AxonJ: Validation of an Automated Axon Counting Method in Rodent Models of Glaucoma

Abstract

Purpose: To validate our publicly-available, novel, automated method (AxonJ) of whole-optic nerve sections in normal and glaucomatous mouse models. Axon counts of whole nerves, though ideal for precise phenotyping, are unfeasible, hence elaborate sampling strategies are required which assume homogeneous distribution of axons across the optic nerve, while semi-automated methods lack validation in rodent models.

Methods: To maximize the strength of the validation, a variety of mouse strains were evaluated at several time points. PPD stained optic nerve sections were imaged at 40x and 100x magnification. Two human experts independently counted sampling regions at 100x magnification. AxonJ was used to determine whole nerve axon counts and axon density on all images. Manual and automated counts were compared using correlation coefficients. Extrapolated and automated 40x counts were also compared. AxonJ inter-section variability was assessed using a correlation.

Results: Counts from the AxonJ method correlated closely with counts obtained from humans at 100x, r² = 0.95, while human-to-human correlation was r² = 0.97. AxonJ whole nerve counts at 40x also correlated with counts extrapolated from human grading of 10% of the optic nerve area, r² = 0.94. Inter-section variability was also low, r² = 0.95.

Conclusions: AxonJ counts all axons in rodent whole optic nerve images with a performance equivalent to human experts counting a small subset of the axons, while inter-section repeatability is high. AxonJ is fully-automated, parameter-free, runs in less than a minute on a standard desktop, and does not make assumptions about the distribution of axons across the optic nerve. It is freely available as an ImageJ plugin.