Airway remodeling occurs across many respiratory diseases, yet routine chest CT tools offer limited, often coarse characterization of the airway tree. The authors introduce Radiomics of the Airway (RadAr), an automated pipeline designed to provide multi-scale, interpretable airway phenotyping from standard chest CT imaging. The stated goal is to capture airway attributes that reflect luminal size, tapering, architectural distortion, and global airway morphology in a form usable for analysis, visualization, and clinical research.
RadAr extracts a set of airway measurements at multiple spatial scales intended to be interpretable. The reported feature domains include luminal dimensions, tapering metrics, indices of architectural distortion, and measures of global airway morphology. The framework also includes an interactive web portal to visualize airway trees and derived metrics. The source describes RadAr as automated and scalable for routine CT datasets. Specific implementation details, feature definitions, and algorithmic parameters were not reported in full in the preprint abstract.
RadAr was evaluated across four distinct clinical and research settings using a mixture of institutional and publicly available datasets. The evaluations reported in the preprint included:
The study combined institutional and public data and reports that institutional cohorts were collected under IRB-approved protocols with informed consent where applicable.
In the fILD cohort, RadAr-derived measures of lower-lobe architectural distortion were associated with 63-week mortality. The reported balanced accuracy for this association was 0.654. The preprint does not provide further model details, confidence intervals, or comparative benchmarks in the abstract.
In the COVID-19 cohort, RadAr identified luminal dilation as independently associated with severe disease. The association metrics reported in the abstract include area under the receiver operating characteristic curve (AUC) of 0.719, an odds ratio of 2.32, and a p value of 0.017. Additional modeling covariates and calibration details were not specified in the abstract.
In the small PPF cohort (N = 9), airway phenotypes derived by RadAr correlated with physiologic and imaging measures of lung function and gas exchange. Specifically, correlations reported were: forced vital capacity (FVC) ρ = 0.83, mid-expiratory flow ρ = 0.87, and 129Xe MRI alveolar gas exchange impairment ρ = 0.70. The preprint abstract does not report statistical significance levels or confidence intervals for these correlations in this summary.
In the pediatric CF cohort (N = 11), RadAr metrics reflecting reduced tapering and increased cylindricity were associated with prior exacerbations and with bronchoalveolar lavage neutrophilia. Reported correlation coefficients ranged from ρ = −0.64 to −0.78 for these associations. The abstract does not include detailed timing, thresholds, or adjustment covariates for the reported associations.
Unsupervised clustering applied to the fILD and COVID-19 airway feature sets identified five distinct airway phenotypes. These phenotypes spanned a spectrum described in the preprint as: extensive, tapered airway trees → sparse, dilated, thick-walled, tortuous trees. The authors report that COVID-19 severity and fILD mortality increased across this phenotype spectrum. Specific cluster definitions, feature thresholds, and prevalence of each phenotype were not detailed in the abstract.
The authors conclude that RadAr identifies interpretable, disease-specific airway signatures that associate with physiologic function and clinical outcomes across restrictive (e.g., fILD, PPF), obstructive, and mixed airway diseases, and across adult and pediatric populations. They propose RadAr as a scalable framework that could support diagnosis, risk stratification, and longitudinal monitoring across pulmonary diseases on routine CT.
Because this work is reported as a preprint, it has not undergone peer review; the preprint explicitly states it should not be used to guide clinical practice. Several important methodological details, model validation procedures, and full statistical reporting are not provided in the abstract and would need review in the full manuscript for clinical interpretation.
The preprint declares that institutional cohorts were acquired under Institutional Review Board (IRB) approvals and that written informed consent was obtained for institutional participants. Emory University and Duke University IRB protocol identifiers are reported in the full declaration (IRB00097352 and Pro00109322). Remaining datasets were publicly available and used according to their data use agreements.
Competing interests are disclosed: Dr. Madabhushi is reported as a Research Career Scientist at the Atlanta Veterans Affairs Medical Center and has equity interests and advisory roles with multiple companies; he also has sponsored research agreements with several pharmaceutical and imaging companies. The source lists multiple funders that supported the work. The abstract does not report adverse events or harms because it is an imaging-methods study.