Artificial intelligence to help identify Parkinson’s

A technical report published in September showed the outcomes of a 12 week project to see if artificial intelligence (the ability of machines to simulate human intelligence and carry out tasks that would usually be done by humans) could be used to make diagnosing Parkinson’s quicker and more accurate. The project was looking to develop an automated tool, using information obtained from brains donated to the Parkinson’s UK Brain Bank. 

The collaboration was made up of an expert team from the Parkinson’s UK Brain Bank at Imperial College London, NHS England, Parkinson’s UK, and the artificial intelligence specialists Polygeist. You can read the full technical report on the bioRxiv website.

What’s the Brain Bank? 

The Parkinson’s UK Brain Bank collects precious brain tissue from people with and without Parkinson’s who have decided to leave their brains to research. This is a vital resource that then sends brain tissue all over the globe to power research into better understanding Parkinson’s and to accelerate the search for better treatments and a cure for the condition. 

Why do you need to check the diagnosis of Parkinson’s? 

There is no simple clinical test to diagnose Parkinson’s. More than a quarter of people with Parkinson's are misdiagnosed at some point, according to our survey. Research is ongoing to improve diagnosis. Alongside this, it is also essential that diagnosis is checked in donated brains before the tissue can then be used in research. 

How is this done? 

From each brain donation, thin sections of tissue are taken from multiple areas of the brain and are carefully mounted onto glass slides. These slides are then stained to look for a troublesome protein called alpha-synuclein. Examination of the stained tissue sections under a microscope is then carried out to identify alpha-synuclein protein clumps in Parkinson’s brains, compared to the absence of protein clumps in donations from people who didn’t have Parkinson’s. 

What’s new? 

This process of confirming a Parkinson’s diagnosis in donated brains is currently done by a neuropathologist who can spend around 6 hours carrying out this task. This project looked at developing a way to automate and speed up this process using an area of artificial intelligence called ‘machine learning’.

What was the outcome? 

The project studied high resolution images of scanned tissue from 300 donated brains with a confirmed diagnosis of Parkinson’s, and 50 brains without Parkinson’s. The machine learning technique was able to identify alpha-synuclein clumps, indicating a diagnosis of Parkinson’s, with greater than 93% accuracy. This process took just 4 minutes. 

The developed method is open source, meaning it will be available for others to use and develop to help with Parkinson’s research, but also other conditions. 

Our Associate Director of Research, Professor David Dexter, said: 

"This is what is known as a proof of concept study that now needs further work to fine tune and explore its potential in Parkinson’s. 

"Being able to confirm which donated tissue is from someone who had Parkinson’s versus someone who didn't, could make the Brain Bank more efficient and help researchers globally to get their hands on precious brain tissue faster, to accelerate research. It may also offer extra information to researchers about the amount and location of alpha-synuclein found in a donated brain, which could be significant to better understanding the condition.

"This study also has wider health care implications because similar staining techniques are used in all NHS diagnostic services, so this machine learning approach could also speed up and improve the accuracy of cancer diagnosis from tissue taken at biopsy."