Pith. sign in

REVIEW 5 major objections 5 minor 67 references

GANORM: Lifespan Normative Modeling of EEG Network Topology based on Multinational Cross-Spectra

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read This paper claims to build an age-dependent normative trajectory of the whole-brain EEG network across the lifespan, using multinational cross-spectral data, and to show that healthy individuals deviate less from it than patients with…

desk verdict A genuinely new EEG lifespan resource wrapped around a validation gap: the headline R² measures reconstruction, not normative prediction, and the MAE/RMSE pair is impossible. read the letter →

arxiv 2506.02566 v1 pith:DMBRVUW2 submitted 2025-06-03 cs.NE

classification cs.NE
keywords normativemodelingEEGfunctionalconnectivitycoherencelifespanGAMLSSencoder-decoderdeviationscore
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to establish that the human EEG functional connectome has a describable, age-dependent normal form: a lifespan trajectory of whole-brain network topology that can be computed from cross-spectral data alone. Drawing on 1,966 subjects aged 5-97 from 9 countries, it uses GAMLSS to chart seven network characteristics across frequency bands, then feeds age and the 50th-percentile network characteristics into a backpropagation network to generate a complete normative coherence matrix for any age. The paper reports the model's test accuracy of $R^2=0.796$ and that deviation scores of healthy controls are smaller than those of disease groups across independent sites. If correct, this would supply a standardized EEG-based reference for individual-level deviation analysis in neuropsychiatric disorders, complementing MRI-based brain charts.

What carries the argument

GANORM is an interpretable encoder-decoder architecture. The encoder performs average reference transformation, global scale-factor correction, coherence computation from EEG cross-spectra, extraction of seven network characteristics, and GAMLSS percentile modeling of those characteristics across age. The decoder is a five-layer backpropagation network with 8 inputs (age plus the seven characteristics), 171 outputs (the upper triangle of the coherence matrix), and hidden layers sized 1368, 325, and 171, chosen from the correlation structure between inputs and outputs. At inference, only chronological age is required: the GAMLSS 50th-percentile characteristics are looked up and fed to the trained network to generate a normative whole-brain network for that age.

What would settle it

Take a held-out cohort with known age and EEG cross-spectra, generate the GANORM normative network using only age and the GAMLSS 50th-percentile network characteristics, and compare the predicted 171 coherence edges against the actual coherence edges of each subject; if the match is no better than a constant-age baseline or degrades markedly away from the training distribution, the claimed lifespan norm fails.

Watch

Extended reading notes

Core claim

The central claim is that GANORM provides the first EEG-based normative model of the complete brain network across the lifespan. In this framework, the encoder is not a learned black box but a pipeline of established EEG processing steps: average referencing, global scale factor correction, coherence-based functional connectivity, and graph-theoretic network characteristics; the decoder is a five-layer backpropagation network that maps age plus seven network characteristics to the 171 upper-triangle coherence values of the 19-electrode network. The age-dependent normative input is obtained by GAMLSS modeling, with the 50th percentile used as the normative network characteristics. The authors report that the resulting networks show the expected developmental arc, with weak connectivity in childhood, strengthening through adolescence, peak in adulthood, and decline in old age, with the alpha band showing the most pronounced changes, and that cross-site validation with healthy and disease cohorts yields significantly smaller deviations for healthy subjects.

Load-bearing premise

The load-bearing premise is that a backpropagation network trained on each subject's own network characteristics can be fed the population median characteristics at inference and still produce a valid normative brain network, a substitution the paper does not directly validate.

Editorial extensions

If this is right

  • Clinicians could compute an individual's deviation from the age-appropriate EEG network norm and use it as a quantitative marker for early detection of brain dysfunction.
  • The alpha-band normative curves give a specific, age-adjusted reference for when network integration peaks and declines, which could anchor studies of cognitive aging.
  • Because the norm is derived from cross-spectra rather than raw EEG, it can be applied to archived or federated datasets that share only spectral summaries.
  • The encoder-decoder formulation turns whole-brain network regression into a tractable generative problem, so the same design could be reused for other parcellations or montages.
  • Cross-site separation between healthy and disease groups suggests the deviation score could stratify patients by severity, not just classify them.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves untested whether substituting the population median network characteristics for a subject's own characteristics at inference preserves the input-output relationship the network learned; a direct check on held-out subjects would settle this.
  • Because the data are cross-sectional, the reported trajectory describes age-group differences, not individual development; longitudinal data could plausibly show shallower or more heterogeneous age effects.
  • The same GANORM structure should transfer to MEG or high-density EEG if cross-spectra are harmonized, offering a modality-comparable functional norm.
  • A stronger clinical test would separate disease subtypes and control for age, sex, and medication, since the current deviation comparison pools several neurodegenerative conditions.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The manuscript proposes GANORM, a normative model for whole-brain EEG functional connectivity across the lifespan. Using the multinational MNCS cross-spectral dataset (1966 subjects, ages 5–97), the authors first fit GAMLSS trajectories to seven network characteristics (NCs) in four frequency bands, then train a five-layer BPNN to map age and NCs to the 171 upper-triangular coherence values. Normative brain networks at each age are generated by feeding the BPNN the chronological age together with the GAMLSS 50th-percentile NCs. The paper claims an age-dependent normative trajectory of the complete brain network, supported by a BPNN test R² of 0.796, and reports cross-site clinical validation in which deviation scores for healthy controls are smaller than those for disease groups.

