The language of the eyes

A novel method for assessing dementia risk could help populate critical drug trials.

Bicycle, velvet, church. I watch over her shoulder as she tries to draw 10 minutes before 11 o’clock on a blank circle. What is today’s date? Who is the Prime Minister of Canada? What did you eat for lunch yesterday? Count backwards from 100 by intervals of seven. Now, what were those three words?

The test is known as the Mini Mental State Examination (MMSE), and the 30-question exam is standard in both clinical and research settings for non-invasive diagnosis of mild cognitive impairment (MCI) and dementias. My 80-year-old mother, a primary school teacher from 1962 to 2002, attempts to answer questions her students might have aced. She fails. Perhaps worst of all, she knows it.

In 2016, her score is 26 out of 30. She is given a diagnosis of MCI, and we leave the clinic shaken but determined to beat this thing. By 2017, it is 17. In 2018, it’s 9. According to the geriatrician, at 9 she should not be mobile, or able to feed or toilet herself. But she is. She is also gardening, exercising, lunching with friends. Right now, though, she is crying. Despite having provided no functionally useful information and no change to the course of treatment, this latest test feels like a TKO.


According to the Alzheimer’s Society of Canada, 44 million people worldwide including over 747,000 Canadians are living with Alzheimer’s or another dementia. The incidence of dementia increases five-fold every year from age 65 onward; one in four seniors aged 85-plus has a dementia diagnosis. Most insidiously, new research shows changes in the brain and body up to 20 years before symptoms appear.

Therapies with the potential to slow or alter dementia’s progression have failed to date in part because by the time a diagnosis is made, neurological changes are too far advanced for the interventions to be effective. Because of disease stealth and persistent social stigma, drug trials are struggling to find enough appropriate candidates, and screening costs can reach $100,000 per person. Already, up to 90 per cent of all potential therapeutics fail in the development stage. But right now, potentially efficacious dementia drugs are failing for lack of trial participants. In 2018, these numbers drove Pfizer out of the Alzheimer’s race altogether.

Enter: CANARY. Ever the emblem of the sensitive early warning system, in this case, CANARY is acronymic for Clinical Data, Natural Language Processing and Eye Tracking for Dementia Risk Stratification. Neurodegeneration in Alzheimer’s disrupts gaze and language pathways, and signs are detectable before any cognitive symptoms appear. A UBC team was first in the world to determine that by tracking the two behaviours in concert and over time, they can identify and assess risk for MCI and Alzheimer’s disease with greater success than by assessing either in isolation.

Dr. Thalia Field leads the CANARY team, a group of multidisciplinary researchers working in natural language processing, machine learning, AI, computer science and neurology. A stroke neurologist and clinician-researcher with UBC Medicine, Field champions the use of technology to improve experiences and outcomes for both clinicians and patients.

Her colleague Dr. Hyeju Jang is a CIHR postdoctoral fellow in Computer Science and the Data Science Institute at UBC and the BC Centre for Disease Control. With a doctorate from the Language Technologies Institute at Carnegie Mellon University, Jang explores humans’ use of language through the lens of computation. 

They built a machine learning model from a 1980s dataset comprising audio recordings and transcriptions of 257 dementia patients and 242 healthy elderly controls. Using a neural network, a relatively new form of machine learning algorithm, and including age as a factor, the model changes as new volunteers enter the study and the dataset grows. By tracking speech patterns and eye movements in concert, the model’s predictions as to whether someone is clinically at risk for dementia or a healthy control has improved from 73 to 80 per cent.

Remarkably, English is not Jang’s first language, and as Field says of the team’s multidisciplinary collaboration, “We are all learning each other’s languages.” This humble exploration across disciplines speaks to the sensibility required for reenvisioning the way the medical field approaches predementia and dementia patients. Because after over a century, there is still so much we don’t know.


With my eyes, I follow a dime-sized red circle around the perimeter of my screen, then I focus on a cross in the screen’s centre. Next, I describe a drawing, then read the paragraph that appears. Finally, I choose a topic from a list and tell a story based on a personal memory. These five exercises form the core of the CANARY test. Although I’m nervous, Field’s research assistants, Caitlin Lewis and Tom Soroski, put me at ease. Now in my mid-50s, I’m apprehensive about what the test might reveal, but since there’s no apparent right or wrong, I find myself enjoying it. Then, after a short survey about the experience, Tom and Caitlin administer the MoCA – the Montreal Cognitive Assessment. I quickly recognize most of its content from the MMSE. I repeat sequences of words and numbers after Caitlin. I draw a clock that says 10 minutes after 11, but my 3D cube is all wonky. I get flustered, feel my face flush and my heart race, wonder if I’m messing up my countdown by sevens. Face, velvet, church, daisy, red, I blurt out at the end, before I can forget.

We say goodbye, the screen goes dark, and all I can see now is my own reflection – a face so much like hers – staring back at me. This one’s for you, Mom, and for everyone looking toward a better outcome.

Call for Volunteers

UBC’s CANARY research compares the speech and eye movements of people with Alzheimer’s disease and mild cognitive impairment (MCI) to those of healthy volunteers to see if there are patterns that can be used to detect Alzheimer’s disease early, and to monitor the progression of Alzheimer’s disease over time.

You may be eligible to participate in this study if: 
• you have a diagnosis of mild to moderate Alzheimer’s disease, mild cognitive impairment, or if you are a healthy volunteer age 70 or over.
• you are 19 years of age or older.
• you are able to speak and understand English.
• you are able to consent to the study.

You may not be eligible to participate if you have had a recent brain injury, or if you have a neurological or psychiatric condition.

CALL 604-822-0768 OR EMAIL