A system based on AI will scan the retina for signs of Alzheimer’s

THE DIFFERENT parts of a health-care system have different focuses. A hospital’s dementia unit keeps records of patients’ mental abilities. The stroke unit monitors blood flow in the brain. The cardiac unit is interested in that same flow, but through and from the heart. Each agglomeration of equipment and data is effective in its own domain, but for the most part has little relevance to other bits of the body and the conditions that plague them. Thus, like the proverbial blind men feeling an elephant, modern health care offers many fragmented pictures of a patient, but rarely a useful cohesive one.

On top of all this, the instruments that doctors use to monitor health are often expensive, as is the training required to wield them. That combined cost is too high for the medical system to scan regularly, for early signs of illness, all patients at risk of dementia, heart disease or a stroke. Rather, doctors work to manage symptoms after a disease has obviously taken hold.

An unusual research project called AlzEye, run from Moorfields Eye Hospital in London, in collaboration with University College, London (UCL), may change this. It is attempting to use the eye as a window through which to detect signals about the health of other organs. The doctors in charge of it, Siegfried Wagner and Pearse Keane, are linking Moorfields’ database of eye scans, which offer a detailed picture of the health of the retina, with information about other aspects of its patients’ health garnered from other hospitals around England. This will allow them to look for telltales of disease in the eye scans.

The data set includes, whether they know it or not, every one of the 300,000 patients who visited Moorfields between 2008 and 2018 and was over the age of 40—though names and other easily identifiable information are not preserved. The idea is to examine changes to people’s eyes within that ten-year period, and correlate these with, say, the emergence of Alzheimer’s disease in the same patient. Building such a data set while respecting privacy and confidentiality has been a challenge. It took the doctors two and a half years to go through a series of ethics-committee approvals at Moorfields, UCL and NHS Digital, the body which handles agglomerated data from English hospitals. In order to create the database without the consent of the patients involved they invoked a special legal provision known as Section 251 assent, which comes with its own review process and, in essence, empowers senior government health officials to give consent on behalf of patients from whom it would be impractical or impossible to acquire such individually. The data sets were linked together on November 1st, and the process of correlation is now under way.

The Moorfields data set has a lot of linked cases to work with—far more than any similar project. For instance, the UK Biobank, one of the world’s leading collections of medical data about individual people, contains 631 cases of a “major cardiac adverse event”. The Moorfields data contain about 12,000 such. The Biobank has data on about 1,500 stroke patients. Moorfields has 11,900. For the disease on which the Moorfields project will focus to start with, dementia, the data set holds 15,100 cases. The only comparable study had 86.

Drs Wagner and Keane are searching for patterns in the eye that betray the emergence of disease elsewhere in the body, and are focusing first, as the name AlzEye suggests, on Alzheimer’s disease. They will seek such patterns with the help of machine-learning algorithms that can crunch through imagery far faster than any human being, and which can spot far tinier variations. (The team has collaborated with DeepMind, a British artificial-intelligence firm co-founded by Mustafa Suleyman, who sits on The Economist’s board.) They may, it should be remembered, never find such patterns. Although there is circumstantial evidence that the back of the eye does change as its owner develops Alzheimer’s, it may be that the changes are too subtle to be detected reliably enough for diagnosis. If such patterns could be recognised reliably, though, the potential impact would be huge. Even in rich countries, between 50% and 80% of Alzheimer’s cases go undiagnosed. Moreover, even if the technique does not work for Alzheimer’s, it might work for something else. Dr Wagner and Dr Keane therefore plan further searches for patterns related to strokes and heart disease. Even one relevant pattern would constitute a remarkable diagnostic leap forward.

Seeing is believing

If it does work, the technique the two researchers are proposing will be cheap to implement. An indication of how cheap is the project’s total budget of just £15,000 (about $19,000). Equipment to perform an eye scan is becoming ubiquitous. Specsavers, which runs a chain of high-street opticians, now routinely offers the same sort of scans as Moorfields’ in half of its 800 branches. Costco, a bulk retailer, offers scans in Britain for £24.99 a pop. An Israeli company called Notal Vision is building an eye-scanning device that is small enough to operate at home. The equipment and algorithms required to run machine learning on an eye scan are available to anyone, through cloud-computing services like Google and Amazon. Dr Keane once joked that he would like to see eye-scanning widgets become so cheap that they could be bundled in boxes of cereal.

This project will also act as a model for linking disparate health data together in a useful way while respecting patients’ rights. Other such endeavours involving information-technology firms handling health data, such as Google’s work with Ascension hospitals in America, or other parts of its subsidiary DeepMind’s work with England’s health service, have generated controversy because they offered no notice of their plans to patients. As well as jumping deftly through all the required legal and ethical hoops, Drs Wagner and Keane also posted notices around the hospital and on Moorfields’ website, informing patients of the impending linkage. Besides explaining what was happening, these also noted the research’s potential benefits. Not one person complained. ■