Development and internal validation of a chest CT radiomics model for identifying pneumoconiosis.

Authors

Abstract

Pneumoconiosis is a major occupational lung disease with an insidious onset, making early diagnosis challenging. We aimed to develop and internally validate a chest computed tomography (CT)-based radiomics model for objective identification of pneumoconiosis and to support subjective visual diagnosis in dust-exposed individuals. This retrospective study consecutively enrolled 474 patients with documented occupational dust exposure who underwent pneumoconiosis evaluation at The Third People's Hospital of Xinjiang Uygur Autonomous Region between January 2020 and December 2024. Eligible participants had high-resolution chest CT images and no history of pulmonary surgery or trauma; those with coexisting respiratory disease, severe systemic dysfunction, or poor image quality were excluded. The reference standard was expert-panel consensus according to GBZ 70-2015 criteria. Lung parenchyma was semi-automatically segmented using a region-growing algorithm in three-dimensional (3D) Slicer, and radiomics features were extracted with the Deepwise Multimodal Research Platform. From 2,153 features, Pearson correlation filtering (r≥0.9 excluded) and analysis of variance (ANOVA) F-test (P<0.05) selected the 10 most discriminative features. Data were randomly divided into training and test sets. A logistic regression model was constructed and evaluated using area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, specificity, calibration, and decision curve analysis (DCA). Of 474 patients, 293 had pneumoconiosis (all stage I) and 181 were dust-exposed controls. The model achieved AUCs of 0.804 [95% confidence interval (CI): 0.754-0.853] in the training set (n=331) and 0.759 (95% CI: 0.680-0.838) in the test set (n=143). At the optimal threshold, sensitivity and specificity were 76.2% and 68.3% in the training set and 61.8% and 67.1% in the test set, respectively. Calibration curves showed good agreement between predicted and observed outcomes, and DCA confirmed net clinical benefit across a range of threshold probabilities. This CT-based whole-lung radiomics model, without manual lesion delineation, shows promise for early pneumoconiosis identification in dust-exposed populations, with favorable diagnostic accuracy and clinical utility. However, the findings are preliminary and require external validation in independent multicenter cohorts before clinical implementation.

Tags

Topics