Accurate Asthma-COPD Overlap Classification via Deep Transfer Learning in Medical Image Segmentation

Differentiating asthma from chronic obstructive pulmonary disease (COPD) remains challenging in clinical practice, and asthma-COPD overlap (ACO) lacks universally accepted diagnostic criteria. In this study, we propose a chest computed tomography (CT) image segmentation framework based on deep transfer learning to support imaging-assisted ACO-related classification as a proof-of-concept approach. Experiments were performed in a single-center cohort of patients with asthma, COPD, and ACO. Model performance was evaluated using classification accuracy and segmentation Dice similarity coefficient against expert-annotated reference masks. In addition, lung function parameters, inflammatory biomarkers, and ACT/CAT scores were summarized to characterize cohort profiles and assist clinical interpretation; these variables were not predicted by the AI model. The proposed approach achieved the highest ACO classification accuracy (93.21%), outperforming NUS-PSL (85.43%) and PRE-1000C (86.92%). These findings suggest potential utility for imaging-assisted ACO-related classification within this internal single-center evaluation. Further multi-center external validation and robustness analyses are warranted before conclusions regarding stability and generalizability can be made.