################################################################################ ##### Dataset which underpins I2MTC paper [1] ##### ################################################################################ ##### Author Details ## Dr Mario E. Giardini (mario.giardini@strath.ac.uk) & Mr Ian Coghill (ian.coghill@strath.ac.uk) Department of Biomedical Engineering University of Strathclyde Glasgow, Scotland ##### Dataset Content ## The data were collected using the equipment and protocol outlined in [1], and briefly summarised below. It is organised in the following folders. A) 100 stereo image pairs of a calibration pattern [6560x2464, 1.165 mm square size, PNG format]* B) A 3D point cloud ground truth for each of the two optic nerve head targets imaged [ply format] C) 4 stereo image pairs of each of the two optic nerve head targets [6560x2464, PNG format] ##### License ## The dataset is licensed under a Creative Commons CC-BY 4.0 license (https://creativecommons.org/licenses/by/4.0/). Please attribute it to [1] ##### Protocol A) Calibration target: The calibration target contained a 10 x 7 checkerboard of squares, created photographically on Rollei RPX 25 black and white film (Rollei, Hamburg, Germany) by photographing a Philips BDM4350UC 43” monitor (Philips, The Netherlands) with 4K resolution displaying the pattern, using a Yashica FX3+ camera (Yashica, Japan) camera with stock 50 mm lens, and then having the film developed by a commercial developing lab. It was glued flat to a standard glass microscope slide using Loctite 4305. A thin white plastic film (from packaging) was taped to the other side of the slide to provide a white backdrop for the film which was transpartent with the black squares of the checherboard. The target was retroilluminated by a 5 mm through-hole white LED (RS Components Ltd, Corby, Northamptonshire, UK). The size of the individual squares within the checkerboard were determined by photogrammetery against a reference scale, in this case the legs of a set of calipers (Mitutoyo Ltd., Andover, Hampshire, UK) set at 10 mm. In this, a length of squares was measured and the value divided by the number of squares measured. This was carried out with 4 different images, measuring different sides of the checkerboard, and the mean square size determined was 1.165 mm (sample standard deviation [SD] = 0.005 mm). B) Ground Truth Data Acquisition: Ground truth scans of each of the targets were obtained using the commercial photogrammetry software 3DF Zephyr Free (3Dflow, Verona, Italy). In turn, 50 JPG images were aquired of the top surface of each target from different and random poses using the camera of a Samsung Galaxy S8 smartphone (Samsung, Seoul, South Korea), set to capture at a resolution of 4032 x 1960 pixels. The images acquired of each target were input into the software to yield proportionally accurate ground truth reconstructions. These were uniformly scaled in three dimensions to the correct size using the mesh processing software CloudCompare (CloudCompare v2.9.1, 2018). The scaling factor for each was determined through comparing measurements made on the surface of the physical targets (using the same measurement method as that used to measure the calibration checkerboard) to those made on the unscaled ground truth reconstructions using CloudCompare. Some speckle features of the targets were used as points where measurements could be made between. Four measurements were made for each target, and the mean scale factors calculated were used for scaling. C) Setup and Image Capture: The magnification setting on the slit lamp was set to 30 and fixed at this for the entirety of the trial. The cameras were focused to infinity and the eye pieces set to their zero-diopter setting. The axial position of the slit lamp lens was adjusted such that the imaging system was focused to infinity – allowing imaging of an emmetropic eye. One of the ONH targets was placed on the retina of the phantom, and on its optical axis. The LEDs that are inside the phantom that provide illumination to the targets from inside, were connected to a power supply and powered. Stereo images were acquired of the ONH target already inside the phantom. The target was then rotated 90° around the phantom’s optical axis and again imaged. This was repeated until 4 images were obtained, each with the target at a different rotation, and then the same was done for the other target after interchanging them, giving 4 repeated measurements of each target. The system was not re-aligned during the imaging session. ##### References [1] Coghill, I & Giardini, ME 2021, Stereo vision based optic nerve head 3D reconstruction using a slit lamp fitted with cameras: performance trial with an eye phantom. 2021 IEEE International Instrumentation and Measurement Technology Conference (I2MTC). IEEE, Piscataway, NJ. At the time of writing, this article is in press. The latest status and a preprint can be downloaded at https://pureportal.strath.ac.uk/en/publications/stereo-vision-based-optic-nerve-head-3d-reconstruction-using-a-sl