Significance. The ambition of this work is worthwhile: a lifespan EEG-based functional connectome norm built from a large, multinational, decentralized dataset would fill a genuine gap relative to MRI-centric normative modeling. The use of GAMLSS to chart NC trajectories, the interpretable encoder–decoder framing, the open code, and the cross-site clinical application are constructive contributions. However, the central quantitative claims are not internally consistent, and the reported test metric does not actually evaluate the normative generation procedure. As presented, the evidence does not support the abstract's claim of a validated whole-brain EEG network norm.

major comments (5)
  1. [Section 3.2, Table 1] The reported MAE and RMSE values are mutually inconsistent. For any set of residuals, RMSE is always greater than or equal to MAE, yet Table 1 reports MAE = 0.081 and RMSE = 0.013 for the test set of Model (and similarly impossible pairs for other rows). This indicates an error in units, computation, or reporting and invalidates the numerical evaluation cited in the Abstract (R² 0.796, MAE 0.081, RMSE 0.013).
  2. [Sections 2.5.2, 2.5.3, and Figure 2] The BPNN is trained and tested on inputs consisting of age plus the subject's actual NCs, with the output being that same subject's coherence upper triangle. At inference the actual NCs are replaced by GAMLSS 50th-percentile values, but no experiment feeds age plus normative NCs to the trained model and compares the output against held-out subjects. Because the GAMLSS curves are fit marginally for each NC, the vector of individually median NCs may lie off the manifold of feasible (age, NC) combinations, especially in sparse or correlated age regions; the paper provides no evidence that the BPNN behaves correctly on such inputs. The test R² = 0.796 therefore does not support the claim of a validated normative brain-network trajectory.
  3. [Sections 2.4 and 2.5] The input NCs are computed from the same coherence matrix that serves as the output target. A network trained to invert NCs back into the upper triangle can achieve high R² even without learning a genuine age-dependent mapping, because the input is a set of summary statistics of the output itself. This circularity is a concern for the interpretation of the reported R². Figure 6, which recomputes NCs from the generated normative matrices and compares them with the input NCs, is a self-consistency check rather than an independent validation, and it cannot detect systematic biases that preserve the input NCs while producing non-normative topologies.
  4. [Section 3.4 and Figure 7] The clinical validation claims that healthy groups have significantly smaller deviations than disease groups, but the paper reports only point estimates of deviations and provides no statistical test, p-value, confidence interval, or effect size. Without such statistics, the claim of significance is unsupported. In addition, Eq. (6) defines the deviation as a mean absolute difference across all entries of the FC matrix, which is not the same as a 'mean functional connection strength' deviation, and the relationship between this quantity and the normative model's predictive uncertainty is not discussed.
  5. [Section 3.2 and Table 1] The ablation conclusion is contradicted by the reported numbers: Model has test R² = 0.796, while Model/2 has test R² = 0.801, so Model is not the best-performing model on the test split. Moreover, the differences among the five configurations are small relative to the reported standard deviations. The claim that the chosen hidden-layer sizes are confirmed as rational by the ablation is not supported by the table.
minor comments (5)
  1. [Section 3.2] Typo: 'Notly' should be 'Notably'.
  2. [Section 2.5.1] 'GL' appears to be a typo for 'GE' (global efficiency); please correct.
  3. [Eq. (3)] The R² formula uses y_i for both the true values and the mean; this is notationally confusing and should use an overbar, e.g., \bar{y}.
  4. [Sections 2.1 and 2.5.1] The sample size is 1966 in Section 2.1 but 1945 in Section 2.5.1; the discrepancy should be explained.
  5. [Figure 7C] The label 'nomative' is a typo for 'normative'; please also ensure the color scheme is legible when the figure is printed in grayscale.

Circularity Check

2 steps flagged · score 6.0 of 10

The whole-brain network norm is validated by a reconstruction metric and a self-consistency check; the reported R2 does not test the age-plus-normative-NC generation pathway.

  1. fitted input called prediction [Section 2.5.2, BPNN module; claimed evidence in Abstract and Section 3.2/Table 1]
    "Our objective is to predict the entire adjacency matrix of the FC using the chronological age of a subject and the age-dependent normative mean NCs. That is, only the chronological age of the subject is required, and the other inputs are the normative mean NCs picked from the NCs trajectories in Section2.4."

    The generated 'normative brain network' is, by construction, the BPNN decode of the GAMLSS 50th-percentile NC vector for that age. The performance advertised as validating this model (Abstract: 'tested performances of BPNN showed the R^2 was 0.796'; Table 1) comes from five-fold cross-validation in which the BPNN input is the subject's actual NCs and the output is that same subject's coherence upper triangle (Section 2.5). No experiment feeds only age plus the normative NCs and compares the output to held-out subjects. The claimed whole-network trajectory therefore reduces to a decoding of the marginal GAMLSS NC curves, and the headline R^2 does not test the deployment pathway.

  2. self definitional [Section 3.3, Evolutionary patterns of brain network, Figure 6]
    "After obtaining the brain network norms, we calculated the NCs of the norm and compared them with the NCs inputted into the BPNN. The results indicate that the NCs of the norms closely match those at the 50th percentile, demonstrating that the norms we constructed are in line with reality (Fig6)."

