Researchers built single-subject cortical morphological networks using six measures: cortical thickness, surface area, cortical volume, mean curvature, sulcal depth and local gyrification index. General linear models were used to compare local cortical morphology and network topology between the generalized anxiety disorder and healthy-control groups. The analysis covered 144 participants in total, all of whom were evaluated in the same study, while the network construction allowed each person’s cortical measurements to be represented as a distinct network.

At the regional level, the generalized anxiety disorder group had increased cortical thickness in the left middle temporal gyrus and a decreased local gyrification index in the left superior frontal gyrus. The clearest network-level findings were reported for cortical volume-based networks. Within those networks, local efficiency and normalized clustering coefficient increased, characteristic path length decreased and small-worldness increased. Together, the findings show both focal differences in cortical structure and broader changes in the modeled organization of cortical networks. However, they do not demonstrate that the observed structural differences cause generalized anxiety disorder or its associated symptoms.

Partial correlation analyses within the patient group connected local morphological abnormalities and network properties with illness duration, age of onset, anxiety severity and personality traits. These analyses controlled for age, sex, years of education and intracranial volume. The reported relationships therefore link particular brain features with clinical characteristics, but the group comparison and correlation analyses do not establish temporal sequence, direction of effect or a causal mechanism. The study’s scope is also limited to the measured population, morphological characteristics and network metrics: it does not establish that the differences are unique to generalized anxiety disorder, that cortical volume networks can be used diagnostically, or that changing the network organization would alter symptoms. Published in Translational Psychiatry in 2026, the accepted research offers a detailed account of cortical structural-network abnormalities and their associations with multiple clinical features of the disorder.

Subject of Research: Cortical morphological-network differences associated with symptoms of generalized anxiety disorder

Article Title: Aberrant single-subject cortical morphological networks and their associations with multiple symptoms in generalized anxiety disorder

Article References: Liu, L., Hu, X., Deng, R., Cao, L., Chai, S., Wang, Y., Li, H., Gao, Y., Gong, Q., Qiu, C., & Huang, X. (2026). Aberrant single-subject cortical morphological networks and their associations with multiple symptoms in generalized anxiety disorder. Translational Psychiatry. https://doi.org/10.1038/s41398-026-04434-3

Image Credits: AI Generated

DOI: 10.1038/s41398-026-04434-3

Keywords: generalized anxiety disorder, cortical morphology, morphological networks, cortical volume, neuroimaging, network topology, cortical thickness, gyrification, anxiety severity, personality traits, Aberrant, single-subject

The research was motivated by a persistent frustration in psychiatry: the effectiveness of psychopharmaceutical treatments has largely stagnated in recent decades, and clinicians still have no reliable method for predicting which class of antidepressant, for example, will help a specific patient. The authors argue that the first step toward truly individualized treatment is the ability to predict psychological attributes from biological markers. If brain morphology can forecast cognitive performance at the individual level, the same logic could eventually extend to predicting treatment outcomes, closing a gap that group-level correlations between brain structure and behavior have never been able to bridge.

To capture as much structural information as possible, the team scanned 74 transdiagnostic patients and 41 healthy controls on a 7-Tesla Magnetom Terra scanner at Forschungszentrum Jülich, one of the most powerful clinical MRI systems available. The MP2RAGE sequence produced T1-weighted anatomical images at an isotropic resolution of 0.75 millimeters, far finer than the 1-millimeter standard of conventional 3-Tesla scanners. The patient group was genuinely heterogeneous: 49 participants had major depressive disorder, with the remainder diagnosed with schizophrenia, bipolar disorder, ADHD, alcohol or drug abuse disorders, and non-schizophrenic psychosis. Healthy controls were screened with the Mini International Neuropsychiatric Interview to exclude any lifetime neurological or psychiatric illness.

