AxonDeepSeg: automatic axon and myelin segmentation from microscopy data using convolutional neural networks

Aldo Zaimi,Christian S Perone,Victor Herman,Pierre-Louis Antonsanti,Maxime Wabartha,Julien Cohen-Adad

doi:10.1038/s41598-018-22181-4

Abstract

Segmentation of axon and myelin from microscopy images of the nervous system provides useful quantitative information about the tissue microstructure, such as axon density and myelin thickness. This could be used for instance to document cell morphometry across species, or to validate novel non-invasive quantitative magnetic resonance imaging techniques. Most currently-available segmentation algorithms are based on standard image processing and usually require multiple processing steps and/or parameter tuning by the user to adapt to different modalities. Moreover, only a few methods are publicly available. We introduce AxonDeepSeg, an open-source software that performs axon and myelin segmentation of microscopic images using deep learning. AxonDeepSeg features: (i) a convolutional neural network architecture; (ii) an easy training procedure to generate new models based on manually-labelled data and (iii) two ready-to-use models trained from scanning electron microscopy (SEM) and transmission electron microscopy (TEM). Results show high pixel-wise accuracy across various species: 85% on rat SEM, 81% on human SEM, 95% on mice TEM and 84% on macaque TEM. Segmentation of a full rat spinal cord slice is computed and morphological metrics are extracted and compared against the literature. AxonDeepSeg is freely available at https://github.com/neuropoly/axondeepseg.

Highlights

Neuronal communication is ensured by the transmission of action potentials along white matter axons
Segmentation of axons and myelin based on deep learning approaches offers significant advantages when compared with traditional image segmentation algorithms: (i) there is no need to hand-select relevant features because the network is able to learn the hidden structural and textural features by itself, (ii) this approach allows to segment both axons and myelin sheaths in two different labels with the same network, without the need of any explicit pre- or post-processing, (iii) the network can be trained for various imaging modalities without significantly changing its architecture and (iv) once trained, the model is relatively fast at the prediction step compared to more traditional image processing methods
We showed that AxonDeepSeg can segment axon and myelin of scanning electron microscopy (SEM) and transmission electron microscopy (TEM) samples of various species with high accuracy

Summary

Introduction

Neuronal communication is ensured by the transmission of action potentials along white matter axons. These axons, which are typically 1–10 μm in diameter, are surrounded by a myelin sheath whose main role is to facilitate the propagation of the electrical impulses along neuronal fibers and increase the transmission speed[1,2]. Naito and collaborators[25] have implemented a two-step process that first performs clustering segmentation of myelinated nerve fibers in optical microscopic images, and discriminates between true and false candidates by using a CNN classification network. This group did not exploit the CNN for the segmentation, but only for discrimination. AxonDeepSeg features: (i) a CNN architecture for semantic segmentation of histological images; (ii) two ready-to-use models for the segmentation of scanning electron microscopy (SEM) and transmission electron microscopy (TEM) samples adapted to a variety of species and acquisition parameters; (iii) a well-documented training pipeline to generate models for new imaging modalities and (iv) free and open source code (https://github.com/neuropoly/axondeepseg)

Methods

Results

Conclusion