    The sole internal network-level validation checks that the generated network reproduces the 50th-percentile NCs that were fed into the decoder. Because the BPNN is trained to map NC vectors to coherence matrices with approximately those NCs, this agreement is a self-consistency check. The 'reality' against which the norms are judged is the GAMLSS input itself, so the validation is satisfied by construction rather than by independent comparison with observed whole-brain networks.

full rationale

The two flagged steps are load-bearing because they are the only internal supports for the paper's central claim of having established a validated lifespan normative trajectory of the complete brain network. The GAMLSS curves of the seven NCs (Section 2.4) are themselves a legitimate univariate norm fitted to a large multinational sample, and the cross-site healthy-versus-disease deviation comparison (Section 3.4) is a genuine external test that gives the norm some independent clinical content. That prevents a score of 8-10. However, because the whole-network trajectory is generated by decoding the same GAMLSS 50th-percentile NCs and is validated only by a reconstruction R^2 and by self-consistency with those inputs, the central 'complete brain network' claim is partially circular; score 6. Separately, Table 1's MAE=0.081 exceeding RMSE=0.013 is arithmetically impossible, a correctness flag unrelated to circularity.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claim rests on a chain of assumptions about data harmonization, connectivity estimation, distributional modeling, and a train/inference distribution shift. The BPNN carries a large fitted parameter inventory whose architecture is tuned on the same cohort used for evaluation, and no invented physical entities are introduced.

free parameters (4)
  • GAMLSS distribution parameters (mu, sigma, nu, tau) for each network characteristic and frequency band = not reported numerically
    These parameters are fit to MNCS data to define the normative percentile curves; the paper gives no fitted values or uncertainty.
  • BPNN hidden layer sizes (1368, 325, 171) = 1368, 325, 171
    Layer sizes are chosen from Pearson correlations between inputs and outputs computed on the same 1,945 subjects later used for cross-validation, so they are tuned to the evaluation data.
  • BPNN weights and biases = trained on MNCS
    All network parameters are fit to the same cohort from which normative outputs are generated; no external pretraining is used.
  • Embedding correlation threshold (r>0.6, p<0.05) = 0.6 and 0.05
    The 325-neuron hidden layer count derives from this threshold, which is selected by the authors and not justified independently.
assumptions (6)
  • domain assumption Multinational EEG cross-spectra are comparable after average reference and global-scale factor correction.
    Invoked in Section 2.2 (AVE and GSF) to merge 14 sites; residual site effects are not tested.
  • domain assumption Coherence is a valid estimator of functional connectivity from cross-spectra.
    Used in Section 2.3 to define the adjacency matrix; coherence depends on stationarity and the available CS and is a standard but assumption-laden measure.
  • domain assumption GAMLSS with Box-Cox-t, cubic smoothing splines, and log(age) adequately describes the age-dependent distribution of every NC.
    Stated in Section 2.4; no model comparison, residual diagnostics, or goodness-of-fit are reported for the chosen family.
  • domain assumption The MNCS sample is a valid healthy reference population despite being skewed toward younger subjects and relying on no stated health-screening criteria.
    The normative curves inherit this representativeness assumption; the paper reports demographic skew but no inclusion or exclusion criteria.
  • ad hoc to paper A BPNN trained on subject-specific NCs can be applied to normative NCs to produce valid normative networks.
    This substitution is the core of GANORM's inference (Section 2.5.2) and is not validated directly.
  • domain assumption Cross-sectional data can be used to infer lifespan trajectories.
    Acknowledged as a limitation in Section 4; the normative curves and deviations are built on cross-sectional observations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GANORM: Lifespan Normative Modeling of EEG Network Topology based on Multinational Cross-Spectra." pith.science (2026). https://pith.science/paper/DMBRVUW2

@misc{pith2026250602566,
  author       = {Pith},
  title        = {Pith review of: GANORM: Lifespan Normative Modeling of EEG Network Topology based on Multinational Cross-Spectra},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DMBRVUW2}},
  note         = {Machine review of arXiv:2506.02566}
}
read the original abstract

Charting the lifespan evolutionary trajectory of brain function serves as the normative standard for preventing mental disorders during brain development and aging. Although numerous MRI studies have mapped the structural connectome for young cohorts, the EEG-based functional connectome is unknown to characterize human lifespan, limiting its practical applications for the early detection of brain dysfunctions at the community level. This work aimed to undertake normative modeling from the perspective of EEG network topology. Frequency-dependent scalp EEG functional networks were constructed based on EEG cross-spectra aged 5-97 years from 9 countries and network characteristics were quantified. First, GAMLSS were applied to describe the normative curves of the network characteristics in different frequency bands. Subsequently, addressing the limitations of existing regression approaches for whole brain network analysis, this paper proposed an interpretable encoder-decoder framework, Generative Age-dependent brain Network nORmative Model (GANORM). Building upon this framework, we established an age-dependent normative trajectory of the complete brain network for the entire lifespan. Finally, we validated the effectiveness of the norm using EEG datasets from multiple sites. Subsequently, we evaluated the effectiveness of GANORM, and the tested performances of BPNN showed the R^2 was 0.796, the MAE was 0.081, and the RMSE was 0.013. Following established lifespan brain network norm, GANORM also exhibited good results upon verification using healthy and disease data from various sites. The deviation scores from the normative mean for the healthy control group were significantly smaller than those of the disease group.

Figures

Figures reproduced from arXiv: 2506.02566 by the authors.

