Traditional neuroimaging analysis often treats the brain as a collection of isolated pixels or independent regions of interest. If a study looks at a disorder like schizophrenia or a trait like intelligence, it typically asks a univariate question: Is the volume or thickness of this specific region significantly different?
However, the human brain is an integrated, interconnected system. Brain regions do not mature or degenerate in isolation; they evolve together through shared genetic blueprints, synchronized developmental timelines, and axon-mediated trophic support. To capture this systemic organization, researchers have shifted from localized measurements to mapping structural networks. By analyzing how physical dimensions covary across populations or within individuals, these methods reveal the hidden structural architecture of the brain.
Population Level Structural Covariance
The earliest and most intuitive way to build a structural network is through Structural Covariance Networks (SCNs). Instead of measuring connectivity via direct white matter tracts (like diffusion tensor imaging) or synchronized firing patterns (like functional MRI), SCNs infer connectivity from population-level statistical correlations of a single anatomical feature, such as cortical thickness or gray matter volume.
Imagine measuring the cortical thickness of 100 individuals across dozens of parcellated brain regions. For any two regions, say the anterior cingulate cortex and the insula, you can plot their thicknesses across all 100 subjects and calculate a Pearson correlation coefficient. If individuals with a thicker anterior cingulate also tend to have a thicker insula, these regions exhibit high structural covariance, suggesting they belong to a shared network.
[Subject 1] -> Region A: 3.2mm, Region B: 2.8mm
[Subject 2] -> Region A: 2.9mm, Region B: 2.5mm
[Subject 3] -> Region A: 3.5mm, Region B: 3.1mm
Result: High correlation (Edge created between Node A and Node B)
This population-level approach has proven highly sensitive to neurodevelopmental and clinical variations:
- Developmental Blueprints: Xiujuan Geng et al. demonstrated in Structural and Maturational Covariance in Early Childhood Brain Development that seed-based structural covariance networks in early childhood closely mirror known functional networks, suggesting that early functional activity may gradually sculpt the brain's physical structure.
- Clinical Biomarkers: SCNs can expose network-level breakdowns that traditional voxel-wise comparisons miss. For instance, Feng Shi et al. used this framework in Altered Modular Organization of Structural Cortical Networks in Children with Autism to reveal a less optimized, less modular structural network in children with autism. Similarly, Mowen Yin et al. in Network analysis on cortical morphometry in first-episode schizophrenia showed that population correlations reveal early topological breakdowns in organizational efficiency during a patient's first episode of schizophrenia.
- Systemic Pathology: In pediatric chronic pain, Zexing Cheng et al. showed in Abnormal structural gray matter and structural covariance networks associated with biopsychosocial characteristics in children with multisite pain that structural thinning is coordinated across the "pain matrix," linking network-level alterations directly to biopsychosocial challenges like poor sleep.
Individual Level Morphometric Similarity
While population-level SCNs are powerful, they suffer from a major limitation: they require an entire cohort to generate a single network graph. You cannot easily extract a structural covariance network for an individual patient to predict their specific clinical outcome.
To solve this, researchers developed Morphometric Similarity Networks (MSNs). Instead of correlating a single feature across many people, MSNs correlate many features within a single person.
For an individual brain scan, researchers extract a multi-dimensional feature vector for each anatomical region. This vector acts as a structural "fingerprint," combining diverse macroscale and microscale metrics such as cortical thickness, surface area, gray matter volume, myelination, and water diffusion properties. By calculating the statistical similarity (correlation) between the feature vectors of Region A and Region B, they can construct a personalized network map for that specific individual.
Individual Subject:
Region A Vector: [Thickness, Surface Area, Volume, Myelination, MD]
Region B Vector: [Thickness, Surface Area, Volume, Myelination, MD]
Result: Correlation between vectors = Individualized Edge Weight
Because MSNs capture a holistic proxy of underlying cytoarchitecture, they track individual differences with high fidelity:
- Lifespan Trajectories: In Imaging lifespan brain structural growth: From region, to connectome, to gradient, Xing Qian mapped how these individual networks reorganize from ages 8 to 89, linking adolescent network maturation directly to cognitive flexibility and neurotransmitter receptor density.
