Ching-Long Lin

Rationale This work aims to enhance multi-disease lung classification and phenotyping from CT scans through the development of a contrastive learning framework that generates disease-discriminative image embeddings. Methods We recruited 1,187 participants across four groups: asthma (n = 315), COPD (n = 355), post-COVID-19 (n = 375), and healthy controls (n = 142). Imaging analyses included 1,003 single scan cases and a longitudinal subgroup of 92 participants (33 asthma, 59 post-COVID-19) with a follow-up scan. We trained a novel contrastive model with an expert-conditioned routing network to encourage disease-aware representation paths and adaptive temperature scaling to stabilize the contrastive learning objective. Classification was assessed by macro area under the ROC curve (macro-AUC). Embedding quality was examined via unsupervised k-means clustering and visualization (t-SNE). Quantitative CT (qCT) metrics differences across clusters were tested with Welch’s ANOVA. We also evaluated whether embeddings could track patient-level progression/regression in the longitudinal subgroup and predict qCT metrics via regression (R²). Results The model achieved a 4-class macro-AUC score of 0.893. Embeddings showed clear between-class separation on t-SNE and yielded clusters with significantly different qCT profiles by Welch’s ANOVA. In the longitudinal subgroup, patient trajectories in embedding space reflected clinical regression/improvement. Embedding-based prediction of qCT metrics was strongest for the Jacobian metric (R²=0.61). Conclusion An expert-conditioned contrastive learning approach produced robust, disease-discriminative CT embeddings that support accurate classification, unsupervised phenotyping, longitudinal tracking, and qCT metrics prediction. This representation may facilitate trial stratification and monitoring of disease course across asthma, COPD, and post-COVID-19 populations.FIGURE. Contrastive learning pipeline for lung disease classification. T and T’ denotes the two types of transformation and X and X’ are the two view of transformed images. The learned embeddings support supervised prediction and downstream analyses, including k-means clustering and feature-space evaluation and are later used for longitudinal trajectory tracking and qCT metrics prediction. This abstract is funded by: NIH Grant R01-HL168116, P30 ES005605, and ED P116S21000

Rationale Long-term cardiopulmonary consequences of post-acute sequelae of SARS-CoV-2 infection (PASC, “Long COVID”) remain poorly characterized beyond two years, particularly using quantitative CT (qCT) and computational fluid and particle dynamics (CFPD) metrics. Methods We prospectively studied 80 adults with prior COVID-19 (81% infected pre-Alpha; unvaccinated at infection) who completed inspiratory and expiratory chest CT and spirometry at approximately 5 months (Visit-1, V1) and again at 3-4 years (Visit-2, V2). Seventy-eight healthy adults served as controls. At V2, post-COVID-19 participants completed the St. George’s Respiratory Questionnaire (SGRQ), Leicester Cough Questionnaire (LCQ), Fatigue Severity Scale (FSS), and modified MRC Dyspnea Scale (mMRC). Quantitative CT (qCT) metrics included airway diameter and wall thickness, functional small-airway disease percentage (fSAD%), ground-glass opacity percentage (GGO%), and an adaptive multiple-feature method (AMFM)-based “Bronchovascular%,” quantifying bronchovascular texture. A whole-lung CFPD model was applied to assess airway resistance. Statistical analyses included ANOVA with Tukey post hoc tests, Mann-Whitney U tests, Spearman correlations, and logistic regression. Results Group-average spirometry values were within normal limits at V1 and V2 and did not differ from controls. Questionnaires at V2 (SGRQ, LCQ, and mMRC) indicated that participants remained symptomatic after 3-4 years. From V1 to V2, fSAD% and GGO% decreased, suggesting resolution of small-airway dysfunction and parenchymal opacities. In contrast, structural remodeling persisted: (1) Bronchovascular% remained elevated at V2 compared with controls. (2) Airway metrics showed smaller normalized hydraulic diameters and thicker walls than controls at both time points, and although airway resistance decreased from V1 to V2, it remained higher than in healthy controls. (3) Vessel metrics demonstrated a redistribution toward larger vessels with a loss of small-vessel volume, which correlated with dyspnea, fatigue, chest pain, and cardiovascular comorbidities. Conclusion In post-COVID-19 subjects followed for 3-4 years after recovery, spirometry normalized, but qCT metrics revealed persistent airway-vascular remodeling, characterized by elevated Bronchovascular% and complementary airway and vessel size shifts. These structural changes were associated with worse symptoms and reduced quality of life. AMFM-based Bronchovascular% may serve as a candidate imaging biomarker for long-term PASC burden. Funding Sources NIH Grant R01-HL168116, P30 ES005605, and ED P116S21000 This abstract is funded by: National Institutes of Health: Department of Education

