Artificial Intelligence and eye disease
Hailed as a “jaw-dropping” breakthrough, computers can use artificial intelligence (AI) to effectively diagnose eye disease such as macular degeneration. Moorfields Eye Hospital in London has been conducting research into the use of AI to assess OCT scans of patients with retinal disease.
AI is the imitation of human cognitive processes by a machine in which computer algorithms learn from data without human direction. It is already in use in various consumer technologies such as Apple’s Siri in the iPhone. Such advances are expected to continue and accelerate.
There are several research projects currently being undertaken internationally that are evaluating the use of AI technology in eye disease. One such recently completed study (De Fauw et al, 2018) used 14,884 high definition OCT retinal scans to teach the computer how to spot features of eye disease. AI technology looks for signs such as haemorrhage and fluid leakage, and has the ability to adapt to different types of retinal scanners.
Scans were then taken from 1000 patients at Moorfields Eye Hospital and given to the computer to analyse. Results were compared to those read by 4 ophthalmologists and 4 specialist optometrists.
Both the computer and clinician groups were asked to decide whether a patient required urgent referral, semi-urgent referral, routine referral or observation. The results were incredibly accurate with the AI system achieving, and in some cases exceeding, expert interpretation by clinicians.
Benefits include earlier time to assessment and diagnosis for patients and the ability to prioritise patients for assessment at a much earlier stage than is currently the case.
AI is not going to replace consultant ophthalmologists. AI will not be used to simply recommend a patient for an injection into their eye on the basis of a machine-generated response. However, it will ensure patients see an appropriate specialist at the earliest possible point, get treatment early and result in better visual outcomes for patients.
AI is also being used to increase the speed and accuracy of diabetic retinopathy screening and will likely result in earlier detection and treatment (Fogel AL, et al. 2018).
Ongoing research into AI technology is necessary to ensure the technology is robust and continues to meet a standard comparable to clinical experts.
References
De Fauw J, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease.
Nat Med. 2018 Aug 13. [Epub ahead of print].
Fogel AL, et al. Artificial intelligence powers digital medicine. npj Digital Med. 2018 Mar 14. www.nature.com/npjdigitalmed