Figure 1
Figure 1. Age distribution of MNCS dataset. appearance, they may differ significantly in overall amplitude in the actual EEG recording, resulted from different amplifiers, recording conditions, and devices. This interference was solved by dividing the CS by the stochastic global scale factor described by Hernandez et al.[50]. To analyze the relation￾ship between signals in different frequency bands and brain activity states, … view at source ↗
Figure 2
Figure 2. Generative Age-dependent brain Network nORmative Model (GANORM). The input is EEG cross-spectra with age. Based on neuroscientific meth￾ods, the encoder performs a series of processing on EEG, including average reference, global factor correction, construction of coherence-based scalp EEG network, calculation of NCs, and GAMLSS modeling of NCs. Embedding is age and normative NCs (50% quantile after GAMLSS modeling).… view at source ↗
Figure 3
Figure 3. Design of the backpropagation neural network (BPNN). (A)Input￾output Correlation analysis. Among the 8 input features × 171 output pairwise FC strengths, 325 show a correlation >0.6 and p-values <0.05. (B) BPNN diagram. A five-layer BP neural network with an input layer of 8 neurons, three hidden layers with 1368, 325, and 171 neurons respectively, and an output layer of 171 neurons. 11 [PITH_FULL_IMAGE:figures/ful… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Trajectory of network characteristics. A, B, and C represent functional in￾tegration, functional segregation and centrality, respectively. Different rows correspond to different network characteristics, while different columns correspond to different frequency bands. C…
Figure 5
Figure 5. Figure 5: Lifespan evolutionary patterns of scalp EEG network topology in α band. Normative network topologies from ages 5 to 85 are depicted, with brain networks for selected ages shown in the figure. The age interval is 3 years before age 17 and 10 years thereafter, with a net…
Figure 6
Figure 6. Figure 6: Comparison of generated normative NCs and actual brain NCs. It illustrates the comparison of seven NCs between the generated norms and the actual network, with the actual depicted in red and the normative in blue. values. The results indicated that our model exhibited …
Figure 7
Figure 7. Figure 7: Cross-site deviation from GANORM. (A) Average absolute deviations for three groups of healthy data (MN-CS, HC, HY-AHU) and four groups of disease data (AD, bvFTD, PD, MS). The horizontal axis represents deviation values, while the vertical axis represents probability d…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

67 extracted references · 67 canonical work pages

  1. [1]

    Mappingthe heterogeneous brain structural phenotype of autism spectrum disorder using the normative model.Biological Psychiatry, 91(11):967–976, 2022

    Xiaolong Shan, Lucina Q Uddin, Jinming Xiao, Changchun He, Zihan Ling, LeiLi, XinyueHuang, HuafuChen, andXujunDuan. Mappingthe heterogeneous brain structural phenotype of autism spectrum disorder using the normative model.Biological Psychiatry, 91(11):967–976, 2022

  2. [2]

    Brain charts for the human lifespan.Nature, 604(7906):525–533, 2022

    Richard AI Bethlehem, Jakob Seidlitz, Simon R White, Jacob W Vogel, Kevin M Anderson, Chris Adamson, Sophie Adler, George S Alexopou- los, Evdokia Anagnostou, Ariosky Areces-Gonzalez, et al. Brain charts for the human lifespan.Nature, 604(7906):525–533, 2022

  3. [3]

    Normative models for neuroimaging markers: Impact of model selection, sample size and evaluation criteria.NeuroImage, 268:119864, 2023

    Jelena Bozek, Ludovica Griffanti, Stephan Lau, and Mark Jenkinson. Normative models for neuroimaging markers: Impact of model selection, sample size and evaluation criteria.NeuroImage, 268:119864, 2023

  4. [4]

    Evidence for embracing normative modeling.Elife, 12:e85082, 2023

    Saige Rutherford, Pieter Barkema, Ivy F Tso, Chandra Sripada, Chris- tian F Beckmann, Henricus G Ruhe, and Andre F Marquand. Evidence for embracing normative modeling.Elife, 12:e85082, 2023

  5. [5]

    Predictive statistical model of early cranial development

    Antonio R Porras, Robert F Keating, Janice S Lee, and Marius George Linguraru. Predictive statistical model of early cranial development. IEEE Transactions on Biomedical Engineering, 69(2):537–546, 2021

  6. [6]

    Conceptualizing mental disorders as deviations from normative functioning.Molecular psychiatry, 24(10):1415–1424, 2019

    Andre F Marquand, Seyed Mostafa Kia, Mariam Zabihi, Thomas Wolfers, Jan K Buitelaar, and Christian F Beckmann. Conceptualizing mental disorders as deviations from normative functioning.Molecular psychiatry, 24(10):1415–1424, 2019

  7. [7]

    Heterogeneity in brain microstructural development following preterm birth.Cerebral Cortex, 30(9):4800–4810, 2020

    Ralica Dimitrova, Maximilian Pietsch, Daan Christiaens, Judit Ciar- rusta, Thomas Wolfers, Dafnis Batalle, Emer Hughes, Jana Hutter, Lu- cilio Cordero-Grande, Anthony N Price, et al. Heterogeneity in brain microstructural development following preterm birth.Cerebral Cortex, 30(9):4800–4810, 2020

  8. [8]

    A norma- tive modelling approach reveals age-atypical cortical thickness in a sub- group of males with autism spectrum disorder.Communications biology, 3(1):486, 2020

    Richard AI Bethlehem, Jakob Seidlitz, Rafael Romero-Garcia, Stavros Trakoshis, Guillaume Dumas, and Michael V Lombardo. A norma- tive modelling approach reveals age-atypical cortical thickness in a sub- group of males with autism spectrum disorder.Communications biology, 3(1):486, 2020. 23

Show all 67 references
  1. [9]

    Mapping the heterogeneous phenotype of schizophrenia and bipolar disorder using normative models.JAMA psychiatry, 75(11):1146–1155, 2018

    Thomas Wolfers, Nhat Trung Doan, Tobias Kaufmann, Dag Alnæs, Torgeir Moberget, Ingrid Agartz, Jan K Buitelaar, Torill Ueland, Ingrid Melle, Barbara Franke, et al. Mapping the heterogeneous phenotype of schizophrenia and bipolar disorder using normative models.JAMA psychiatry, ...

