Absolute and lung volume-normalized deep learning-derived pulmonary vascular measures in idiopathic pulmonary fibrosis: distinct clinical associations and the role of lung volume in prognosis

Abstract

Introduction Idiopathic pulmonary fibrosis (IPF) has an unpredictable course and poor outcome, and lung function-based risk stratification remains incomplete. Whether absolute and lung volume-normalized deep learning-derived CT pulmonary vascular measurements show distinct associations with disease severity and mortality in patients with IPF remains unclear. Methods In this dual-center retrospective study, 118 adults with newly diagnosed IPF, including 57 from the Italian cohort and 61 from the French cohort, underwent baseline chest CT analyzed with a deep learning-based quantitative pipeline combining lung parenchyma (whole lung volume, WLV; fibrosis extent) and pulmonary vascular metrics (total blood volume, TBV; and derived BV5 and BV>10 indices representing vessels with cross-sectional areas of <5 mm² and >10 mm²). Twenty subjects without lung disease served as a comparison group. Associations with fibrosis were assessed by Pearson correlation, mortality by multivariable Cox hazards models, and longitudinal lung function by linear mixed-effects models. Additional models examined the prognostic associations of vascular metrics after accounting for WLV. Results Compared with controls, IPF patients had lower WLV and higher absolute TBV. Relative TBV (normalized to WLV) correlated strongly with fibrosis, while absolute TBV showed weaker associations. Relative vascular metrics inversely related to baseline lung function, whereas absolute TBV showed no significant association. Baseline TBV, BV5, and BV>10 did not predict forced vital capacity (FVC) or diffusing capacity of the lung for CO trajectories, but higher WLV predicted better FVC over time. Over a median 7.8-year follow-up, 72 deaths occurred. In multivariable models, higher absolute TBV, lower-lobe BV, BV5, BV>10, and WLV, but not normalized TBV, were associated with reduced mortality after adjustment for established clinical variables. These findings suggest that preservation of lung volume and pulmonary vascular volume may be associated with improved survival. However, the associations of absolute vascular metrics with mortality were substantially attenuated after additional adjustment for WLV. Conclusion Normalized and absolute pulmonary vascular metrics capture different aspects of IPF, with normalized measures associated with fibrosis burden and functional impairment, while absolute vascular volumes were associated with mortality, although this association was substantially attenuated after accounting for WLV. Absolute vascular volume may therefore partly reflect preserved lung volume rather than an independent prognostic effect. Prospective validation is warranted.