Imaging-based Assessment Of Glucocorticoid Sensitivity In Patients With Idiopathic Pulmonary Hemosiderosis

Objective To investigate the sensitivity to glucocorticoids in patients with idiopathic pulmonary hemosiderosis based on CT imaging radiomic features.Materials and methods Retrospective analysis was performed on 104 patients with Idiopathic pulmonary hemosiderosis(IPH)diagnosed by sputum smear,alveolar lavage fluid or gastric fluid and clinical features.104 IPH patients were divided into hormone active group(training dataset,n=60,test dataset,n=16)and hormone ineffective group(training dataset,n =22,test dataset,n=6).CT imaging morphological changes,clinical symptoms,partial laboratory results and laboratory tests’ results of the two groups were collected.P< 0.05 was considered statistically significant between the 2 groups.Using Radcloud software to establish image radiomic features model 1.1409 image features were extracted,and using the Least Absolute Shrinkage and Selection Operator(Lasso)algorithm to select the highest correlated features.The prediction model was established by KNN,SVM and LR and tested through the cross validation algorithm.The diagnostic efficacy of prediction model 1 was evaluated by Receiver Operating Characteristic(ROC)Curve,Area Under Curve(AUC),sensitivity and specificity.Meanwhile.The predictive model 2combined with radiomics and major clinical features was further established to evaluate the diagnostic efficacy of model 2.Results Compared with the hormone effective group and the hormone ineffective group,the proportion of lesions in the hormone effective group was significantly higher than that that in the hormone ineffective group on CT scan,with statistical significance(P< 0.05).Compared with the main clinical symptoms of the 2 groups after treatment,the proportion of patients with cough and hemoptysis in the hormone ineffective group was higher than that in the hormone effective group,with statistical significance(P< 0.05).The diagnostic accuracy of cough and hemoptysis were 69.3% and 66.7%,respectively.The AUC of hemoglobin and erythrocyte count were 0.662 and 0.671 respectively.Imaging radiomic features model 1 was imported into KNN,SVN and LR models for prediction.In the KNN prediction model,the AUC of the test set was 0.896,and the AUC of the training set was 0.827.In the SVM prediction model,the AUC of the test set was 0.906,and the AUC of the training set was0.898.In the LR prediction model,the AUC of the test set was 0.813,and the AUC of the training set was 0.917.Imaging radiomic features combined with the main clinical features to establish prediction model 2,which was imported into the SVM model for prediction,the AUC of the test set was 0.820,and the AUC of the training set was 0.953.Conclusion(1)Reality changes were more common in CT findings in the hormone effective group.For clinical symptoms after treatment,cough and hemoptysis were less common in the hormone effective group.(2)The predictive model based on CT imaging radiomic features can better identify patients with glucocorticoid-sensitive type,and the predictive model based on imaging radiomic features combined with clinical symptoms also has better diagnostic efficacy.