Using AI to spot Alzheimer's early - it's just a matter of time

Alzheimer’s disease is a devastating disease that currently has no cure and even the best treatments have a limited effect in slowing its progression. We know that the disease begins developing decades before the obvious signs such as memory loss appear. 

But if we could pinpoint when a patient is in the very early stages of Alzheimer’s, it could open the door to new treatments that might slow or even halt the progression of the disease before irreversible damage has been done to the brain. At the very least it would give patients more time to plan for their future with the disease.

Artificial intelligence, or AI, that catch-cry of the modern world, isn’t just about robots and self-driving cars. It is increasingly being used in medical research to analyse patient data in order to determine an optimal treatment plan, develop a vaccine or search for early markers of complex diseases, including Alzheimer’s.

"AI can see patterns in patient data that a human can’t, due to the size and complexity of the datasets involved."

To help us understand when Alzheimer’s disease begins, we need to be able to follow the progression of patients through the disease and so we need to be able to analyse patient data collected at intervals over time. But most statistical and AI methods can only be used to analyse data collected at a single time point. Given this reality, we developed a new AI method to help with this challenging task.

Have you ever bought something from an online store, such as Amazon, and found that they recommended other items that you might be interested in? These recommendations are based on other peoples’ purchasing history, and are determined using an AI technique called pattern mining. Pattern mining is used by businesses in their marketing of products but can also be adapted to search for patterns occurring over time in patient data, and that is exactly what we did.

If we think of our patient data as a bridge through time that the AI model must cross in order to search for patterns, then naturally the model will be most effective if the bridge is easy to cross i.e. if it is well maintained, with the planks neatly abutted and none loose or missing. But patient data are typically messy, making the bridge rickety and difficult to cross, with missing planks (missing data values), large gaps between the planks (times between data collection points) and loose or skewed planks (different sources, types and statistical distributions of the data). Our new pattern mining method allows the AI model to cross even the most rickety bridge. 

Using this method and data from CHeBA’s Sydney Memory and Ageing study we found some interesting patterns. For example, one common pattern in those who developed Alzheimer’s disease showed visual acuity declining early on, followed at a later stage by a significant decline in mental activity as well as other illnesses. So make sure you keep up those challenging puzzles and crosswords and maintain a healthy lifestyle throughout life.

More work is needed now to see if similar patterns are found in other groups of patients and to determine how early they appear. 

"They say time heals all wounds, but not so for Alzheimer’s disease. Time is the enemy. But with this new AI method we hope to give patients and clinicians alike more time: time for treatment, time for planning and precious time to spend with loved ones."

Further information

This research analysed data from CHeBA’s Sydney Memory and Ageing Study and Older Australian Twins Study: https://link.springer.com/article/10.1186/s12859-024-06018-8