Machine Learning–Driven Cognitive Screening Tool Aims to Improve Early Dementia Detection: Adrian Owen, OBE, PhD

Cognitive decline represents a progressive deterioration in cognitive abilities such as memory, attention, executive function, and language, and may occur as part of normal aging or as an early manifestation of neurodegenerative disease. Mild cognitive impairment (MCI), often considered an intermediate stage between healthy aging and dementia, is associated with an increased risk of progression to Alzheimer disease (AD) and related dementias.¹ Early identification of cognitive changes has become increasingly important as emerging disease-modifying therapies and supportive technological interventions may offer the greatest benefit when implemented during the earliest stages of impairment.²

One of the researchers helping to advance technology-driven approaches for earlier cognitive assessment is Adrian Owen, professor of cognitive neuroscience and imaging at the University of Western Ontario in Canada. Owen, who is also the co-founder and chief scientific officer at Creyos, has been involved in the development of a digital cognitive assessment platform designed to evaluate domains such as memory, attention, and executive functioning through brief computerized tasks. More recently, he and colleagues have explored how machine learning can be integrated into these digital tools to create rapid, scalable screening approaches capable of identifying subtle signs of cognitive decline and dementia risk.

NeurologyLive sat down with Owen discuss new research focused on improving the early detection of cognitive decline and dementia risk. In the conversation, he outlined the rationale behind developing a rapid, machine learning–driven cognitive screener designed to identify individuals along the AD continuum earlier and more efficiently than many traditional tools currently used in practice. Owen emphasized the ongoing challenges surrounding underdiagnosis of mild cognitive impairment, particularly in primary care settings, and discussed how limitations of legacy screening instruments continue to create barriers to timely identification.

NeurologyLive: Can you provide a general overview of the research?

Adrian Owen, PhD: This study was really about addressing a very practical gap in dementia care: how do we identify people who are beginning to show signs of cognitive decline early enough for it to matter? We know that mild cognitive impairment sits on a continuum between healthy aging and dementia, but it’s often missed in routine clinical practice because existing screening tools are either too time-consuming, require specialist training, or simply aren’t sensitive enough at the earliest stages.

What we did was take a step back and first review the current landscape of cognitive screening, and then develop a new, much more streamlined approach. Using a large, well-characterised dataset from our Creyos platform, we applied machine learning to identify a pair of short cognitive tasks that target working memory and attention. Together they can flag individuals at risk of progressing along the Alzheimer’s disease continuum. The key idea was to create something that is rapid, fully automated, and practical for use in real-world settings, while still maintaining the level of sensitivity and specificity you’d expect from much longer, traditional assessments.

Why was it important to conduct research of this nature? What current gaps in AD/Dementia detection does this study aim to highlight?

This kind of research is important because, despite decades of work in Alzheimer’s disease and related dementias, we are still largely diagnosing people too late for any of the interventions that are coming along to have their full impact. Mild cognitive impairment (that is the stage where intervention is most likely to be effective) is both common and surprisingly under-detected, particularly outside specialist clinics. In primary care, clinicians are often relying on brief observation or subjective complaints, and in many cases no formal cognitive testing is done at all, meaning a substantial proportion of individuals are never assessed in the early stages of their disease.

Even when tests are used, the field tends to rely heavily on a small number of legacy tools like the mini mental state (MMSE) or the Montreal Cognitive Assessment (MoCA), which have well-known limitations, including cultural and educational biases, ceiling effects, and the need for trained administration. The gap we’re highlighting is not just about accuracy, but about practicality. There’s a real need for tools that are sensitive to early changes, but also fast, scalable, and usable in routine care. Without that, we miss the window where detection can actually change outcomes.

Can you explain the results and their clinical significance?

The results were extremely encouraging, particularly given how simple the screener is. Using just two short Creyos tasks, we were able to distinguish between cognitively healthy individuals and those at risk of impairment with a level of sensitivity and specificity that compares favorably with many established tools. In our preliminary validation, the screener correctly identified all clinically diagnosed Alzheimer’s patients in the sample, including a case that would have been missed using a standard measure like the MMSE, while maintaining good specificity in matched control groups.

We also saw a clear, expected increase in positive screening rates with age in a large population sample, which gives us confidence that the tool is capturing meaningful variation in cognitive risk rather than noise. Clinically, the significance lies in the balance we’ve achieved: this is not a replacement for full neuropsychological assessment, but as a rapid, first-line screener it has the potential to flag individuals earlier and more reliably, prompting timely referral and intervention.

In a system where time and resources are limited, that kind of triage tool could make a substantial difference to how and when we identify people on the dementia continuum. On a personal note, it’s extremely gratifying to see 30 years of research into cognitive measurement and dementia come to fruition like this.

What are some limitations of this study?

As with any early-stage study, there are a few important limitations to keep in mind. First, although the performance metrics are encouraging, they’re based on relatively small clinical samples, particularly the Alzheimer’s group, so the estimates of sensitivity and specificity should be viewed as preliminary. We really need to see how the screener performs in larger, more diverse populations and in real-world clinical settings and that’s exactly what we’re doing right now.

Second (and this is really a minor limitation), while the screener is effective at flagging cognitive impairment, and notably did not produce significantly more positive results in individuals with other health conditions that impact cognitive function, it’s not designed to distinguish between different causes or stages along the dementia continuum. It’s a first-line screener, not a diagnostic instrument.

What needs to be studied further in the AD/Dementia space?

The most immediate priority is large-scale, real-world validation. We need to know how the Creyos tool performs across diverse populations, healthcare systems, and clinical contexts, not just in controlled datasets but in everyday practice. That includes longitudinal work, following individuals over time to determine how well early screening results actually predict progression along the dementia continuum. As I’ve said above, that’s one of the things we’re working on right now. Beyond that, there’s a real opportunity to integrate cognitive screening with emerging biomarkers (whether blood-based markers or other digital health signals) to improve both sensitivity and specificity at the earliest stages.

Creyos has solved the problem of how we embed these tools into healthcare workflows and electronic records in a way that is useful rather than burdensome, but more data is always needed to make sure we stay ahead of technical developments as they happen. And finally, there are important ethical and health system questions. Who should be screened, when, and what happens after a positive result? Identifying people earlier is only valuable if it leads to meaningful intervention and support. While the results are promising, they really represent the beginning rather than the end of the story.

REFERENCES
1. Petersen RC. Mild cognitive impairment. Continuum (Minneap Minn). 2016;22(2 Dementia):404-418. doi:10.1212/CON.0000000000000313
2. Jack CR Jr, Bennett DA, Blennow K, et al. NIA-AA research framework: toward a biological definition of Alzheimer disease. Alzheimers Dement. 2018;14(4):535-562. doi:10.1016/j.jalz.2018.02.018