Rationale This study introduces a new segmentation model for expiratory CT images by integrating a novel attention gate with a U-Net framework to improve airway continuity and completeness. Methods We analyzed data from 120 subjects across four cohorts: asthma, COPD, post-COVID-19, and healthy controls. Each cohort includes 25 subjects with residual volume (RV) scans, which were used for training, validation, and independent testing. We proposed a 3D segmentation framework incorporating an attention gate model, termed Averaged Multi-Gaussian Response (AMGR), integrating with the U-Net architecture, specifically optimized for expiratory airway segmentation. The proposed model was compared against two state-of-the-art methods, which are Fuzzy Attention Neural Network (FANN) and standard U-Net, to evaluate its performance. Quantitative assessments were conducted using Dice, Precision, and Recall metrics. Results The proposed model achieved a Dice score of 0.9629. It outperformed two baseline models: FANN (0.9084) and U-Net (0.9477). In terms of Precision, the proposed model had 0.9621, indicating superior segmentation performance compared with FANN (0.8370) and U-Net (0.9397). While FANN reported the highest Recall (0.9947), the proposed model showed a balanced trade-off between Precision and Recall (0.9647). Conclusion The AMGR approach stabilizes feature fusion by suppressing peripheral noise and enhancing distal airway continuity. The proposed model achieved a high Dice score, while maintaining balanced Precision and Recall for the multi-disease cohort. Overall, the model demonstrated robust noise suppression and improved airway continuity in the RV scans. This abstract is funded by: NIH

Rationale Accurately predicting bronchodilator performance remains a major challenge in inhaled therapeutics. We developed a coupled computational fluid and particle dynamics (CFPD) and physiologically based pharmacokinetics/pharmacodynamics (PBPK/PD) framework that leverages subject-specific, CT-derived airway geometries and regional ventilation to predict lung exposure and bronchodilator response, quantified as the change in Forced Expiratory Volume in 1 second (ΔFEV1). Methods Airway trees were extracted from CT scans at full inspiration and propagated distally using a volume-filling algorithm to generate a generation-resolved 1D airway tree, including tracheobronchial and alveolar regions. Two cohorts were included in this study: asthmatic (n = 6) and healthy (n = 10) subjects. CFPD simulations predicted subject-specific airflow and drug deposition by generation (G) for inhalation of 400 μg albuterol under two device/inhalation profiles: metered-dose inhaler (MDI)/(slow and deep, SD) and dry powder inhaler (DPI)/(quick and deep, QD). Deposited doses were input into the PBPK model at the epithelial lining fluid (ELF), coupled to whole-body distribution and blood clearance, to predict plasma and tissue concentrations. The PD model was then implemented to predict ΔFEV1 at the sub-epithelium effect site. Results Validations against literature data showed good agreement for plasma and ELF concentrations following albuterol inhalation via MDI, as well as for the ΔFEV1 time course over 24 hours. Two key patterns emerged: (i) SD inhalations shifted the dose to the alveolar region, sustaining high drug concentrations in the distal ELF, whereas QD inhalation deposited albuterol in lung generations G2-G6; (ii) asthmatic subjects exhibited higher resistance vs. healthy subjects in G5-G10, amplifying the effect of short-acting β2-agonist (SABA) dilation and explaining larger early ΔFEV1. PD model parameters, fitted by least-squares to the observed ΔFEV1 experimental data, resulted in a half-maximal effective concentration (EC50 =1.1±0.7 nM) and Hill slope (n = 1.6±0.1) with R2= 0.98. This framework mechanistically links inhaler type to clinical benefit and supports patient-specific treatment in asthmatics. Conclusion A subject-specific CFPD model integrated with a PBPK/PD framework was developed to predict inhaled albuterol exposure in plasma and ELF, as well as clinical response (ΔFEV1), based on CT-derived airway structure, lung ventilation, and inhalation profile. This approach has the potential to guide inhaler selection, dosing schedules, and disease-informed inhalational treatment planning across a range of respiratory diseases. Funding Sources NIH Grant R01-HL168116, P30 ES005605, and ED P116S21000 This abstract is funded by: NIH and Department of Education