Deep learning-based meta-classifier approach for COVID-19 classification using CT scan and chest X-ray images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• Convolutional Neural Network (CNN)-based models are common for COVID-19 classification using chest X-ray (CXR) and computed tomography (CT) scans.
• Existing methods often use limited datasets, potentially hindering generalization to unseen COVID-19 cases.
• Model generalization is crucial for reliable performance on diverse, real-world medical data.

Purpose of the Study:

• To propose a large-scale learning approach for improved COVID-19 classification.
• To develop a stacked ensemble meta-classifier with deep learning-based feature fusion for enhanced accuracy.
• To address the generalization limitations of current COVID-19 diagnostic models.

Main Methods:

• Feature extraction from EfficientNet-based models' penultimate layer (global average pooling).
• Dimensionality reduction using kernel principal component analysis (PCA).
• Feature fusion followed by a two-stage stacked ensemble meta-classifier (Random Forest, SVM, and Logistic Regression).

Main Results:

• The proposed model demonstrated superior performance compared to existing CNN-based pretrained models.
• The method achieved high accuracy in classifying COVID-19 from large-scale CT and CXR datasets.
• The model showed strong generalization capabilities on unseen medical imaging data.

Conclusions:

• The developed stacked ensemble and feature fusion model offers a robust solution for COVID-19 classification.
• This approach significantly improves upon existing methods, particularly in handling unseen data.
• The model shows potential as a valuable tool for point-of-care diagnosis in clinical settings.