ONHSD – Optic Nerve Head Segmentation Dataset
The optic nerve head is a key landmark in the retina. In this project we have developed
accurate methods based on active contours that exploit the particular appearance
properties of the ONH to accurately segment.
This dataset contains 100 fundal images taken from 50 patients randomly sampled
from a diabetic retinopathy screening programme; 96 images have discernable ONH.
The subjects are from various ethnic backgrounds (Asian 20%, Afro-Caribbean 16%,
Caucasian 50%, Unknown 14%); 19 have type 2 diabetes mellitus, while the diabetes
status was unavailable for the remaining 31. The images were acquired using a Canon
CR6 45MNf fundus camera, with a field angle lens of 45 degrees, resolution 640 x
480. Images were converted to grey-scale by extracting the Intensity component from
the HSI representation. There is considerable quality variation in the images, with
many characteristics that can affect segmentation algorithms. The ONH centre has
been marked up a clinician. Then, four clinicians marked the ONH edge where it intersects
with radial spokes (at 15 degree angles) radiating from the nominated centre. These
multiple nominations of the edge can be used to characterize the degree of subjective
uncertainty in the edge position.
The attached dataset allows for comparative study of alternative algorithms, and
includes MATLAB code for our method, and for performance evaluation.
The datasets can be used, free of charge, for research and educational purposes.
Copying, redistribution, and any unauthorized commercial use is prohibited. The
use of these datasets is restricted to those individuals or organizations that obtained
the datasets directly from this website. Any researcher reporting results which
use one of these datasets must acknowledge us. We request you to do so by citing
its relevant publication. Please also inform us of your results so that we can provide
a citation and record comparative results on the site.
Click here to download this dataset