Axon and glial cell segmentation in optic nerve images utilizing deep learning approaches

The optic nerve is located in the back of the eye and serves as a communication cable between the eye and brain. The optic nerve contains axons, which are the neuron extensions carrying visual information from the eyes to the brain enabling the person to see. Due to the diseases such as glaucoma, axons in the optic nerve degenerate and can cause irreversible vision loss. To assess the severity of the damage to axons and potentially prevent vision loss in humans, there is a need to determine axon density in the optic nerve. However, this metric requires invasive methods to be determined. Therefore, studies are undertaken to find the correlation between noninvasively derived metrics and axon loss to evaluate optic nerve damage in humans. Nevertheless, because of the hardship in accessing human tissues, mice models are usually used in these studies.

A necessary part of these studies requires counting axons in the optic nerve images of mice models. However, conventional manual axon counting approaches can be highly time-consuming and labor-intensive. In this work, we used automated models to label and count axons in the optic nerve images. These automated approaches are fast and require little effort and time from the users. Using the same structure in these models, we also developed other models that can label glial cells (non-neuronal cells in the optic nerve). These models are useful because they provide a secondary metric to assess the optic nerve damage. We tested and compared our models to existing methods in the field and showed that our models have higher accuracy.