A New Retinal OCT-Angiography Diabetic Retinopathy Dataset(ROAD)
--------------------------------------------------------------------------------------------------------------------------------------------
A New Retinal OCT-Angiography Diabetic Retinopathy Dataset (ROAD). Projective maps attention-based convolutional neural network (PACNet). 2023.2.1.,
Permission to use copy, or modify this dataset, tool and codes for educational and research purposes.
E-mail : mafei0603@163.com
Homepage : https://mip2019.github.io/ROAD
--------------------------------------------------------------------------------------------------------------------------------------------
1. Dataset Description
Diabetic retinopathy (DR) is one of the major causes of vision loss even blindness in adults suffering from diabetes. With the large number of retina-disease patients, there is a urgent need for automatic recognition of retinal diseases. The wide-field optical coherence tomography angiography (WF-OCTA) is non-invasive imaging technique and convenient to diagnosis diabetic retinopathy. OCTA has the potential efficacy in the analysis of common ophthalmologic diseases, such as the clinical capability of delineating pathology information and obtaining the high-resolution fundus vessel. A new retinal OCT-Angiography diabetic retinopathy dataset(ROAD) is imaged by OCTA. The newly reconstructed ROAD dataset includes 1200 no DR images, 1440 DR images and 1440 ground truth of segmentation for DR images, which is based on DRAC in 2022. The ROAD dataset consists of three grades: no DR, mild non-proliferative diabetic retinopathy (NPDR), and proliferative diabetic retinopathy (PDR). The ground truth of lesion region for DR images in ROAD has also been labeled by professional ophthalmologists. The new DR dataset will be helpful for development of the early detection of DR field and the future researches.
2.1 The making ground-truth tool developed by our team can be downloaded with URL：MakeGroundtruthTool_v1.01 (windows desktop app at .netframework2.0).This software is a specialized tool to make the ground truth from original samples under complex scene. The ground-truth images can be obtained by this tool with the help in Fig.2, which is developed by our team. This application is run under .netframework2.0(win-x64) with windows 10 (x86 or x64).
2.2 The ground-truth tool developed by our team can be downloaded with URL：MakeGroundtruthTool_v1.0 (windows desktop app at .net 5.0).This software is a specialized tool to make the ground truth from original samples under complex scene. The ground-truth images can be obtained by this tool with the help in Fig.2, which is developed by our team. This application is run under .net 5.0 runtime(win-x86) with windows 10 (x86 or x64).
2.3 The key code for PACNet can be downloaded here (Pytorch).
--------------------------------------------------------------------------------------------------------------------------------------------
（For the reviewers）FIGURE 1 The thumbnail view of 400 samples in the ROAD dataset
The normal photos:
The DR images for Grade 1:
The DR images for Grade 2:
The groundtruth for Grade 1:
The groundtruth for Grade 2:
--------------------------------------------------------------------------------------------------------------------------------------------
FIGURE 2 The help for the ground-truth tool
This tool can be used to make many kinds of fields for the groundtruth. The following illustrations are the typical applications but not limited to these.
