Purchase this article with an account.
RIMA GHASHUT, Tomoaki Murakami, Akihito Uji, Kiyoshi Suzuma, Masahiro Fujimoto, Shin Yoshitake, Yoko Dodo, Rina Yoza, Nagahisa Yoshimura; Automatic detection of hyperreflective foci on en face images of swept-source optical coherence tomography in diabetic retinopathy. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):4697.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Optical coherence tomography (OCT) delineates hyperreflective foci which may correspond to lipid-laden macrophages or the precursor of hard exudates in diabetic retinopathy (DR). Hyperreflective foci represent the extravasated blood components as well as have clinical relevance in diabetic macular edema. We thus investigated a novel method for automatic detection of hyperreflective foci on en face images of swept-source (SS)-OCT in DR.
We retrospectively investigated 7 eyes of 7 patients suffering from DR (2 eyes with moderate nonproliferative diabetic retinopathy [NPDR], 3 eyes with severe NPDR, 2 eyes with proliferative diabetic retinopathy) on whom 3x3 mm volume scans of sufficient quality were acquired using SS-OCT (DRI OCT-1, Topcon). After B-scan images were aligned by the flattening using the Bruch’s membrane, en face OCT images were constructed. In order to emphasize three-dimensional spherical objects, ‘Laplacian of Gaussian (LoG) 3D filter’, a plugin function of ImageJ, was applied to consecutive 6 slices of en face images, followed by ‘Projection’ function of ‘Volume viewer’ in ImageJ.
The combined methods of LoG3D and Volume viewer function provided better segmentation of hyperreflective foci on en face OCT images in eyes with DR. We compared the raw images to the processed one in three slices from each eye, and found that the novel methods for automatic detection provided higher sensitivity ([true positive]/[true positive + false negative] = 0.832±0.119) and moderate positive predictive value ([true positive]/[true positive + false positive] = 0.621±0.184). The number of ‘true positive’ hyperreflective foci was significantly correlated to the number in the raw image (R=0.995, R<0.001) and the total number detected automatically (R=0.988, P<0.001).
We demonstrated a novel method for automatic detection of hyperreflective foci in DR, whose sensitivity and positive predictive values may guarantee the objective evaluation of hyperreflective foci to some extent.
This PDF is available to Subscribers Only