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

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Differentiating asthma from chronic obstructive pulmonary disease (COPD) is clinically challenging.
• Asthma-COPD overlap (ACO) lacks standardized diagnostic criteria, complicating patient management.
• Current diagnostic methods may not fully capture the complexities of these overlapping respiratory conditions.

Purpose of the Study:

• To develop and evaluate a deep transfer learning framework for chest CT image segmentation.
• To support imaging-assisted classification of asthma-COPD overlap (ACO).
• To assess the performance of the proposed framework in a single-center cohort.

Main Methods:

• A deep transfer learning model was employed for chest CT image segmentation.
• The framework was trained and tested on a cohort of patients with asthma, COPD, and ACO.
• Model performance was quantified using classification accuracy and segmentation Dice similarity coefficient.
• Clinical data including lung function, biomarkers, and symptom scores were summarized for cohort characterization.

Main Results:

• The proposed framework achieved a high ACO classification accuracy of 93.21%.
• Performance surpassed existing methods like NUS-PSL (85.43%) and PRE-1000C (86.92%).
• Segmentation accuracy was evaluated against expert annotations.

Conclusions:

• The developed imaging-assisted classification framework shows promise for identifying ACO.
• This proof-of-concept study highlights the potential utility of AI in respiratory disease diagnosis.
• Further multi-center validation is necessary to confirm generalizability and stability.