Application of Imaging Omics to Predict the Stage of Pneumoconiosis Patients Based on Plain CT Scan
Objective To investigate the application value of chest CT imaging features in screening and staging of pneumoconiosis patients.Methods The clinical data and chest CT imaging data of 108 patients with clinically diagnosed pneumoconiosis in our hospital were retrospectively analyzed,including 69 cases of stage Ⅰ,34 cases of stage Ⅱ,and 5 cases of stage Ⅲ.The patients were randomly divided into the training set(n=70)and the verification set(n=38)according to the ratio of 7:3.MitkWorkbench software was used to conduct semi-automatic manual segmentation of ROI on chest CT images.3D Slicer was used to extract the features of the image omics map,the intra-group correlation coefficient(ICC)calculation and correlation analysis were used to remove the redundancy between features,and L1 regularization method was used for feature screening.The Random Forest classifier was used to establish the stage prediction model of pneumoconiosis,and the accuracy,sensitivity,specificity and receiver operating characteristic(ROC)curve were used to evaluate the efficiency of the model.Results There was no significant difference in age and service among patients with different disease stages(P＞0.05).The AUC for the identification of pneumoconiosis from stage Ⅰ to Ⅲ in the training set was 0.827,0.820 and 0.962,respectively,and the AUC for the identification of pneumoconiosis from stageⅠ to Ⅲ in the validation set was 0.823,0.817 and 1.000,respectively,with the accuracy,sensitivity and specificity higher than 75％.Conclusion The image omics features extracted from chest CT can accurately diagnose and screen the stage of pneumoconiosis patients,which has high clinical value.
万方数据
维普
