2017/07/24 by Sripad Krishna Devalla, Jean Martial Mari, Devalla, Sripad Krishna +11
Engineering · Medicine · #Optical Coherence Tomography Applications #Glaucoma and retinal disorders #Retinal Imaging and Analysis
paper · pdf · doi:10.48550/arxiv.1707.07609
Purpose: To develop a deep learning approach to digitally-stain optical\ncoherence tomography (OCT) images of the optic nerve head (ONH).\n Methods: A horizontal B-scan was acquired through the center of the ONH using\nOCT (Spectralis) for 1 eye of each of 100 subjects (40 normal & 60 glaucoma).\nAll images were enhanced using adaptive compensation. A custom deep learning\nnetwork was then designed and trained with the compensated images to digitally\nstain (i.e. highlight) 6 tissue layers of the ONH. The accuracy of our\nalgorithm was assessed (against manual segmentations) using the Dice\ncoefficient, sensitivity, and specificity. We further studied how compensation\nand the number of training images affected the performance of our algorithm.\n Results: For images it had not yet assessed, our algorithm was able to\ndigitally stain the retinal nerve fiber layer + prelamina, the retinal pigment\nepithelium, all other retinal layers, the choroid, and the peripapillary sclera\nand lamina cribrosa. For all tissues, the mean dice coefficient was 0.84 \±\n0.03, the mean sensitivity 0.92 \± 0.03, and the mean specificity 0.99 \±\n0.00. Our algorithm performed significantly better when compensated images\nwere used for training. Increasing the number of images (from 10 to 40) to\ntrain our algorithm did not significantly improve performance, except for the\nRPE.\n Conclusion. Our deep learning algorithm can simultaneously stain neural and\nconnective tissues in ONH images. Our approach offers a framework to\nautomatically measure multiple key structural parameters of the ONH that may be\ncritical to improve glaucoma management.\n