Optimization-Based Methods for Reconstruction and Structural Quantification of Neurons and Glia from Microscopic Images

Files

TR Number

Date

2025-06-02

Journal Title

Journal ISSN

Volume Title

Publisher

Virginia Tech

Abstract

Structural analysis of neurons and glia based on light and electron microscopic imaging is indispensable for the understanding of the central nervous system and drug development for brain diseases. It requires the accurate reconstruction of the interesting structures and the design of quantitative metrics to describe the morphology. Structural quantification of myelin sheaths from in vivo light microscopy imaging gives insights into multiple sclerosis pathology and supports drug screening. However, accurate reconstruction of myelin sheaths from light microscopes (LM) remains difficult due to the low signal-to-noise ratio, inhomogeneous intensity, and anisotropic resolution. As a result, most studies still depend on labor‑intensive manual annotation. To address these limitations, we propose GOMS3D and MMS3D, the first fully automated pipelines for reconstructing myelin sheaths in zebrafish and mice, respectively. Both frameworks begin with statistics-driven adaptive thresholding to achieve accurate foreground detection in noisy data. Then, GOMS3D formulates an optimization problem to project centerlines detected in 2D to 3D lines, thereby compensating for anisotropy, while MMS3D employs a front‑propagation algorithm that maximizes angular coherence along each trajectory to bridge discontinuities. Both methods outperform existing tools designed for tracing similar tubular structures. Electron microscopy (EM) imaging of neurons and astrocytes reveals the detailed tripartite structures formed by dendritic spines, axon terminals, and astrocyte processes, which are crucial for understanding and predicting the functions of neural circuits. However, most methods for segmenting dendritic spines from electron microscopy images focus solely on surface information and yield non-ideal results. To address this issue, we propose VSOT, a graph‑theoretic optimization framework that combines both surface and volume cues to deliver a more robust result than competing methods. Based on VSOT and our newly designed quantification scores, we can, for the first time, quantify the structural differences of dendritic spines and tripartite structures across different cortical layers of the mouse visual cortex in a large-scale EM dataset.

Description

Keywords

Morphological quantification, Neuron, Glia, Microscopy, Segmentation

Citation