Identification of gastroenteric viruses by electron microscopy using higher order spectral features

Area of Science:

• Medical Imaging
• Virology
• Computational Biology

Background:

• Paediatric illnesses, such as acute gastroenteritis, are frequently caused by viral agents.
• Electron microscopy (EM) is a powerful tool for visualizing viral particles but requires specialized expertise for interpretation.
• Current EM diagnostic methods are time-consuming and demand high levels of skill.

Purpose of the Study:

• To introduce a semi-automated method for identifying viruses from electron microscopy images.
• To overcome the limitations of manual interpretation in EM-based viral diagnostics.
• To enhance the efficiency and accessibility of viral agent identification.

Main Methods:

• A novel method utilizing bispectral features to analyze contour and texture information of viral particles.
• The method demonstrates robustness to variations in size, rotation, noise, and image shifts.
• Viral particles are segmented and classified using a Gaussian Mixture Model (GMM) for probability density estimation.

Main Results:

• Achieved an equal error rate (EER) of 2% for Rotavirus verification using averaged features from 15 images.
• Reduced EER to less than 0.2% by averaging scores from two independent tests.
• Demonstrated an EER of less than 2% for Astrovirus verification with 20 particles and two tests.

Conclusions:

• Bispectral features combined with GMM provide an effective means for virus identification in EM images.
• The proposed method significantly enhances diagnostic capabilities for viral infections.
• Full automation is achievable with the integration of digital imaging technology in electron microscopy.