LUNet: Deep Learning for the Segmentation of Arterioles and Venules in High Resolution Fundus Images

The retina is the only part of the human body in which blood vessels can beaccessed non-invasively using imaging techniques such as digital fundus images(DFI). The spatial distribution of the retinal microvasculature may change withcardiovascular diseases and thus the eyes may be regarded as a window to ourhearts. Computerized segmentation of the retinal arterioles and venules (A/V)is essential for automated microvasculature analysis. Using active learning, wecreated a new DFI dataset containing 240 crowd-sourced manual A/V segmentationsperformed by fifteen medical students and reviewed by an ophthalmologist, anddeveloped LUNet, a novel deep learning architecture for high resolution A/Vsegmentation. LUNet architecture includes a double dilated convolutional blockthat aims to enhance the receptive field of the model and reduce its parametercount. Furthermore, LUNet has a long tail that operates at high resolution torefine the segmentation. The custom loss function emphasizes the continuity ofthe blood vessels. LUNet is shown to significantly outperform twostate-of-the-art segmentation algorithms on the local test set as well as onfour external test sets simulating distribution shifts across ethnicity,comorbidities, and annotators. We make the newly created dataset open access(upon publication).