Machine Learning for Microglia Morphology Classification

Table of Contents

Abstract

Presentation

01. Introduction 02. Microglial Classification 03. Machine Learning Model 04. Activation Score 05. Alpha-Synuclein Mouse Model 06. Microglia Density 07. Microglia Activation 08. Activation vs. Density 09. Mouse Model Activation Score 10. Clustering 11. Summary

https://opt003stagmediafiles.blob.core.windows.net/image/a132bd5458244aa8b3941de504b3e2c8

Examples of segmented microglia. The red outline indicates the cell body and the yellow outline indicates the processes.

https://opt003stagmediafiles.blob.core.windows.net/image/74650f0339dd426397f7503da6894151

This plot shows, for each cell, how much the value of each parameter, such as the cell soma perimeter, influences the decision to classify cells as activated or not.

The Shapley (SHAP) values shown on the graph represent the additive effects of the different features, and can be used to explain the classification decision for each particular cell. Each point represents the value of the parameter of a cell (indicated by the color, where red corresponds to values higher than the average) and its effect on the decision, where higher SHAP values tend towards classifying cells as activated.

For example, cells with high soma perimeter and area, low soma circularity, and high number of branches in their skeleton are more likely to be classified as activated.

--:--

Our machine learning model uses morphological features to distinguish non-activated from activated microglia. The model makes explainable decisions that are consistent with the literature. As an example, cells with higher soma perimeter and area, higher number of branches and branching points, and lower soma circularity are more likely to be classified as activated.