Software aids monitoring of eye disease in diabetics
Because it affects the entire vascular system, including the vasculature of the eye, diabetes can lead to retinopathy — damage to the retina often associated with blood supply problems — and even to blindness. Clinicians can monitor patients for retinopathy by measuring retinal thickness, which generally ranges from 100 to 500 μm in healthy people and from 300 to 600 μm in those at risk for diabetic macular edema. Scanning laser ophthalmoscopes offer resolution on the order of 300 to 500 μm, however, and therefore cannot measure retinal thickness directly.
Lickenbrock Technologies LLC of Troy, N.Y., recently announced a software package that will address this limitation. Developed in collaboration with Heidelberg Engineering GmbH in Germany, and investigators from the University of Waterloo in Ontario, Canada, the software is sold as a component of Heidelberg’s scanning laser ophthalmoscope, the Heidelberg retinal tomograph — a clinical instrument that allows examination of the posterior segment of the human eye.
New software determines the thickness of a patient’s retina using images imported from a scanning laser ophthalmoscope. Signs of thickening may indicate the early stages of diabetic retinopathy; therefore, the software may aid in monitoring patientsfor the disease.
Previously, university researchers John Flanagan, Christopher Hudson and Natalie Hutchings had developed and tested software for the device to evaluate edema in the retina, particularly for diabetic eye disease. The software identified geographic areas of fluid in the retina in arbitrary units. Soon, Hutchings said, “it became obvious that being able to map edematous areas and concurrently identify retinal thickness (in microns) would be clinically advantageous.”
In 1998, Flanagan and Gerhard Zinser, a developer and one of the owners of Heidelberg Engineering, approached Tim Holmes of Lickenbrock (then AutoQuant Imaging) and asked whether the company could develop software that would pull the necessary information from retinal scans. AutoQuant was working on a number of projects related to image processing. “We thought we were going to adapt existing software,” Holmes recalled, “but it just didn’t work. We really had to start over from scratch.”
The problem, he said, was that the existing software was geared toward microscope images of cells. “When applied to the retina in live human subjects, the variability you get is very different from that in cells under a microscope.”
The software now imports images of the retina from the scanning laser ophthalmoscope. From these, it calculates the thickness of the retina and then produces a complete map of this thickness across the retina. If a patient’s retina shows signs of thickening, the doctor recognizes it as possibly being the early stages of diabetic retinopathy and knows to run additional tests.
One of the advantages of the software is that it provides a high-resolution lateral map of the retina; other techniques provide good resolution axially, but not laterally. “As with all optical measures of the retina,” Flanagan said,” there are confounding effects of the clinical consequences of the disease process — aside from the increase in retinal thickness —such as hemorrhage and exudate. This is where the advantage of high-resolution lateral maps has been most useful.”
The researchers reported the algorithm used to record retinal thickness — and thus to monitor for retinopathy in diabetics — in a paper presented at the SPIE medical imaging conference in February (and to be published by SPIE soon). Using a technique called “maximum likelihood estimation,” the algorithm takes an image from the scanning laser ophthalmoscope and finds the positions of the internal limiting membrane and the retinal pigment epithelium — the two boundaries of the fundus that delimit the retina. Then it calculates retinal thickness by subtraction. The researchers view this approach as a trade-off between tractability of the solution — allowing computation speeds of about 1 min on a 2.5-GHz PC — and mathematical rigor.
They tested the algorithm using images from 50 subjects and comparing the retinal thickness with that measured using optical coherence tomography — a technique that offers superior axial resolution and therefore served as a good standard for comparison. They observed good agreement between the measurements, thus validating the algorithm for use in determining retinal thickness.
The researchers continue to develop the software and are constantly making improvements to make it more robust with respect to variations in thickness in different types of subjects, Holmes explained. They note, for example, that there are likely better basis functions than that currently used in one of the equations, although they initially tested a number of others. They hope to explore other incremental improvements to the algorithm as well.
Contact: Tim Holmes, Lickenbrock Technologies LLC; e-mail: email@example.com; Natalie Hutchings, University of Waterloo; e-mail: firstname.lastname@example.org; John Flanagan, University of Waterloo; e-mail: email@example.com.
- 1. The photosensitive membrane on the inside of the human eye. 2. A scanning mechanism in optical character generation.
MORE FROM PHOTONICS MEDIA