HN Summary

• Innovative AI Model: Researchers at Baycrest, the University of Toronto, and the Toronto Dementia Research Alliance have developed an AI model that analyzes hand-drawn clock images from the Clock Drawing Test (CDT), enabling accurate, scalable, and low-cost dementia screening both in-person and virtually — even in remote or under-resourced areas.

• Scientific Breakthrough: Using over 50,000 CDT images, the AI leverages a vision transformer (ViT) to assess global and relational drawing features with 76.5% accuracy, outperforming human scoring and existing deep learning methods while handling variable image quality and orientations.

• Global Impact and Future Potential: By democratizing access to cognitive screening, this technology could transform early dementia detection worldwide. Ongoing research aims to integrate health and demographic data to further refine accuracy and expand the tool’s applicability across diverse populations and healthcare settings.

A simple pen, a blank sheet of paper and a hand-drawn clock may soon become the foundation for a powerful new way to screen for dementia using artificial intelligence (AI). 

I along with Drs. Michael Bone and Bradley Buchsbaum at Baycrest and University of Toronto, and in collaboration with the Toronto Dementia Research Alliance (TDRA), have developed an AI model that can accurately analyze clock drawing test (CDT) images drawn by individuals with cognitive problems, potentially transforming how dementia is screened around the world. This AI model can be applied to the CDT administered both virtually and in-person. The virtual capability enables the reach of this quick screening test to rural communities.

Dementia is a global health crisis, with millions of new cases each year and rising care costs. In many regions, especially low- and middle-income countries, limited access to screening tools leaves countless people undiagnosed and without care. 

The CDT is a brief screening tool widely used to evaluate cognitive function. Often employed as part of larger cognitive assessments (like Toronto Cognitive Assessment or TorCA and the Behavioural Neurology Assessment-Short Form or BNA-SF), it has the capability to identify cognitive impairment since it engages many cognitive skills. Typically taking less than one minute to complete, individuals are asked to draw a clock face, put in all of the numbers and set the hands to a specific time, commonly 10 after 11. The nature of the errors provides insight into a breakdown of different brain functions. 

Historically, the CDT has required trained clinicians to score and analyze the qualitative aspects of the of the drawings, making sophisticated interpretation difficult to easily obtain outside of specialized clinics. However, at Baycrest, we, along with our academic partners, have been able to develop a novel approach that could help close the diagnostic gap between specialized clinics and under-resourced communities both within Canada and beyond. 

The novel approach for automated dementia diagnosis from CDT images leverages a pretrained vision transformer (ViT) to better capture global features (such as positioning of numbers on the clock face, shape and completeness of the circle and placement/orientation of clock hands) and relational features (which can include spacing between numbers and the angle between clock hands).  

This method, which was trained using more than 50,000 CDT drawings from the National Health and Aging Trends Study (NHATS) by Johns Hopkins Bloomberg School of Public Health dataset and more than 800 from the TDRA, has the unique capability to identify clocks as being distinct from other images. It may sound simple, but doing so requires the model to automatically handle challenging image quality issues like shadows, irrelevant markings and improper cropping. 

The potential impact of this is hard to overstate. It means that the model can make sense of a CDT image no matter how big the drawing is, where it appears on the page or how it’s oriented. All it takes is a pen and paper, opening the door to low-cost, scalable screening almost anywhere. 

Accuracy is not compromised. In fact, this model outperformed human scoring and existing deep learning methods with a 76.5 per cent accuracy in dementia detection.  

In areas where medical resources are stretched thin, AI offers a way to extend the reach of care without sacrificing quality.  

“What makes AI so powerful in healthcare isn’t just its speed or accuracy,” says Dr. Buchsbaum, Senior Scientist at Baycrest’s Rotman Research Institute and senior co-author of the paper titled ‘A vision transformer approach for fully automated and scalable dementia screening using clock drawing test images’ which was published recently in Alzheimer’s and Dementia: Diagnosis, Assessment & Disease Monitoring. “It’s its potential to democratize access to diagnostic and decision-making support.” 

AI is rapidly becoming one of the most transformative forces in healthcare. With the ability to process large volumes of complex data, recognize subtle patterns and continuously learn from new inputs, it is helping clinicians make faster, more accurate decisions. But perhaps the most profound potential lies in AI’s ability to scale, helping ensure that no one is left behind due to geography or cost. 

There is still more work to be done in analyzing the different types of errors in CDT drawings and how they relate to cognitive impairment and dementia as a whole. In future studies, our priorities will be to add other health and background information to improve accuracy, making sure the tool works well for different people in different healthcare settings. 

As AI’s capacity continues to increase and its accuracy improves, it may soon help us go beyond diagnosis to offer insights into how different patterns of impairment cluster across individuals and disease types. 

By: Dr. Morris Freedman, Head, Baycrest’s Division of Neurology, Medical Director, Pamela and Paul Austin Centre for Neurology and Behavioural Support and Scientist, Baycrest’s Rotman Research Institute. An expert in the diagnosis of dementia and the treatment of its symptoms, Dr. Freedman has developed innovative cognitive assessment tools, co-created Baycrest’s game changing Virtual Behavioural Medicine (VBM) program and champions global distance learning in neurology.