Seniors Profile: Scientist Frank Rudzicz talks about his new tool that can detect Alzheimer's

One computer science specialist has been developing a tool that is capable of detecting Alzheimer’s and other cognitive disorders through speech analysis.

Frank Rudzicz recently spoke to SeniorsZen about the tool and its implication for the health industry.

SZ) I understand that you wear many hats but what is your main position?

I am a scientist at the University Health Network, which is a group of four and a half hospitals, and an assistant professor in Computer Science at the University of Toronto. Recently, I joined the Vector Institute, which is an organization for the development of artificial intelligence, and I am the co-founder and president of WinterLight Labs, which is a spin off company from the University of Toronto.


SZ) Can you tell us about this tool that can detect Alzheimer’s?

We consider speech to be very indicative of somebody’s mental or cognitive states. We all know that we reveal a lot about how we are feeling and how we are thinking based on how we speak. So, if somebody is really talking very fast they are likely stressed or if they are having trouble remembering words they might be distracted or have cognitive issues. At the same time, we know that in Alzheimer’s Disease in particular, language is often the second faculty that is affected chronologically after memory. Language is one of the early markers.

Our team were all basically people who worked in computational linguistics, which is a sub-area of artificial intelligence that uses machine learning on speech and text data. So, we all had an awareness of what tools exist to measure different aspects of language and we used those tools on spontaneous speech. If people are speaking off the top of their head, they reveal a lot more about their mental states than if they are just reading something off of a page. If you are talking spontaneously, you have to rely on your own memory for words, your own context. You can get a lot more about someone’s mental state from listening to how they speak in conversations.

Our assessment usually lasts about 45 seconds to a minute.  We can use signal processing to see when people are pausing between words and how long those pauses are. That is pretty straight-forward to measure and pretty straight-forward to interpret because if the person is pausing a lot more then we expect. There are also other tools that computational linguistics give us that are harder to interpret in a clinical sense.

In the end, we have hundreds of features that we feed into machine learning algorithms, which basically takes examples of healthy and unhealthy people describing these images. It looks at all of these features and determines, okay, pausing before a word combined with the ratios of nouns and pronouns, in combination with certain grammatical features all add up to a picture of Alzheimer’s Disease. 


SZ) If your tests find signs of cognitive issues what can the person do?

In a pharmaceutical setting, basically the software can determine someone’s probability of having the disease and the pharmaceutical company would use that information just to enroll or not enroll people who are more or less likely to have the disease. 

We are also doing some pilots in long term care facilities, or retirement homes. In that case, you might have a retirement home with dozens or hundreds of people living in it. People’s cognition changes over the course of their stay. The idea is you could run this task every week, every few weeks or even every month, and it could be used as an early warning marker if one’s cognition is changing.

This can be another powerful tool in the tool kit used by doctors. So, usually there are a battery of tests that can last hours. If we can reduce that time somewhat, or the stress level associated with clinical diagnosis, by combining our information with traditional paper-and-pen assessments, then we can make some serious improvements to well-being.


SZ) Is this something that will be available to the general public or just practitioners?

The goal has always been to have it in the hands of doctors and the hands of the general public. For business reasons, and because there was sort of an opportunity to work with Pharma, the company is focusing on that direction first, a business-to-business sort of thing.  We will be using our software to conceivably reduce the cost of clinical trials, in some cases by millions of dollars.

Getting it in the hands of doctors requires FDA or Health Canada approval, and those are things the company is looking into. It is a long-term goal.

My research lab has a project called Talk2Me, which is an online resource; a website and a telephone application that people can sign up for. They do picture descriptions, but they also do other typical behavioural language tasks. People can devote their voice to that now. It is more of a research project than a tool that officially diagnoses somebody.


Find Senior Housing

Blog Subscription


Checklists and Forms

Need specific form or publication? Click here to access our Checklists and Forms Page.