Differentiating COPD and asthma using quantitative CT imaging and machine learning.

Area of Science:

• Pulmonary Medicine
• Radiology
• Medical Imaging Analysis

Background:

• Chronic obstructive pulmonary disease (COPD) and asthma share some computed tomography (CT) imaging features, necessitating advanced methods for differentiation.
• Accurate distinction between COPD and asthma is crucial for appropriate patient management and treatment strategies.

Purpose of the Study:

• To identify an optimal subset of CT imaging features for machine learning-based differentiation of COPD and asthma.
• To evaluate the performance of machine learning models in classifying COPD and asthma using selected CT features.

Main Methods:

• Recruitment of COPD and asthma patients from Heidelberg University Hospital.
• Extraction of 93 CT imaging features, including low-attenuating area (LAA₉₅₀), low-attenuation cluster (LAC) count, airway parameters (Pi10, TAC), and airway dimensions.
• Application of hybrid feature selection and support vector machine learning for classification.

Main Results:

• A model using all CT features achieved 80% accuracy and 81% F1 score, with key features being LAA₉₅₀, airway perimeters, TAC, and LAC count.
• A model focused solely on CT airway features yielded 66% accuracy and 68% F1 score.
• The most discriminative features included LAA₉₅₀, outer and inner airway perimeters, TAC, and LAC total hole count.

Conclusions:

• Machine learning, utilizing a select set of seven CT features, can differentiate between COPD and asthma with moderate to high accuracy.
• This CT-based machine learning approach offers a promising tool for distinguishing these two respiratory diseases.