- Predicting Development: Daphna Fenchel et al. applied this to newborns in Development of Microstructural and Morphological Cortical Profiles in the Neonatal Brain (2020) and Neonatal multi-modal cortical profiles predict 18-month developmental outcomes, proving that multi-feature structural networks are already present at birth and can successfully predict social-emotional outcomes at 18 months.
- Pathological Remodeling: When the brain experiences trauma, MSNs can map the resulting structural disorganization. Yizhen Pan et al. demonstrated in Cortical Morphometric Similarity Remodeling in Traumatic Brain Injury Links Cognitive Impairments with Transcriptional Changes and Type‐Specific Cells that mild traumatic brain injury alters MSN connectivity patterns in ways that correlate with cognitive deficits and underlying gene expression profiles.
Source Based Morphometry
Both SCNs and MSNs rely on pre-defined anatomical boundaries (parcellations) to define their network nodes. However, pathology and development do not always respect these arbitrary borders. Source-Based Morphometry (SBM) solves this by using a data-driven, multivariate decomposition technique—typically Spatial Independent Component Analysis (ICA)—to discover naturally co-varying structural networks directly from whole-brain images.
Instead of calculating region-by-region correlations, SBM takes whole-brain structural maps (like gray matter density or concentration) from a large population and decomposes them into statistically independent spatial components. Each component represents a network of voxels that vary in volume or concentration together across the cohort.
Population Gray Matter Maps -> [ Spatial ICA ] -> Independent Structural Components (Networks)
This approach bypasses localized region-of-interest constraints and provides a clearer picture of systemic diseases:
- Disentangling Confounding Metrics: In Spatially Covarying Patterns of Gray Matter Volume and Concentration Highlight Distinct Regions in Schizophrenia, Kelly Rootes-Murdy et al. used SBM to prove that gray matter volume and gray matter concentration, though often conflated, yield distinct spatial networks that degrade differently in schizophrenia.
- Early Detection of Neurodegeneration: Jennifer Ciarochi leveraged SBM in Patterns of Co-Occurring Gray Matter Concentration Loss across the Huntington Disease Prodrome to identify distinct networks of gray matter loss that degrade in tandem years before the onset of clinical motor symptoms, distinguishing true network-level atrophy from isolated regional degradation.
- Tracking Lifespan Phenotypes: In Gray matter networks associated with attention and working memory deficit in ADHD across adolescence and adulthood, Kuaikuai Duan et al. showed that while localized gray matter deficits in ADHD often appear to normalize in adulthood, the underlying SBM-derived networks remain robustly linked to cognitive deficits across the entire lifespan.
Why Structural Covariance Works
Why do these physical dimensions correlate so tightly? The phenomenon relies on three core biological mechanisms:
- Axon-Mediated Trophic Support: Brain regions that are heavily wired together via white matter tracts constantly exchange neurotrophic factors. If an injury or disease disrupts a pathway, the loss of trophic support causes coordinated thinning or volume loss across the entire connected circuit.
- Coordinated Neurodevelopment: Regions that serve shared functional roles mature along a synchronized timeline. In Evidence in cortical folding patterns for prenatal predispositions to hallucinations in schizophrenia, Colleen Rollins et al. looked at cortical folding patterns established during the second trimester of gestation. They found that prenatal structural synchrony between the salience and auditory networks predicts whether a patient will experience hallucinations later in life, demonstrating that structural covariance can capture early developmental trajectories.
- Shared Genetic Control: The physical scaling of specific functional networks is heavily driven by shared genetic influences. As shown in Heritability of Multivariate Gray Matter Measures in Schizophrenia, the authors found that the multivariate gray matter networks extracted via SBM are highly heritable, acting as reliable endophenotypes that reflect genetic vulnerability.
By shifting the neuroimaging paradigm from isolated regional measurements to population-level covariance, individual morphometric similarity, and data-driven source decomposition, these techniques allow researchers to read the structural network organization of the human brain directly from standard structural scans.
Key Foundational Work
Network-level structural covariance in the developing brain (2010) — contributed 3 concept elements to this direction.
Structural Angle and Power Images Reveal Interrelated Gray and White Matter Abnormalities in Schizophrenia (2011) — contributed 1 concept elements to this direction.