Towards Automatic Glaucoma Assessment: An Encoder-decoder CNN for Retinal Layer Segmentation in Rodent OCT images

Abstract

Optical coherence tomography (OCT) is an important imaging modality that is used frequently to monitor the state of retinal layers both in humans and animals. Automated OCT analysis in rodents is an important method to study the possible toxic effect of treatments before the test in humans. In this paper, an automatic method to detect the most significant retinal layers in rat OCT images is presented. This algorithm is based on an encoder-decoder fully convolutional network (FCN) architecture combined with a robust method of post-processing. After the validation, it was demonstrated that the proposed method outperforms the commercial Insight image segmentation software. We obtained results (averaged absolute distance error) in the test set for the training database of 2.52 ± 0.80 μm. In the predictions done by the method, in a different database (only used for testing), we also achieve the promising results of 4.45 ± 3.02 μm.