1、Comparison of AI Techniques for Prediction of Liver Fibrosisin Hepatitis Patients,Journal of Medical SystemJiajun Shi,Some explanations,Fibrosis - 纤维化Hepatitis - 肝炎Hepatitis B/C 乙肝/丙肝Cirrhosis 肝硬化Liver biopsies - 活组织检查Non-invasive techniques 无创技术Serum markers 血清标记,Outline,IntroductionBackground: AI
2、and CDSSNave Bayes Classifier (NBC) & Logistics RegressionHepatitis and Fibrosis StageAI Assisted Web-based Clinical Decision Support SystemFour MethodsResults and Diagnostic AccuracyConclusion,Introduction,One in twelve people have the Hepatitis B or Hepatitis C virusDiagnosis and treatment of this
3、 disease is guided by liver biopsies where a small amount of tissue is removed by a surgeon and examined by a pathologistDetermine the fibrosis stage from F0 (no damage) to F4 (cirrhosis),Risk and costly,Non-invasive techniques, with serum markers, imaging test, and genetic studies,Accuracy not achi
4、eved sufficient acceptance,Introduction,Non-invasive techniques, with serum markers, imaging test, and genetic studies,AI & CDSS,Knowledge of the level of liver damage in a patient withliver disease (particularly Hepatitis B and Hepatitis C) is acritical factor in determining the optimal course of t
5、reatmentand to measure the effectiveness of alternative treatments inpatients.,Not accurate,Background of AI and CDSS,Artificial Intelligence and Data Mining techniques Include Neural Networks, Fuzzy Logic, Decision Trees, Bayesian Classifiers, Support Vector Machines, Genetic Algorithms and Hybrid
6、System,Clinical and Medical Decision Support SystemsSupport the process of discovering useful information in large clinical repositories They had done the system designed with neural networks and decision tree methods because of their successful application in similar problem domains,Hepatitis and F
7、ibrosis Stage,One in twelve people have the Hepatitis B or Hepatitis C virus,AI Assisted Web-based Clinical Decision Support System,AI Assisted CDSS,AI techniquesResulting knowledge base,AI Assisted Web-based Clinical Decision Support System,Variables：,SerumMarkers,PatientsInfo,AI Assisted web-based
8、 Clinical Decision Support System,System inputs& Outputs:,Four Methods,Paper Advanced Decision Support for Complex Clinical Decisions Neural Networks, Decision TreesThis paperNaive Bayes and Log Regression,Method inputs:,Four Methods Nave bayes classifier,The variation in mean values for two paramet
9、ers (ABL and G-GL) are shown by fibrosis stage in the Figure.With this model, we can calculate the combined probability of each fibrosis stage then select the highest probable as our predicted result.,Four Methods - Logistics regression,Cross Validation and Diagnostic Accuracy,Cross Validation and D
10、iagnostic Accuracy,Accuracy of Fibrosis Stage Predictions (424 patients),Predictive Sensitivity and Specificity,Conclusion,The four artificial intelligence methods presented in this study showed some significant variability in accuracy, sensitivity, and specificity in predicting fibrosis stage in da
11、ta on 424 hepatitis patients. Although neural network methods showed the highest sensitivity and specificity, their role is predicting the exact fibrosis stage was relatively poor. Logistic regression and nave bayes methods were the best in identifying the exact fibrosis stage. Logistic regression had the highest Positive Predictive Value, and neural network methods had the highest negative Predictive Value.,THANKS,
