New spectral-domain optical coherence tomography algorithms can provide objective retina exam

November 1, 2008

Use of spectral-domain optical coherence tomography with automated image processing seems to provide more objective and potentially more reproducible assessments of the optic nerve head when compared with ophthalmologists' assessments of conventional fundus photos.

Key Points

Boston-Use of new spectral-domain optical coherence tomography (SD-OCT) algorithms seem to provide a more objective assessment of the optic nerve head compared with physician assessment of conventional fundus photos, according to Wei Sun, MA, a PhD candidate at Boston University, and a graduate research assistant at the Wellman Center for Photomedicine, Massachusetts General Hospital, Boston.

Use of the automated SD-OCT image-processing algorithm is more objective and does not have interobserver variability compared with the standard method in which an ophthalmologist interprets the disc borders in fundus photographs, she said.

"Subjective ophthalmic assessment of the optic nerve head remains standard for the diagnosis of optic nerve diseases," Sun said. "In glaucoma, there are qualitative measurements that describe the thinning of the neuroretinal rim and also quantitative measurements that evaluate optic nerve head parameter that is often referred to as the cup-to-disc ratio. However, these subjective methods suffer from high interobserver variability."

"SD-OCT allows us to do quantitative and objective examinations of the optic nerve head, with an automated image-processing algorithm," she said.

Led by Johannes F. de Boer, PhD (formerly of the Massachusetts General Hospital and Harvard Medical School, Boston, and now of the VU University, Amsterdam, Netherlands) and Teresa C. Chen, MD, of the Massachusetts Eye and Ear Infirmary and Harvard Medical School, Boston, Sun, and her colleagues conducted a study to correlate the manual disc border estimated by physicians based on disc photos, with the anatomic structures visualized using the automated algorithm in the SD-OCT system.

The study consisted of 20 eyes; Dr. Chen recruited the patients from the Massachusetts Eye and Ear Infirmary. Seven eyes had no eye conditions, five had ocular hypertension, four had glaucoma, and four were glaucoma suspects. The patients with glaucoma substantially were older than the control subjects. All participants underwent optic disc photography and SD-OCT scan.

Sun explained that because the SD-OCT en face images and the disc photos do not always perfectly overlap with each other, a scientific computing software (MATLAB, The MathWorks) was used to register the disc photos to overlap SD-OCT en face images by using the branching and turning corners of blood vessels as landmarks. Two glaucoma specialists then used a tablet PC to draw the optic cup and disc borders on the registered disc photos. The manual determinations then were projected on the en face images and the SD-OCT cross-sectional images.

An automated algorithm was run independently to determine the location of the disc border, and then the vertical disc diameter and disc area were calculated.

Statistical software (MedCalc) was used to analyze the data. The inter-rater agreement test (Cohen's κ) was used to compare the degree of agreement between two glaucoma specialists. After finding good agreement between the two, the average was used as the manual estimation to be compared with the automated methods with concordance correlation coefficient and Bland-Altman plots.

The concordance correlation coefficient test result indicated a good agreement between the automated and the manual methods for determining the vertical disc diameter and the disc areas. The Bland-Altman plots confirmed the previous results.

Sun reported that the mean disc diameter was about 1.7 mm and the disc area was 2.084 mm2 . The standard deviation of the two was about 10% (for disc diameter) to 20% (for disc area) of the average.

"This study indicated that there is substantial agreement between the automated detection method and the manual estimation," she said. "The automated method is more objective and does not suffer from interobserver variability, and [it] has the potential for greater reproducibility than the standard method."

"The goal of this study was to obtain an objective measurement of the structures seen in SD-OCT images and to correlate these measurements with known glaucoma diagnosis parameters such as the optic disc diameter, disc area, and the cup-to-disc ratio," Sun concluded.

Related Content: