Classification of Activated Microglia by Convolutional Neural Networks

Area of Science:

• Neuroscience
• Immunology
• Computational Biology

Background:

• Microglia, the brain's immune cells, activate and change morphology during injury.
• Assessing microglial activation is crucial for understanding central nervous system disorders.
• Current methods for analyzing microglial morphology can be labor-intensive.

Purpose of the Study:

• To develop and evaluate a Convolutional Neural Network (CNN) model for detecting activated microglia in immunohistochemistry images.
• To compare the performance of different CNN architectures and a Support Vector Machine (SVM) classifier for this task.

Main Methods:

• Acquisition of 2D Iba1 immunohistochemistry images from rat brains subjected to cardiac arrest.
• Training and testing CNN models (Resnet18, Resnet50, Resnet101) and an SVM classifier on over 54,000 single-cell images.
• Evaluation of model performance based on classification accuracy.

Main Results:

• The Resnet18 CNN architecture achieved the highest performance, with a classification accuracy of 95.5% after 120 training epochs.
• CNN models demonstrated superior performance compared to the SVM classifier.
• Significant differences in microglial morphology were observed between control and injured brain regions.

Conclusions:

• CNNs provide a highly accurate and efficient tool for automated detection and quantitative analysis of activated microglia.
• This approach holds promise for large-scale analysis of microglial morphology across different brain regions and injury conditions.
• The findings support the application of deep learning in neuroinflammation research.