AI biomarker built to measure brain age

University of Oslo researchers develop AI biomarker for tracking neurological treatment over time.

Finding timely therapies for age begins with finding a way to measure how old a thing is. Now, in a groundbreaking new study, researchers at the University of Oslo have been able to build an algorithm to accurately predict brain age — all from a quick glance at a patient’s MRI scan.

And what’s more, their research, appearing last week in Nature Neuroscience, shows that the brains of patients with a slew of neurological disorders, from dementia to schizophrenia, are aging at an accelerated rate — and the causes are at least partly genetic.

“We reveal that genes involved in apparent brain aging in healthy individuals overlap with genes that we know to be associated with common brain disorders,” says study author Tobias Kaufmann.

Longevity.Technology: How quickly we can find treatments for the conditions of aging is fully dependent upon our ability to find effective biomarkers to track the effects of prospective therapies. The news of another prospective biomarker will be greeted warmly by many longevity researchers, not in the least by the Buck Institute’s CEO Eric Verdin, who told us last week that he sees the development of “a series of comprehensive biomarkers,” as the most immediate challenge for the aging field. This breakthrough sits among others, such as the work on gut bacteria by Insilico and Professor Steve Horvath’s methylation clock, at the forefront of promising new biomarkers, a comprehensive summary of which can be seen inside the Aging Analytic Agency’s Biomarkers report, which is slated for release next month. As it stands, the study is a promising first step on a long road for this technique, with improvements to its ability to draw accurate judgements from single scans needed before it can be deployed in hospitals.

The TRL score for this Longevity.Technology domain is currently set at:

The TRL score for the technology addressed in this article is: “Late proof of concept demonstrated in real life conditions.”

The study is the first of its kind to make these connections, and did so in two ambitious steps. The first was made by using a machine learning algorithm on roughly 45,000 MRI scans gathered from individuals between the ages of 3 and 96. 34,474 of these scans were taken from healthy volunteers and given to one of two models, depending on sex, to train them on the brain features visible from an MRI scan — cortical thickness, brain area, and regional volume.

Once the models were trained, they were validated with data from 4,353 healthy brains before being put to work to predict the brain age of 5,788 people with various brain disorders, including multiple sclerosis, schizophrenia, major depressive disorder, mild cognitive impairment and dementia. The results showed that the conditions that created the biggest ‘brain age gaps’ between an individual’s chronological age and the biological age of their brains were schizophrenia, multiple sclerosis and dementia; with developmental disorders like autism and ADHD having little to no effect.

Next, the researchers took the second step of teasing out the genetic connections to accelerated brain aging. Using human genomic data from the UK Biobank, the team were able to find a small group of genes that were connected to brain disorders and the symptoms of accelerated aging in the brain — showing that an individual’s brain age gap is at least partially inheritable.

The method is promising, but far from watertight at the moment. It is still tough for models trained on average results to make good predictions from individuals scans, as MRI readings tend to be noisy and affected by a lot of inter-individual variability. The researchers hope to find solutions to these problems with further research.

Eventually, and with enough refinement, though, they hope to see their AI biomarker method employed as a neurological biomarker that will be able to predict brain age based upon genetics, as well as tracking the brain’s response to treatment. It joins a number of hopeful new AI based biomarkers with the same goals for other parts of the body, the development of which will see a dramatic acceleration to the speed of drug trials targeting aging.