Explainable COVID-19 Detection Using Chest CT Scans and Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Infectious Diseases

Background:

• Accurate and rapid diagnosis of COVID-19 is crucial for effective patient management and public health.
• Chest CT imaging shows promise for COVID-19 detection, but automated diagnostic tools are needed.
• Deep learning offers potential for analyzing complex medical image data.

Purpose of the Study:

• To evaluate the efficacy of deep learning models for automated COVID-19 diagnosis using chest CT images.
• To develop and optimize deep learning architectures and transfer learning strategies for improved diagnostic performance.
• To enhance model interpretability through visualization techniques.

Main Methods:

• Utilized advanced deep network architectures with a transfer learning strategy.
• Customized input sizes for different deep architectures to maximize performance.
• Conducted experiments on two distinct COVID-19 CT image datasets (SARS-CoV-2 CT-scan and COVID19-CT).
• Applied feature visualization techniques for model interpretability and prediction explanation.

Main Results:

• Achieved superior performance compared to previous studies on both datasets.
• Best models demonstrated high accuracy (up to 99.4% on SARS-CoV-2, 92.9% on COVID19-CT), precision, sensitivity, specificity, and F1-score.
• Visualizations confirmed model ability to differentiate COVID-19 from non-COVID-19 cases and accurately localize affected regions.

Conclusions:

• Deep learning models, optimized with transfer learning, are highly effective for automated COVID-19 diagnosis from chest CT scans.
• The developed models offer a fast, accurate, and interpretable solution for COVID-19 detection.
• Visualization techniques provide insights into model decision-making, correlating with expert radiologist assessments.