Lung radiomics features for characterizing and classifying COPD stage based on feature combination strategy and...

Area of Science:

• Pulmonary Medicine
• Medical Imaging
• Artificial Intelligence

Background:

• Computed tomography (CT) is crucial for characterizing chronic obstructive pulmonary disease (COPD).
• Radiomics features from CT scans are underutilized for COPD characterization and classification.
• Accurate COPD staging is essential for effective patient management.

Purpose of the Study:

• To investigate the utility of lung radiomics features for COPD stage classification.
• To compare the performance of machine learning (ML) models with deep learning (CNNs) for COPD classification.
• To develop and validate a radiomics-based strategy for improved COPD diagnosis.

Main Methods:

• Extracted 1316 radiomics features from chest CT images.
• Selected 19 optimal features using Lasso regression.
• Developed two radiomics combination features: Radiomics-FIRST and Radiomics-ALL.
• Utilized a multi-layer perceptron (MLP) classifier for COPD stage classification.

Main Results:

• ML methods based on selected radiomics features outperformed classic CNNs on high-resolution CT images.
• The MLP classifier with 19 selected radiomics features and Radiomics-ALL achieved 0.83 accuracy, 0.83 precision, 0.83 recall, 0.82 F1-score, and 0.95 AUC.
• The proposed radiomics combination strategy improved classifier performance by 12% overall.

Conclusions:

• ML methods using lung radiomics features are more suitable and interpretable for COPD classification than classic CNNs.
• Feature selection using Lasso and radiomics combination strategies enhance classification accuracy.
• Radiomics analysis offers a promising, data-driven approach for objective COPD staging and diagnosis.