Radiomics-Guided Machine Learning Models for Differentiating and Predicting Prognosis in Interstitial Lung Diseases

Le résumé fourni par la source

Background: While radiomics-guided machine learning (ML) holds significant promise, its potential for both classifying and predicting prognosis in interstitial lung diseases (ILD) has yet to be fully explored. Methods: A retrospective cohort of 154 idiopathic pulmonary fibrosis (IPF) and 189 connective tissue disease-associated ILD (CTD-ILD) patients (aged 50–80) was from the "Disease-Specific Research Database" at Gansu Provincial Hospital, China. Radiomic features via HRCT images were extracted via PyRadiomics, and feature selection (t-test + LASSO) identified disease-specific signatures. Support vector machine (SVM), stochastic gradient descent (SGD), and random forest (RF) models were trained for classification and prognosis prediction. Results: A total of 1,037 radiomic features were reduced to 11 (classification: 9 texture, 2 first-order), 12 (IPF prognosis: 9 texture, 2 first-order, 1 shape), and 13 (CTD-ILD prognosis: 9 texture, 4 first-order). For classification, SVM achieved the highest AUC (training/validation: 0.847/0.792), outperforming RF (0.873/0.739) and SGD (0.740/0.714). For prognosis prediction, patients were stratified into high- and low-risk groups by median survival (IPF: 32 months, CTD-ILD: 59 months). ML-based prognostic models successfully differentiated risk groups in IPF, ML-based prognostic models successfully differentiated risk groups in IPF, with RF showed optimal IPF risk prediction (AUC: 0.909/0.807), while SVM excelled in CTD-ILD (AUC: 0.858/0.842). Conclusion: This study establishes a radiomics-ML framework for ILD classification and prognosis, offering insights for precision management.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.