Summary: New research challenges the long-standing idea that better navigators have visibly larger brain regions. Using advanced deep learning applied to MRI scans, researchers found no clear link between macroscopic brain structure—including the hippocampus—and real-world spatial navigation performance in healthy young adults.
Researchers applied state-of-the-art deep convolutional neural networks and graph convolutional neural networks to T1-weighted MRI scans, aiming to detect subtle structural patterns that might predict navigation skill. Despite these powerful tools, the models did not reliably predict individual differences in navigation ability from brain shape or size alone.
Key Facts
- AI vs. traditional measures: Earlier studies that relied on coarse measures like regional volume sometimes suggested links between brain structure and navigation. But deep learning models designed to detect complex, spatial patterns in MRI data failed to find a consistent “navigation signal” in a sample of healthy young adults.
- Hippocampus and control regions: The study compared the hippocampus—commonly linked to spatial memory—and the thalamus as a control. Structural differences between these regions did not predict navigation performance.
- Sample: Ninety healthy participants (mean age 23.1) learned and recalled routes within a realistic virtual environment. Objective tests measured their mapping and spatial memory abilities.
- Behavior vs. disease prediction: While AI models have proven effective at identifying disease-related changes (for example, in Alzheimer’s), mapping everyday behavioral skills like navigation appears more challenging using macroscopic structural MRI alone.
- Implications for brain function: The results suggest that neural function and connectivity—how neurons fire and communicate—may matter more for navigation than gross anatomical size or shape.
Source: UT Arlington
Steven Weisberg, now at The University of Texas at Arlington, led the project and reports that current deep learning methods could not find a robust structural marker of navigation ability in this population, challenging popular interpretations of classic findings.
For decades, influential work—most famously studies of London taxi drivers—suggested that extensive navigation experience could be reflected in increased volume or altered shape of the hippocampus. Those results helped establish the hippocampus as a key brain region for spatial memory. The current study revisits these ideas using more powerful, data-driven methods that search for complex structural signatures beyond simple volumetric comparisons.
Weisberg, together with coauthors including Ashish Sahoo, trained and evaluated convolutional networks (3D CNNs) and graph convolutional neural networks on the MRI dataset to predict participants’ performance on a validated virtual navigation test. Although models fit training data reasonably well, their predictions generalized poorly to held-out test sets, producing weak predictive value overall.
The findings do not rule out structural changes after prolonged, intensive navigation training (such as years of taxi driving). Rather, they indicate that, in typical young adults, individual navigation skill may not be reflected in large-scale anatomical differences detectable by current MRI and machine-learning pipelines.
“With the resolution and quality of these MRI scans, and in a healthy young sample, we did not find a detectable structural signal,” Weisberg said. He emphasized that behavior and functional connectivity—how brain regions interact during tasks—are promising directions for future study.
Key Questions Answered:
A: Not necessarily. Extensive, long-term training may still produce structural changes that are detectable. This study indicates that ordinary variation in routing experience among young adults is unlikely to create large, detectable anatomical differences; performance likely depends more on neural activity patterns and connectivity.
A: At a macroscopic level captured by standard MRI, this study suggests there may be no large differences. Experts may differ in microscopic wiring, synaptic strength, or functional dynamics—features current structural MRI and many AI models cannot reliably measure.
A: Spatial disorientation is often an early sign of Alzheimer’s disease. If healthy navigation ability lacks a clear structural baseline, early detection efforts should emphasize behavioral tests and functional measures rather than relying solely on hippocampal size.
Editorial Notes:
- This article was edited by a Neuroscience News editor.
- The original journal paper was reviewed in full by staff.
- Additional context was provided by the editorial team.
About this AI and neuroscience research news
Author: Drew Davison
Source: UT Arlington
Contact: Drew Davison, UT Arlington
Image credit: Neuroscience News
Original Research: Open access. “Deep learning approaches to map individual differences in macroscopic neural structure with variations in spatial navigation behavior” by Ashish K. Sahoo et al., published in Neuropsychologia. DOI: 10.1016/j.neuropsychologia.2025.109352
Abstract
Deep learning approaches to map individual differences in macroscopic neural structure with variations in spatial navigation behavior
Linking brain structure to individual behavioral differences remains difficult because the brain’s organization is highly complex. Earlier studies that used coarse structural measures—such as regional volume or cortical thickness—sometimes reported associations between hippocampal anatomy and navigation ability, particularly in older adults or in individuals with extreme training. However, larger pre-registered studies in typical younger adults often find no such association.
This study takes a data-driven approach, developing and comparing graph convolutional neural networks and 3D convolutional networks trained on T1-weighted MRI data (N = 90) to predict spatial navigation performance measured in a virtual reality mapping task. Across models, predictive power on held-out data was weak despite reasonable fits on training sets. These results suggest that much larger datasets, broader behavioral assessments, or different imaging modalities (such as functional or microstructural measures) may be necessary to link neural structure reliably to navigation ability in healthy young adults. Alternatively, hippocampal macroscopic structure may simply be a limited predictor of everyday spatial behavior in this population.