Page 267 - Computational Retinal Image Analysis

5 Datasets 265

AMD trials, with 128 participants on the AMD-only trial and 3512 on the
AMD and Cataract Trial. Color stereoscopic fundus images were obtained from
mydriatic eyes with a Zeiss FF series 30 camera (Carl Zeiss AG, Oberkochen,
Germany) in field 2 (centered above the macula). A grading system for AMD
was developed and participants were followed at 6-month intervals [100].
5. A dataset of 45 volumetric OCT acquisitions was developed at Duke
University, Harvard University, and the University of Michigan [101]. The
dataset was acquired using the Heidelberg Spectralis SD-OCT (Heidelberg
Engineering Inc.) and included normal eyes, eyes showing dry AMD, and
DME, with 15 subjects per class. These were used for develop a support vector
machine-based classifier, which was able to correctly identify 100% of AMD
and DME cases, and 86.7% of the normal cases. The set can be downloaded
from http://www.duke.edu/~sf59/Srinivasan_BOE_2014_dataset.htm.
6. A dataset was developed with 148 macular OCT images along with SLO
at Noor Eye Hospital in Tehran [102]. In the dataset, 50 normal eyes were
included, as well as 48 showing dry AMD and 50 with DME. The dataset can
be downloaded from http://www.biosigdata.com/?download=macular-oct-data-
from-heidelberg-imaging-device. The authors used the dataset to develop a
multi-scale convolutional mixture of expert model along with the Duke Dataset.
7. OCTID is a dataset of over 500 OCT images, classified with a variety of
conditions, including normal, AMD, diabetic retinopathy, central serous
chorioretinopathy, and macular holes [103]. The images were acquired using a
Cirrus HD-OCT (Carl Zeiss Meditec, Inc., Dublin, CA) at Sankara Nethralaya
eye hospital, Chennai, India The dataset can be downloaded at https://
dataverse.scholarsportal.info/dataverse/OCTID.
8. A large dataset of 1200 retinal fundus images from non-AMD and AMD
subjects was collected by Baidu Research (Sunnyvale, CA, United States) for
the REFUGE challenge. The image set, which can be downloaded at https://
refuge.grand-challenge.org/, includes diagnosis classifications (AMD/non-
AMD), disc boundaries, fovea locations and the boundaries of lesions to allow
for the training of AI models for AMD diagnosis.
9. The Kooperative Gesundheitsforschung in der Region Augsberg study has
been following participants at 5-year intervals since 1984, focusing on
research into old age since 2008. Images of the full macular region and optic
disc of nonmydriatic eyes were acquired from participants with a 45 degrees
Topcon TRC-NW5S fundus camera (Topcon Corporation, Tokyo, Japan).
Researchers can register to access this dataset at https://epi.helmholtz-
muenchen.de/.
10. The UK Biobank is a national and international resource, which follows
the health of 500,000 volunteers, with the aim of improving the prevention,
diagnosis and treatment of serious and life threatening illnesses, including eye
disorders such as AMD. The collected health information can be provided to
approved researchers in academia and industry, in the United Kingdom and
overseas. Researchers can register at https://www.ukbiobank.ac.uk/.