From each scan, the researchers extracted three families of morphological features: grey matter volume, cortical thickness, and gyrification, the degree of cortical folding. The brain was parcellated into 100 regions using the Schaefer Atlas, which defines areas according to functional connectivity patterns rather than anatomical landmarks alone. In total, 304 features fed into the model. The authors deliberately avoided any a priori hypotheses about which brain areas matter, adopting a strictly data-driven approach in which every atlas region contributed volumetric and surface parameters for every participant.

The classification model was a feed-forward neural network trained with a triple loss function, an objective well suited to small datasets because it embeds information in a compact latent space and carries a relatively low risk of overfitting. Participants were labeled as above- or below-median performers on the Trail-Making Test, with the A and B versions combined to double the number of data points per subject. Training ran for up to 200 epochs using the Adam optimizer with a learning rate of 0.2, a scheduler that reduced the rate when validation loss plateaued, and gradient clipping to stabilize updates. Crucially, the team employed a strict Leave-Subject-Out splitting scheme, ensuring that no individual’s data appeared in both training and validation sets during any iteration.

To guard against overly optimistic results, the entire pipeline was wrapped in an outer hold-out loop executed ten times with different random splits. The mean and median accuracy across all ten iterations was 0.795, with a standard deviation of only 0.068; the worst run scored 0.682 and the best 0.891. Validation loss tracked training loss closely, and accuracy reached a stable plateau by roughly epoch 30 to 40, indicating that the network learned generalizable patterns rather than memorizing noise. The low variance across independent runs was presented as key evidence that the model’s performance reflects true predictive power rather than a lucky data split.

Interpretability came from Shapley Additive Explanations, or SHAP, a technique that quantifies each feature’s contribution to the model’s output. Averaged across all ten hold-out iterations, the SHAP analysis revealed that gyrification was by far the most decisive feature, while cortical thickness did not appear among the important variables at all. Grey matter volume in frontal, parietal, temporal, and occipital regions also contributed meaningfully. The influential regions clustered within four functional networks: the sensorimotor network, the limbic network, the default mode network, and the dorsal attention network.

The anatomical pattern aligns remarkably well with decades of independent neuroscience. Gyrification changes in the sensorimotor network, including the auditory cortex, make intuitive sense because completing the Trail-Making Test requires fine motor coordination and speed control. The limbic contribution came from the left orbitofrontal cortex, a region known for temporal cognition, sequencing, and stimulus-reward processing. The default mode network was represented by the dorsal prefrontal cortex and temporal cortex, while the dorsal attention network contributed through the postcentral gyrus. Previous lesion-mapping studies have shown that damage to the dorsolateral prefrontal cortex and anterior cingulate cortex impairs Trail-Making performance, and functional imaging work has repeatedly implicated these same circuits, lending independent validation to the data-driven findings.

The dominance of gyrification carries a deeper biological message. Unlike grey matter volume and cortical thickness, which change measurably across adulthood in response to stress, aging, and disease, cortical folding is largely fixed during ontogenesis and early childhood development. The model’s heavy reliance on gyrification suggests that predisposing structural patterns shaping cognitive performance are set early in life, while the concurrent relevance of grey matter volume in temporal and parietal-occipital cortex shows that modifiable features still matter. This distinction between static and dynamic morphology may prove important for deciding which biomarkers are useful for tracking change over time versus characterizing stable individual traits.

Perhaps the most consequential implication concerns how psychiatric illness should be classified. A sizable proportion of both patients and controls fell on the slow side of the median test times, meaning that diagnosis alone was a poor predictor of cognitive function. The authors argue this supports a transdiagnostic approach, in which structural brain features, psychological test results, and demographic data inform clinical decisions rather than broad diagnostic categories. Notably, the model drew on the orbitofrontal cortex, a region more often associated with affective disease than with executive function, hinting that psychiatric conditions as different as bipolar disorder, unipolar depression, and the negative symptoms of schizophrenia may share overlapping neurobiological substrates. Symptom prediction from neuroimaging, the study concludes, appears to be an achievable goal, and establishing that brain structure can forecast cognitive performance is a crucial first step toward identifying which biological interventions are most likely to help each individual patient.

