Purchase this article with an account.
B. L. Petrig, M. S. Muller, J. A. Papay, A. E. Elsner; Fixation Stability Measurements Using the Laser Scanning Digital Camera. Invest. Ophthalmol. Vis. Sci. 2010;51(13):2275.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To determine whether an automatic method of alignment of retinal images from the Laser Scanning Digital Camera (LSDC) provides fixation stability estimates comparable to manual alignment for patients without macular scotomas. To develop equipment and software that accurately quantifies eye motion and identifies patients with poor fixation.
The fixation stability of each eye of 27 patients with diabetes or documented macular pathology was computed from retinal images. The LSDC scans a slit of light across the retina, using 1 scanning element, and detects the light return with a 2-dimensional CMOS detector array. The 850 nm illumination is limited to 2 mW at the cornea. While we have acquired 1 Mp images at rates from 1 - 36 Hz, 11 - 19 Hz provides high quality retinal images. Using a smaller than 1 Mp region of interest, image series at rates up to 250 Hz are accurately digitized. This report is on 36 deg fovea-centered images, comparing automatic vs. manual alignment methods for assessing fixation stability. We used a series of 20 images for each eye, acquired at 11 Hz and 1 Mp while the patient viewed a fixation target. For the automatic method, a suitable region of interest for registration was found by performing a zero-lag cross-correlation between the base image with and without median filtering. The remaining 19 images were registered by cross-correlation. Images that did not meet the criteria for successful alignment were dropped from the average. Each image series was also registered manually, excluding only frames with blinks or large, intra-frame motion. The S.D. of the Euclidean distance between a retinal feature and its centroid was used as the metric for fixation stability.
The difference between the automatic and manual methods was not statistically significant, nor between left and right eyes (2-way ANOVA, p = .123, p = .907, respectively). The automatic method gave, on average, slightly lower values than the manual method (S.D. = 55 vs. 66 microns, respectively). This difference and outliers on Bland-Altman analysis were driven by data sets with the fewest images successfully aligned or large mean errors.
This automatic method works well to assess fixation stability in patients with good fixation and identifies those with poorer fixation.
This PDF is available to Subscribers Only