The Research For Computer-aided Diagnosis Of Kawasaki Disease Based On Data Mining

Objectives: Kawasaki disease is an acute self-limited vasculitis of childhood that is characterized by fever,rash,bilateral nonexudative conjunctivitis,erythema of the lips and oral mucosa,changes in the extremities,and cervical lymphadenopathy.The etiology of Kawasaki disease remains unknown and the highest incidence rate was in children below 5 years old.In the absence of a specific diagnostic test,the diagnosis of Kawasaki disease rests upon clinical criteria that are shared by other common pediatric illnesses.Clinical confusion can lead to a missed or delayed diagnosis,which increases the risk of coronary artery aneurysms.Therefore,it's a challenge to diagnose Kawasaki disease quickly and accurately.To find a convenient,reliable Kawasaki disease diagnosis method,We develop diagnostic model-based data mining algorithm to differentiate Kawasaki disease from other pediatric febrile illnesses using clinical and laboratory data.Methods: Demographic,clinical,laboratory data and discharge diagnosis records of Kawasaki disease and other febrile illnesses were collected as the study subject.The sample database was established after data preprocessing and feature selection.The diagnostic model was established using Logistic regression,BP neural network,bayesian network and decision tree respectively and was evaluated by separate test dataset.We selected the best model by comparing the diagnostic performance of four methods.Finally,we build an Kawasaki disease computer-assisted diagnostic system based on CGI script programming.Results: The results showed that BP neural network had higher classification accuracy than other classification algorithm(Logistic regression,bayesian network and decision tree)and its accuracy is about90%.37 variables were selected as the input of models from 51 clinical indexes using feature selection based on univariate analysis.Our Kawasaki disease computer-assisted diagnostic system can automatically achieve diagnosis by entering demographic information,laboratory data and clinical symptoms for reference.Conclusions: We screened out clinical data related to the diagnosis of Kawasaki through the analysis of medical record data of Kawasaki disease and other febrile illnesses and it provides reference for follow-up clinical study.BP neural network can accurately differentiate Kawasaki disease from other febrile illnesses.Kawasaki disease computer-assisted diagnostic system can provide reference for clinical doctor.