  2. [10]

    Normative modeling with the predictive clinical neuroscience toolkit (pcntoolkit)

    Saige Rutherford and Andre F Marquand. Normative modeling with the predictive clinical neuroscience toolkit (pcntoolkit). InMethods for Analyzing Large Neuroimaging Datasets, pages 329–364. Springer US New York, NY, 2024

  3. [11]

    Mariam Zabihi, Marianne Oldehinkel, Thomas Wolfers, Vincent Frouin, David Goyard, Eva Loth, Tony Charman, Julian Tillmann, Tobias Ba- naschewski, Guillaume Dumas, et al. Dissecting the heterogeneous cor- tical anatomy of autism spectrum disorder using normative models.Bi- olog...

  4. [12]

    Individual differ- ences v

    Thomas Wolfers, Christian F Beckmann, Martine Hoogman, Jan K Buitelaar, Barbara Franke, and Andre F Marquand. Individual differ- ences v. the average patient: mapping the heterogeneity in adhd using normative models.Psychological medicine, 50(2):314–323, 2020

  5. [13]

    Neurocognitive growth charting in psychosis spectrum youths.JAMA psychiatry, 71(4):366–374, 2014

    Ruben C Gur, Monica E Calkins, Theodore D Satterthwaite, Kosha Ruparel, Warren B Bilker, Tyler M Moore, Adam P Savitt, Hakon Hakonarson, and Raquel E Gur. Neurocognitive growth charting in psychosis spectrum youths.JAMA psychiatry, 71(4):366–374, 2014

  6. [14]

    Growth chart- ing of brain connectivity networks and the identification of attention impairment in youth.JAMA psychiatry, 73(5):481–489, 2016

    Daniel Kessler, Michael Angstadt, and Chandra Sripada. Growth chart- ing of brain connectivity networks and the identification of attention impairment in youth.JAMA psychiatry, 73(5):481–489, 2016

  7. [15]

    Imaging patterns of brain development and their relationship to cogni- tion.Cerebral cortex, 25(6):1676–1684, 2015

    Guray Erus, Harsha Battapady, Theodore D Satterthwaite, Hakon Hakonarson, Raquel E Gur, Christos Davatzikos, and Ruben C Gur. Imaging patterns of brain development and their relationship to cogni- tion.Cerebral cortex, 25(6):1676–1684, 2015

  8. [16]

    Closing the life-cycle 24 of normative modeling using federated hierarchical bayesian regression

    Seyed Mostafa Kia, Hester Huijsdens, Saige Rutherford, Augustijn de Boer, Richard Dinga, Thomas Wolfers, Pierre Berthet, Maarten Mennes, Ole A Andreassen, Lars T Westlye, et al. Closing the life-cycle 24 of normative modeling using federated hierarchical bayesian regression. P...

  9. [17]

    Charting brain growth and aging at high spatial precision.elife, 11:e72904, 2022

    Saige Rutherford, Charlotte Fraza, Richard Dinga, Seyed Mostafa Kia, Thomas Wolfers, Mariam Zabihi, Pierre Berthet, Amanda Worker, Ser- ena Verdi, Derek Andrews, et al. Charting brain growth and aging at high spatial precision.elife, 11:e72904, 2022

  10. [18]

    Shuting Sun, Xiaowei Li, Jing Zhu, Ying Wang, Rong La, Xuemin Zhang, Liuqing Wei, and Bin Hu. Graph theory analysis of functional connectivity in major depression disorder with high-density resting state eeg data.IEEE Transactions on Neural Systems and Rehabilitation En- ginee...

  11. [19]

    Topological network anal- ysis of early alzheimer’s disease based on resting-state eeg.IEEE Trans- actions on Neural Systems and Rehabilitation Engineering, 28(10):2164– 2172, 2020

    Feng Duan, Zihao Huang, Zhe Sun, Yu Zhang, Qibin Zhao, Andrzej Ci- chocki, Zhenglu Yang, and Jordi Sole-Casals. Topological network anal- ysis of early alzheimer’s disease based on resting-state eeg.IEEE Trans- actions on Neural Systems and Rehabilitation Engineering, 28(10):2...

  12. [20]

    Alberto Nogales, Álvaro J García-Tejedor, Pedro Chazarra, and Ar- turo Ugalde-Canitrot. Discriminating and understanding brain states in children with epileptic spasms using deep learning and graph met- rics analysis of brain connectivity.Computer Methods and Programs in Biome...

  13. [21]

    The human connectome project: a retrospective.NeuroImage, 244:118543, 2021

    Jennifer Stine Elam, Matthew F Glasser, Michael P Harms, Stamatios N Sotiropoulos, Jesper LR Andersson, Gregory C Burgess, Sandra W Cur- tiss, Robert Oostenveld, Linda J Larson-Prior, Jan-Mathijs Schoffelen, et al. The human connectome project: a retrospective.NeuroImage, 244:...

  14. [22]

    Mapping neurophysiological subtypes of major depressive disorder using normative models of the functional connectome.Biological Psychiatry, 94(12):936–947, 2023

    Xiaoyi Sun, Jinrong Sun, Xiaowen Lu, Qiangli Dong, Liang Zhang, Wenxu Wang, Jin Liu, Qing Ma, Xiaoqin Wang, Dongtao Wei, et al. Mapping neurophysiological subtypes of major depressive disorder using normative models of the functional connectome.Biological Psychiatry, 94(12):93...

