Researchers have found that the physical wiring patterns of a one-year-old child’s brain can predict their intelligence scores several years later. By analyzing specific brain imaging maps using artificial intelligence, the study suggests that the foundations of cognitive ability are established in early infancy. The findings were published in the journal Frontiers in Human Neuroscience.
Early childhood is widely recognized as a major period for the development of lifelong cognitive abilities and behaviors. Identifying biological indicators of brain development allows professionals to potentially predict and track cognitive trajectories over a person’s lifespan. Identifying these biological markers might also allow for timely interventions to optimize learning outcomes.
The human brain depends on a vast network of nerve fibers called white matter. This material acts like physical cables, transmitting electrical signals between different biological processing centers. The complete map of these neurological connections is known as a structural connectome.
Because a connectome contains an overwhelming amount of raw data, researchers often use mathematical tools to create simplified models called gradients. A connectome gradient represents how brain connectivity gradually changes across different spatial dimensions. This provides a topographical map of the organ’s physical layout.
Previous studies have looked closely at functional gradients, which track how different brain areas communicate with one another in real time. Functional gradients often show how the brain handles tasks ranging from primary senses, like vision and touch, to advanced reasoning. However, less attention has been paid to structural gradients.
Structural gradients represent the actual physical nerve pathways that dictate where and how those real-time communications can travel. They represent the anatomic scaffolding of the mind.
Yoonmi Hong, a researcher in the Department of Psychiatry at the University of North Carolina at Chapel Hill, and her colleagues wanted to know if these physical scaffolding gradients look fully formed in toddlers. They also designed a study to test whether a child’s structural connectome at age one could predict their general cognitive abilities throughout the rest of their early childhood. Cognitive performance plays a major role in how well children adjust academically and socially once they reach school age.
The research team suspected that networks associated with advanced thought processing would be particularly relevant to their predictions. Areas like the frontal and parietal lobes are known to govern executive functions, which include problem-solving, attention, and working memory. By measuring how well these regions are physically wired together early in life, the researchers hoped to capture a baseline snapshot of future cognitive development.
To investigate this idea, the research team analyzed brain scans from a long-term infant development project. They focused on imaging data from around one hundred children who underwent brain evaluations at one year of age. The specific type of scan used, called diffusion magnetic resonance imaging, tracks the microscopic movement of water molecules in the brain.
Because water travels more easily along the length of a nerve fiber than across it, scientists can use this water movement to map out the direction and thickness of white matter tracts. Using these scans, the investigators calculated two main structural gradients for each child.
The primary gradient measured the connectivity patterns running from the left side of the brain to the right side. This axis is heavily defined by the relative lack of physical connections between the two hemispheres, meaning each half forms its own distinct networking architecture.
The secondary gradient measured the pathways running from the front of the brain to the back. This typically reflects a transition from basic sensory regions in the rear to complex executive centers in the front.
The children in the study later completed standardized cognitive assessments at ages four, six, and eight. To link the early brain scans with these later intelligence scores, Hong and her team developed a particular type of machine learning model designed to analyze complex networks. They trained an artificial neural network to find computational relationships between the one-year-old structural connectome gradients and the subsequent childhood intelligence evaluations.
Typically, traditional network analysis looks at specific points in the brain in isolation. This approach can miss the broader, highly distributed topographical patterns that the structural gradients are designed to capture. By utilizing an artificial neural network, the model could evaluate the brain’s entire connective architecture simultaneously. It learned to compare how slight variations in spatial organization correlated with variations in the final intelligence scores.
The computer model successfully forecast the children’s later cognitive abilities based entirely on their physical brain maps at age one. Even though the original scans were taken in infancy, the computer’s predictions remained highly consistent across the intelligence evaluations at ages four, six, and eight.
The researchers point to the stability of white matter maturation as the reason for this success. When they examined additional scans taken at ages two, four, and six, the structural gradients looked very similar to the ones originally measured at twelve months.
By analyzing the inner workings of their artificial intelligence model, the researchers identified which brain regions contributed most heavily to the prediction. The mapping tool relied almost exclusively on regions within the frontoparietal network and the executive control network. These are the areas of the brain involved in managing attention, guiding cognitive flexibility, and integrating basic sensory information. The model’s reliance on these specific regions aligns with existing theories about where human intelligence is localized in adults.
While the computer model successfully predicted cognitive scores, the researchers outlined some limitations to their approach. Reconstructing the brain’s physical wiring by counting nerve pathways is an imperfect science. Minor head movements during an infant’s brain scan can distort the imaging data, resulting in an artificially low count of connecting fibers. Complex intersections where multiple distinct nerve fibers cross paths can also confuse the tracking software, creating potential inaccuracies in the map.
The choice of software mapping definitions, known as cortical parcellations, also influences the boundaries used to define network nodes. Using a different mapping program could alter the shape of the gradient topography, potentially changing which specific brain regions the artificial intelligence flags as predictive.
The early intelligence prediction model also did not include demographic details, such as maternal education levels. During statistical testing, the researchers noted that demographic variables actually predicted child intelligence scores better than the biological brain imaging features did. Maternal education is heavily associated with later cognitive outcomes, likely due to differences in household learning resources, early language exposure, and general environmental support.
Future research will examine how demographic advantages and physical brain development might biologically relate to one another. The team hopes to determine if the development of physical white matter networks acts as a mediating bridge between environmental factors completely outside a child’s brain, like early education, and biological intelligence. They will also test whether a computer model that incorporates both brain imaging and demographics yields the highest predictive accuracy.
Additional studies are also needed to explore specific types of cognitive tasks. Instead of grouping all childhood abilities into a single intelligence quotient score, researchers plan to track specific skills. The team intends to investigate whether separate brain connectivity gradients can predict distinct educational outcomes, isolating things such as verbal fluency from nonverbal visual memory. Mapping multiscale structural gradients could completely open up new possibilities for uncovering the comprehensive principles of organizational brain development.
The study, “Structural connectome gradients and their relationship to IQ in childhood,” was authored by Yoonmi Hong, Emil Cornea, Jessica B. Girault, Rebecca L. Stephens, Maria Bagonis, Mark Foster, Sun Hyung Kim, Juan Carlos Prieto, Martin A. Styner, and John H. Gilmore.