Automated quantification of optic nerve axons in primate glaucomatous and normal eyes--method and comparison to semi-automated manual quantification.

Purpose

To describe an algorithm and software application (APP) for 100% optic nerve axon counting and to compare its performance with a semi-automated manual (SAM) method in optic nerve cross-section images (images) from normal and experimental glaucoma (EG) nonhuman primate (NHP) eyes.

Methods

ON cross sections from eight EG eyes from eight NHPs, five EG and five normal eyes from five NHPs, and 12 normal eyes from 12 NHPs were imaged at 100×. Calibration (n = 500) and validation (n = 50) image sets ranging from normal to end-stage damage were assembled. Correlation between APP and SAM axon counts was assessed by Deming regression within the calibration set and a compensation formula was generated to account for the subtle, systematic differences. Then, compensated APP counts for each validation image were compared with the mean and 95% confidence interval of five SAM counts of the validation set performed by a single observer.

Results

Calibration set APP counts linearly correlated to SAM counts (APP = 10.77 + 1.03 [SAM]; R(2) = 0.94, P < 0.0001) in normal to end-stage damage images. In the validation set, compensated APP counts fell within the 95% confidence interval of the SAM counts in 42 of the 50 images and were within 12 axons of the confidence intervals in six of the eight remaining images. Uncompensated axon density maps for the normal and EG eyes of a representative NHP were generated.

Conclusions

An APP for 100% ON axon counts has been calibrated and validated relative to SAM counts in normal and EG NHP eyes.