  15. [23]

    Abdulyekeen T Adebisi, Ho-Won Lee, and Kalyana C Veluvolu. Eeg- based brain functional network analysis for differential identification of 25 dementia-related disorders and their onset.IEEE Transactions on Neu- ral Systems and Rehabilitation Engineering, 2024

  16. [24]

    Eeg sleep stages iden- tification based on weighted undirected complex networks.Computer methods and programs in biomedicine, 184:105116, 2020

    Mohammed Diykh, Yan Li, and Shahab Abdulla. Eeg sleep stages iden- tification based on weighted undirected complex networks.Computer methods and programs in biomedicine, 184:105116, 2020

  17. [25]

    Diagnosis of autism spectrum disorder based on complex network features.Computer methods and programs in biomedicine, 177:277–283, 2019

    Ghasem Sadeghi Bajestani, Mahboobe Behrooz, Adel Ghazi Khani, Mostafa Nouri-Baygi, and Ali Mollaei. Diagnosis of autism spectrum disorder based on complex network features.Computer methods and programs in biomedicine, 177:277–283, 2019

  18. [26]

    Human connectomics across the life span.Trends in cognitive sciences, 21(1):32–45, 2017

    Xi-Nian Zuo, Ye He, Richard F Betzel, Stan Colcombe, Olaf Sporns, and Michael P Milham. Human connectomics across the life span.Trends in cognitive sciences, 21(1):32–45, 2017

  19. [27]

    The development of brain network hubs.Developmental cognitive neuroscience, 36:100607, 2019

    Stuart Oldham and Alex Fornito. The development of brain network hubs.Developmental cognitive neuroscience, 36:100607, 2019

  20. [28]

    Development of large-scale func- tional networks from birth to adulthood: A guide to the neuroimaging literature.Neuroimage, 160:15–31, 2017

    David S Grayson and Damien A Fair. Development of large-scale func- tional networks from birth to adulthood: A guide to the neuroimaging literature.Neuroimage, 160:15–31, 2017

  21. [29]

    Ap- plication of graph theory for identifying connectivity patterns in human brain networks: a systematic review.frontiers in Neuroscience, 13:585, 2019

    Farzad V Farahani, Waldemar Karwowski, and Nichole R Lighthall. Ap- plication of graph theory for identifying connectivity patterns in human brain networks: a systematic review.frontiers in Neuroscience, 13:585, 2019

  22. [30]

    A cross-disorder connec- tome landscape of brain dysconnectivity.Nature reviews neuroscience, 20(7):435–446, 2019

    Martijn P van den Heuvel and Olaf Sporns. A cross-disorder connec- tome landscape of brain dysconnectivity.Nature reviews neuroscience, 20(7):435–446, 2019

  23. [31]

    Imaging struc- tural and functional brain development in early childhood.Nature Re- views Neuroscience, 19(3):123–137, 2018

    John H Gilmore, Rebecca C Knickmeyer, and Wei Gao. Imaging struc- tural and functional brain development in early childhood.Nature Re- views Neuroscience, 19(3):123–137, 2018

  24. [32]

    Developmental changes in organization of struc- tural brain networks.Cerebral Cortex, 23(9):2072–2085, 2013

    Budhachandra S Khundrakpam, Andrew Reid, Jens Brauer, Felix Car- bonell, John Lewis, Stephanie Ameis, Sherif Karama, Junki Lee, Zhang Chen, Samir Das, et al. Developmental changes in organization of struc- tural brain networks.Cerebral Cortex, 23(9):2072–2085, 2013. 26

  25. [33]

    Effects of aging on functional and structural brain connectivity.Neuroimage, 160:32–40, 2017

    Jessica S Damoiseaux. Effects of aging on functional and structural brain connectivity.Neuroimage, 160:32–40, 2017

  26. [34]

    Topological organization of the human brain functional connectome across the lifespan.Developmental cognitive neuroscience, 7:76–93, 2014

    Miao Cao, Jin-Hui Wang, Zheng-Jia Dai, Xiao-Yan Cao, Li-Li Jiang, Feng-Mei Fan, Xiao-Wei Song, Ming-Rui Xia, Ni Shu, Qi Dong, et al. Topological organization of the human brain functional connectome across the lifespan.Developmental cognitive neuroscience, 7:76–93, 2014

  27. [35]

    Hamish A Deery, Robert Di Paolo, Chris Moran, Gary F Egan, and Sharna D Jamadar. The older adult brain is less modular, more in- tegrated, and less efficient at rest: A systematic review of large-scale resting-state functional brain networks in aging.Psychophysiology, 60(1):e1...

  28. [36]

    Combining magnetoencephalography with magnetic resonance imaging enhances learning of surrogate-biomarkers.Elife, 9:e54055, 2020

    Denis A Engemann, Oleh Kozynets, David Sabbagh, Guillaume Lemaître, Gael Varoquaux, Franziskus Liem, and Alexandre Gramfort. Combining magnetoencephalography with magnetic resonance imaging enhances learning of surrogate-biomarkers.Elife, 9:e54055, 2020

  29. [37]

    Dcnet: A self-supervised eeg classification framework for improving cog- nitive computing-enabled smart healthcare.IEEE journal of biomedical and health informatics, 2024

    Yiyang Zhang, Le Sun, Deepak Gupta, Xin Ning, and Prayag Tiwari. Dcnet: A self-supervised eeg classification framework for improving cog- nitive computing-enabled smart healthcare.IEEE journal of biomedical and health informatics, 2024

  30. [38]

    Learn- ing interpretable brain functional connectivity via self-supervised triplet network with depth-wise attention.IEEE Journal of Biomedical and Health Informatics, 2024

    Yunbo Tang, Weirong Huang, Rongchang Liu, and Yuanlong Yu. Learn- ing interpretable brain functional connectivity via self-supervised triplet network with depth-wise attention.IEEE Journal of Biomedical and Health Informatics, 2024

  31. [39]

    Abdul Qayyum, Imran Razzak, M Tanveer, Moona Mazher, and Ban- dar Alhaqbani. High-density electroencephalography and speech sig- nal based deep framework for clinical depression diagnosis.IEEE/ACM Transactions on Computational Biology and Bioinformatics, 20(4):2587– 2597, 2023

