Generalizable Representation Learning for fMRI-based Neurological Disorder Identification

Anand A. Joshi; Haleh Akrami; Richard M. Leahy; Wenhui Cui

REVIEW 5 major objections 4 minor 25 references

Generalizable Representation Learning for fMRI-based Neurological Disorder Identification

T0 review · 5 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read MeTSK, a meta-learned self-supervised fMRI representation, transfers to four scarce neurological-disorder tasks and beats both fMRI foundation models and connectivity-feature classifiers.

desk verdict Solid engineering paper combining meta-learning and contrastive SSL for fMRI; the four-dataset claim is plausible but the Neurocon split must be subject-level, and the stats are weaker than the headline. read the letter →

arxiv 2412.16197 v2 pith:MPBNESPB submitted 2024-12-16 eess.IV cs.CEcs.CVcs.LG

Wenhui Cui , Haleh Akrami , Anand A. Joshi , Richard M. Leahy This is my paper · ORCID

classification eess.IVcs.CEcs.CVcs.LG

keywords fMRIrepresentationlearningmeta-learningself-supervisedneurologicaldisorderclassificationfunctionalconnectivityST-GCNdomaingeneralization

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

This paper sets out to show that deep learning can be made useful for neurological-disorder classification from fMRI even when clinical training data are scarce, by first learning general features from abundant healthy-control scans. The proposed strategy, MeTSK, combines contrastive self-supervised learning on the HCP healthy-control dataset with a bi-level meta-learning loop that adapts a target head to a small clinical dataset while updating a shared feature extractor for both domains. Evaluated by linear probing on four held-out clinical datasets—post-traumatic epilepsy, Alzheimer's disease, and two Parkinson's disease cohorts—the MeTSK features give classification AUCs above those of functional-connectivity features and above two recent fMRI foundation models. If the claim holds, MeTSK is a reusable feature extractor for low-data clinical tasks where end-to-end deep learning typically overfits.

What carries the argument

The load-bearing machinery is MeTSK (Meta Transfer of Self-supervised Knowledge), built on a spatio-temporal graph convolutional network (ST-GCN) whose graph nodes are 116 AAL brain regions with Pearson-correlation edges. Training alternates two loops: an inner loop fast-adapts only the target head on a meta-training split of the small clinical dataset, and an outer loop updates the source head and the shared feature extractor by minimizing a source-domain graph-contrastive loss plus the target head's loss on a held-out meta-validation split. The graph-contrastive loss treats two randomly sampled fMRI sub-sequences from the same subject as positive pairs and views different subjects as negatives, so the feature extractor must encode subject-specific but temporally invariant functional patterns. This nested optimization is what carries the transfer argument: the target head's fast adaptation defines what the extractor must preserve, and the source loss keeps those features from collapsing to the source task.

What would settle it

Train the same MeTSK pipeline without the source domain—removing the HCP self-supervised loss and keeping only the bi-level target-head adaptation, as in the paper's MeL ablation—and run the four held-out linear-probing tasks; any dataset on which MeL matches or exceeds MeTSK would falsify the claim that healthy-control self-supervision drives the generalization.

Watch

Extended reading notes

Core claim

The central discovery claimed is that the features learned by MeTSK—pre-trained with a self-supervised contrastive loss on healthy controls and shaped by meta-learning on one small clinical target—transfer to unseen clinical populations. On all four held-out datasets, linear classifiers trained on MeTSK features achieved the best mean AUC among the compared methods, exceeding both the frozen and fine-tuned versions of the two fMRI foundation models and beating classifiers trained directly on functional-connectivity features. The authors attribute this to the combination: self-supervision keeps the representation tied to intrinsic functional patterns rather than to a supervised source task, while the inner-loop adaptation of the target head forces the feature extractor to preserve information useful for the clinical domain. The direct consequence is that a model pretrained once on HCP plus a single small clinical site can serve as an off-the-shelf feature extractor for other disorder-classification tasks.

Load-bearing premise

The method assumes that self-supervised features learned from resting-state scans of healthy young adults capture intrinsic functional brain patterns that remain invariant enough to transfer to patient populations with different disorders, scanners, and age ranges.

