Efficacy of high-resolution CT imaging radiomics classification for diagnosis of rheumatoid arthritis-associated interstitial lung disease

Abstract

Objective To investigate the efficacy of high-resolution computed tomography (HRCT) imaging radiomics for the classification and diagnosis of rheumatoid arthritis associated interstitial lung disease (RA-ILD). Methods A total of 261 patients diagnosed with RA-ILD admitted in our hospital from January 2019 to July 2023 were recruited in this study.There were 143 cases of usual interstitial pneumonia (UIP) and 118 cases of nonspecific interstitial pneumonia (NSIP).All the patients underwent HRCT.A U-net deep learning lung segmentation model was applied to obtain HRCT images for automatic lung segmentation, and 1 688 imaging histologic features were extracted from each lung segmentation.Variance thresholding, univariate feature selection, and least absolute shrinkage and selection operator (LASSO) were used for feature dimensionality reduction step by step, and various machine learning algorithms were conducted to construct the RA-ILD diagnostic histology model.The diagnostic value of each model was compared using receiver operating characteristic (ROC) curve and area under curve (AUC), and the accuracy, sensitivity and specificity of the models were evaluated. Results Feature screening finally identified 18 best features from the HRCT images of RA-ILD patients.Statistical difference was found in the distribution of Radiomics score (Radscore) between the UIP and NSIP patients in both the training and test sets (P < 0.01).Among the 5 imaging histology models, the support vector machine (SVM) algorithm had an AUC value of 0.943(95%CI: 0.916~0.966), a sensitivity of 0.787 and a specificity of 0.912, respectively for the training set, and an AUC value of 0.909(95%CI: 0.849~0.969), a sensitivity of 0.625 and a specificity of 0.897, respectively for the testing set. Conclusion Our constructed RA-ILD classification and diagnosis model performs well, and the model based on SVM algorithm shows the best potential in classifying and diagnosing RA-ILD.

Keywords

• radiomics
• machine learning
• rheumatoid arthritis
• interstitial lung disease