A new composite approach for COVID-19 detection in X-ray images using deep features

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Traditional COVID-19 detection methods like PCR and serological tests have limitations.
• Medical imaging, including X-rays, offers a complementary approach for disease detection.

Purpose of the Study:

• To develop and evaluate a novel combined artificial intelligence approach for COVID-19 detection using deep features from X-ray images.
• To compare the performance of single layer-based (SLB) and feature fusion-based (FFB) models for COVID-19 classification.
• To establish new benchmarks for accuracy in COVID-19 detection using medical imaging and AI.

Main Methods:

• Development of two main model variances: single layer-based (SLB) and feature fusion-based (FFB).
• Implementation of pre-processing, deep feature extraction, and post-processing phases within each model.
• Creation of four SLB and six FFB models varying in layer combinations for feature extraction.
• Conducting binary and multi-class classification experiments using a five-fold cross-validation strategy.

Main Results:

• The proposed feature fusion-based (FFB3) model achieved 99.52% accuracy for binary classification (COVID-19 vs. no-findings), surpassing the literature's best at 98.08%.
• For multi-class classification, the FFB3 model reached 87.64% accuracy, outperforming the previous best of 87.02%.
• The FFB3 model demonstrated superior performance across sensitivity, specificity, precision, and F1-score metrics compared to existing methods.

Conclusions:

• The developed composite models (SLBs and FFBs) are effective for COVID-19 detection using X-ray images.
• Feature extraction, pre-processing, post-processing, and hyperparameter tuning are critical for achieving high diagnostic success.
• The study highlights the potential of advanced AI techniques in medical imaging for pandemic response and future research.