μGlia-Flow, an automatic workflow for microglia segmentation and classification

Abstract

Background

Microglia are important immune cells in the central nervous system, playing a key role in various pathological processes. The morphological diversity of microglia is closely linked to the development of brain diseases, yet accurate segmentation and automatic classification of microglia remain challenging.

New method

We proposed a workflow, μGlia-Flow, which integrates both segmentation and classification for microglia analysis. The Frangi filtering algorithm was employed for branch segmentation, and an edge-guided attention TransUNet (EGA-Net) was used for soma segmentation. A Vision Transformer (ViT) network was applied to classify different morphologies.

Results

The Frangi filtering algorithm produces more complete branches with smoother edges and clearer structures. The EGA-Net improves Dice and IoU scores by 4.02 % and 6.75 %, respectively. ViT achieves over 99 % precision in classification. Post-processing reveals decreasing complexity during activation, validating the accuracy of μGlia-Flow.

Comparison with existing methods

μGlia-Flow introduces deep learning, significantly improving segmentation accuracy and addressing the parameter dependency of existing classification methods.

Conclusion

we present an automatic workflow for segmenting and classifying microglia, providing a powerful tool for different morphology analysis.