The NHS AI Lab has developed a model to support research into Parkinson’s disease by detecting signs of the conditions in images of brain slices.
It has worked on the project with software company Polygeist and research organisation Parkinson’s UK, using the latter’s Brain Bank dataset.
Guisseppe Sollazzo, deputy director of the NHS AI Lab, said the project has created a “minimum viable data pipeline” based on a specific slice of the brain to detect the disease in images.
“We have proven that an AI model could detect Parkinson's in pathology slides, which will help speed up a very slow and painstaking process,” he said. “This will ultimately help progress research into Parkinson's that can support living patients.”
The model developed so far can be used solely for research. Solazzo said there are hopes to ultimately take it to the stage of supporting diagnosis but this would be some way in the future.
Parkinson’s disease – a degenerative disorder of the central nervous system – is currently subject to clinical evaluation and often identified only when the symptoms are advanced.
“Of course, the benefits of this three-month project are limited,” Solazzo said. “But we have plans to explore what's next. We are working to launch a second phase, which will look at other areas of the brain as well as the possibility of implementing the algorithms on microscope hardware, making diagnosis even more rapid as it can be performed while digitising the pathology slides.
“If we are able to prove that – and it's a big ‘if’ – we could then ask can this process be adapted to work on other pathology challenges? If it could, that would allow the development of tests for live patients (e.g. through biopsies) across a range of diseases.”
He added: “This project captures a lot of what I want AI to be: a tool that, under control and in the appropriate circumstances, could help our NHS staff deliver faster, safer diagnosis for patients. No magic, no hype, no obfuscation; a transparent open source process empowering humans.”
Parkinson’s UK Brain Bank uses brains donated by people upon their death. For the project these were cut into numbered slices for specific areas and scanned in ultra-high definition of 0.93 sensitivity and 0.88 specificity.