A multidisciplinary team of cosmologists, neuroscientists and engineers from the University of La Laguna, the Institute of Astrophysics of the Canary Islands and the University of Messina (Italy) has recently published in the journal NeuroImage an article This paper describes how analytical techniques originally developed to study the large-scale structure of the Universe have been adapted to investigate how the human brain ages. By bridging cosmology and neuroscience, this work demonstrates how these statistical tools, commonly used in astronomy, can also shed light on the structure and aging of the human brain.
This method allows us to detect scales at which brain anatomy differs between different age groups, providing information about brain aging processes. Like the evolution of cosmic structures, brain structure also evolves naturally, but it exhibits contrasting behaviors at different scales.
At larger scales, there is a loss of structure with age, possibly due to ventricular expansion, while at smaller scales there is an increase in structure, probably related to the decrease in cortical thickness and gray and white matter volume. Thus, the study suggests that complex genetic or lifestyle-related factors may influence brain aging.
Statistical methods
Both the cosmic web—the vast network of galaxies that spans the Universe—and the human brain share a key characteristic: they are complex systems whose structures do not follow a random distribution dictated by chance. To characterize these intricate patterns, scientists turn to higher-order statistical methods that go beyond conventional one- and two-point analyses.
Directed by Francisco-Shu Kitaura, principal investigator of the project cosmic-brain.org, The team has introduced a spatial frequency analysis of magnetic resonance imaging (MRI) data of the human brain that incorporates both the power spectrum (two-point statistics) and the bispectrum (three-point statistics), the latter being sensitive to structural features that deviate from Gaussian distributions.
“After a couple of years of research calculating thousands of bispectral configurations, we discovered that the combination of the power spectrum and only four bispectral configurations is enough to achieve an average absolute error in estimating the age of the human brain of about three years using the OASIS-3 database —thus identifying a new biomarker,” explains Aurelio Carnero Rosell, first author of the study.
“It’s fascinating to see how tools developed to study the evolution of the Universe can contribute to the study of the human mind. Detecting neurodegenerative diseases before symptoms appear could be just one example of the significant advances that can arise from interdisciplinary collaborations like this one,” says Antonella Maselli of the University of Messina.
Another of the study's authors, Marc Huertas Company, points out that neural networks were necessary in the final regression analysis: "Unlike cosmology, we still do not have predictive models for two- and three-point statistics of the human brain."“
The project also involved careful preparation of the magnetic resonance imaging data, as he recalls Niels Janssen"I found it quite entertaining to watch cosmologists translating megaparsecs - about 3.26 million light-years - into millimeters so they could apply their cosmological tools to neuroimaging data."“
For his part, another of the co-authors, Ernesto Pereda, highlights the broader potential of the method: “The usefulness of this novel approach in neuroscience is that it not only allows us to obtain estimates of brain age, as machine learning methods do, but also to interpret the results, since different configurations of the bispectrum plot different scales and show specific aging trends.”
Francisco-Shu Kitaura offers a final assessment of the origin and relevance of this study: “In the field of modeling the distribution of galaxies that trace the underlying dark matter field, we discovered that the same power spectrum can be reproduced with different galactic bias models. To differentiate between models and overcome degeneracies, we must resort to higher-order statistics, particularly the bispectrum. This idea inspired me to propose the application of these methods to neuroscience.”
Spanish 
English 
French