Purchase this article with an account.
Maciej Szkulmowski, Daniel Ruminski, Pawel Liskowski, Bartosz Wieloch, Krzysztof Krawiec, Bartosz Sikorski, Maciej D Wojtkowski; OCT retinal angiography using neural networks. Invest. Ophthalmol. Vis. Sci. 2016;57(12):454.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To demonstrate noninvasive visualization of retinal microcapillary network (RMN) in retinal diseases with deep convolutional neural network (CNN) using data from recently developed device combining Scanning Laser Ophthalmoscope (SLO) and Spectral Optical Coherence Tomography (SOCT). SLO system provides fast eye tracking system while SOCT delivers 3D data for knowledge-free vessel segmentation technique.
The study was performed with ultra high resolution and high speed SOCT laboratory setup (100,000 Ascans/sec, 4.5 um axial resolution, 91 dB detection sensitivity). Constant 30 Hz retinal preview is provided by the SLO device and is used to guide the SOCT scanning beam to the region of interest. To increase RMN visualization area mosaic protocols were used. In case of eye blink or data corruption SOCT guided by SLO can still provide good quality RMN maps as corrupted B-scans are rejected and reacquired. RMS maps are created from 3D SOCT data using supervised machine learning algorithm using deep convolutional neural network (CNN) trained on data acquired from a set of 10 eyes (from both healthy volunteers and patients with retinal diseases) with vessels labeled by three independent skilled specialists. Each labeled voxel with a small cube of neighboring voxels forms an example. Examples are split into disjoint training and test sets. The CNN consists of 7 layers and is trained by stochastic gradient descent with batch updates and momentum (equivalent to multinomial logistic regression).
Trained CNNs provide sensitivity and specificity for RMN detection in training sets between 0.95 and 0.98 depending on training algorithm. We will show RMN maps obtained using CNNs with both proposed approach for 4 healthy volunteers and 12 patients with diabetic retinopathy, branch retinal vein occlusion and central retinal vein occlusion. We will compare the maps obtained using CNN with maps obtained using standard phase-variance angiographic algorithms.
Our results shows that CNN approach to RMN visualization provides accurate vessel detection incorporating a priori knowledge of skilled specialists and allows for increased sensitivity and specificity of SOCT based angiography.
This is an abstract that was submitted for the 2016 ARVO Annual Meeting, held in Seattle, Wash., May 1-5, 2016.
This PDF is available to Subscribers Only