Adaptive Feature Selection Guided Deep Forest for COVID-19 Classification With Chest CT.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Chest computed tomography (CT) is crucial for diagnosing coronavirus disease-19 (COVID-19).
• The global COVID-19 pandemic necessitates efficient computer-aided diagnosis (CAD) tools to support clinicians.
• Automated classification of COVID-19 from CT images can reduce diagnostic workload.

Purpose of the Study:

• To propose an Adaptive Feature Selection guided Deep Forest (AFS-DF) model for COVID-19 classification using chest CT images.
• To leverage deep forest models for high-level feature representation from limited data.
• To enhance classification accuracy by adaptively incorporating feature selection.

Main Methods:

• Extraction of location-specific features from chest CT images.
• Utilizing a deep forest model to learn high-level feature representations.
• Implementing an adaptive feature selection method integrated with the classification model to reduce feature redundancy.

Main Results:

• The AFS-DF model achieved high performance metrics: 91.79% accuracy, 93.05% sensitivity, 89.95% specificity, 96.35% AUC, 93.10% precision, and 93.07% F1-score.
• Evaluated on a dataset of 1495 COVID-19 and 1027 community-acquired pneumonia (CAP) cases.
• Demonstrated superior performance compared to four other widely used machine learning methods for COVID-19 vs. CAP classification.

Conclusions:

• The proposed AFS-DF model offers a robust and effective approach for classifying COVID-19 from chest CT images.
• Adaptive feature selection integrated with deep forest learning improves classification performance.
• This AI-driven method shows significant potential in assisting the clinical diagnosis of COVID-19.