Interactive Deep Learning-Based Retinal OCT Layer Segmentation Refinement by Regressing Translation Maps

Retinal layer segmentation in optical coherence tomography (OCT) is essential for the diagnosis and follow-up of several diseases. Despite the success of deep learning approaches for this task, their clinical applicability is limited, since they neither account for pathologies other than those present in the training set nor for the specialists’ subjectivity. Thus, we propose an interactive layer segmentation approach that allows to obtain an initial segmentation and, more importantly, to interactively correct those segmentations. Our deep learning-based approach predicts the translation required to correct layer boundary segmentations by regressing pixel-wise translation maps that account for the user input. The method is designed to allow for segmentation correction by interactions with point-clicks or line-scribbles. Additionally, the system outputs a coordinate-wise confidence, allowing to automatically identify regions of possible segmentation failure that may require user attention. We extensively validate our approach on multiple private and public datasets with different pathomorphological complexities, achieving state-of-the-art performance, while allowing for a simple and efficient user interaction.