Unsupervised image-to-image translation plays a crucial role in medical imaging, enabling data harmonization and crossmodality synthesis without paired data. Existing approaches, including CycleGAN, UNIT, and CUT, often fail to preserve fine anatomical details, limiting their applicability for clinical use. In this work, we develop a novel one-sided unsupervised image-to-image translation method - Adaptive Normalized Gradient Field (AdaNGF) - which enforces structural consistency throughout the translation process. We adapt the normalization of the gradient field of the generated image to match the target gradient magnitude distribution, encouraging the generator towards anatomically-sound outputs with appropriate preservation or suppression of local gradient alignment. We evaluate our approach on intra-modality (BraTS: T1 → T2) and cross-modality (Gold Atlas: MRI → CT) datasets. Quantitative and qualitative results show that our method outperforms state-of-the-art unsupervised translation frameworks in terms of SSIM, PSNR, and structural similarity metrics. Code will be publicly shared on GitHub upon acceptance.

Deep-learning-based segmentation algorithms have gained considerable accuracy for processing biological images. In particular, the introduction of large foundation models, novel architectures, and semantically varied datasets now allows for deployment of state-of-the-art models for clean image cohorts with limited re-training or, in the best of cases, in an out-of-the-box fashion. Biological imaging, however, is liable to corruptions that can hinder their deployment. While some methods document their robustness to the most common corruptions, a systematic robustness analysis of the state of the art to the expansive gamut of corruptions in biological imaging remains to be done. We perform this benchmarking by simulating 36 corruption types with varying degradation severity on images sampled from 30 different datasets. Our benchmark accounts both for the variety in biological images and the nature of corruptions. Among other things, our study reveals that performance on clean images does not correlate with overall robustness to image corruptions. In fact, we find that a decade-old method, StarDist, is more robust than many of its more recent foundation-model-based counterparts. We also show in a dedicated representation analysis that the performance of segmentation models collapses in the early layers of the encoding phase.

Associations between exposure to ambient air pollution and progression of emphysema have been identified in longitudinal observational studies. However, previous work has not used statistical causal inference methods tailored to address bias from time-varying confounding. The objective of this study is to propose an analytical approach for estimating longitudinal health effects of air pollution while accounting for time-varying confounding using marginal structural models and to re-analyze data on air pollution and emphysema progression from the Multi-Ethnic Study of Atherosclerosis using this analytical approach. We estimate weights for continuous exposure levels using two techniques: quantile binning of the exposure and a semiparametric model for the requisite conditional densities. The latter approach incorporates flexible machine learning methods. We find evidence for the harmful effects of ambient ozone pollution during study follow-up on the progression of emphysema, consistent with previously reported results. We find no evidence of effects of NOx during study follow-up. This investigation demonstrates that analyses based on marginal structural models are feasible in studies of the health effects of air pollution and may address possible sources of bias that traditional regression-based methods fail to address. Further investigation is warranted to understand differences between our findings and previously published results.

Recent works have established diffusion models as viable candidates for unpaired medical image translation, by incorporating supervision through non-diffusive models in an end-to-end fashion.. Existing models typically rely on cycleconsistent, two-sided, non-diffusive modules of relatively high computational cost where structure preservation is not explicit. In this work, we propose SPAD, a novel one-sided approach that explicitly preserves the source domain firstorder (gradient) orientations, while enabling flexible contrast adaptation between source and target domains. Our approach provides direct edge guidance for the diffusive module, leading to better structural preservation. Experiments performed on image synthesis tasks using BraTS, CERMEP, and Gold Atlas datasets demonstrate that our model improved translation fidelity and reduced training time by over 50% compared to SynDiff while keeping the same diffusion architecture. The reduction in computational cost allowed us to train 3D variants, which achieved consistently higher quantitative scores. Code will be publicly shared on GitHub upon acceptance.

Multiple Instance Learning (MIL) has been widely applied in histopathology to classify Whole Slide Images (WSIs) with slide-level diagnoses. While the ground truth is established by expert pathologists, the slides can be difficult to diagnose for non-experts and lead to disagreements between the annotators. In this paper, we introduce the notion of Whole Slide Difficulty (WSD), based on the disagreement between an expert and a non-expert pathologist. We propose two different methods to leverage WSD, a multi-task approach and a weighted classification loss approach, and we apply them to Gleason grading of prostate cancer slides. Results show that integrating WSD during training consistently improves the classification performance across different feature encoders and MIL methods, particularly for higher Gleason grades (i.e. worse diagnosis).

