Morphometric Analysis of Optic Nerve Axons Using Automated ArcGIS Feature Extraction

Abstract

Purpose : Manual histological light microscopy image analysis of optic nerve axons is an excellent outcome measure in animal studies and human histopathologic evaluations, but extremely labor intense with additional challenges of reproducibility and analyst bias. Previous publications have demonstrated methods for automating axonal morphometric analysis using electron microscopy but this is less useful or desirable compared to light microscopy. The present study explores the feasibility of using ArcGIS to automate feature extraction from optic nerve cross-sections acquired with light microscopy.

Methods : We performed manual, semi-automated and automated analysis on optic nerves from human controls and patients with Lebers Hereditary Optic Neuropathy (LHON) stained with P-phenylenediamine (PPD) and imaged with light microscopy. Manual analysis was done on ImageJ: considered the gold standard for image analysis of optic nerves. Automation was achieved using Matlab for image enhancement and ArcGIS for feature extraction developed in-house with batch processing ability. This semi-automated method involved manual review and editing of the automation results using ArcGIS. Statistics using excel was performed to benchmark semi-automation and Automation against manual analysis. Processing time was also assessed for all three methods.

Results : A comparative analysis showed that semi-automation produced 102% concordance with gold standard manual analysis for control nerves and 81% with LHON nerves, compared with full automation at 95% concordance for control nerves and 85% concordance for LHON nerves. Both semi-automation and automation were much faster than manual analysis: full automation was 30 to 40 fold faster than manual analysis, compared to semi-automation which was 3-4 times faster.

Conclusions : Our study found a 102% and 81% concordance with semi-automated axon morphometric analysis and 95% and 85% with full automation, compared to the gold standard manual analysis in control and LHON nerves. With an improved segmentation algorithm and application of fuzzy logic image contrast enhancement, we anticipate that ArcGIS automation may approach the accuracy of semiautomation and manual analysis and being much faster, provide a more feasible outcome measure of optic neuropathies in experimental and pathological analyses.

This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.