Subject of Research: Predicting cognitive test performance from 7-Tesla structural brain MRI using deep learning

Article Title: Predicting trail-making test performance using 7-Tesla structural cerebral MRI

Article References: Schnellbächer, G. J., Rajkumar, R., Veselinović, T., Ramkiran, S., Hagen, J., Collee, M., Shah, N. J., Neuner, I., & Dammers, J. (2026). Predicting trail-making test performance using 7-Tesla structural cerebral MRI. Translational Psychiatry, 16(1), Article 572. https://doi.org/10.1038/s41398-026-04402-x

Image Credits: AI Generated

DOI: 10.1038/s41398-026-04402-x

Keywords: 7-Tesla MRI, deep learning, Trail-Making Test, gyrification, grey matter volume, cortical thickness, transdiagnostic psychiatry, SHAP analysis, executive function, neuroimaging biomarkers, Translational Psychiatry, cognitive prediction

The framework, described in Nature Computational Science by Yingjie Zhao and Zhiping Xu of Tsinghua University together with Yicheng Song and Fan Xu of Fudan University, is called physics-transfer learning, and its central insight is deceptively simple: the governing equations of nonlinear elasticity do not care whether they are describing a rubber sheet, an engineered shell or living brain tissue. If a neural network can first learn the mechanics of folding in simple, analytically tractable geometries where the mathematics is fully understood, it can then carry that knowledge across a carefully staged sequence of increasingly complex systems, ultimately arriving at the brain itself. Each step along this chain of physics-anchored domains reinforces the same underlying mechanical principles, so the network never drifts far from the physical laws that actually govern morphogenesis.

The starting point of this transfer chain is deliberately humble. The researchers began with elastic structures whose buckling behavior can be derived in closed form, embedding the classical theory of nonlinear elasticity into the network’s architecture and training objectives as explicit constraints rather than leaving the model to infer them from examples. From there, the team constructed what they call chains of physics-anchored domains, a progression of model systems that gradually increases in complexity and fidelity while preserving the shared mechanical substrate. This staging matters because transfer learning in the standard machine learning sense has long been criticized for moving knowledge between domains with no guarantee of success. By anchoring every step to consistent governing laws, the researchers could go further than intuition and actually prove something remarkable: a theoretical generalization bound that explains why and when the transferred knowledge remains reliable on the unfamiliar target domain.

That theoretical foundation sets the work apart from the crowded field of physics-informed machine learning. Since the introduction of physics-informed neural networks, researchers have sought to weave partial differential equations into deep learning, producing impressive results in fluid dynamics, materials discovery and beyond. But these approaches often face what practitioners describe as an accuracy-performance dilemma, and a 2024 analysis in Nature Machine Intelligence warned that weak baselines and reporting biases have inflated optimism in the field. The new study confronts that skepticism head-on by deriving its generalization bound from the physics-transfer theory itself, turning what is usually an empirical claim into a mathematical statement about how closely the source and target domains share their governing laws. In effect, the bound quantifies the value of physical consistency, providing a principled answer to the question every applied scientist asks: how well will my model work on data it has never seen?

To test the framework against reality, the researchers turned to clinical data, validating their models against magnetic resonance imaging atlases of the developing human brain, including the spatio-temporal surface atlas of the fetal brain generated by the Developing Human Connectome Project. The models were asked to perform two distinct tasks. The first was descriptive: characterizing geometric features of the brain surface, such as curvature, that encode the signature of the folding process. The second was predictive: forecasting the progression of morphological development, essentially predicting how the folding landscape will evolve over developmental time. Compared with purely supervised learning models trained on the same limited data, the physics-transfer approach demonstrated markedly stronger performance on both tasks, a result the generalization bound anticipated. The comparison makes clear that when labeled examples are scarce, physics is worth more than data.

