Multimodal data-driven multitask learning for enhanced identification and classification of chronic obstructive...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Chronic obstructive pulmonary disease (COPD) is a leading global cause of mortality.
• Accurate COPD identification and phenotyping are crucial for effective patient management.
• Current diagnostic methods may benefit from advanced computational approaches.

Purpose of the Study:

• To develop and evaluate a multimodal multi-task learning framework for automated COPD detection and classification.
• To integrate computed tomography (CT) imaging with clinical data for enhanced diagnostic performance.
• To assess the model's efficacy in differentiating COPD from non-COPD and various COPD subtypes.

Main Methods:

• A retrospective multi-task learning model was developed, fusing CT scans and clinical data from 2320 patients.
• Model performance was evaluated using metrics such as concordance correlation coefficient (CCC), mean absolute error (MAE), area under the receiver operating characteristic curve (AUC), accuracy (ACC), precision, recall, and F1-score.
• Generalizability was tested by replicating experiments on three distinct backbone networks.

Main Results:

• The optimal model achieved a maximum CCC of 0.75 for forced vital capacity (FVC) and 0.77 for forced expiratory volume in one second (FEV1).
• For binary COPD classification, the multi-task learning model reached an AUC of 0.88 and an ACC of 0.83.
• In ternary classification (COPD/PRISm/Normal), the highest AUC was 0.87 with an ACC of 0.79.

Conclusions:

• Multitask-learning models integrating CT images and clinical variables surpass single-task models for COPD identification and classification.
• This approach supports evidence-based clinical decision-making in respiratory medicine.
• The findings advance the delivery of precision medicine for COPD management.