Comparison of Feature Selection Methods and Machine Learning Classifiers for Predicting Chronic Obstructive Pulmonary...

Area of Science:

• Radiology and Medical Imaging
• Pulmonology
• Machine Learning in Healthcare

Background:

• Texture-based radiomics analysis of lung CT images shows promise in predicting chronic obstructive pulmonary disease (COPD) status.
• Current machine learning approaches vary, leading to uncertainty regarding optimal performance.

Purpose of the Study:

• To compare common feature selection and classification methods for COPD prediction.
• To identify the optimal machine learning models for classifying COPD status in a mild, population-based cohort.

Main Methods:

• CT images from the Canadian Cohort Obstructive Lung Disease (CanCOLD) study were pre-processed.
• 95 texture features were extracted, and 17 feature selection methods combined with 9 classifiers were tested.
• Data cleaning, including outlier and highly correlated feature removal, was evaluated. Model performance was assessed using the area under the curve (AUC).

Main Results:

• A total of 1204 participants (602 COPD, 602 no COPD) were analyzed.
• The highest AUC of 0.78 (0.73, 0.84) was achieved after data cleaning and feature selection using Elastic Net with the Linear-SVM classifier.
• No significant differences in sex or BMI were observed between groups.

Conclusions:

• Elastic Net and Linear-SVM represent the optimal combination for radiomics-based COPD prediction in this population-based cohort.
• This study provides a validated optimal model for COPD status classification using lung CT radiomics.