Virus Particle Detection by Convolutional Neural Network in Transmission Electron Microscopy Images

Area of Science:

• Computational Biology
• Microscopy Imaging

Background:

• Accurate virus particle detection in transmission electron microscopy (TEM) images is crucial for virology research.
• Existing methods often rely on handcrafted features, leading to false positives and extensive postprocessing.

Purpose of the Study:

• To develop and evaluate a new computational method for automated virus particle detection in TEM images.
• To leverage machine learning for improved accuracy and efficiency compared to traditional approaches.

Main Methods:

• Utilized a convolutional neural network (CNN) to transform TEM images into probabilistic maps indicating virus particle locations.
• The CNN automatically learned discriminative features and a classifier for virus particle detection through supervised learning.

Main Results:

• The proposed CNN-based method demonstrated state-of-the-art performance in detecting feline calicivirus particles in TEM images.
• Achieved superior detection accuracy compared to several existing methods, with significantly fewer false positives.

Conclusions:

• The developed computational method offers a highly effective and automated solution for virus particle detection in TEM imagery.
• While trained on known viruses, the CNN's flexibility allows adaptation to detect novel virus particles with annotated datasets.