Hyperreflective foci are poorly understood transient elements seen on optical coherence tomography (OCT) of the retina in both healthy and diseased eyes. Systematic studies may benefit from the development of automated tools that can map and track such foci. The outer nuclear layer (ONL) of the retina is an attractive layer in which to study hyperreflective foci as it has no fixed hyperreflective elements in healthy eyes. In this study, we intended to evaluate whether automated image analysis can identify, quantify and visualize hyperreflective foci in the ONL of the retina.Methods: This longitudinal exploratory study investigated 14 eyes of seven patients including six patients with optic neuropathy and one with mild non-proliferative diabetic retinopathy. In total, 2596 OCT B-scan were obtained. An image analysis blob detector algorithm was used to detect candidate foci, and a convolutional neural network (CNN) trained on a manually labelled subset of data was then used to select those candidate foci in the ONL that fitted the characteristics of the reference foci best. Results:In the manually labelled data set, the blob detector found 2548 candidate foci, correctly detecting 350 (89%) out of 391 manually labelled reference foci. The accuracy of CNN classifier was assessed by manually splitting the 2548 candidate foci into a training and validation set. On the validation set, the classifier obtained an accuracy of 96.3%, a sensitivity of 88.4% and a specificity of 97.5% (AUC 0.989). Conclusion:This study demonstrated that automated image analysis and machine learning methods can be used to successfully identify, quantify and visualize hyperreflective foci in the ONL of the retina on OCT scans.

We address a subclass of segmentation problems where the labels of the image are structured in layers. We propose applying autoregressive CNNs which, when given an image and a partial segmentation of layers, complete the segmentation. Initializing the model with a user-provided partial segmentation allows for choosing which layers the model should segment. Alternatively, the model can produce an automatic initialization, albeit with some performance loss. The model is trained exclusively on synthetic data from our data generation algorithm. It yields impressive performance on the synthetic data and generalizes to real data it has never seen.