelligent diagnostic methods to discover the patient's condition and develop personalized diagnosis or treatment plan has become the direction of reform and development of the medical industry and is also an important part of the application of medical big data.In China,with the widespread use of medical service software such as EMR,HIS,and PACS,a large amount of text and image data has been deposited in the hospital information system,laying a foundation for data for the popularization of intelligent diagnosis.With the acceleration of the pace of life and the improvement of the nutritional level,the obese population in China has significantly expanded,and obesity is one of the important risk factors for diabetes,which increases the number of people with diabetes in China.Diabetic complications are the main cause of death in diabetic patients.How to effectively predicting complications is the key to improving the quality of life of patients.However,with the ever-increasing scale of electronic medical records,how to effectively and easily discover diabetes complications from complex multimodal case data has become a challenging task.This article first defines an improved topic discovery model.Based on this,it discovers the potential topics recorded in different disease courses.Then it uses the time series feature extraction and multi-label classification algorithm model to perform complications discovery for diabetic patients,and finally completes the comparison test.The verification of the algorithm,the specific research content is as follows:(1)This paper proposes an enhanced LDA model based on medical records similarity.It is analyzed that there are certain similarities in the course records of similar patients during hospitalization based on the mining of latent topics.In view of the similarity between the medical records of patients with diabetes,an improved LDA algorithm is proposed.And the discovery of disease history records is the basis for subsequent complications.(2)We designed an algorithmic process for the discovery of multiple-label diabetes complications based on time series.Due to the multi-dimensional temporal characteristics of the records during hospitalization,this paper uses singular value decomposition to extract multi-dimensional time-series topic features and uses Ensembles of Classifier Chains to combine classifier chain ideas for multi-label classification data processing of diabetic complications.Through the comparison of multiple classification algorithms,a classification model suitable for predicting diabetes complications is obtained.