Search for: All records

Abstract Microglia, the brain’s resident macrophages, participate in development and influence neuroinflammation, which is characteristic of multiple brain pathologies. Diverse insults cause microglia to alter their morphology from “resting” to “activated” shapes, which vary with stimulus type, brain location, and microenvironment. This morphologic diversity commonly restricts microglial analyses to specific regions and manual methods. We introduce StainAI, a deep learning tool that leverages 20x whole-slide immunohistochemistry images for rapid, high-throughput analysis of microglial morphology. StainAI maps microglia to a brain atlas, classifies their morphology, quantifies morphometric features, and computes an activation score for any region of interest. As a proof of principle, StainAI was applied to a rat model of pediatric asphyxial cardiac arrest, accurately classifying millions of microglia across multiple slices, surpassing current methods by orders of magnitude, and identifying both known and novel activation patterns. Extending its application to a non-human primate model of simian immunodeficiency virus infection further demonstrated its generalizability beyond rodent datasets, providing new insights into microglial responses across species. StainAI offers a scalable, high-throughput solution for microglial analysis from routine immunohistochemistry images, accelerating research in microglial biology and neuroinflammation. 

Microglia are the macrophages resident in the central nervous system. Brain injuries, such as traumatic brain injury, hypoxia, and stroke, can induce inflammatory responses accompanying microglial activation. The morphology of microglia is notably diverse and a prominent manifestation of activation. In this study, we propose to classify activated microglia using a convolutional neural network (CNN). Iba1 images were acquired from a control and cardiac arrest Long-Evans rat brain with a bright-field microscopy. The training data of 54,333 single-cell images were collected from the cortex and midbrain areas and curated by experienced neuroscientists. Results were compared between CNNs with different architectures, including Resnet18, Resnet50, Resnet101, and support vector machine classifiers. The highest model performance was found by Resnet18, trained after 120 epochs with a classification accuracy of 95.5-98.8 percent. The findings indicate a potential application for using CNN in the quantitative analysis of microglial morphology over regional differences in a large brain section.