Automatic Segmentation of Choroid Layer Using Deep Learning on Spectral Domain Optical Coherence Tomography

The purpose of this article is to evaluate the accuracy optical coherence tomography (OCT) measurement choroidal thickness in healthy eyes using a deep-learning method with Mask R-CNN model. Thirty EDI-OCT thirty patients were enrolled. A mask region-based convolutional neural network (Mask R-CNN) model composed deep residual (ResNet) and feature pyramid networks (FPNs) standard convolution fully connected heads for box prediction, respectively, was used automatically depict choroid layer. average subfoveal measured. results study showed that ResNet 50 layers (R50) 101 (R101). R101 U R50 (OR model) demonstrated best an error 4.85 pixels 4.86 pixels, respectively. ? (AND took least time execution 4.6 s. Mask-RCNN models good prediction rate layer rates 90% 89.9% thickness, In conclusion, provides faster accurate thickness. Comparing manual delineation, it better effectiveness, which feasible clinical application larger scale research on choroid.