Abstract
Background Airway remodeling is a convergent feature across respiratory diseases, yet current quantitative CT tools quantify a small number of prespecified structural abnormalities of the airway tree. Radiomics of the Airway (RadAr) derives interpretable, multi-scale airway measurements from chest CT to characterize airway deformation and discover quantitative imaging biomarkers from routine chest CT.
Methods RadAr extracts over 400 multi-scale, interpretable airway measurements across lobes or generations capturing luminal dimensions, tapering, architectural distortion, and global morphology. It was evaluated in N=1331 patients across four settings: 63-week mortality prediction in fibrotic interstitial lung disease (fILD), COVID-19 severity prediction, structure-function association in progressive pulmonary fibrosis (PPF) and structure-inflammation markers in pediatric cystic fibrosis (CF). Unsupervised clustering was used to identify airway phenotypes across the fILD and COVID-19 cohorts.
Results In fILD, lower-lobe architectural distortion was associated with mortality (balanced accuracy 0·654). In COVID-19, severe disease was independently associated with luminal dilation (AUC 0·719, odds ratio 2·32, p=0·017). In PPF, airway phenotypes correlated with forced vital capacity (ρ=0·83), mid-expiratory flow (ρ=0·87), and ¹²⁹Xe MRI alveolar gas exchange impairment (ρ=0·70). In pediatric CF, reduced tapering and increased cylindricity were associated with prior exacerbations and bronchoalveolar lavage neutrophilia (ρ=-0·64 to -0·78). Five phenotypes were identified from extensive, tapered airway trees to sparse, dilated, thick-walled, tortuous trees, with increasing COVID-19 severity and fILD mortality across this spectrum.
Conclusion RadAr identified interpretable, disease-specific airway phenotypes associated with function and outcomes across restrictive, obstructive, and mixed lung diseases in adult and pediatric settings. These findings establish a framework for discovery and development of quantitative airway biomarkers for patient stratification, disease monitoring, or imaging endpoint development in pulmonary trials.
Competing Interest Statement
The corresponding author, Dr. Anant Madabhushi (AM), is a Research Career Scientist at the Atlanta Veterans Affairs Medical Center. AM is an equity holder in Picture Health, Elucid Bioimaging, and Inspirata Inc. He currently serves on the advisory board of Picture Health and consults for Takeda Inc. and Johnson and Johnson. He has sponsored research agreements with AstraZeneca and Bristol Myers-Squibb, and his technology has been licensed to Picture Health and Elucid Bioimaging. Dr. Bastiaan Dreihuys (BD) is a shareholder in Polarean Imaging, BD and Dr. David Mummy consult for Polarean Imaging without any fiduciary obligations. The remaining authors declare no relevant conflicts of interest.
Author Declarations
I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.
Yes
The details of the IRB/oversight body that provided approval or exemption for the research described are given below:
This retrospective study included both institutional and publicly available datasets. The institutional progressive pulmonary fibrosis and cystic fibrosis cohorts were acquired under protocols approved by the Institutional Review Boards (IRB), and written informed consent was obtained from all participants. IRB of Emory University and Duke University gave ethical approval for this work under protocol IRB00097352 and Pro00109322, respectively. The remaining datasets were obtained from publicly available repositories and were used in accordance with their respective data use agreements and institutional policies.
I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals.
Yes
I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).
Yes
I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable.
Yes
Footnotes
The introduction, methods, and discussion were all tightened to reduce word count, with numerous supporting details, statistics, examples, and section headings removed or moved to supplement to streamline the manuscript. No major changes to the results or figures were made. The abstract is updated to better reflect RadAr as a quantitative imaging-biomarker discovery and development framework. The abstract quantifies the scope stating that RadAr extracts "over 400" measurements "across lobes or generations" and that evaluation spanned "N=1331 patients." The introduction is condensed. The explicit hypothesis statement and the enumerated (a)-(d) list of evaluation settings were removed in favor of prose describing discovery of "disease-specific and cross-disease structural phenotypes." Methods were similarly tightened: the RAI-reorientation preprocessing sentence and its subsection heading were moved to supplement, mild/severe count breakdowns were simplified to severe counts only, and the statistical-analysis text lists AUC and average precision among the general performance metrics while moving the note about their unavailability for the fILD portal to the supplement. A new opening paragraph was added to the Results, framing the analyses explicitly around imaging-biomarker validity (repeatability, prognostic, structure-function, structure-inflammation, and phenotype discovery) and additional supplementary figures and a table are now cited. In the Discussion, the dedicated paragraph describing the interactive web portal was removed. The GLP-1 receptor-agonist example was deleted, and the concluding sentence now explicitly positions the measurements as potential "novel surrogate endpoints in pulmonary trials." Terminology was standardized from "lingula" to "left middle lobe (LML)" and a correction from "UIP-pattern" to "IPF-pattern" fibrosis. A declaration of generative-AI use specifying the tools used for figure creation, language editing, and portal development was added. Funding acknowledgements were also updated.
Data Availability Statement
The COVID-19 datasets analyzed in this study are publicly available. The STOIC2021 COVID-19 cohort can be accessed through the STOIC2021 COVID-19 AI Challenge (https://stoic2021.grand-challenge.org), and the Stony Brook University COVID-19 cohort is available from The Cancer Imaging Archive (https://www.cancerimagingarchive.net/collection/covid-19-ny-sbu/). The fibrotic interstitial lung disease cohort was obtained from the AIIB23 challenge (https://codalab.lisn.upsaclay.fr/competitions/13256). All public datasets were used in accordance with their respective data use agreements. The progressive pulmonary fibrosis and pediatric cystic fibrosis datasets are institutional and are not publicly available owing to privacy and ethical restrictions, but de-identified data may be available from the corresponding author on reasonable request and subject to institutional review board approval and a data use agreement.
Funder Information Declared
