Purchase this article with an account.
H. L. Klein, I. Kozak, S. Cheng, A. Gamst, L. Cheng, D. Bartsch, W. R. Freeman; A Correlation Between Manual and Automated Drusen Segmentation on Sections of Spectral Domain Optical Coherence Tomography (SDOCT) Images. Invest. Ophthalmol. Vis. Sci. 2010;51(13):2773.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To evaluate methods of drusen segmentation on image stabilized raster SD-OCT scans. Currently, there is no gold standard for quantification of abnormal drusen material in the living human eye. We therefore sought to assess inter-observer reproducibility of manual drusen area and volume measurements, and to develop an automated algorithm for drusen volume.
11 eyes of 7 patients with drusen associated with dry AMD were studied. SD-OCT using image averaging and eye tracking (Heidelberg engineering, Vista CA) was used. A raster protocol yielded 97 B-scan images in an area 6x6 mm. Manual drusen segmentation was performed by two independent observers. Image J software was used to delineate drusen area. For automated segmentation, sections were converted to a pixmap library. An algorithm created a two-dimensional map of image intensity and gradient information to isolate drusen. Factors such as reflectance and similarity to surrounding tissue were used to define drusen boundaries.
236 scans were scored by two different graders. There was a high concordance correlation coefficient between the two graders with concordance of 0.98, 95% CI -0.974-0.99. Evaluation of two measurement methods, every other vs. every raster scan, showed no difference (1.83%, p=0.38) suggesting that manual segmentation time can be reduced by analyzing every other scan. Preliminary analysis of automated segmentation (below) shows good correlation between the automated algorithm (top figure) and the manual segmentation (lower figure).
Heidelberg Spectralis image stabilized raster scan protocol allows determination of drusen volume. Refinement of automated algorithms to determine volume without manual segmentation will be extremely useful in the study of AMD.
This PDF is available to Subscribers Only