Abstract
Diagnosing and classifying the imaging patterns of idiopathic inflammatory myopathies-associated interstitial lung disease (IIM-ILD) is a crucial but challenging task requiring specialized physicians' expertise. This study aims to develop and validate a deep-learning model to assist in classifying the IIM-ILD imaging patterns. The study retrospectively collected 629 patients with IIM-ILD and split them into a training set (361 subjects), an internal testing set (156 subjects) from January 2015 to December 2019, and a temporal external validation set (112 subjects) from January 2020 to August 2022. A deep-learning model was developed and validated to categorize IIM-ILD imaging patterns using HRCT images. Class activation mapping and label smoothing strategy were utilized to enhance the interpretability and performance. Model performance was assessed by the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, and F1-score. The proposed deep-learning model achieved an average AUC of 0.885, accuracy of 0.724, and F1-score of 0.706 in the internal testing set, and an AUC of 0.835, accuracy of 0.795, and F1-score of 0.727 in the temporal external validation set. In summary, the deep-learning model can effectively classify the multiple imaging patterns of IIM-ILD, showing potential as a valuable radiological diagnostic support system in clinical practice.