Related Experiment Video
Updated: Oct 9, 2025
09:29
Induction of Paralysis and Visual System Injury in Mice by T Cells Specific for Neuromyelitis Optica Autoantigen Aquaporin-4
Published on: August 21, 2017
11.6K
AxonDeep: Automated Optic Nerve Axon Segmentation in Mice With Deep Learning
Wenxiang Deng1,2, Adam Hedberg-Buenz2,3, Dana A Soukup2,3
1Department of Electrical and Computer Engineering, The University of Iowa, Iowa City, IA, USA.
Translational Vision Science & Technology
|December 21, 2021
Summary
Deep learning models now accurately segment and quantify optic nerve axons, improving upon manual methods. This automated approach offers faster, more objective analysis for glaucoma research.
Area of Science:
- Neuroscience
- Ophthalmology
- Artificial Intelligence
Background:
- Optic nerve damage is a key feature of glaucoma, leading to vision loss.
- Current methods for assessing nerve health in animal models are manual, time-consuming, and variable.
- Existing automated approaches have limitations.
Purpose of the Study:
- To develop and evaluate deep learning methods for segmenting and quantifying axons in mouse optic nerve cross-sections.
- To improve the accuracy and efficiency of optic nerve damage assessment.
Main Methods:
- Developed and compared two deep learning approaches: a supervised fully convolutional network and a semi-supervised generative adversarial network framework.
- Trained models using labeled and unlabeled image data of optic nerve cross-sections.
Main Results:
- Both deep learning methods outperformed a baseline automated approach.
- The semi-supervised deep learning approach achieved performance comparable to human experts.
- The superior semi-supervised method was implemented into a tool named AxonDeep.
Conclusions:
- AxonDeep enables automated, expert-level quantification and segmentation of optic nerve axons.
- This method overcomes the variability and time constraints of manual analysis.
- Deep learning provides rapid, objective, and high-throughput optic nerve analysis.
