AI and Parkinson's - a sprint and a marathon
In the year before I was diagnosed with Parkinson’s in January 2019, it seemed that excitement about the potential of artificial intelligence to transform healthcare was reaching new heights. Indeed, just a couple of months before my diagnosis was confirmed I reported on a collaboration between the Chinese tech giant Tencent and the London medtech firm Medopad to use AI to spot Parkinson’s more quickly.
But since then we’ve all calmed down a bit about what AI can do and how quickly it can have an impact - four years on I haven’t heard much lately about the progress of that Tencent project or many others. And in any case, people with Parkinson’s learn to be patient because with a slow moving disease measuring the success of any new treatment takes time.
So when I talked recently to Parkinson’s UK research chief David Dexter about two major AI projects the charity is undertaking I knew not to expect him to be trumpeting major breakthroughs.
Still, the first project we discussed, a collaboration with the NHS X AI Skunkworks, was designed to move fast - that’s the nature of this kind of model of innovation. The idea was to create algorithms which could speed up the diagnosis of Parkinson’s in a brain donated to the brain bank at Hammersmith Hospital. At this centre of Parkinson’s research, which I visited in 2020, scientists examine brains donated by people who have the disease and some who do not.
But any Parkinson’s diagnosis on a live patient is a somewhat haphazard affair - while scans are getting more sophisticated, it’s only when the brain is dissected after death that the extent of the damage becomes clear. “In the brain bank,” explained Dr Dexter, “we have to diagnose that it's somebody with Parkinson's who's donated into the bank rather than some other Parkinson's-like disorder or whether they're a healthy donor. And it takes the neuropathologist about six to eight hours to diagnose a case.”
The pitch to NHS X was to test whether AI could be used to diagnose Parkinson’s by examining cross sections of the brain and detecting alpha-synuclein, the “bad protein” thought to be at the root of the disease. And the researchers wanted to go further, developing an algorithm which could work out how far the disease had progressed. “We're not very good at actually quantifying things visually,” says David Dexter. “So one of the other things that would be a benefit is could we actually quantify the level of pathology in each brain.”
The Skunkworks process gives researchers just 12 weeks - a “sprint” in the jargon - to come up with a result. In collaboration with Polygeist, an AI company that works on defence and law enforcement applications as well as healthcare, the researchers have been training their algorithms on 300 brains from the Brain Bank. Dr Dexter says that after just three weeks they were confident that their AI could. successfully detect Parkinson’s and next week when the sprint ends they’ll evaluate their efforts to measure the progress of the disease.
Sounds great, I say, before asking as gently as possible why people with Parkinson’s should get excited about a way of detecting the condition once they’re dead. “Working on the brains of people who've had Parkinson's is the only way that we're going to find out what causes it,” he explains. “Speeding up the new pathological diagnosis and quantifying the level of pathology in the brains is really going to be quite helpful in research.”
Another AI project - this time involving the living - may offer more immediate hope. In work partly funded by the Michael J.Fox Foundation researchers from Parkinson’s UK are looking at a large dataset of people with the condition who have been tracked over a long period. Parkinson’s is a disease with a very broad range of symptoms and widely differing rates. of progression for different patients, and the project aims to develop algorithms which can detect the various subtypes of the condition.
With certain drugs only working with specific types of Parkinson’s, this research could have an application in making clinical trials more effective. Dr. Dexter says that the AI could also have a very real impact on treatment because knowing what subtype of Parkinson’s a patient had would tell doctors what their trajectory might be.
He gives as an example the tendency of some Parkinson’s patients to have problems with balance and suffer falls, something which apparently costs the NHS half a billion pounds a year: “With these algorithms, you will be able to spot who is going to develop balance and falls problems many years before they do. You'd be able to put in physiotherapy and exercise programmes to ward off the problems.”
So - two AI projects that could one day help us to understand Parkinson’s better and perhaps improve the treatment of it. A few years ago artificial intelligence seemed to promise miracle drugs, robot surgeons, and a productivity revolution in healthcare. That dream hasn’t quite died, but we are learning once more that it’s one thing to develop brilliant new ideas, quite another to put them into practice.