gACSON software for automated segmentation and morphology analyses of myelinated axons in 3D electron microscopy

TL;DR

gACSON addresses the challenge of automatic segmentation and morphometric analysis of myelinated axons in large 3D-EM volumes. It introduces a MATLAB-based GUI that performs BM4D denoising, semantic myelin segmentation, BVG-based instance segmentation of intra-axonal spaces, and myelin instance segmentation, followed by 3D morphometry including the g-ratio defined as $g = \frac{d_e}{d_e + d_m}$, where $d_e$ is the equivalent intra-axonal diameter and $d_m$ is myelin thickness. The authors demonstrate the pipeline on six SBEM volumes from rat somatosensory cortex, reporting that ipsilateral equivalent diameter decreases in TBI animals and providing a freely available tool under the MIT license. The framework supports interactive proofreading, segmentation evaluation, and cross-sectional morphometry, offering a practical platform for quantitative ultrastructural analyses relevant to conduction properties and brain injury studies.

Abstract

Background and Objective: Advances in electron microscopy (EM) now allow three-dimensional (3D) imaging of hundreds of micrometers of tissue with nanometer-scale resolution, providing new opportunities to study the ultrastructure of the brain. In this work, we introduce a freely available Matlab-based gACSON software for visualization, segmentation, assessment, and morphology analysis of myelinated axons in 3D-EM volumes of brain tissue samples. Methods: The software is equipped with a graphical user interface (GUI). It automatically segments the intra-axonal space of myelinated axons and their corresponding myelin sheaths and allows manual segmentation, proofreading, and interactive correction of the segmented components. gACSON analyzes the morphology of myelinated axons, such as axonal diameter, axonal eccentricity, myelin thickness, or g-ratio. Results: We illustrate the use of the software by segmenting and analyzing myelinated axons in six 3D-EM volumes of rat somatosensory cortex after sham surgery or traumatic brain injury (TBI). Our results suggest that the equivalent diameter of myelinated axons in somatosensory cortex was decreased in TBI animals five months after the injury. Conclusions: Our results indicate that gACSON is a valuable tool for visualization, segmentation, assessment, and morphology analysis of myelinated axons in 3D-EM volumes. It is freely available at https://github.com/AndreaBehan/g-ACSON under the MIT license.

Figures (10)

• Figure 1: gACSON segmentation pipeline. The pipeline takes 3D-EM volume as input and 1) denoises the EM volume using the BM4D method, 2) segments the myelin in the denoised volume using a user-defined method (thresholding, seeded region growing (SRG) or an ML-based approach), 3) performs the instance segmentation of the intra-axonal space of myelinated axons, and 4) performs the instance segmentation of myelin. The numbers in parentheses refer to subsections where the steps are detailed.
• Figure 1: The morphology analysis of myelinated axons in the ipsilateral (a-f) and contralateral (g-l) somatosensory cortex of sham-operated and TBI rats: (a, g) equivalent diameter, (b, h) minor axis length (c, i) major axis length, (d, j) axonal eccentricity, (e, k) myelin thickness, and (f, l) g-ratio. On each box, the central mark indicates the median, and the bottom and top edges of the box indicate the $25^{th}$ and $75^{th}$ percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers, and the outliers are plotted individually using the ‘o’ symbol. The asterisk in (a) shows that there is a significant reduction in the equivalent diameter of the intra-axonal space of myelinated axons, comparing the sham-operated and TBI groups ipsilaterally.
• Figure 2: a A representative SBEM image from the somatosensory cortex of a sham-operated rat. (b) BM4D denoising of a. The magnification factor of the zoomed-in images in the orange boxes is 300 %.
• Figure 3: (a) Semantic segmentation of myelin by a user-defined threshold. (b) Machine learning (ML)-based semantic segmentation of myelin. Left panel: user specifies myelin (magenta) against other ultrastructures (green) by drawing a scribble on an EM image. These scribbles provide data to train a random forest model. Right panel: the semantic segmentation of myelin (magenta) against other ultrastructures (green). The thresholding-based segmentation of myelin incorrectly includes cell membranes and is therefore noisier than the ML-based segmentation. Both methods segment mitochondria as myelin.
• Figure 4: The workflow of instance segmentation of myelin. (a) Semantic segmentation of myelin. (b) Instance segmentation of the intra-axonal space of myelinated axons. (c) Watershed transformation using the segmented intra-axonal spaces partitions the image space into disjoint subvolumes. (d) Instance segmentation of myelin. (e) Instance segmentation of myelin and the intra-axonal spaces. Colors of distinct instances in the figure have been randomly generated and they are not comparable between different panels.