Estimating lung function from computed tomography at the patient and lobe level using machine learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonary Medicine

Background:

• Computed Tomography (CT) can aid in screening, diagnosis, and staging of restrictive pulmonary diseases.
• Estimating lung function per lobe from CT is crucial for surgical risk assessment and lung volume reduction procedures.

Purpose of the Study:

• To automatically estimate Pulmonary Function Test (PFT) results from CT scans.
• To disentangle the individual contribution of pulmonary lobes to a patient's lung function.

Main Methods:

• Proposed I3Dr, a deep learning architecture for estimating global image measures and individual part contributions.
• Applied I3Dr to CT scans, utilizing lobe-level and patient-level models trained with patient lung function data.
• Trained and evaluated I3Dr on a large dataset of 8,433 CT volumes for training, 1,775 for validation, and 1,873 for testing.

Main Results:

• Demonstrated model viability by showing implicit learning of individual digit values from image sums.
• Successfully estimated lobe-level quantities like COVID-19 severity, pulmonary volume (PV), and functional pulmonary volume (FPV) from CT.
• Achieved mean absolute errors of 0.377 L for FEV1, 0.297 L for FVC, and 2.800 mL/min/mm Hg for DLCO estimates.

Conclusions:

• I3Dr effectively estimates global image measures and individual component contributions.
• Offers a promising method for PFT estimation from CT and lobe-specific lung function analysis.
• Potential to enhance CT's role in diagnosing and managing restrictive lung diseases and in surgical planning.