No-Reference Image Quality Assessment (NR-IQA) aims to predict perceptual image quality without reference images, a task made even more challenging in the Opinion-Unaware setting, where no ground-truth scores are available for supervision. In this work, we propose a simple yet effective Opinion-Unaware NR-IQA approach based on distortion functions capable of simulating the degradations observed in a target dataset. Instead of relying on pristine images to model ideal quality features, as in existing Opinion-Unaware NR-IQA methods, our method introduces a similarity function to explicitly relate the severity of distortions to their impact on perceived quality. By exploiting this relationship, we guide the learning of a quality assessment network by a relative constraint that is complemented by a natural ranking constraint induced by increasing distortion severity levels. On the LDCT-IQA dataset, our method achieves superior performance over other Opinion-Unaware NR-IQA approaches across all widely-used correlation metrics.

Thoracic Computed Tomography (CT) scans offer detailed insights into the intricate branching network of the airway tree, which is essential for understanding various respiratory diseases. Airway bifurcations, where airway branches split, are crucial landmarks for understanding lung physiology, disease mechanisms and lesion localization. Despite the significance of bifurcation analysis, a notable lack of datasets annotated for this task hinders the development of advanced automated specialized detection or segmentation tools. In this paper, we introduce BifDet, the first publicly-available dataset specialized for 3D airway bifurcation detection, filling a critical gap in existing resources. Our dataset comprises carefully annotated CT scans from the ATM22 open-access cohort with bifurcation bounding boxes covering the parent and daughter branches. As a use-case for demonstrating the potential of BifDet, we fine-tune and evaluate RetinaNet and DETR for 3D airway bifurcations detection on CT scans. We provide detailed pipelines, including preprocessing steps and specific implementation design choices. Results are detailed over various categories of minimal bounding box sizes to serve as baseline to benchmark future research.

Automated airway segmentation from lung CT scans is vital for diagnosing and monitoring pulmonary diseases. Despite advancements, challenges like leakage, breakage, and class imbalance persist, particularly in capturing small airways and preserving topology. We propose the Boundary-Emphasized Loss (BEL), which enhances boundary preservation using a boundary-based weight map and an adaptive weight refinement strategy. Unlike centerline-based approaches, BEL prioritizes boundary voxels to reduce misclassification, improve topology, and enhance structural consistency, especially on distal airway branches. Evaluated on ATM22 and AIIB23, BEL outperforms baseline loss functions, achieving higher topology-related metrics and comparable overall-based measures. Qualitative results further highlight BEL's ability to capture fine anatomical details and reduce segmentation errors, particularly in small airways. These findings establish BEL as a promising solution for accurate and topology-enhancing airway segmentation in medical imaging.

Accurately tracking neuronal activity in behaving animals presents significant challenges due to complex motions and background noise. The lack of annotated datasets limits the evaluation and improvement of such tracking algorithms. To address this, we developed SINETRA, a versatile simulator that generates synthetic tracking data for particles on a deformable background, closely mimicking live animal recordings. This simulator produces annotated 2D and 3D videos that reflect the intricate movements seen in behaving animals like Hydra Vulgaris. We evaluated four state-of-the-art tracking algorithms highlighting the current limitations of these methods in challenging scenarios and paving the way for improved cell tracking techniques in dynamic biological systems.

