AxoNet 2.0: A Deep Learning-Based Tool for Morphometric Analysis of Retinal Ganglion Cell Axons
Vidisha Goyal1, A Thomas Read2, Matthew D Ritch2
1School of Electrical and Computer Engineering, Georgia Institute of Technology, Atlanta, GA, USA.
Translational Vision Science & Technology
|March 14, 2023
Summary
Automated deep learning tool AxoNet 2.0 rapidly quantifies retinal ganglion cell (RGC) axon loss and morphology in animal models. This non-subjective method enhances glaucoma research by providing accurate RGC axon assessment.
Area of Science:
- Neuroscience
- Ophthalmology
- Biomedical Engineering
Background:
- Manual assessment of retinal ganglion cell (RGC) axonal loss in animal models is time-consuming and labor-intensive.
- Accurate quantification of RGC axonal damage is crucial for evaluating glaucomatous optic neuropathy.
- Developing automated tools can improve the efficiency and rigor of preclinical studies.
Purpose of the Study:
- To develop AxoNet 2.0, an automated deep learning tool for quantifying RGC axonal loss and morphologic changes from light micrographs.
- To enable rapid and non-subjective assessment of glaucomatous damage in animal models.
- To facilitate basic science research on RGC axon protection and regeneration.
Main Methods:
- A deep learning algorithm was trained to segment axoplasm and myelin sheath of normal-appearing axons using annotated rat optic nerve micrographs.
- Performance was validated using metrics such as soft-Dice coefficient and mean absolute percentage error.
- Axon counts, density, and size distributions were quantified and compared between hypertensive and control rat eyes.
Main Results:
- AxoNet 2.0 demonstrated high accuracy in counting and quantifying RGC axon morphometry compared to manual annotations (R2 = 0.92).
- The tool showed excellent generalization across different animal models, including mice and non-human primates (R2 = 0.97-0.98).
- AxoNet 2.0 detected decreased axon density and preferential loss of large axons in hypertensive rat eyes (P ≪ 0.001).
Conclusions:
- AxoNet 2.0 provides a fast, non-subjective, and accurate method for quantifying RGC axon counts and morphology.
- This tool assists in assessing axonal damage in animal models of glaucomatous optic neuropathy.
- The deep learning approach enhances the rigor of studies investigating RGC axon protection and regeneration.
