Lobar Lung Density Embeddings with a Transformer encoder (LobTe) to predict emphysema progression in COPD

Area of Science:

• Pulmonary Medicine
• Radiology
• Artificial Intelligence in Healthcare

Background:

• Emphysema, characterized by alveolar wall destruction, leads to heterogeneous lung destruction and decreased gas exchange.
• Emphysema progression in Chronic Obstructive Pulmonary Disease (COPD) is linked to worsening symptoms and increased mortality.
• Current CT scan diagnostics for emphysema lack predictive capabilities for disease evolution.

Purpose of the Study:

• To develop and validate a novel prognostic lobe-based transformer (LobTe) model for predicting emphysema progression.
• To capture the spatial heterogeneity and complexity of emphysema using CT imaging data.
• To enhance understanding of COPD by predicting lung density evolution based on %LAA-950 measurements.

Main Methods:

• Utilized a transformer encoder with lobe embedding fingerprints to maintain global attention based on lobe positions.
• Trained and tested the LobTe model on a dataset of 4,612 smokers (COPDGene participants) with 5-year follow-up data.
• Evaluated model performance using %LAA-950 measurements for predicting lung density changes.

Main Results:

• The LobTe model demonstrated effectiveness in predicting lung density evolution over five years.
• Achieved a Root Mean Squared Error (RMSE) of 2.957%, a correlation coefficient (ρ) of 0.643, and R² of 0.36 on test data.
• Image embeddings from baseline CT scans successfully forecasted emphysema progression by quantifying lung tissue loss.

Conclusions:

• The LobTe model shows potential for early identification of patients at risk of emphysema progression.
• Predicting lung density changes from baseline CT scans offers a novel approach to managing COPD.
• The study highlights the utility of AI-driven analysis of medical imaging for disease prognostication.