Normative modeling is a paradigm shift in neuroimaging: instead of comparing group averages (patients vs. controls), it characterizes individual deviation from a normative range — akin to growth charts for the brain.
-
Normative Models: Review — Marquand et al., 2019
- Key idea: Definitive review establishing normative modeling as a framework. Maps brain features against age/sex norms to produce individual deviation scores (z-scores). Captures heterogeneity within diagnostic categories.
- Breakthrough: Shifted the field from "average patient" to "this specific patient" — each brain is compared against the expected trajectory, not the group mean.
- 🔗 Download
-
Conceptualizing Mental Disorders as Deviations from Normative Functioning — Marquand et al., 2019
- Key idea: Articulated the theoretical foundation — psychiatric disorders are not discrete categories but positions in a continuous space of brain variation. Normative models map this space.
- 🔗 Download
-
Normative Methods — Review/tutorial
- Key idea: Technical overview of normative modeling approaches — Gaussian process regression, hierarchical Bayesian models, and warped Bayesian regression for non-Gaussian features.
⚠️ No verified link available
-
Normative Modelling with Scalable Multi-Task Gaussian Processes — Kia et al., 2018
- Key idea: Scaled normative modeling to large datasets using multi-task GPs that share information across brain regions. Handles the high dimensionality of neuroimaging (100K+ voxels).
- 🔗 Download
-
Brain Morphology Normative Modelling Platform (Brain Monocle) — Little et al., 2025
- Key idea: An open-source platform for computing normative centile scores (like pediatric growth charts) for brain morphological features. Enables clinicians to assess whether an individual brain is developing/aging normally.
- 🔗 Download
-
An Enhanced Conditional VAE-Based Normative Model — Mai Ho, 2025
- Key idea: Used conditional VAEs for normative modeling — conditioning on age/sex produces expected brain patterns; deviations in reconstruction error and latent space indicate abnormality.
- 🔗 Download
-
Normative Modeling on Alzheimer's with Adversarial Autoencoders — Pinaya (Walter) et al., 2020
- Key idea: AAEs provide sharper normative distributions than VAEs. Adversarial training enforces a prior distribution on the latent space more effectively than KL regularization.
- 🔗 Download
-
Deep Normative Modelling for Multi-Modal Neuroimaging — Aguila et al., 2025
- Key idea: Extended normative models to jointly model multiple MRI modalities (structural, diffusion, functional). Multi-modal deviations capture complementary aspects of pathology.
- 🔗 Download
-
Gaussian Process-Based Prediction and Detection of Gray Matter Abnormalities in Alzheimer's — Ziegler et al., 2014
- Key idea: Early normative modeling work using GPs to predict gray matter density from age, then detecting AD as deviations from the predicted norm.
- 🔗 Download
- Growth charts for the brain: Normative models establish what's "normal" at each age/sex, then quantify individual deviation.
- Psychiatric diagnoses are heterogeneous — two patients with the same diagnosis can have opposite brain deviations. Normative models reveal this hidden structure.
- The evolution: GP-based (interpretable, exact uncertainty, limited scale) → deep learning-based (VAEs, AAEs — scalable, multi-modal, less interpretable).
- Centile scores (Brain Monocle) make results clinically interpretable — a brain region in the 2nd centile is clearly abnormal.
-
Normative Model to Map the Heterogeneity in ADHD — Wolfers et al., 2019
- Key idea: Applied normative modeling to ADHD, revealing that the disorder is not a single entity but a collection of distinct neurobiological subtypes. Different patients show deviations in different brain regions.
- Breakthrough: Demonstrated that group-level differences in ADHD virtually disappeared when individual heterogeneity was accounted for — the "average ADHD brain" is a misleading concept.
- 🔗 Download
-
Normative Model to Detect Schizophrenia from Longitudinal MRI — Alexander-Bloch et al., 2014
- Key idea: Used normative brain aging trajectories to detect schizophrenia as abnormal cortical growth patterns targeting normative modules of synchronized development.
- 🔗 Download
-
Computational Approaches for Stratifying Psychiatric Disorders — Marquand et al., 2016
- Key idea: Proposed using computational methods (clustering, normative models) to stratify patients within diagnostic categories rather than comparing diagnostic groups.