BACKGROUND:COPD is traditionally associated with pulmonary hypertension, but treatments targeting elevated pulmonary artery pressure in COPD have largely failed, possibly due to an incomplete understanding of subphenotypes of the disease. RESEARCH QUESTION:Are novel machine-learned CT emphysema subtypes associated with specific cardiac hemodynamic profiles? STUDY DESIGN AND METHODS:The Multi-Ethnic Study of Atherosclerosis (MESA) COPD Study recruited participants with and without COPD aged 50 to 79 years with ≥ 10 pack-years of smoking and without clinical cardiovascular disease, predominantly from MESA and a lung cancer screening cohort. COPD and COPD severity were defined by standard spirometric criteria. CT emphysema subtypes were defined by unsupervised machine learning in an independent study and labeled on chest CT scans. Hemodynamics were estimated on cardiac MRI using validated equations. Linear regression models were weighted by the inverse probability of sampling and adjusted for potential confounders. RESULTS:The mean age of the 300 participants was 68 ± 7 years, 60% were male, 28% currently smoked, and 47% had COPD (45% of mild and 41% of moderate severity). More severe COPD was associated with lower estimated pulmonary arterial wedge pressure (ePAWP; P trend = .02) and greater estimated pulmonary vascular resistance (ePVR; P trend = .03) but not estimated pulmonary artery pressure (ePAP; P trend = 0.83). Only the combined bronchitic-apical emphysema subtype was associated with greater ePAP (1.08 mm Hg/10%; 95% CI, 0.40-1.75). The diffuse emphysema subtype was associated with lower ePAWP (-0.49 mm Hg/10%; 95% CI, -0.75 to -0.24) and greater ePVR (0.36 Wood units/10%; 95% CI, 0.10-0.61). INTERPRETATION:In this case-control study of predominantly mild-moderate COPD, greater ePAP was specific to the combined bronchitic-apical emphysema subtype whereas the diffuse emphysema subtype and COPD severity were associated with lower ePAWP and greater ePVR. The CT emphysema subtype findings suggest more precise avenues to therapeutic interventions in cardiopulmonary dysfunction.

INTRODUCTION: Airway-to-lung ratio (ALR) is a quantitative index of airway tree caliber relative to lung size that has been used as an imaging biomarker of dysanapsis. Assessment of ALR on full-lung CT is repeatable and associated with incident chronic obstructive pulmonary disease and respiratory hospitalizations later in life. C4R is examining risk of severe COVID-19 and Long COVID, and includes five cohorts with cardiac CT but not full-lung CTs. This study evaluated the agreement of cardiac CT-and full-lung CT-assessed ALR, repeatability of cardiac CT-assessed ALR and associations of cardiac CT-assessed ALR with dysanapsis endpoints in one C4R cohort. METHODS: The Multi-Ethnic Study of Atherosclerosis (MESA) Lung Study included adults 48-85 years old free of clinical cardiovascular disease that underwent repeated cardiac CT scans at Exam 1 and paired cardiac and full-lung CT scans at Exam 5 (Figure).Full-lung ALR was computed as the mean of airway lumen diameters measured at 19 standard anatomic locations (trachea-to-subsegments) divided by the cube-root of total lung volume. Cardiac CT ALR was inferred using deep-learning models trained to estimate full-lung ALR from projections of lungs and airways segmented on cardiac CT. Agreement between cardiac CT and full-lung CT ALR, and repeatability of cardiac CT ALR were both assessed by intra-class correlation (ICC). Validation of cardiac CT ALR with dysanapsis endpoints consisted of computing the variance in forced expired volume in 1-second divided by forced vital capacity (FEV1/FVC) explained in the Exam 5 test-set, and Exam 1 associations with respiratory mortality and genotype-derived dysanapsis genetic risk. Regression models were adjusted for age, sex, height, race-ethnicity, income, educational attainment, cigarette smoking status, pack-years, second-hand smoke exposure (hours per week) and asthma diagnosis. RESULTS: Among n=377 Exam 5 test-set participants, the ICC of paired cardiac CT and full-lung CT ALR was 0.849. Among n=6,310 Exam 1 participants with repeated cardiac CT-assessed ALR, the ICC was 0.951. The cardiac CT ALR increment in FEV1/FVC variance explained was +8.7% (vs. full-lung CT ALR increment: +8.8%). A 1-SD decrement in cardiac CT ALR was associated with higher respiratory mortality (adjusted hazard ratio: 2.22, 95%CI 1.68-2.92) and higher dysanapsis genetic risk score (+0.14 SD dysanapsis genetic risk score [95%CI: 0.11-0.16]). CONCLUSION: Cardiac CT-assessed ALR is repeatable, agrees well with full-lung CT-assessed ALR and associates with airflow obstruction, respiratory mortality and dysanapsis genetic risk. Assessing ALR in cohorts with cardiac CT in C4R may facilitate research into the role of dysanapsis in COVID-19.

