Researchers have found that an artificial intelligence (AI) system is able to recommend patient referrals for over fifty eye conditions as accurately as world-leading doctors. The project is part of a collaboration between Moorfields Eye Hospital in England and Deepmind, a company that specialises in AI. For more information, you can read the source article here, at Deepmind’s website. To read the original research paper in full, scroll to the bottom of Deepmind’s article where there is a link to an open-access copy of the text.
The Current Process
At the moment, doctors typically use 3D scans of the back of a patient’s eye, called OCT scans, to diagnose conditions. However, this is an expensive and time-consuming process that can lead to delays for patients who may require a fast referral.
Only highly trained specialists are able to interpret the scans, and, at Moorfields Hospital, they typically review over 1,000 OCT scans per day. The time needed to analyse OCT scans combined with the large number that are taken can lead to slow patient referral, with potentially serious consequences if a condition is urgent.
How an AI System Could Help
Moorfields and Deepmind’s collaboration explored the possible benefit of using an AI system to streamline this process, which could reduce the burden on healthcare workers and speed up the process for patients.
The experimental AI system the researchers used is based on two ‘neural networks’ – one that analyses the OCT scan for features of a disease, such as lesions or haemorrhages, and a second ‘classification’ network that uses these features to make recommendations for patient referral. These recommendations are given in a percentage form so that doctors can see how confident the system is in its conclusions. Importantly, this format allows doctors to understand the system’s reasoning process and why it makes certain recommendations. The patients that appear to have the most urgent problems are automatically prioritised by the system.
In addition, the researchers designed the technology to be applied to various types of eye scanners. Crucially, this means that it could be used in many hospitals around the world, as well as in Moorfields. It also means that as eye scanners are replaced and updated the AI system could continue to be used.
Future Use
However, it is still early days for this experimental technology. To be used in hospitals, it would need to be made into a product, tested in clinical trials, and approved by healthcare regulators. Despite this, Deepmind is optimistic that it could eventually be used to improve healthcare for eye conditions. Deepmind says of the technology, “we don’t just want this to be an academically interesting result – we want it to be used in real treatment.”
