ROSE: A Retinal OCT-Angiography Vessel Segmentation Dataset and New Model

Retinal OCT-Angiography Vessel Segmentation Dataset (ROSE) is now open source. It includes two subsets: ROSE-1 and ROSE-2. ROSE-1 subset consists of a total of 117 OCTA images from 39 subjects (26 with disease and the rest are healthy control) and was split into 90 images for training and 27 images for testing. Each subject has en face angiograms of superficial (SVC), deep (DVC), and the inner retinal vascular plexus that includes both SVC and DVC (SVC+DVC) respectively. All the OCTA scans were captured by the RTVue XR Avanti SD-OCT system (Optovue, USA) equipped with AngioVue software, with image resolution of 304 × 304 pixels. The scan area was 3×3 mm2 area centred at the fovea. The manual annotations of these vascular network were graded by image experts and clinicians, and the consensus of them was then used as the groundtruth. Two different types of vessel annotations are available: centerline-level and pixel-level annotation. ROSE-2 subset contains a total of 112 OCT-A images of 112 eyes acquired by Heidelberg OCT2 system with Spectralis software (Heidelberg Engineering, Heidelberg, Germany) and was split into 90 images for training and 22 images for testing. All the images in this dataset are the en face angiograms of SVC within a 3×3 mm2 area centred at the fovea, and each image was resized into a grayscale image with 840 × 840 pixels. All the visible retinal vessels were manually traced using an in-house program written in Matlab (Mathworks R2018, Natwick) by an experienced clinical ophthalmologist.视网膜光学相干断层扫描血管造影（Optical Coherence Tomography Angiography, OCT-A）血管分割数据集（Retinal OCT-Angiography Vessel Segmentation Dataset, ROSE）现已开源。该数据集包含两个子数据集：ROSE-1与ROSE-2。 ROSE-1子数据集共包含来自39名受试者的117幅OCT-A图像（其中26名患有疾病，其余为健康对照），并被划分为90幅训练集与27幅测试集。每名受试者分别拥有浅层视网膜血管丛（Superficial Vascular Complex, SVC）、深层视网膜血管丛（Deep Vascular Complex, DVC）以及包含SVC与DVC的内层视网膜血管丛（SVC+DVC）的正面扫描（en face）血管造影图像。所有OCT-A扫描均由搭载AngioVue软件的RTVue XR Avanti光谱域光学相干断层扫描（Spectral-Domain Optical Coherence Tomography, SD-OCT）系统（美国Optovue公司）采集，图像分辨率为304×304像素，扫描区域为以黄斑中心凹为中心的3×3 mm²区域。该数据集的血管网络手动标注由影像专家与临床医师完成，经协商一致后作为基准真值（groundtruth）。标注包含两种类型：中心线级标注与像素级标注。 ROSE-2子数据集共包含112幅来自112只眼睛的OCT-A图像，由搭载Spectralis软件的Heidelberg OCT2系统（德国海德堡工程公司，海德堡）采集，并被划分为90幅训练集与22幅测试集。该数据集内所有图像均为以黄斑中心凹为中心的3×3 mm²区域内的SVC正面扫描血管造影图像，且所有图像均被调整为840×840像素的灰度图像。所有可见视网膜血管均由一名经验丰富的临床眼科医师使用Matlab（MathWorks R2018, Natwick）编写的自研程序进行手动追踪标注。