Deep transfer learning for COVID-19 detection and infection localization with superpixel based segmentation.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Disease Diagnosis
• Radiology and Pulmonary Medicine

Background:

• The COVID-19 pandemic necessitates rapid and accurate diagnostic tools.
• Chest X-rays are preferred for COVID-19 screening due to accessibility, but exhibit lower sensitivity than CT scans.
• Enhancing the diagnostic value of Chest X-rays is crucial for early detection and management.

Purpose of the Study:

• To develop a deep learning framework to improve the diagnostic accuracy of Chest X-ray images for COVID-19 detection.
• To create a model that can classify COVID-19, viral pneumonia, and normal cases with high precision.
• To integrate image segmentation with classification to highlight regions of interest for better clinical interpretation.

Main Methods:

• A modified SqueezeNet classifier with segmentation capabilities, termed COVID-19 Super pixel SqueezNet (COVID-SSNet), was developed.
• The model was trained on deep features extracted from a standard Chest X-ray dataset.
• Super pixel segmentation was applied to activation maps to identify and overlay regions of interest.

Main Results:

• The binary classifier achieved 99.53% accuracy in distinguishing COVID-19 from Non-COVID-19 images.
• The multi-class classifier accurately classified COVID-19, Viral Pneumonia, and Normal cases with 99.79% accuracy.
• The COVID-SSNet model provides an overlay of relevant image regions, aiding in integral examination.

Conclusions:

• The proposed COVID-SSNet framework significantly enhances the diagnostic value of Chest X-rays for COVID-19.
• This deep learning approach offers a highly accurate and potentially faster method for COVID-19 diagnosis.
• The model's ability to highlight diagnostic regions aids clinicians in expedited treatment decisions.