- 🔗 Download
-
Pattern Recognition for Neuroimaging-Based Psychiatric Diagnostics — Wolfers et al., 2015
- Key idea: Review of machine learning for psychiatric diagnosis from neuroimaging. Highlighted the problem of small effect sizes and the need for individual-level prediction rather than group statistics.
- 🔗 Download
-
Promises and Pitfalls of Deep Neural Networks in Neuroimaging-Based Psychiatric Research — Eitel et al., 2023
- Key idea: Critical review of DL in psychiatric neuroimaging. Warns about overfitting, confounds (head motion, scanner effects), and the gap between classification accuracy and clinical utility.
- 🔗 Download
-
Deep Learning Frameworks for MRI-Based Diagnosis of Neurological Disorders: Meta Review — Ali et al., 2025
- Key idea: Meta-review of DL for neurological diagnosis — Alzheimer's, Parkinson's, MS, epilepsy. Identifies common methodological weaknesses: small datasets, lack of external validation, information leakage.
- 🔗 Download
-
Deep Learning for Classification of Psychiatric Disorders Using Neuroimaging Data: Meta Review — Quaak et al., 2021
- Key idea: Systematic meta-review showing that reported classification accuracies for psychiatric disorders are often inflated due to methodological issues (leakage, small samples, lack of replication).
- 🔗 Download
- Heterogeneity is the rule: Within any psychiatric diagnosis, patients show vastly different brain profiles. This explains why group-average comparisons yield small effect sizes.
- Brain age gap (chronological age minus predicted brain age) is a powerful biomarker — accelerated aging appears in schizophrenia, AD, and other disorders.
- Methodological rigor is critical — many published DL results in psychiatry don't replicate due to data leakage, small samples, or confounds.
-
Universality in Human Cortical Folding in Health and Disease — Wang et al., 2016
- Key idea: Discovered universal scaling laws in cortical folding — the relationship between cortical thickness, surface area, and folding degree follows power laws that are conserved across species and disrupted in disease.
- Breakthrough: Revealed fundamental physical constraints on brain morphology — cortical folding obeys the same mathematical laws across mammals.
- 🔗 Download
-
Neuro-Evolutionary Evidence for a Universal Fractal Primate Brain Shape — Wang et al., 2024
- Key idea: Showed that primate brains have a conserved fractal geometry across evolution — the fractal dimension of cortical surfaces is remarkably stable despite vast differences in brain size.
- 🔗 Download
-
Cortical Complexity Estimation Using Fractal Dimension: A Systematic Review — Meregalli et al., 2022
- Key idea: Reviews methods for computing fractal dimension of cortical surfaces — box-counting, spherical harmonic analysis, and cortical complexity indices. Shows fractal dimension changes in aging and neurodegenerative disease.
- 🔗 Download
-
Local Cortical Surface Complexity Maps from Spherical Harmonic Reconstructions — Yotter et al., 2011
- Key idea: Proposed computing vertex-wise cortical complexity using spherical harmonic reconstructions, enabling spatial maps of local folding complexity.
- 🔗 Download
-
Independent Components of Human Brain Morphology — Wang, Little et al., 2021
- Key idea: Applied ICA to brain morphological features (thickness, area, volume, curvature), discovering independent modes of variation that are shared across populations and relate to genetic and clinical factors.
- 🔗 Download
-
Multiscale Cortical Morphometry: Regional and Scale-Dependent Variations Across the Lifespan — Leiberg & Little, 2024
- Key idea: Analyzed cortical morphometry at multiple spatial scales, showing that fine-scale morphological features (local gyrification) change differently with age than coarse-scale features (regional volume).
- 🔗 Download
- Fractal geometry of the cortex reflects the physical mechanics of cortical folding — the brain literally folds according to mathematical scaling laws.
- Cortical complexity (fractal dimension, gyrification index) decreases with aging and is altered in neuropsychiatric disorders.
- Multi-scale analysis reveals that different spatial scales of brain morphology are affected by different biological processes.
- The universality findings (Wang et al.) suggest fundamental biophysical constraints that transcend species — a remarkable convergence of neuroscience and physics.
← Previous: Brain Segmentation | → Next: Neurodegenerative Disease