A Novel Method for COVID-19 Diagnosis Using Artificial Intelligence in Chest X-ray Images

Area of Science:

• Medical Imaging and Artificial Intelligence
• Deep Learning in Healthcare
• Infectious Disease Diagnostics

Background:

• Current COVID-19 detection methods (temperature checks, nasal swabs) are resource-intensive, intrusive, and face kit limitations.
• Developing non-intrusive, AI-driven technologies for rapid COVID-19 patient identification is a critical global health challenge.
• Acquiring research datasets is difficult due to patient consent limitations for COVID-19 studies.

Purpose of the Study:

• To propose a novel Artificial Intelligence (AI) method for automatic and non-intrusive diagnosis of Coronavirus disease 2019 (COVID-19) patients.
• To develop a deep learning (DL) algorithm, CoVIRNet, utilizing chest X-rays for efficient COVID-19 assessment.
• To address the challenges of limited datasets and intrusive testing in COVID-19 diagnostics.

Main Methods:

• Developed CoVIRNet, a novel deep learning model employing inception residual blocks for multi-scale feature extraction from chest X-rays.
• Utilized various layers and concatenation of feature maps at different scales within the classification blocks.
• Integrated deep learning features with machine learning models, including a random-forest classifier, for validation.

Main Results:

• The proposed CoVIRNet model achieved an accuracy of 95.7% in diagnosing COVID-19 from chest X-rays.
• The CoVIRNet feature extractor combined with a random-forest classifier yielded the highest accuracy of 97.29%, outperforming existing state-of-the-art deep learning methods.
• The model demonstrated effective use of regularization techniques to mitigate overfitting on a small COVID-19 dataset.

Conclusions:

• CoVIRNet offers an automatic and efficient solution for the assessment and classification of COVID-19 using chest X-rays.
• The combination of deep learning feature extraction and machine learning classification shows significant promise for improved diagnostic accuracy.
• The proposed method is predicted to demonstrate superior performance compared to current state-of-the-art COVID-19 diagnostic techniques.