Chronic Obstructive Pulmonary Disease: Thoracic CT Texture Analysis and Machine Learning to Predict Pulmonary...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Fixed airflow limitation and ventilation heterogeneity are hallmarks of chronic obstructive pulmonary disease (COPD).
• Conventional CT scans offer structural data but lack direct functional lung assessment.
• Accurate assessment of lung function is crucial for managing COPD.

Purpose of the Study:

• To develop and validate a CT texture analysis and machine learning algorithm for predicting lung ventilation heterogeneity in COPD.
• To establish a non-invasive method for assessing ventilation defects using CT.
• To correlate CT-derived ventilation predictions with established MRI and pulmonary function measures.

Main Methods:

• A prospective study involving CT texture analysis and machine learning model development.
• Hyperpolarized helium-3 MRI served as the ground truth for ventilation mapping.
• Quantitative imaging features were extracted from CT, optimized, and used to train a quadratic support vector machine classifier.

Main Results:

• The developed machine learning model achieved 88% accuracy in predicting ventilation maps compared to HP 3He MRI.
• Model-predicted ventilation defect percentage showed strong correlation with MRI (r=0.90, P<.001).
• Predictions correlated significantly with FEV1, FEV1/FVC ratio, diffusing capacity, and quality of life scores.

Conclusions:

• CT texture analysis combined with machine learning can reliably predict lung ventilation heterogeneity in COPD patients.
• This AI-driven approach offers a promising non-invasive tool for assessing lung function in COPD.
• The model's strong correlation with MRI and clinical measures supports its clinical utility.