| Objective:1.To preliminarily explore the diagnostic value of artificial intelligence robot pre-diagnosis risk assessment for chronic obstructive pulmonary disease(COPD).2.To explore the diagnostic value of chest CT-based radiomics machine learning model for chronic obstructive pulmonary disease.3.To construct a nomogram prediction model and evaluate its diagnostic efficacy for COPD.Methods:1.From November 2020 to May 2022,the patients who met the pre-diagnosis screening criteria in Subei People’s hospital were surveyed by the pre-diagnosis risk assessment artificial intelligence robot.At the same time,the pre-diagnosis risk assessment questionnaire of respiratory chronic diseases and the completed COPD-SQ questionnaire were collected.A total of 700 subjects who met the screening criteria and underwent pulmonary function tests were included,of which 339 patients were diagnosed as COPD by pulmonary function tests.The basic clinical data,pulmonary function test and chest CT report of these 700 patients were collected.The risk factors of COPD were analyzed by univariate analysis and binary Logistic regression.Sensitivity,specificity and AUC were used to evaluate and compare the value of the pre-diagnosis risk assessment artificial intelligence robot and COPD-SQ in the diagnosis of COPD.2.Radiomics analysis was performed in 246 patients who had completed chest high-resolution CT scans,including 182 patients diagnosed with COPD by pulmonary function test,and 64 subjects with normal pulmonary function were randomly selected as controls.ITK-SNAP was used to delineate the whole lung tissue in chest CT,and then the radiomics features were extracted.Pearson correlation coefficient and LASSO algorithm were used to reduce the feature dimension,and the radiomics features with greater correlation with COPD were determined to obtain the radiomics score.Six machine learning models(LR,SVM,KNN,Random Forest,Extra Trees,XGBoost)were trained,and the accuracy,sensitivity,specificity,AUC and other indicators were used to evaluate the recognition effect of each model for COPD,and the best model was obtained by comparison.3.The risk factors of COPD obtained from pre-diagnosis screening,the pre-diagnosis risk assessment and prediction results of artificial intelligence robot,and the radiomics score obtained in the above process were combined to establish a nomogram diagnosis and prediction model to identify COPD.The AUC and calibration curve were used to evaluate its diagnostic value,and the decision curve was used to evaluate its clinical practicality.Finally,the value of artificial intelligence machine pre-diagnosis risk assessment,radiomics machine learning model and nomogram diagnosis prediction model for COPD diagnosis was compared.Results:1.Binary Logistic regression analysis showed that old age,male,and smoking history were independent risk factors for COPD.2.The sensitivity of artificial intelligence robot pre-diagnosis risk assessment for the diagnosis of COPD was 76.11%,the specificity was 84.76%,the Youden index was60.87%,the missed diagnosis rate was 23.89%,the misdiagnosis rate was 15.24%,the positive likelihood ratio was 4.99,and the negative likelihood ratio was 0.28.The AUC was 0.804.The sensitivity,specificity,Youden index,missed diagnosis rate,misdiagnosis rate,positive likelihood ratio and negative likelihood ratio of COPD-SQ for the diagnosis of COPD were 78.17%,70.36%,48.53%,21.83%,29.64%,2.64 and0.31,respectively.The AUC was 0.829.In contrast,the pre-diagnosis risk assessment of artificial intelligence robot has a higher diagnostic value for COPD.3.Six radiomics features were obtained by feature dimension reduction,and six machine learning models for COPD diagnosis were trained.The accuracy of LR model in training group and test group were 0.71 and 0.68,and AUC were 0.80 and 0.66,respectively.The accuracy of SVM model in training group and test group were 0.69 and 0.73,and the AUC were 0.78 and 0.65,respectively.The accuracy of KNN model in training group and test group were 0.70 and 0.65,and AUC were 0.80 and 0.70,respectively.The accuracy of Random Forest model in training group and test group were 0.97 and 0.57,and the AUC were 0.99 and 0.65,respectively.The accuracy of Extra Trees model in the training group and the test group were 1.00 and 0.68,respectively,and the AUC were 1.00 and 0.68,respectively.The accuracy of XGBoost model in training group and test group were 0.96 and 0.91,respectively,and the AUC were 0.99 and 0.86,respectively.Among them,the XGBoost model performs well in diagnosing COPD.4.The nomogram prediction model established by combining COPD risk factors,pre-diagnosis risk assessment results of artificial intelligence robot and radiomics score showed good predictive ability,with an AUC of 0.83,a sensitivity of 0.73 and a specificity of 0.80,and showed good clinical practicability.By contrast,the value of radiomics machine learning model for COPD diagnosis was higher than that of artificial intelligence robot pre-diagnosis risk assessment and nomogram prediction model.Conclusions:1.Male,old age,and smoking history are risk factors for COPD.COPD is more likely to occur when combined with chronic respiratory symptoms such as chronic cough and sputum,dyspnea,etc.2.Artificial intelligence robot pre-diagnostic risk assessment shows good screening performance in people aged 40 years and over,and can be used for COPD screening.3.Chest CT-based radiomics machine learning model is an effective method to identify COPD.As a non-invasive prediction tool,it can make up for the shortcomings of pulmonary function tests.4.The nomogram diagnosis and prediction model established by combining COPD risk factors,AI robot pre-diagnosis risk assessment and radiomics score can effectively identify COPD and provide a scoring system for COPD diagnosis. |
PDF Full Text Request