Modeling And Analysis Of Auxiliary Diagnosis Of Intestinal Diseases In Children Based On Clinical Data

With the continuous development of information technology,the medical data generated by smart medical instruments has shown an explosive growth every year.In most cases,people’s research on numerical and text-based medical data is only limited to simple statistical analysis,and the hidden information behind these data is not deeply explored.This thesis introduces some machine learning algorithms to mine the relationship between these numerical,text-based clinical data and diseases,so as to provide decision support for the diagnosis and treatment of diseases.The specific research content and results are as follows:1)Construct a decision-making model for the next stage of diagnosis and treatment for patients with pediatric intestinal obstruction.This thesis uses deep neural network algorithm to train the blood routine examination and liver and kidney function test indicators of children with intestinal obstruction.Aiming at the problem of incomplete data samples,different interpolation methods were tried to train the model.Finally,the "category mean" interpolation method is used to fill in these vacant data,which can not only avoid the loss of sample data information,but also make the filled data conform to the original true value to a greater extent.At the same tine,the feature importance evaluation index based on Gini coefficient is used to score the importance of each feature,and selected as indicators for the doctor’s diagnosis.In order to evaluate the stability of the model,this thesis uses 5-fold cross-validation to test the model.Experimental results show that the classification performance of the model can provide doctors with auxiliary decision-making suggestions.2)To construct a diagnostic model of complications associated with intestinal necrosis in children with inguinal hernia.In this thesis,based on the characteristics of the collected numerical clinical data,the blood routine examination and liver and kidney function test indicators were modeled and analyzed.Aiming at the problem of data imbalance in the experiment,this thesis adopts an integrated method based on voting to deal with the problem.First,the sample data of the majority category is divided equally into multiple sub-samples,and each sub-sample is formed with a minority sample to form a balanced data set.Then train the models on the balanced data sets respectively,combined the many trained models into an integrated learning model,and use the voting results between the models as the classification results of the integrated learning model.After that,this thesis also made further optimization work on the model features,using the importance score of the features as the basis of iteration to add feature values and train the model successively,and selecting the best feature combination according to the model performance.After experimental verification,the accuracy of the model trained using the feature optimization data set is 86.43%,and the performance is better than the model trained with the original inspection parameters.Therefore,the feature-optimized model can provide auxiliary decision support when doctors diagnose whether pediatric inguinal hernia patients are accompanied by complications of intestinal necrosis.In summary,this thesis uses statistical analysis,interpolation,and ensemble learning methods to deal with these common problems in medical clinical data when faced with data redundancy,missing,and imbalance.The auxiliary decision-making model constructed in this thesis based on numerical clinical data can provide reasonable medical explanations for the classification results while ensuring accuracy.It makes the application of auxiliary decision-making model to this type of data of great significance.