Deep Learning-Based CT Analysis Identifies Risk Factors in PF-ILD
Overview
This study utilized deep learning-based quantitative CT (QCT) analysis to identify risk factors and prognostic markers for progressive fibrosing interstitial lung disease (PF-ILD). It demonstrated that automated segmentation and quantification of fibrosis extent on CT scans can aid in predicting disease progression and outcomes in PF-ILD patients.
Background
Progressive fibrosing interstitial lung disease (PF-ILD) is characterized by increasing fibrosis on CT, declining lung function, and worsening symptoms leading to early mortality. While idiopathic pulmonary fibrosis (IPF) is the prototypical progressive fibrosing ILD, non-IPF ILD subtypes such as nonspecific interstitial pneumonia (NSIP), connective tissue disease-associated ILD (CTD-ILD), and fibrotic hypersensitivity pneumonitis (HP) also exhibit progressive fibrosis. Chest CT plays a critical role in diagnosis, monitoring, and prognostication, but visual assessment is limited by inter-reader variability. Quantitative CT methods, particularly those leveraging deep learning, offer objective and reproducible evaluation of ILD extent and progression.
Data Highlights
The study retrospectively analyzed 2864 patients with ILD-related CT scans from 2015 to 2021, applying exclusion criteria to yield a final cohort with baseline and follow-up CT scans at least 24 months apart and corresponding pulmonary function tests (PFTs). Deep learning software segmented ILD features including emphysema, consolidation, ground-glass opacity, reticular opacity, and honeycombing, quantifying these as percentages of total lung volume. Fibrosis extent was defined as the sum of reticular opacity and honeycombing, while total ILD extent included ground-glass opacity as well. Visual assessments of ILD patterns and fibrosis progression were performed by experienced radiologists following current guidelines.
Key Findings
Deep learning-based QCT successfully segmented and quantified ILD features, enabling objective measurement of fibrosis extent and total ILD involvement.
Fibrosis extent on baseline CT correlated with PF-ILD progression and decline in forced vital capacity (FVC) over time.
Visual CT patterns, including usual interstitial pneumonia (UIP) and probable UIP, were associated with progressive fibrosis but showed variability in interpretation, underscoring the value of quantitative analysis.
One-year follow-up CT scans analyzed by deep learning QCT provided early indicators of disease progression before the 24-month endpoint.
Non-IPF ILD subtypes with progressive fibrosis exhibited similar short-term FVC decline and mortality risk as IPF, highlighting the importance of early identification across ILD categories.
Clinical Implications
Deep learning-based quantitative CT analysis offers a reproducible and objective tool to identify patients at risk for PF-ILD progression, potentially enabling earlier intervention. Incorporating automated fibrosis quantification into routine clinical practice may improve prognostication and guide treatment decisions beyond traditional visual CT assessment. Early detection of progression at 1-year follow-up CT could facilitate timely therapeutic adjustments.
Conclusion
This study supports the integration of deep learning-based QCT in the evaluation of PF-ILD, providing valuable risk stratification and prognostic information that complements clinical and visual radiologic assessments. Such approaches may enhance management strategies for patients with progressive fibrosing ILD.
References
Koh et al 2023 -- Identifying Risk Factors and Prognostic Markers for PF-ILD Using Deep Learning CT Analysis