  32. [40]

    Modern views of machine learning for precision psy- chiatry.Patterns, 3(11), 2022

    Zhe Sage Chen, Isaac R Galatzer-Levy, Benedetta Bigio, Carla Nasca, Yu Zhang, et al. Modern views of machine learning for precision psy- chiatry.Patterns, 3(11), 2022. 27

  33. [41]

    Unsupervised seizure identification on eeg.Computer methods and pro- grams in biomedicine, 215:106604, 2022

    İlkay Yıldız, Rachael Garner, Matthew Lai, and Dominique Duncan. Unsupervised seizure identification on eeg.Computer methods and pro- grams in biomedicine, 215:106604, 2022

  34. [42]

    Segnet: A deep convolutional encoder-decoder architecture for image segmenta- tion.IEEE transactions on pattern analysis and machine intelligence, 39(12):2481–2495, 2017

    Vijay Badrinarayanan, Alex Kendall, and Roberto Cipolla. Segnet: A deep convolutional encoder-decoder architecture for image segmenta- tion.IEEE transactions on pattern analysis and machine intelligence, 39(12):2481–2495, 2017

  35. [43]

    Cnn-based encoder-decoder networks for salient object detection: A comprehen- sive review and recent advances.Information Sciences, 546:835–857, 2021

    Yuzhu Ji, Haijun Zhang, Zhao Zhang, and Ming Liu. Cnn-based encoder-decoder networks for salient object detection: A comprehen- sive review and recent advances.Information Sciences, 546:835–857, 2021

  36. [44]

    Thorir Mar Ingolfsson, Xiaying Wang, Upasana Chakraborty, Simone Benatti, Adriano Bernini, Pauline Ducouret, Philippe Ryvlin, Sandor Beniczky, Luca Benini, and Andrea Cossettini. Brainfusenet: Enhanc- ing wearable seizure detection through eeg-ppg-accelerometer sensor fu- sion...

  37. [45]

    Review of brain encoding and decoding mechanisms for eeg-based brain–computer interface.Cognitive neurodynamics, 15:569–584, 2021

    Lichao Xu, Minpeng Xu, Tzyy-Ping Jung, and Dong Ming. Review of brain encoding and decoding mechanisms for eeg-based brain–computer interface.Cognitive neurodynamics, 15:569–584, 2021

  38. [46]

    Driver drowsiness estimation using eeg signals with a dynamical encoder–decoder modeling framework.Scientific reports, 12(1):2650, 2022

    Sadegh Arefnezhad, James Hamet, Arno Eichberger, Matthias Früh- wirth, Anja Ischebeck, Ioana Victoria Koglbauer, Maximilian Moser, and Ali Yousefi. Driver drowsiness estimation using eeg signals with a dynamical encoder–decoder modeling framework.Scientific reports, 12(1):2650, 2022

  39. [47]

    Normative modeling using multimodal variational autoencoders to identify abnor- mal brain volume deviations in alzheimer’s disease

    Sayantan Kumar, Philip RO Payne, and Aristeidis Sotiras. Normative modeling using multimodal variational autoencoders to identify abnor- mal brain volume deviations in alzheimer’s disease. InProceedings of SPIE–the International Society for Optical Engineering, volume 12465. N...

  40. [48]

    Unsupervised au- tomatic seizure detection for focal-onset seizures recorded with behind- 28 the-ear eeg using an anomaly-detecting generative adversarial network

    Sungmin You, Baek Hwan Cho, Soonhyun Yook, Joo Young Kim, Young-Min Shon, Dae-Won Seo, and In Young Kim. Unsupervised au- tomatic seizure detection for focal-onset seizures recorded with behind- 28 the-ear eeg using an anomaly-detecting generative adversarial network. Computer...

  41. [49]

    The applied principles of eeg analysis methods in neuroscience and clin- ical neurology.Military Medical Research, 10(1):67, 2023

    Hao Zhang, Qing-Qi Zhou, He Chen, Xiao-Qing Hu, Wei-Guang Li, Yang Bai, Jun-Xia Han, Yao Wang, Zhen-Hu Liang, Dan Chen, et al. The applied principles of eeg analysis methods in neuroscience and clin- ical neurology.Military Medical Research, 10(1):67, 2023

  42. [50]

    Harmonized-multinational qeeg norms (harmnqeeg).NeuroImage, 256:119190, 2022

    Min Li, Ying Wang, Carlos Lopez-Naranjo, Shiang Hu, Ronaldo César García Reyes, Deirel Paz-Linares, Ariosky Areces-Gonzalez, Aini Ismafairus Abd Hamid, Alan C Evans, Alexander N Savostyanov, et al. Harmonized-multinational qeeg norms (harmnqeeg).NeuroImage, 256:119190, 2022

  43. [51]

    The statistics of eeg unipolar references: derivations and properties.Brain topography, 32:696–703, 2019

    Shiang Hu, Dezhong Yao, Maria L Bringas-Vega, Yun Qin, and Pedro A Valdes-Sosa. The statistics of eeg unipolar references: derivations and properties.Brain topography, 32:696–703, 2019

  44. [52]

    Which reference should we use for eeg and erp practice?Brain topography, 32:530–549, 2019

    Dezhong Yao, Yun Qin, Shiang Hu, Li Dong, Maria L Bringas Vega, and Pedro A Valdés Sosa. Which reference should we use for eeg and erp practice?Brain topography, 32:530–549, 2019

  45. [53]

    How do reference montage and electrodes setup af- fect the measured scalp eeg potentials?Journal of neural engineering, 15(2):026013, 2018

