An XAI approach for COVID-19 detection using transfer learning with X-ray images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• COVID-19 poses significant global health challenges, with pneumonia as a primary complication.
• Chest X-rays (CXRs) are crucial for diagnosing lung diseases but require expert interpretation.
• The pandemic accelerated healthcare digitalization and the adoption of Artificial Intelligence (AI).

Purpose of the Study:

• To develop and evaluate an AI model for detecting COVID-19 positive CXR images.
• To utilize Explainable Artificial Intelligence (XAI) for interpreting AI-driven CXR analysis.
• To enhance model performance through lung segmentation and transfer learning techniques.

Main Methods:

• Proposed an AI model incorporating Explainable Artificial Intelligence (XAI).
• Employed transfer learning and data augmentation for efficient model training.
• Implemented lung segmentation to improve diagnostic accuracy.
• Compared pre-trained networks, selecting ResNet for optimal performance.

Main Results:

• The AI model achieved a high F1-Score of 98% in classifying COVID-19 positive CXR images.
• XAI techniques were used to analyze and visualize the impact of positive CXR findings via heatmaps.
• Lung segmentation and transfer learning significantly enhanced model performance.

Conclusions:

• The developed AI model demonstrates high accuracy in detecting COVID-19 from CXR scans.
• Explainable AI provides valuable insights into the model's decision-making process for CXR interpretation.
• This approach offers a promising tool to support radiologists in diagnosing COVID-19 pneumonia.