As the population grows older in industrialized societies some challenges to healthcare problems increase. But, also the potential to solve such challenges in healthcare increases. Hasnain Ali Shah, a Doctoral Candidate in Eyes4ICU at the University of Eastern Finland, is currently researching a non-invasive method to detect Mild Cognitive Impairment by combining modern eye-tracking equipment with deep learning methods. Ali talks about his research and its potential in the interview.
What is Mild Cognitive Impairment (MCI) and why is researching early detection so important?
Mild Cognitive Impairment (MCI) is a transitional phase between normal aging and more severe forms of dementia, like Alzheimer’s. People with MCI have memory problems, but they can still function independently. The problem is, many cases go unnoticed until they progress. Early detection matters because interventions whether lifestyle changes or medical are most effective before significant brain damage occurs.
Why do you focus on eye tracking and what makes it so unique and promising for early diagnosis?
Our eyes don’t just see, they reflect how our brain processes information. Changes in attention, reading patterns, fixation duration, and pupil size can indicate cognitive difficulties. What makes our approach unique is that we use machine learning and deep learning to analyze subtle patterns across groups, such as how MCI participants read compared to controls. We also try to keep the method non-invasive and scalable.
Can you walk us through what a typical test looks like for someone being assessed using your methods?
In our setup, participants read sentences or texts on a screen while their eye movements are recorded with high-resolution eye trackers. We then analyze metrics like fixation durations, skipping behavior, and even changes in pupil size. The data is anonymized and processed to identify whether their gaze behavior deviates from typical patterns.
What are your core findings so far?
In our research, we found that pupil size features offer a surprisingly strong signal for identifying MCI. In one study, we showed that adding pupil-derived metrics such as dilation patterns and variability significantly improved the accuracy of machine learning models beyond traditional eye movement features like fixations and saccades. Building on this, our second study combined eye-tracking data with speech-based features using a multimodal fusion network. This approach achieved over 79% accuracy for MCI and over 82% for Alzheimer’s disease, outperforming single-modality models. Interestingly, pupil size and speech timing emerged as the most important features, with reading tasks revealing more about cognitive health than rapid number-naming. Together, these findings support the use of simple, non-invasive tasks like reading and speaking to develop future screening tools for cognitive decline
How close are we to a future in which this technology could be used in regular doctor visits or even at home?
Yes, I think it’s realistic to imagine this being used in routine doctor visits or even at home in the future. The methods we’re working on don’t require complex medical procedures just reading or speaking tasks paired with eye-tracking. If the hardware becomes more affordable and user-friendly, and the models keep improving, there’s no reason this couldn’t be part of basic cognitive check-ups. We’re not fully there yet, the biggest hurdle is making it work reliably outside controlled lab environments but the progress so far is encouraging. It feels like we’re getting closer to that kind of practical application.
What are the biggest technical or clinical challenges that need to be solved before this Method for Diagnosing MCI can be widely adopted?
There are a few big challenges that need to be addressed before this kind of technology can be used widely. Technically, one of the main issues is making the eye-tracking data reliable in everyday environments. Another major challenge is data diversity, most current models are trained on relatively small and homogeneous datasets. To build something that works globally, we need more participants across different age groups, languages, and cultural backgrounds. On the clinical side, the biggest challenge is validation. We need larger studies and clinical trials to prove that this method is as reliable as traditional screening tools. And even if it works, it needs to fit into real healthcare systems.
Do you see this technology being used for detecting other brain-related disorders beyond MCI and Alzheimer’s?
Yes, definitely. While our current focus is on MCI and Alzheimer’s, the potential applications go far beyond that. Eye movements and pupil dynamics are closely tied to brain activity, so any condition that affects cognition, attention, or neurological function could potentially be studied using similar methods. each disorder has its own complexity, and the models would need to be tailored accordingly. But the core idea remains the same: the eyes offer a window into how the brain is functioning, and with the right tools and datasets, we can start to pick up on signals that aren’t obvious otherwise. So yes, I see this technology evolving into a broader platform not just for dementia, but for a range of brain-related disorders.
Are you interested in more details of Ali‘s work?
Eye tracking based detection of mild cognitive impairment: A review