Artificial intelligence can detect the early stages of Parkinson’s disease from brain scans. The hope is that this will lead to earlier diagnosis and treatments to slow the progression of the disease.
Parkinson’s disease is usually first noticed when people start to show visible tremors and lose some control of their motor movements. The disease is then confirmed with further tests involving injecting radioactive tracers into the body.
To see if artificial intelligence could do a better job, researchers at two Italian universities trained a machine learning algorithm to distinguish between brain scans of people with and without Parkinson’s.
Advertisement
The AI saw hundreds of different structural MRI scans. Such scans pick up the brain’s anatomy and any changes from disease in great detail. Around two thirds of the subjects were of people diagnosed with Parkinson’s, many with early signs of the disease, such as mild cognitive impairment and REM sleep disorders, and the remainder were from healthy controls.
During tests on previously unseen images, including brain scans of people diagnosed with Parkinson’s, the AI could identify the disease with 94 per cent accuracy.
“We hope we can use these results to develop drugs or disease-modifying therapies before the late phase appears,” says Nicola Amoroso, a member of the team from the University of Bari and the National Institute of Nuclear Physics, both in Italy.
No cure currently exists for Parkinson’s, but efforts are ongoing to develop therapies to delay the progression of the disease.
Stefan Klöppel, at the University of Bern in Switzerland, says that the new system could be accurate enough to be identify people who should receive potential preventative therapies that don’t have severe side-effects.
But for therapies, such as vaccines that target the immune system, and have more severe side-effects you would need a higher level of accuracy. “You would not want to give that to somebody who does not have the disease,” says Klöppel.
Last year, the same Italian team showed that AI could identify signs of Alzheimer’s in brain scans of people almost a decade before they could be clinically diagnosed.
The hope is that by understanding how the AI picks up the different diseases may lead to a better understanding. “What we learn today will be useful in the next year or in a decade,” says Amoroso.
Medical Image Analysis
