Uncertain-CAM: Uncertainty-Based Ensemble Machine Voting for Improved COVID-19 CXR Classification and Explainability

Area of Science:

• Artificial Intelligence in Medicine
• Medical Imaging Analysis
• Deep Learning for Diagnostics

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Distinguishing COVID-19 from other respiratory infections like pneumonia is challenging.
• Explainability of deep learning models is crucial for clinical adoption.

Purpose of the Study:

• To develop a highly accurate deep learning model for COVID-19 classification from chest X-rays.
• To enhance the explainability and reliability of AI-driven medical diagnostics.
• To address the cost and early detection challenges in pandemic management.

Main Methods:

• Training deep learning models with an uncertainty-based ensemble voting policy.
• Integrating cyclic cosine annealing, cross-validation, and uncertainty quantification (PICP) for model training.
• Proposing the Uncertain-CAM technique for improved deep learning explainability.

Main Results:

• Achieved 99% accuracy in classifying COVID-19 chest X-rays against normal and pneumonia cases.
• Demonstrated enhanced explainability using the novel Uncertain-CAM technique.
• Validated a new image processing method for explainability measurement.

Conclusions:

• The proposed method offers a reliable and accurate AI solution for COVID-19 detection.
• Enhanced explainability through Uncertain-CAM increases trust in AI diagnostic systems.
• This approach can aid in cost-effective and early identification of COVID-19 patients.