Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Seeing Cells Clearly: Evaluating Machine Vision Strategies for Microglia Centroid Detection in 3D Images

Youjia Zhang

TL;DR

This work addresses the problem of robustly locating microglia centroids in 3D confocal images by comparing three segmentation strategies: ilastik (pixel-based classification), 3D Morph (thresholding with morphology-focused features), and Omnipose (deep-learning flow-based segmentation). Using a denoised 3D microglia dataset, the study evaluates how each method counts cells, identifies centroids, and reports morphological metrics, revealing that algorithm design heavily biases the results. Key findings show substantial differences in object counts, centroid dispersion, and available features across methods, underscoring that segmentation choice directly shapes downstream analyses and biological interpretations. The work highlights the need to align segmentation strategy with the research goal and data characteristics, especially when quantitative morphology and spatial patterns drive conclusions in neuroimaging. The insights have practical impact for researchers selecting automated tools for microglia analysis, guiding expectations about centroid accuracy, noise tolerance, and the richness of extracted features.

Abstract

Microglia are important cells in the brain, and their shape can tell us a lot about brain health. In this project, I test three different tools for finding the center points of microglia in 3D microscope images. The tools include ilastik, 3D Morph, and Omnipose. I look at how well each one finds the cells and how their results compare. My findings show that each tool sees the cells in its own way, and this can affect the kind of information we get from the images.

Seeing Cells Clearly: Evaluating Machine Vision Strategies for Microglia Centroid Detection in 3D Images

TL;DR

This work addresses the problem of robustly locating microglia centroids in 3D confocal images by comparing three segmentation strategies: ilastik (pixel-based classification), 3D Morph (thresholding with morphology-focused features), and Omnipose (deep-learning flow-based segmentation). Using a denoised 3D microglia dataset, the study evaluates how each method counts cells, identifies centroids, and reports morphological metrics, revealing that algorithm design heavily biases the results. Key findings show substantial differences in object counts, centroid dispersion, and available features across methods, underscoring that segmentation choice directly shapes downstream analyses and biological interpretations. The work highlights the need to align segmentation strategy with the research goal and data characteristics, especially when quantitative morphology and spatial patterns drive conclusions in neuroimaging. The insights have practical impact for researchers selecting automated tools for microglia analysis, guiding expectations about centroid accuracy, noise tolerance, and the richness of extracted features.

Abstract

Microglia are important cells in the brain, and their shape can tell us a lot about brain health. In this project, I test three different tools for finding the center points of microglia in 3D microscope images. The tools include ilastik, 3D Morph, and Omnipose. I look at how well each one finds the cells and how their results compare. My findings show that each tool sees the cells in its own way, and this can affect the kind of information we get from the images.
Paper Structure (15 sections, 4 figures, 2 tables)

This paper contains 15 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background (Related Work)
  3. Methods
  4. Dataset
  5. Segmentation Tools
  6. Experiment
  7. Data Processing
  8. Results
  9. ilastik
  10. 3D Morph
  11. Omnipose
  12. Centroid Position Comparison
  13. Conclusion
  14. Implications
  15. References

Figures (4)

  • Figure 1: Example of microglia segmentation on a single confocal image. Green contours outline detected cell boundaries. This illustrates how segmentation algorithms isolate individual microglia from dense, noisy data.
  • Figure 2: Z-stack structure from the 12hrMD_denoised dataset. Each z-stack is composed of 2D confocal slices taken at different depths to form a 3D image of in situ mouse microglia.
  • Figure 3: ilastik segmentation result on z-stack 79. Top: original confocal image. Bottom: semantic segmentation output showing detected cells in blue against a yellow background. Some over-segmentation is visible in densely packed regions.
  • Figure 4: 3D Morph segmentation result on z-stack 79. Each identified microglia is color-coded by volume. Larger structures appear in red and orange, while smaller ones are shown in blue. The panel on the left lists volumes of detected objects in decreasing order.