Deep learning-based quantitative CT assessment of interstitial lung abnormalities: prognostic risk thresholds in a health screening population

초록

Quantitative risk thresholds for interstitial lung abnormalities (ILAs) and their association with long-term outcomes remain unclear in health screening populations. In this retrospective study, individuals aged >= 50 years who underwent chest CT between 2007 and 2013 were analyzed. Baseline CT scans were independently reviewed by two chest radiologists and categorized as none, equivocal ILA, or ILA, with consensus adjudication. ILA extent was quantified using a deep learning-based approach, including total ILA% and fibrotic ILA% across the whole lung. Multivariable Cox proportional hazards models were used to assess associations between ILA extent and clinical outcomes, including interstitial lung disease diagnosis, lung cancer diagnosis, and all-cause mortality. Optimal thresholds were determined using the minimum P-value method. Among 3,363 participants, 73 (2.2%) had ILA and 124 (3.7%) had equivocal ILA. The optimal cutoffs for all-cause mortality were 2.89% for total ILA and 0.26% for fibrotic ILA. Participants with total ILA >= 2.89% (HR, 5.15; P < 0.001) or fibrotic ILA >= 0.26% (HR, 2.71; P < 0.001) had significantly higher mortality risks compared with those below these thresholds. Deep learning-based quantitative ILA assessment was independently associated with long-term mortality, with prognostic thresholds of 3% for total ILA and 0.3% for fibrotic ILA in a health screening population.

키워드