A Multi-task Network to Detect Junctions in Retinal Vasculature

Abstract

Junctions in the retinal vasculature are key points to be able to extract its topology, but they vary in appearance, depending on vessel density, width and branching/crossing angles. The complexity of junction patterns is usually accompanied by a scarcity of labels, which discourages the usage of very deep networks for their detection. We propose a multitask network, generating labels for vessel interior, centerline, edges and junction patterns, to provide additional information to facilitate junction detection. After the initial detection of potential junctions in junctionselective probability maps, candidate locations are re-examined in centerline probability maps to verify if they connect at least 3 branches. The experiments on the DRIVE and IOSTAR showed that our method outperformed a recent study in which a popular deep network was trained as a classifier to find junctions. Moreover, the proposed approach is applicable to unseen datasets with the same degree of success, after training it only once.

A Maximum Entropy Deep Reinforcement Learning Neural Tracker

“…Furthermore, the need for a smaller labelled dataset of only 2000 images increases the likelihood of training the tracker directly on biomedical datasets in cases where artificial models cannot be built easily. Future work could include training in synthetic multi-axon images and accounting for branching of neurons, potentially by combining the network with an automatic junction detection algorithm [22], or extending to subvoxel tracking through extending the environment to 3D, and introducing highly anisotropic spatial sampling into the environment-a common challenge in confocal imaging.…”

clDice -- a Novel Topology-Preserving Loss Function for Tubular Structure Segmentation

“…In stark contrast to previous works, where segmentation and centerline prediction has been learned jointly as multi-task learning [37,34], we are not interested in learning the centerline. We are interested in learning a topologypreserving segmentation.…”

Deep multi-instance heatmap regression for the detection of retinal vessel crossings and bifurcations in eye fundus images

“…The most commonly followed strategy is to split the problem into two different tasks: the general detection of vessel junctions, and the later classification of the detected junctions as crossings or bifurcations [12,13]. Additionally, there are several works that only tackle the first task, without facing the complex and difficult distinction between both types of landmarks [14,15].…”