Experts from the Accelerated Capability Environment (ACE) and the NHS’s AI Skunkworks have developed an AI tool to speed up Parkinson’s disease diagnosis.
Current Parkinson’s disease diagnosis is a time-consuming process that limits the number of cases specialists can deal with. This can put patients at risk of getting their diagnosis too late, meaning their disease progresses before treatment can be administered.
Currently, brain changes can only be assessed manually, taking between four to six hours. Moreover, to further understand pathological causes and develop potential new treatments, manual brain tissue grading is required after death, which also takes a significant amount of time.
Now, researchers have collaborated to develop an AI-powered Parkinson’s disease diagnosis tool that promises to expedite this process significantly to improve care and treatment for the condition.
Parkinson’s disease cases are rising
Parkinson’s disease is the most prominent and fastest-growing neurological disorder in the world, with more than 10 million diagnoses globally.
This figure is expected to increase dramatically, with the number of Parkinson’s patients forecasted to double over the next 50 years.
Ageing is the biggest risk factor for this neurodegenerative disease, and advanced treatments and diagnosis will be essential in combatting it.
Advancing Parkinson’s disease diagnosis with Artificial Intelligence
For their study, ACE and AI Skunkworks collaborated with Parkinson’s UK, the largest membership-based charity in the world.
During the 12-week project, the researcher utilised the charity’s brain bank at Imperial College London – which has more than 1,300 brains from Parkinson’s patients and healthy donors.
The charity also provided a dataset containing 401 digitised images of brain sections immunostained to detect alpha-synuclein (a-syn), the protein that is a pathological marker of the disease. This also included 100 control cases from health donors.
Next, Polygeists from ACE’s Vivace community repurposed existing technology to exclude different types of brain matter that are not involved in this process. They then synthetically stained slides of brain tissues using the iDeepColour neural network, which highlights areas of the brain affected by a-syn.
These images are processed, with areas of interest turning bright green, making them easily identifiable. The images could also be chopped up and sliced into squares, meaning the green areas could reveal disease density.
This allowed Polygeist to create a proof-of-concept classifier capable of 92% Parkinson’s disease diagnosis accuracy with no false alarms.
The new AI process was able to assess one brain in mere minutes, significantly cutting diagnosis times. This enables neurologists to focus on more complex cases.
Next stages of development
The AI tool is nearly ready to be used in real-world applications. The team is now working to optimise the process to differentiate stages of the disease and determine if other proteins can be identified. In the future, it may also be possible to use this technique with brain scans on live patients.
