What better time to take a good look at the state of ophthalmology research than the year 2020? The proverbial expression for good eyesight has its roots in a standard eye test, where being able to see 20/20 means that you can see objects at 20 feet away just as well as a person with normal vision sees them at that distance.
Not everyone has 20/20 vision, but advances in eye imaging techniques are making it possible for experts to study and treat vision-limiting eye conditions.
For 215 years, Moorfields Eye Hospital in London has been at the forefront of ophthalmic imaging for research and patient care. The images in this gallery highlight some of the advances made over the years by the hospital’s researchers and clinicians.
The first glimpse at the back of the eye
When physicist and physiologist Hermann von Helmholtz invented the first ophthalmoscope in 1851, ophthalmologists finally got a good look at the back of the living eye. With this new technology, they were able to study the retina in detail. Helmholtz made four copies of his new instrument and gave one of them to his colleague William Bowman at Moorfields Eye Hospital.
Early ophthalmoscopes used reflected light to allow study of the back of the eye without dazzling the ophthalmologist.
Soon other ophthalmologists started using similar devices. “Some of the ophthalmologists made adaptations or designed their own ophthalmoscopes,” says Debbie Heatlie, librarian of the Joint Library of Ophthalmology at Moorfields Eye Hospital and the UCL Institute of Ophthalmology. “They drew and published their own images so that others could see what diseases of the eye looked like.”
Just a few years after the development of the ophthalmoscope, Moorfields started hiring professional artists to create realistic depictions of the fundus, the back of the eye. These images were used in teaching or to accompany research publications and case studies.
An illustration by Mary Boole from 1883, showing abnormal blood vessels near the optic disk in the eye of a syphilis patient.
The practice of ophthalmological illustration to depict different conditions of the eye was still in use well into the twentieth century. A notable illustrator at Moorfields was Terry Tarrant, whose detailed and bold watercolour paintings of the retina have been used in many publications and textbooks.1 “He would take along colour pencils when examining a patient and make a quick sketch of what he saw,” says Heatlie. “Then he would come back to his drawing board and paint up the image.” The full process from sketch to water colour could take up to a day.
Camera technology wasn’t good enough to reliably image the back of the eye until the 1980s.
One of Terry Tarrant’s many water colour images. This painting from 1976 shows retinal detachment in the eye of a diabetic patient, with a mass of tissue obscuring the optic nerve.
Seeing the eye in three dimensions
Since the 1990s, ophthalmologists have been able to peer underneath the retina thanks to optical coherence tomography (OCT). This method uses light-scattering detection to image different layers of biological tissues. OCT is now in widespread use in ophthalmology.
Initially, OCT showed a single cross-section of the retinal layers, but further advances in the technology made it possible to recreate the three-dimensional form of the retina and its blood supply by assembling stacks of these cross-sections imaged at different locations in the eye. Ophthalmologists usually analyse these images by looking at them on a two-dimensional computer screen. That was until Moorfields’ research fellow Peter Maloca together with Adnan Tufail and Philippe Cattin - in collaboration with Rick Spaide - developed a way to explore these 3D data in a virtual reality environment.
The retinal vessels of a healthy eye, 3D printed in resin and gold-plated for durability. The area without vessels is the foveal avascular zone - the site of sharpest vision.
With these virtual reality retinal representations ophthalmologists are able to explore the 3D environment of a patient’s eye before surgery. They’re also used for teaching ophthalmology students, engaging patients or collaboration with experts who work at distant sites. For Maloca, who is also one of the group leaders at the Institute of Molecular and Clinical Ophthalmology in Basel, this means he can share images from Switzerland directly with his colleagues at Moorfields and have everyone interact with the same 3D model of a patient’s eye.
Using a head-mounted display, patients can explore the VR model of their eye, and manipulate it with the hand grips to rotate or enlarge the model. The physician is able to interact within this same VR environment and explain the model.
AI for the eye
OCT machines are becoming more prominent, and have started to make their way into high street optometrists. Researchers are also working on future models of the equipment, which will even be able to measure both eyes simultaneously to gather more information.4
Representation of a binocular optical coherence tomography system, in which both eyes can be viewed at the same time, allowing for the collection of additional types of measurements. Such machines are still in the early stages of testing at the moment, but may eventually replace current OCT machines that measure eye at a time.
An early and accurate diagnosis based on OCT images taken at an optometrist appointment would ensure that patients get the best care, but interpreting these OCT images requires a trained eye to recognise unusual features that suggest underlying eye conditions. A team at Moorfields is collaborating with Google Health to study whether this type of image analysis can be carried out by artificial intelligence (AI).
“We train the algorithm first to segment the images,” says research optometrist Reena Chopra. Here, the AI learns which features of the image are important. “Then it goes through a classification network to give a diagnosis.” Thanks to the AI’s feature segmentation the clinician is able to get an indication of the reasoning behind the AI’s decision.
In initial tests the AI algorithm was able to diagnose common eye conditions from OCT images alone.5 It performed equally as well as expert ophthalmologists and optometrists who not only had the images, but also the patient’s clinical notes to consult.
An AI algorithm trained to interpret OCT images first assigns different segments to relevant parts of the image, which it will then use those to make a diagnosis prediction.
Chopra hopes that AI analysis can address a bottleneck in patient referrals at Moorfields. At the moment, too many patients are unnecessarily given an urgent referral, which delays treatment for those patients who do need more urgent care.
It will be a while until this system is ready to roll out at a larger scale. But if clinical trials are successful, AI image analysis is a promising tool for telemedicine -- using phone and internet to diagnose and advise patients wherever they may be.