Various diseases including laminopathies and certain types of cancer are associated with abnormal nuclear mechanical properties that influence cellular and nuclear deformations in complex environments. Recently, microgroove substrates designed to mimic the anisotropic topography of the basement membrane have been shown to induce 3D nuclear deformations in various adherent cell types. Importantly, these deformations are different in myoblasts derived from laminopathy patients from those in cells derived from normal individuals. Here we assess the ability of a Variational Autoencoder (VAE) and a Gaussian Mixture Model (GMM) to cluster patches of nuclei of both wildtype myoblasts and myoblasts with laminopathy-associated mutations cultured on microgroove substrates, and we explore the impact of image processing parameters on clustering performance. We show that a standard VAE with GMM is able to cluster nuclei based on their morphologies and degrees of deformations and that these clusters correspond to either wildtype myoblasts or myoblasts with LMNA mutations. The current results suggest that combining deep learning techniques with microgroove substrates enables automatic classification of nuclear deformations and thus provides a promising approach for easy and rapid diagnosis of pathologies that involve abnormalities in nuclear deformation.

RATIONALE: Central airway tree topology is established in utero and varies in the general population. Visual airway tree assessment restricted to lower lobe segmental anatomy has identified absent and accessory airways in approximately 25% of adults and these variants are associated with higher chronic obstructive pulmonary disease (COPD) prevalence later in life, including in non-smoking study participants. This study applied unsupervised machine-learning to the full CT-resolved airway tree to discover novel quantitative airway tree subtypes (QATS) and evaluated the association of QATS with COPD endpoints. METHODS: The Multi-Ethnic Study of Atherosclerosis (MESA) Lung Study includes community-dwelling adults from six U.S. sites that underwent full-lung CT at suspended inspiration, spirometry, symptom questionnaires, genotyping and follow-up for respiratory hospitalization and/or death defined by primary diagnostic code. QATS were identified by clustering participant-level deep-learned representations of airway tree structures segmented from CT. We repeated clustering on five subsets of MESA Exam 5 (80% each) and quantified QATS reproducibility using the Rand Index. QATS differences in lung structure (mean airway lumen diameter, percent wall area, branch density, and geometric complexity), and COPD endpoints (forced expired volume in 1-sec / forced vital capacity [FEV1/FVC]<0.70, Medical Research Council [MRC] dyspnea rating >1, and rate of primary respiratory-related hospitalization or death) were assessed among all participants and among never smokers using regression models to adjust for age, sex, height, body mass index, race-ethnicity, total lung volume, cigarette smoking status and pack-years. RESULTS: Among 2,588 participants (mean±SD age: 69.4±9.3 years, 54% female, 51% ever smokers with 19±24 pack-years), four QATS clusters were identified (Table) and were 95% reproducible. Compared with QATS A (prevalence 30.3%; reference cluster), QATS B (prevalence 26.4%) was associated with higher prevalence of airflow obstruction, dyspnea and a higher rate of CLRD hospitalization or death among all participants and among never-smokers; QATS C (prevalence 22.6%) was associated with higher airflow obstruction prevalence but similar prevalence of dyspnea and CLRD events; and QATS D (prevalence 20.8%) was associated with higher airflow obstruction prevalence and percent of emphysema-like lung but similar dyspnea prevalence and rate of CLRD events. CONCLUSION: Unsupervised machine-learning identified four common quantitative airway tree subtypes (QATS) that differed with respect to COPD endpoints, including, as previously, participants who never smoked. These findings suggest that central airway tree topology varies considerably in the general population and is associated with distinct COPD phenotypes.

Despite advances with deep learning (DL), automated airway segmentation from chest CT scans continues to face challenges in segmentation quality and generalization across cohorts. We address this by integrating Curriculum Learning (CL) into segmentation networks, using ad-hoc complexity scores derived from CT scans and ground-truth tree features. We specifically investigate few-shot domain adaptation, targeting scenarios where manual annotation of a full fine-tuning dataset is prohibitively expensive. Results are reported on two large open-cohorts (ATM22 and AIIB23) with high performance using CL for full training (Source domain) and fewshot fine-tuning (Target domain), but also some insights on potential detrimental effects when using a classic Bootstrapping scoring function or indadequate scan sequencing.

