Purchase this article with an account.
Ali Fard, Homayoun Bagherinia, Simon Stock, Jochen Straub; Automatic Detection of the Optic Nerve Head in Line Scanning Ophthalmoscope Images in CIRRUS™ HD-OCT. Invest. Ophthalmol. Vis. Sci. 2017;58(8):4010.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Localization of optic nerve head (ONH) center is beneficial for tracking and registration purposes during optical coherence tomography (OCT) scan alignment. Detecting this landmark is particularly crucial for imaging and tracking patients with poor fixation, where manual tracking by the operator is required to ensure full capture of the peripapillary region. Here we present an automated method for detection of ONH center in healthy and glaucoma eyes using the line scanning ophthalmoscope (LSO) available on OCT instruments.
Our method relies on detecting contrast change and feature detection in the LSO fundus image in and around the optic disc. It combines linear and nonlinear image filtering techniques to determine the boundary of optic disc. It then uses cross correlation using a customized ONH template to accurately find the center coordinates. The ONH template is created by averaging 20 images selected from a database of over 100 LSO fundus images from healthy and diseased eyes. In order to evaluate the performance of our algorithm, we used another 41 LSO fundus images captured over an area of 10.3 mm x 8.6 mm from healthy (n=20), glaucoma (n=11), and pre-perimetric (n=10) eyes. ONH centers were independently marked by an expert grader and the results were compared with those of the automated algorithm to determine the success rate and accuracy. The images used for development and validation were acquired using a CIRRUS™ HD-OCT 5000 instrument (ZEISS, Dublin, CA) under an institutional review board protocol.
The mean and standard deviation of error between manual grading and automated algorithm was found to be 0.255mm and 0.635mm for all 50 test cases, respectively. Moreover, the automated algorithm detected a location within 0.5mm of the manual grading for 92% all test cases, respectively. The processing time was found to be sufficiently low to enable real time detection and tracking during OCT scan alignment.
Our results suggest that accurate detection of ONH in LSO fundus images is feasible with high success rate. The presented approach provides a way to reduce the dependency of the OCT scan quality on operator and patient fixation by enabling fully automated tracking of this landmark both in healthy and diseased eyes.
This is an abstract that was submitted for the 2017 ARVO Annual Meeting, held in Baltimore, MD, May 7-11, 2017.
This PDF is available to Subscribers Only