Penn study shows variability among children's neural anatomy that may inform personalized treatments for psychiatric disorders

Machine learning is helping Penn Medicine researchers identify the size and shape of brain networks in individual children, which may be useful for understanding psychiatric disorders. In a new study published today in the journal Neuron, a multidisciplinary team showed how brain networks unique to each child can predict cognition. The study--which used machine learning techniques to analyze the functional magnetic resonance imaging (fMRI) scans of nearly 700 children, adolescents, and young adults--is the first to show that functional neuroanatomy can vary greatly among kids, and is refined during development.

The human brain has a pattern of folds and ridges on its surface that provide physical landmarks for finding brain areas. The functional networks that govern cognition have long been studied in humans by lining up activation patterns--the software of the brain--to the hardware of these physical landmarks. However, this process assumes that the functions of the brain are located on the same landmarks in each person. This works well for many simple brain systems, for example, the motor system controlling movement is usually right next to the same specific fold in each person. However, multiple recent studies in adults have shown this is not the case for more complex brain systems responsible for executive function--a set of mental processes which includes self-control and attention. In these systems, the functional networks do not always line up with the brain's physical landmarks of folds and ridges. Instead, each adult has their own specific layout. Until now, it was unknown how such person-specific networks might change as kids grow up, or relate to executive function. {module INSIDE STORY}

"The exciting part of this work is that we are now able to identify the spatial layout of these functional networks in individual kids, rather than looking at everyone using the same 'one size fits all' approach," said senior author Theodore D. Satterthwaite, MD, an assistant professor of psychiatry in the Perelman School of Medicine at the University of Pennsylvania. "Like adults, we found that functional neuroanatomy varies quite a lot among different kids--each child has a unique pattern. Also like adults, the networks that vary the most between kids are the same executive networks responsible for regulating the sorts of behaviors that can often land adolescents in hot water, like risk-taking and impulsivity."

To study how functional networks develop in children and supports executive function, the team analyzed a large sample of adolescents and young adults (693 participants, ages 8 to 23). These participants completed 27 minutes of fMRI scanning as part of the Philadelphia Neurodevelopmental Cohort (PNC) a large study that was funded by the National Institute of Mental Health. Machine learning techniques developed by the laboratory of Yong Fan, Ph.D., an assistant professor of Radiology at Penn and co-author on the paper, allowed the team to map 17 functional networks in individual children, rather than relying on the average location of these networks.

The researchers then examined how these functional networks evolved over adolescence and were related to performance on a battery of cognitive tests. The team found that the functional neuroanatomy of these networks was refined with age, and allowed the researchers to predict how old a child with a high degree of accuracy.

"The spatial layout of these networks predicted how good kids were at executive tasks," said Zaixu Cui, Ph.D., a post-doctoral fellow in Satterthwaite's lab and the paper's first author. "Kids who have more 'real estate' on their cortex devoted to networks responsible for executive function in fact performed better on these complex tasks." In contrast, youth with lower executive function had less of their cortex devoted to these executive networks.

Taken together, these results offer a new account of developmental plasticity and diversity and highlight the potential for progress in personalized diagnostics and therapeutics, the authors said.

"The findings lead us to interesting questions regarding the developmental biology of how these networks are formed, and also offer the potential for personalizing neuromodulatory treatments, such as brain stimulation for depression or attention problems," said Satterthwaite. "How are these systems laid down in the first place? Can we get a better response for our patients if we use neuromodulation that is targeted using their own personal networks? Focusing on the unique features of each person's brain may provide an important way forward."

The automatic identification of complex features in images has already become a reality thanks to artificial neural networks. Some examples of software exploiting this technique are Facebook's automatic tagging system, Google's image search engine, and the animal and plant recognition system used by iNaturalist. We know that these networks are inspired by the human brain, but their working mechanism is still mysterious. New research, conducted by SISSA in association with the Technical University of Munich and published for the 33rd Annual NeurIPS Conference, proposes a new approach for studying deep neural networks and sheds new light on the image elaboration processes that these networks are able to carry out.

Similar to what happens in the visual system, neural networks used for automatic image recognition analyze the content progressively, through a chain of processing stages. However, to date, it is not completely clear which mechanisms allow deep networks to reach their extraordinary levels of accuracy.

"We have developed an innovative method to systematically measure the level of complexity of the information encoded in the various layers of a deep network - the so-called intrinsic dimension of image representations," said Davide Zoccolan and Alessandro Laio, respectively neuroscientist and physicist at SISSA, a research institution in Trieste, Italy. "Thanks to a multidisciplinary work that has involved the collaboration of experts in physics, neurosciences and machine learning, we have succeeded in exploiting a tool originally developed in another area to study the functioning of deep neural networks." {module INSIDE STORY}

SISSA scientists, in association with Jakob Macke, of the Technical University of Munich, have examined how the information acquired from neural networks used for image classification is processed: "We have found that image representations undergo a progressive transformation. In the early processing stages, image information is faithfully and exhaustively represented, giving rise to rich and complex representations. In the final processing stages, the information is radically simplified, producing image representations that are supported by a few dozen parameters" explain the two scientists. "Surprisingly we found that the classification accuracy of a neural network tightly depends on its ability to simplify: the more it simplifies the information, the more accurate it is."

This is an especially important result for SISSA that has recently launched a new research program in Data Science, with the goal of studying and developing innovative algorithms for the processing of complex and large data sets.

This study has been published in the proceedings of the 33rd Annual NeurIPS (Neural Information Processing Systems) Conference, the key appointment dedicated to artificial intelligence and to machine learning, in Vancouver from 8 to 14 December 2019. On that occasion, it will be presented by Alessio Ansuini, the first author of the study and the scientist who was responsible for conducting the experiments during his post-doctoral research activity at SISSA.