xViTCOS: Explainable Vision Transformer Based COVID-19 Screening Using Radiography.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• The rapid spread of COVID-19 strains healthcare systems globally.
• Accurate and timely COVID-19 identification is crucial for containment.
• Traditional RT-PCR tests have long turnaround times, necessitating faster screening methods.

Purpose of the Study:

• To propose and evaluate vision transformers for COVID-19 screening using X-ray and CT images.
• To address limitations of Convolutional Neural Networks (CNNs) in capturing global context for disease detection.
• To develop an explainable AI model for improved diagnostic accuracy.

Main Methods:

• Utilized vision transformers, a type of deep learning model, for analyzing chest radiography (X-ray and CT).
• Employed a multi-stage transfer learning technique to overcome data scarcity challenges.
• Demonstrated the explainability of features learned by the transformer networks.

Main Results:

• The proposed vision transformer method quantitatively outperformed existing benchmarks for COVID-19 detection.
• The model accurately focused on clinically relevant regions in images, confirmed by radiologists.
• Achieved accurate diagnosis and localization of COVID-19 infected areas.

Conclusions:

• Vision transformers offer a promising alternative to CNNs for medical image analysis in COVID-19 screening.
• The method facilitates timely identification of COVID-19, improving healthcare resource utilization.
• Explainable AI features enhance trust and understanding in diagnostic AI models.