AI tool could speed up dementia diagnosis

A new AI tool that could help doctors assess the early signs of dementia and Alzheimer's more quickly and efficiently, has been developed by researchers at the University of Sheffield.

The system, known as CognoSpeak, uses a virtual agent displayed on a screen to engage a patient in a conversation. It asks memory-probing questions inspired by those used in outpatient consultations and conducts cognitive tests, such as picture descriptions and verbal fluency tests.

The tool then uses artificial intelligence and speech technology to analyze language and speech patterns to look for signs of dementia, Alzheimer's disease and other memory disorders.

Researchers behind the technology say CognoSpeak could play a key role in reducing the burden on dementia assessment services, once further testing in GP and secondary care memory clinics across the U.K. is complete.

The system is being designed to work in between primary and secondary care. This means that once fully rolled out, a GP could refer a person with memory complaints to use the technology. CognoSpeak would send the test results back to the GP and then they would decide whether they need to refer the patient to a memory clinic for further assessment.

CognoSpeak can be accessed through a web browser—meaning patients are able to take the test in the comfort of their home via a computer, laptop or tablet, rather than having to wait for a hospital appointment to take a pen-and-paper-based assessment, which can often cause undue stress and anxiety.

Early trials have shown the technology is as accurate at predicting Alzheimer's as the current pen-and-paper-based tests used to assess or screen for cognitive, memory or thinking impairments. The team has demonstrated accuracies of 90 percent for distinguishing people with Alzheimer's from people that are cognitively healthy.

Developed by Dr. Dan Blackburn and Professor Heidi Christensen from the University of Sheffield's Departments of Neuroscience and Computer Science, the CognoSpeak system is still in the research phase, but thanks to a grant, the technology is being trialed more widely. The researchers are recruiting 700 participants from memory clinics across the U.K. to help develop the system further.

Dr. Dan Blackburn, from the University of Sheffield's Department of Neuroscience, said, "Waiting for a possible diagnosis of dementia can be a very anxious time for patients and their families. This tool could help patients start treatments sooner, reduce waiting times and give people certainty earlier.

"The CognoSpeak system could transform how dementia and other memory disorders are diagnosed by speeding up assessments. This would also free-up clinicians' valuable time and mean that those who need specialist care get access to it as quickly as possible."

Professor Heidi Christensen from the University of Sheffield's Department of Computer Science, said, "The way a person speaks can tell us a great deal about their cognitive health and emotional well-being, and give us a very early indication of any signs of cognitive decline that may not otherwise have been detected. The system we've developed here at Sheffield uses speech technology to automatically extract these signs and the automation means we can provide a consistent, accurate and fast assessment for everyone.

"CognoSpeak is advanced, high tech and based on world leading research in this field. We have the biggest collection of data for this type of assessment anywhere in the world, which we're using to advance the technology and improve its accuracy."

The CognoSpeak tool has been developed in collaboration with Therapy Box—a company specializing in speech and language technology—and the National Institute for Health and Care Research (NIHR) Devices for Dignity MedTech Cooperative (D4D), who are leading work with patients, families and ethnic minority community groups to ensure that the AI is acceptable, reliable and accessible for all future users.

The research team is also working to make the tool accessible to patients from ethnic minority communities who are less likely to engage with dementia services and who may speak English as an additional language.

Lise Sproson, Patient and Public Involvement Lead at D4D said, "We are working closely with a broad range of community groups, including Sheffield's Israac Somali Community Association, to co-develop the look and feel of the CognoSpeak system, to ensure it is acceptable, relatable and easy to use.

"We are training the AI with a range of regional UK accents and those speaking English as an additional language, in order to maintain the accuracy it has previously demonstrated on tests with native English speakers across the population as a whole."

Dr. Blackburn, who is also an Honorary Consultant Neurologist at Sheffield Teaching Hospitals NHS Foundation Trust and researcher at NIHR Sheffield BRC, added, "There is a real clinical need for this kind of technology. There are long waiting lists for memory clinics across the UK, but there are also inequalities in accessing the memory clinics service. The CognoSpeak tool can reduce these inequalities and help make the service more efficient."

There are currently around 900,000 people in the UK living with dementia, and this is projected to almost double by 2040, according to the Alzheimer's Society. Referrals for assessment are increasing rapidly and memory clinics often have long waiting lists.

Professor Mike Lewis, Director of NIHR's i4i Programme, said "Cognospeak is an innovative example of how digital health technology can transform the way we tackle conditions like dementia, helping to find ways to make it easier to find and assess patients to ensure they get access to the right treatment and support"

Patients are being recruited to the CognoSpeak trial through memory clinics across the UK. To take part in the trial, visit www.cognospeak.com

Provided by University of Sheffield