Learning from imbalanced COVID-19 chest X-ray (CXR) medical imaging data

Area of Science:

• Medical image analysis
• Artificial intelligence in healthcare
• Radiology

Background:

• Digital medical image analysis is crucial for research and deployment.
• Chest X-rays (CXR) can serve as a screening tool for COVID-19 diagnosis.
• Publicly available CXR datasets for COVID-19 are essential for developing diagnostic tools.

Purpose of the Study:

• To present a systematic approach for learning from imbalanced CXR datasets for COVID-19 detection.
• To develop a robust diagnostic model for discerning COVID-19 from other conditions using CXR.
• To provide a publicly available CXR dataset on Kaggle for research.

Main Methods:

• Utilized a systematic approach to train a model on imbalanced CXR image data.
• Leveraged publicly available COVID-19 CXR images.
• Developed a case study on a Kaggle platform dataset.

Main Results:

• The proposed methodology achieved superior performance compared to top finishers in a related Kaggle challenge.
• Demonstrated the effectiveness of the approach in learning from limited, imbalanced COVID-19 CXR data.
• Achieved high accuracy in distinguishing COVID-19 from other conditions.

Conclusions:

• The developed methodology offers a reliable tool for medical personnel to diagnose COVID-19.
• The approach addresses the challenge of imbalanced datasets in medical image analysis.
• Publicly available datasets and robust models are key to advancing AI in medical diagnostics.