Published:

An artificial-intelligence algorithm trained on brain images and movement data from 20,000 UK Biobank participants can spot early signs of Alzheimer’s and Parkinson’s disease.

Summary

An artificial-intelligence (AI) program can spot early signs of Alzheimer’s and Parkinson’s disease many years before people are diagnosed. The algorithm, trained on nearly 20,000 UK Biobank participants’ information, combines brain scans and activity-tracker data for the first time. Eventually, AI technology could help clinicians to identify and treat neurodegenerative conditions in people who haven’t developed any symptoms yet.

An artificial-intelligence (AI) algorithm can spot early signs of brain conditions, including Alzheimer’s and Parkinson’s disease, many years before people are diagnosed. The algorithm combined nearly 20,000 UK Biobank participants’ brain scans and activity data, which came from a watch-like device some participants wore for a week.

An early diagnosis is important to living well with neurodegenerative conditions. It allows people to adjust to the changes they’re experiencing, and access support and treatments.

“[Brain scans] are more important for tracking cognitive deficits, and [activity tracking] is more important for motor deficits.

Marirena Bafaloukou, UK Dementia Research Institute

Marirena Bafaloukou who works, independently of the study team, on medical AI at the UK Dementia Research Institute, explains that using multiple sources of health information makes it possible to find which one is best at predicting diseases – and what would be most useful for population-wide screening.

“[Brain scans] are more important for tracking cognitive deficits, and [activity tracking] is more important for motor deficits,” Bafaloukou says. Changes to thinking and memory as well as motor impairments – such as slow and unsteady walking – can be signs of Parkinson’s or Alzheimer’s disease.

A new diagnostic tool

The new AI program analyses both brain scans and activity data for the first time, to detect brain conditions at a very early stage – possibly even before people notice any symptoms.

The algorithm’s accuracy was “probably comparable, if not slightly better, than the standard diagnostic tools for predicting these [conditions]”, says study leader team Vince Vardhanabhuti from the University of Hong Kong. Accurately diagnosing conditions at the stage where people notice mild symptoms is already very difficult, he adds, “but to predict them when patients are ‘asymptomatic’ is even more difficult – I think you have to view the performance with that mindset”.

Studies like this push us forward towards adopting [AI prediction] in the near future, when we have a body of evidence that they can predict things early, help to diagnose early and can contribute to improvement of patient care.

Anastasia Ilina, Imperial College London, UK

Brain images seemed to be key to the algorithms’ performance. “Potentially what one could do is to use a cheaper method, like [activity tracking], as a first line of screening for those who are potentially at risk, and then only certain select people could go on to have the brain [magnetic resonance imaging] assessment,” Vardhanabhuti suggests.

The AI program would need to be tested with a large, diverse group of people to see if its predictive power holds up. Fewer than 100 of the participants included in the study were diagnosed with dementia in the years that followed their brain imaging. “Generally, the bigger the group, the better, and the more you can argue that the results of the [AI algorithm] are generalisable,” explains healthcare AI specialist Anastasia Ilina from Imperial College London, UK.

Nevertheless, Ilina says, “studies like this push us forward towards adopting [AI prediction] in the near future, when we have a body of evidence that they can predict things early, help to diagnose early and can contribute to improvement of patient care”.

Related publication