Chronic obstructive pulmonary disease (COPD) and pulmonary emphysema were the third-leading cause of death globally in 2019. Besides alpha-1 antitrypsin deficiency, emphysema lacks disease-modifying medications due to its heterogeneity and there remains a need to further investigate emphysema subtypes to advance its diagnosis and treatment. Recent unsupervised machine learning of the texture and anatomical location of emphysema on computed tomography (CT) images of the lung defined six quantitative CT emphysema subtypes uniquely associated with specific gene variants, environmental exposures, symptoms, physiology and clinical events. We aim to identify associations of plasma proteins with each CT emphysema subtype cross-sectionally in a multi-ethnic general population sample. MESA is a prospective cohort study that recruited 6,814 participants from six communities who self-identified as White, Black, Hispanic, or Asian race/ethnicity, ages 45-84 years old and free of clinical cardiovascular disease in 2000-02. The MESA Lung Study performed CT scans among 3,137 participants in 2010-12, which were labeled with previously described (PMID: 37268414) CT emphysema subtypes. Proteomics was assessed using SomaScan version 4.1 assay (7,596 plasma proteins). Linear regression models were adjusted for age, height, weight, sex, race, educational attainment, smoking status, pack years, urine cotinine level, body-mass index (BMI), estimated glomerular filtration rate (eGFR), CT scanner manufacturer and other CT emphysema subtypes. Results were adjusted to reflect a False Discovery Rate (FDR) p-value of less than 5% using the Benjamini-Hochberg method. Among 2,343 participants with both CT and proteomic data (mean age 69.5 years, 52.3% women, 23.7% with COPD), we identified statistically significant associations of proteins with the vanishing lung subtype (42 proteins; top hit Natural cytotoxicity triggering receptor 3 ligand 1, q = 2.9[asterisk]10-9), restrictive combined pulmonary fibrosis/emphysema (CPFE) subtype (11 proteins; top hit Integrin alpha-2/b1, q=1.0[asterisk]10-5), the combined bronchitis-apical emphysema subtype (9 proteins; top hit Desmoglein-2, q=8.8[asterisk]10-13), and the diffuse emphysema subtype (4 proteins; top hit Pulmonary surfactant-associated protein D, q=5.7[asterisk]10-3). No significant proteins were associated with the obstructive CPFE or senile subtypes after FDR correction. Most proteins were uniquely associated with one CT emphysema subtype, while a few proteins were associated with more than one subtype. This study identified specific proteins for different CT emphysema subtypes, many of which have been related less specifically to emphysema, including pulmonary surfactant production, anti-inflammatory pathways, and mitochondrial deregulation, amongst others. Validation of these proteomic associations can provide insight into further pathway assessment, clinical biomarker discovery, and targeted prevention of emphysema in the general population.

Introduction: Breast cancer is the most common malignancy and the second-leading cause of cancer mortality in women. Breast density, defined as the extent of fibroglandular tissue within the breast, is assessed semi-quantitatively on a four-grade scale, from mostly fatty (grade 1) to extremely dense (grade 4), with a severalfold increase in prospective cancer risk between the lowest and highest grades. Millions of chest CT studies are performed annually in patients eligible for routine mammographic screening, representing a significant opportunity for incidental density reporting and risk stratification. While automated density grading on mammography is robust and widely deployed, assessment on CT remains limited by comparison. To facilitate the efficient and standardized assessment of breast density on chest CT, we train and evaluate a 3D convolutional neural network (CNN) model to predict breast density grade from chest CT volumes using a retrospective dataset of patients at Columbia University Irving Medical Center.Methods: A retrospective, IRB-approved chart review identified female patients at CUIMC who underwent both mammography and chest CT acquired less than one year apart from January 2010 through March 2019. All cases were reviewed by a faculty radiologist, recording both CT breast density and whether the entire breast was present within the scanner field of view (FOV). All CT scans with or without IV contrast, with axial soft-tissue reconstruction and no breast malignancy, prosthesis or prior mastectomy were included. CT volumes were downsampled to isotropic 2.5 mm resolution and 80x80x80 voxel volumes of interest (VOIs) were extracted containing each imaged breast. Scans were randomly allocated to training, validation, and test sets in a 3:1:1 ratio with stratification by density grade.Our 3D CNN model consists of five blocks containing 3D convolutional layers with ReLU activation, convolutional block attention modules (CBAM), and batch normalization, interleaved with pooling operations. Rank-consistent ordinal regression (CORN) was applied to the final fully-connected layer to predict the density grade for each image. The model was trained to minimize ordinal cross-entropy with L2 regularization, using training batches balanced across density grades and train-time augmentation, over 60,000 training steps with early stopping conditioned on validation ordinal cross-entropy. Patient-level predictions were generated on the test cases by averaging predictions over the right- and left-breast VOIs.Results: Of 503 scans which met the inclusion criteria, 86 (17.1%) were assigned breast density grade 1, 244 (48.5%) were grade 2, 131 (26.0%) were grade 3, and 42 (8.3%) were grade 4. One or both breasts were partially outside the FOV in 134 scans (26.6%). On prediction of high (grades 3-4) versus low breast density, the optimized model achieved test-set accuracy of 90.1%, receiver-operating characteristic AUC of 0.935, and Cohen kappa of 0.787 with respect to radiologist-assessed ground truth, comparable to reported values for inter-radiologist agreement. One-vs-all accuracy for breast density grade was 67.3%, with 98.0% of predicted grades at most one level off from the ground truth. ROC AUC and Cohen kappa improved to 0.955 and 0.873, respectively, on the subset of test cases (N=80) where both breasts were contained within the FOV.Conclusions: We have introduced a 3-D CNN model for automated ordinal regression of breast density on chest CT using a retrospective clinical CT dataset, with performance relative to ground-truth comparable to published inter-reader agreement among trained radiologists. Estimating breast density grade independently of threshold-based volumetric percent density is a direct precursor to deep-learning on chest CT to improve on image-based breast cancer risk stratification. Citation Format: Artur Wysoczanski, Elsa D. Angelini, Sachin S. Jambawalikar, Andrew F. Laine, Mary M. Salvatore. Automated Breast Density Assessment on Chest CT with a Deep-Learned 3D Ordinal Regression Model [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2024; 2024 Dec 10-13; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(12 Suppl):Abstract nr P2-06-29.