    Shiang Hu, Yongxiu Lai, Pedro A Valdes-Sosa, Maria L Bringas-Vega, and Dezhong Yao. How do reference montage and electrodes setup af- fect the measured scalp eeg potentials?Journal of neural engineering, 15(2):026013, 2018

  46. [54]

    Unified bayesian es- timator of eeg reference at infinity: rrest (regularized reference electrode standardization technique).Frontiers in Neuroscience, 12:297, 2018

    Shiang Hu, Dezhong Yao, and Pedro A Valdes-Sosa. Unified bayesian es- timator of eeg reference at infinity: rrest (regularized reference electrode standardization technique).Frontiers in Neuroscience, 12:297, 2018

  47. [55]

    Shiang Hu, Zhihao Zhang, Xiaochu Zhang, Xiaopei Wu, and Pedro A Valdes-Sosa.ξ-π: a nonparametric model for neural power spectra de- composition.IEEE Journal of Biomedical and Health Informatics, 2024

  48. [56]

    Guido Nolte, Edgar Galindo-Leon, Zhenghan Li, Xun Liu, and An- dreas K Engel. Mathematical relations between measures of brain con- nectivity estimated from electrophysiological recordings for gaussian dis- tributed data.Frontiers in Neuroscience, 14:577574, 2020. 29

  49. [57]

    Coherence based graph convolution network for motor imagery-induced eeg after spinal cord injury.Frontiers in Neuroscience, 16:1097660, 2023

    Han Li, Ming Liu, Xin Yu, JianQun Zhu, Chongfeng Wang, Xinyi Chen, Chao Feng, Jiancai Leng, Yang Zhang, and Fangzhou Xu. Coherence based graph convolution network for motor imagery-induced eeg after spinal cord injury.Frontiers in Neuroscience, 16:1097660, 2023

  50. [58]

    Complex network measures of brain connectivity: uses and interpretations.Neuroimage, 52(3):1059–1069, 2010

    Mikail Rubinov and Olaf Sporns. Complex network measures of brain connectivity: uses and interpretations.Neuroimage, 52(3):1059–1069, 2010

  51. [59]

    The brainlat project, a multimodal neuroimaging dataset of neurodegen- eration from underrepresented backgrounds.Scientific Data, 10(1):889, 2023

    Pavel Prado, Vicente Medel, Raul Gonzalez-Gomez, Agustín Sainz- Ballesteros, Victor Vidal, Hernando Santamaría-García, Sebastian Moguilner, JhonyMejia, AndreaSlachevsky, MariaIsabelBehrens, etal. The brainlat project, a multimodal neuroimaging dataset of neurodegen- eration fr...

  52. [60]

    Design of a novel biomedical signal processing and analysis tool for functional neuroimaging.Computer Methods and Programs in Biomedicine, 125:46–57, 2016

    Sezgin Kaçar and Ünal Sakoğlu. Design of a novel biomedical signal processing and analysis tool for functional neuroimaging.Computer Methods and Programs in Biomedicine, 125:46–57, 2016

  53. [61]

    Understanding geometry of encoder-decoder cnns

    Jong Chul Ye and Woon Kyoung Sung. Understanding geometry of encoder-decoder cnns. InInternational Conference on Machine Learn- ing, pages 7064–7073. PMLR, 2019

  54. [62]

    The encoder-decoder framework and its applications.Deep learning: Concepts and architectures, pages 133–167, 2020

    Ahmad Asadi and Reza Safabakhsh. The encoder-decoder framework and its applications.Deep learning: Concepts and architectures, pages 133–167, 2020

  55. [63]

    A novel explainable machine learning approach for eeg-based brain-computer interface systems.Neural Computing and Applications, 34(14):11347–11360, 2022

    Cosimo Ieracitano, Nadia Mammone, Amir Hussain, and Francesco Carlo Morabito. A novel explainable machine learning approach for eeg-based brain-computer interface systems.Neural Computing and Applications, 34(14):11347–11360, 2022

  56. [64]

    Towards interpretable machine learning in eeg analysis

    Maged Mortaga, Alexander Brenner, and Ekaterina Kutafina. Towards interpretable machine learning in eeg analysis. InGerman Medical Data Sciences 2021: Digital Medicine: Recognize–Understand–Heal, pages 32–38. IOS Press, 2021

  57. [65]

    Map- 30 ping human brain charts cross-sectionally and longitudinally.Proceed- ings of the National Academy of Sciences, 120(20):e2216798120, 2023

    Maria A Di Biase, Ye Ella Tian, Richard AI Bethlehem, Jakob Seidlitz, Aaron F Alexander-Bloch, BT Thomas Yeo, and Andrew Zalesky. Map- 30 ping human brain charts cross-sectionally and longitudinally.Proceed- ings of the National Academy of Sciences, 120(20):e2216798120, 2023

  58. [66]

    Neuroimag- ing brain growth charts: A road to mental health.Psychoradiology, 1(4):272–286, 2021

    Li-Zhen Chen, Avram J Holmes, Xi-Nian Zuo, and Qi Dong. Neuroimag- ing brain growth charts: A road to mental health.Psychoradiology, 1(4):272–286, 2021

  59. [67]

    Using normative models pre-trained on cross-sectional data to evaluate intra-individual longitudinal changes in neuroimaging data.eLife, 13, 2025

    Barbora Rehák Bučková, Charlotte Fraza, Rastislav Rehák, Marián Kolenič, Christian Beckmann, Filip Španiel, Andre Marquand, and Jaroslav Hlinka. Using normative models pre-trained on cross-sectional data to evaluate intra-individual longitudinal changes in neuroimaging data.eL...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.