Application Of Machine Learning On Renal Disease Related Clinical Decision Support System

Nowadays,the medical field of China is experiencing a revolution driven by artificial intelligence.The core of the revolution is WIT120 which is short for Wise Information Technology of 120.WIT120 seeks to build an informationization online platform that allows patients to obtain safe,convenient and high-quality medical services with a shorter waiting time and basic expenses.In this system,the clinical decision support system is an important component.It is an interactive expert system which assists the doctors in making treatment decision.Based on the actual needs of the Second Hospital of Dalian Medical University,this paper studies two common problems about renal diseases which are involved in nephrology department,and constructs the corresponding clinical decision support system in the end.The main work are summarized as follows:The prediction of Glomerular Filtration Rate(GFR)is studied.GFR is a main measure of kidney function which plays an important role in staging chronic kidney disease and instructing treatment of it.While for the current detection methods,some are harmful to human's body,and some has an low precision.In view of this,this article designs a model which is based on XGBoost technology,through comprehensively using of patient's relevant test indicators to get a prediction result.The model belongs to ensemble learning and its theoretical basis is decision tree,which has the advantages of high accuracy,strong anti-noise ability and good interpretability.Compared with the current methods,the XGBoost model proposed in this paper gets an at least 10% accuracy improved at the same test set.After the model is built,it is embedded in the backstage of an application system.The front-end of the system integrates an user interface to complete the system construction.The suitable dosage of Heparin used in the process of Continual Renal Replacement Therapy(CRRT)is studied.CRRT is an important means to deal with chronic and acute renal failure.While the appropriate amount of Heparin directly affects the outcome of the surgery,the current usage is mainly given by the chief physician based on the experience.Obviously,the decision always contains subjectivity and uncertainty.Using the collected dataset,after the procedures of feature extraction and dataset balancing,the prediction work is done by firstly using support vector machine to do a binary classification,and then the exact value of the first and additional dose are accurately predicted based on combining ensemble learning model and least squares regression model.The experiments show that the models built in this paper could give a reasonable and reliable dosage for the surgery.Also a system is constructed for the practical application.