Establishment Of Model For Predicting Pathological Types Of Acute Appendicitis Based On Artificial Intelligence

Objective:Acute appendicitis is still one of the common acute abdominal diseases in the world.Some studies have found that the misdiagnosis of acute appendicitis is mostly related to the clinical manifestations of patients,and its adverse events(perforation,abscess formation)are also mostly related to medical misdiagnosis and missed diagnosis.The current clinical experience has shown that conservative treatment or surgical resection is still controversial,acute appendicitis treatment choice still lies in its pathological types.Based on the data mining of the real world,we forecast preoperative pathological diagnosis of acute appendicitis,acute appendicitis treatment scheme selection is dedicated to provide you with a new train of thought.Methods:Data extraction,data cleaning,data analysis and data visualization were carried out based on the Nanfang Hospital medical big data platform.According to histopathology,the cases were randomly divided into the training set and the test set in a ratio of 7:3.By entering the clinical features,biological markers and imaging log variables to Logistic Regression,SGD Classifier,Gaussian Process Classifier,Gradient Boosting Classifier,MLP Classifier,K Neighbors Classifier,Random Forest Classifier,Ada Boost Classifier and other algorithms to evaluate the performance of the model using accuracy,receiver operating characteristic curve(ROC curve)and recall rate.Results:1373 cases of acute appendicitis were retrospectively collected,including acute simple appendicitis(N=40),acute suppurative appendicitis(N=912),acute gangrenous appendicitis(N=254)and perforated appendicitis(N=167).In this study,the optimal subset algorithm was used to screen the optimal number of variables K value,and the optimal subset was screened according to the same K value.Among the binary prediction models established by different algorithms,the random forest model was the best(the highest accuracy rate was 97.309%;The lowest accuracy rate was 69.167%)and the AUC value(the highest AUC was 0.984;The lowest AUC 0.437).At the same time,the study found that the percentage of neutrophil,percentage of lymphocyte and duration of onset were very important for the prediction of simple appendicitis,complex appendicitis,perforated appendicitis and recurrent appendicitis.Conclusion:Based on data mining in the real world,this study uses machine learning to establish a variety of different algorithm prediction models for rapid diagnosis of acute appendicitis pathological types by clinical characteristics and biological markers of patients,among which the random forest model is the best evaluation.At the same time,it was found that the percentage of neutrophil,percentage of lymphocyte and duration of onset were very important to predict the types of simple appendicitis,complex appendicitis,perforated appendicitis and recurrent appendicitis.In addition,the new text feature variables can also significantly improve the accuracy and AUC value of the model.