Beyond raw predictive accuracy, the framework delivers something arguably more valuable for scientists: understanding. The trained networks yield reduced-dimensional evolutionary representations, compact mathematical descriptions that distill the essential physics of brain morphogenesis into a low-dimensional space. In this compressed representation, the bewildering complexity of a folding cortex collapses onto a small set of coordinates that trace the trajectory of development, in much the same way that principal component analysis reveals dominant patterns in high-dimensional data. The researchers’ analysis of the internal structure of the neural networks shows that these reduced representations are not statistical artifacts but carry genuine mechanical meaning, connecting the information bottleneck of deep learning with the physical instabilities, such as buckling and wrinkling, that mechanicians have long studied in shrinking and growing shells.

The implications reach well beyond developmental neurobiology. The same group had previously applied physics-transfer learning to discover high-strength alloys, demonstrating that the framework generalizes across materials science, and the new work extends the recipe to soft matter and biomechanics. For clinicians, the prospect of physics-aware digital twins of the developing brain is particularly compelling. A digital twin in this context is a patient-specific computational model that evolves in lockstep with its biological counterpart, offering a virtual laboratory in which development can be monitored, diagnosed and, in principle, intervened upon. Malformations of cortical development, a class of congenital conditions arising when the folding process goes awry, are currently classified largely by descriptive criteria; a predictive mechanical model could identify deviations from normal development earlier and connect them to their underlying biomechanical causes.

The team has been unusually generous with the products of the research, releasing digital libraries of brain models and simulation data through figshare, the fetal brain surface atlas data through G-Node, and all scripts for reproducing the results through GitHub, Zenodo and Code Ocean. This openness matters because reproducibility has been a persistent sore point in machine learning for the physical sciences, and it invites the broader community to stress-test the generalization bounds and probe the limits of the domain transfer. It also lowers the barrier for groups in resource-limited settings, where large labeled datasets of fetal imaging are precisely the kind of resource that is hardest to obtain. A method that achieves strong performance from scarce data, grounded in physics rather than brute-force annotation, democratizes access to high-fidelity developmental modeling.

The work also speaks to a deeper conversation about how machine intelligence should engage with the physical world. Data-driven approaches have achieved spectacular successes, but critics note that pure pattern recognition can fail catastrophically outside its training distribution, an unacceptable risk in medicine. Theory-grounded approaches, by contrast, trade some flexibility for reliability, encoding invariants that hold regardless of the examples on hand. Physics-transfer learning offers a bridge between these poles: it exploits data where data exist, leans on physical law where they do not, and, crucially, comes with a mathematical guarantee that quantifies the trade. As Zhao, Song, Xu and Xu suggest, that combination of accuracy, interpretability and theoretical assurance may be exactly what physics-aware digital-twin technologies need to move from promising demonstrations to clinical tools for understanding, diagnosing and ultimately intervening in the developing and diseased brain.

For a species whose intelligence is written in the folds of its cortex, there is a satisfying symmetry in the fact that it took a fusion of physics and machine learning to finally read that script. The folding of the brain is neither pure genetics nor pure mechanics but the inseparable entanglement of the two, and any model that captures only one side of the story was always destined to fall short. By teaching a neural network the elastic grammar that all folding matter obeys, and then letting it climb the ladder of complexity from simple shells to living tissue, the Tsinghua and Fudan team has shown that the shortest path to understanding nature’s most intricate forms may run through her simplest ones. If the framework continues to validate against clinical imaging, the folding patterns on a fetal scan could one day speak not just of anatomy, but of mechanics, prognosis and possibility.

Subject of Research: Physics-transfer learning for predicting brain morphogenesis and cortical folding from limited data

Article Title: Predicting brain morphogenesis via physics-transfer learning

Article References: Zhao, Y., Song, Y., Xu, F., & Xu, Z. (2026). Predicting brain morphogenesis via physics-transfer learning. Nature Computational Science. https://doi.org/10.1038/s43588-026-01040-7

Image Credits: AI Generated

DOI: 10.1038/s43588-026-01040-7

Keywords: physics-transfer learning, brain morphogenesis, cortical folding, gyrification, nonlinear elasticity, physics-informed machine learning, neural networks, generalization bound, digital twin, fetal brain imaging, Developing Human Connectome Project, computational mechanics