Key Takeaways

  • The difference between chronological and brain age can be measured by examining the hippocampus. The challenge? Human analysis isn't accurate enough.
  • Using multiple instanced learning (MIL), artificial intelligence (AI) tools were able to accurately determine the brain age of deceased patients.
  • Brain age analysis demonstrates key capabilities of AI, such as identifying complex and subtle patterns in large bodies of data.

Artificial intelligence (AI) in medicine may evoke images of robots examining patients or even performing surgery — and that would be somewhat accurate. For example, AI technology has already been put to work performing some surgical tasks, according to the American College of Physicians. But AI is also impacting the medical world in other crucial ways, including research tools now being developed and used in brain research.

How is AI helping humans understand the brain and, in particular, how the brain ages? According to the Icahn School of Medicine at Mount Sinai, an AI tool called HistoAge is able to determine "brain age" by identifying subtle cellular patterns in a region of the brain called the hippocampus. The tool gets its name from medical histology, which per the National Institutes of Health is the study of biological structures on a microscopic scale.

Brain Age Acceleration

The hippocampus region of the brain is known to be associated with deteriorating brain function with age, resulting in conditions such as Alzheimer's disease. As reported by Neuroscience News, by determining the difference between brain age and the chronological age of a patient, brain age acceleration can be determined. This comparison, in turn, yields insights into both normal brain structure and function and also how the brain ages, including age-related decline as well as changes that may indicate the onset of brain disorders.

Brain researchers believed that close comparative examination of structures in the hippocampus would yield key insights into how the brain ages, but there was a hitch. The Mount Sinai research team suspected that close examination of sectioned tissue sliced from the hippocampus could be used to determine brain age acceleration, but as noted by Neuroscience News, this was "a task that is impossible for a human observer to perform with any degree of accuracy."

The problem is the structural patterns that reveal brain age are both subtle and numerous — meaning more correlations working together and interacting than anyone could hold in mind to make the comparison. This is not unlike the challenge of large-scale language translation, as reported by MIT News. Linguistics researchers sought for years to automate translation by sussing out the underlying grammatical logic that allows different human languages to express ideas. But grammar is just too complicated to fully work out by hand.

Then, AI researchers discovered that if they just gave a computer an enormous amount of text to chew over (e.g., millions or billions of pages), they didn't need to fully work out the grammar in advance. The computer could patiently work through the mass of material, find correlations — such as how often one particular word follows some other word — and use all those subtle relationships to determine the best translation for a series of words. And it worked!

Multiple Instance Learning

The brain researchers at Mount Sinai use the same approach but with the preserved brains of nearly 700 elderly brain donors instead of a corpus of text. However, this raised a further challenge. Fully detailing all these preserved brains would lead the researchers right back to the problem of too many subtle patterns to follow. The research team could only work with limited information, such as the chronological age and medical diagnoses of the donors at death.

The solution to this problem was an AI technique known as multiple instance learning (MIL). This technique is suited to "weakly supervised" machine learning. As described by Jonathan Glaser, "Whereas traditional machine learning techniques rely on feature extraction by domain experts, deep learning algorithms learn high-level features from data on their own."

In MIL, as reported at arXiv.org, data is initially sorted into "bags," or assorted collections of individual data instances. A bag may be positive if its data includes at least one instance of a specified characteristic or negative if it has no instances with that characteristic.

Repeated passes through the data allowed HistoAge to build up its own detailed model of hippocampus characteristics based on its training data. The process was complex, but the goal was simple: The AI was to determine an estimated brain age at death for each brain studied. And, as with machine translation, it worked. Per Neuroscience News, HistoAge was able to estimate the average age at death within 5.45 years of the donors' actual ages at death.

The Future of AI Brain Study?

Mount Sinai researchers found that when stacked up against previous existing brain age determination techniques, such as DNA methylation, the age acceleration estimations derived from HistoAge reliably determined brain age and discovered insights into the driving factors of clinical and pathological outcomes related to cerebrovascular disease, cognitive impairment and abnormal levels of Alzheimer's-type degenerative protein aggregation. This offers a preliminary answer to the question, "How is AI helping humans understand the brain?"

Note that this line of research into AI in neuroscience is not aimed at the goal of AI examining individual (living) patients but at building up a research model that human brain specialists can then use as a baseline tool for evaluating an individual patient's brain health. This exemplifies the wide range of applications in which AI technology is enabling transformative progress.

From a big-picture perspective, HistoAge is a demonstration of AI capabilities: identifying complex and subtle patterns in a large body of data — patterns that humans without AI tools would never be able to detect, let alone work with. And this innovative use of AI could pave the way forward for brain study in the years to come.

AI is a key component of Northrop Grumman digital transformation efforts. See how we're applying AI to discover new insights.