The ratio of airway tree lumen to lung size (ALR), assessed at full inspiration on high resolution full-lung computed tomography (CT), is a major risk factor for chronic obstructive pulmonary disease (COPD). There is growing interest to infer ALR from cardiac CT images, which are widely available in epidemiological cohorts, to investigate the relationship of ALR to severe COVID-19 and post-acute sequelae of SARS-CoV-2 infection (PASC). Previously, cardiac scans included approximately 2/3 of the total lung volume with 5-6x greater slice thickness than high-resolution (HR) full-lung (FL) CT. In this study, we present a novel attention-based Multi-view Swin Transformer to infer FL ALR values from segmented cardiac CT scans. For the supervised training we exploit paired full-lung and cardiac CTs acquired in the Multi-Ethnic Study of Atherosclerosis (MESA). Our network significantly outperforms a proxy direct ALR inference on segmented cardiac CT scans and achieves accuracy and reproducibility comparable with a scan-rescan reproducibility of the FL ALR ground-truth.

Rationale: Increased risk of coronavirus disease (COVID-19) hospitalization and death has been reported among patients with clinical lung disease. Objective: To test the association of objective measures of prepandemic lung function and structure with COVID-19 outcomes in U.S. adults. Methods: Prepandemic obstruction (FEV1/FVC < 0.70) and restriction (FEV1/FVC ⩾ 0.7, FVC < 80%) were defined based on the most recent spirometry exam conducted in 11 prospective U.S. general population-based cohorts. Severe obstruction was classified by FEV1 < 50%. Percentage emphysema, percentage high-attenuation areas, and interstitial lung abnormalities were defined on computed tomography in a subset. Incident COVID-19 was ascertained via questionnaires, serosurvey, and medical records from 2020 to 2023 and classified as severe (hospitalized or fatal) or nonsevere. Cause-specific hazard models were adjusted for sociodemographics, anthropometry, smoking, comorbidities, and COVID-19 vaccination status. Measurements and Main Results: Among 29,323 participants (mean age, 67 yr), there were 748 severe incident COVID-19 cases over median follow-up of 17.3 months from March 1, 2020. Greater hazards of severe COVID-19 were associated with severe obstruction (vs. normal; adjusted hazard ratio [aHR], 2.11; 95% confidence interval [CI], 1.02-1.27), restriction (vs. normal; aHR, 1.40; 95% CI, 1.12-1.76), and percentage emphysema (highest vs. lowest quartile; aHR, 1.64; 95% CI, 1.03-2.61), but not greater high-attenuation areas or interstitial lung abnormalities. COVID-19 vaccination provided greater absolute risk reduction in these groups. Results were similar in participants without smoking, obesity, or clinical cardiopulmonary disease. Conclusions: Prepandemic severe spirometric obstruction, spirometric restriction, and greater percentage emphysema lung on computed tomography were associated with risk of severe COVID-19. These findings support enhanced COVID-19 risk mitigation for individuals with impaired lung health and warrant further mechanistic studies on interactions of lung function, structure, and vulnerability to acute respiratory illnesses.