Deep learning based CT images for lung function prediction in patients with chronic obstructive pulmonary disease

Area of Science:

• Pulmonary Medicine
• Radiology
• Artificial Intelligence

Background:

• Chronic obstructive pulmonary disease (COPD) is a leading cause of death and morbidity globally.
• Pulmonary function tests (PFTs) are crucial for COPD diagnosis but can be time-consuming.
• Deep learning (DL) methods for COPD detection using CT scans require improved predictive performance due to disease heterogeneity.

Purpose of the Study:

• To develop a DL-based multimodal feature fusion model for accurate PFT parameter estimation from chest CT images.
• To validate the performance of the developed model in predicting key PFT metrics.

Main Methods:

• A retrospective study utilized chest CT scans and PFT data from 3108 participants.
• A multimodal feature fusion model based on a multilayer perceptron (MLP) was developed and validated using 10-fold cross-validation.
• Evaluation metrics included Mean Absolute Error (MAE), Mean Squared Error (MSE), and Pearson correlation coefficient (r); Bland-Altman plots assessed consistency.

Main Results:

• The MLP model demonstrated strong correlations between predicted and measured PFT parameters (FEV1, FVC, FEV1/FVC, FEV1%).
• MAE, MSE, and r values indicated good predictive accuracy for FEV1 (0.34, 0.20, 0.84) and FVC (0.42, 0.31, 0.81).
• Bland-Altman analysis confirmed good consistency between estimated and actual PFT values.

Conclusions:

• The MLP-based multimodal feature fusion model shows promise for real-time PFT parameter prediction in COPD patients.
• Pre-bronchodilator measurements were used, which may influence results; future studies should use post-bronchodilator measurements.
• Further research is needed to align with clinical standards and enhance diagnostic capabilities.