Automatic COVID-19 Detection Using Exemplar Hybrid Deep Features with X-ray Images

Area of Science:

• Medical imaging and artificial intelligence
• Computer-aided diagnosis of respiratory diseases

Background:

• Accurate detection of COVID-19 and pneumonia from medical images is crucial.
• Existing methods often require complex feature engineering or lack generalizability.

Purpose of the Study:

• To propose a novel COVID-19 detection system using exemplar and hybrid fused deep features from X-ray images.
• To enhance the accuracy of respiratory disorder detection through advanced machine learning techniques.

Main Methods:

• Developed the Exemplar COVID-19FclNet9 system involving deep feature generation, iterative feature selection, and classification.
• Utilized three pre-trained Convolutional Neural Networks (CNNs)—AlexNet, VGG16, and VGG19—for feature extraction.
• Fused features from the top three performing CNNs and employed an iterative selector with a Support Vector Machine (SVM) classifier.

Main Results:

• The proposed system achieved high classification accuracies: 97.60% (DB1), 89.96% (DB2), 98.84% (DB3), and 99.64% (DB4).
• The hybrid feature fusion and iterative selection significantly improved image classification performance.
• The model demonstrated robust performance across datasets with varying class numbers (two, three, and four classes).

Conclusions:

• The Exemplar COVID-19FclNet9 model effectively detects COVID-19 and pneumonia using X-ray images.
• The proposed hybrid deep feature generation and selection approach enhances diagnostic accuracy.
• The system shows promise for real-world clinical deployment in diagnosing respiratory infections.