Editorial extensions

If this is right

MeTSK trained once on HCP plus one small clinical site can be frozen and reused as a feature extractor for new disorder-classification tasks without retraining the encoder.
Linear probing on MeTSK features outperforms direct classifiers on functional-connectivity features, so the learned representation captures disorder-relevant signal that raw connectivity misses.
Training MeTSK with the full multi-site ADHD dataset yields no significant gain over a single site, implying that target-data scarcity is not the bottleneck once the representation is learned.
Frozen MeTSK features beat frozen features from two large fMRI foundation models, indicating that massive-scale pretraining is not necessary for low-data clinical transfer.
The SVM feature-importance map for PTE highlights temporal, parietal, and occipital regions, consistent with known epilepsy localization and with an independent PTE study.

Reading between the lines

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

Beyond the paper, a testable extension is to replace the ST-GCN backbone with a foundation-model encoder while keeping MeTSK's bi-level adaptation; the paper's fine-tuning results suggest foundation models are not necessarily improved by the same meta-adaptation unless their scale is handled carefully.
Beyond the paper, the domain-similarity metric used in the appendix could be turned into a screening tool: compute the earth mover's distance between source and candidate target features to decide in advance whether MeTSK will transfer before collecting a clinical dataset.
Beyond the paper, the method's success across epilepsy, Alzheimer's, and Parkinson's suggests the recipe could extend to other fMRI classification targets such as psychiatric disorders or stroke recovery, although the paper does not test these.
Beyond the paper, because the inner loop re-initializes the target head and re-samples meta-train and meta-validation splits each iteration, the reported stability may depend on the small meta-validation set; a sensitivity analysis over target-split size is an implicit next test.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Bottom line: this is a competent engineering paper, not a conceptual breakthrough. The new piece is the combination: meta-learning with a contrastive self-supervised source task on a ST-GCN, transferring healthy-control features to scarce clinical datasets. The authors evaluate on four held-out datasets, include a useful ablation (self-supervision beats sex classification as a source task; meta-learning beats fine-tuning and MTL), and release code. That is enough to take seriously.

Where it gets soft: the statistics do not support the strength of the claims. Tables 1 and 2 report mean AUC with standard deviations, and the error bars overlap with several baselines. No significance tests are run against the two foundation models in the main table. And the 'best on all four datasets' sentence in Sec 5.2.2 is not uniformly true: for PTE and Neurocon, BrainLM's RF AUC is higher than MeTSK's RF AUC (0.5589 vs 0.5392; 0.6774 vs 0.6230). The statement only holds if you aggregate across the three linear probes in a particular way, and the paper should say how.

The sharpest concern is Neurocon. That dataset is described as 27 PD patients and 16 controls with two replicate scans per subject. The paper never states that the 5-fold CV split is at the subject level. If folds are split by scan, replicate scans from the same subject can appear in both training and test folds, which would inflate MeTSK's 0.7529 SVM AUC. This is a specific, checkable issue. It does not kill the paper's overall point -- the held-out evaluation design is sound and the other three datasets do not have this leakage problem -- but the Neurocon result should be redone or explicitly justified with a subject-level split.

Minor: lambda=30 and tau=30 were chosen by watching loss convergence, with no sensitivity analysis. For a method that is mainly an engineering recipe, that is a small gap, but worth noting. The paper is otherwise transparent: it reports the overfitting of deep baselines, includes a convergence analysis, and fine-tunes foundation models as an extra fairness check. The citation pattern is appropriate -- the meta-learning and contrastive SSL ideas are attributed to their sources.

Verdict: worth a serious referee. If I were editing, I would send it out and ask for subject-level Neurocon splits, a definition of what 'best' means across classifiers, and either significance tests or paired bootstrap intervals. The code and ablations make this a useful baseline for anyone working on low-data clinical fMRI.

Referee Report

5 major / 4 minor

Summary. The paper proposes MeTSK, a representation learning strategy that combines meta-learning with contrastive self-supervised learning for fMRI-based neurological disorder identification. MeTSK is trained on HCP healthy-control data (source) and a small clinical target dataset (ADHD-Peking or ABIDE-UM), then evaluated by linear probing on four held-out clinical datasets (PTE, OASIS-3, TaoWu, Neurocon) for binary classification. The central claim is that MeTSK achieves the best linear-probing AUC on all four held-out datasets, outperforming two fMRI foundation models (Thomas et al. 2022 and BrainLM). The paper also includes comparisons with traditional classifiers, ablations of meta-learning and self-supervised components, fine-tuning experiments for foundation models, and a domain-similarity analysis.

Significance. The proposed method addresses a real and important problem: learning transferable fMRI representations from abundant healthy-control data to scarce and heterogeneous clinical datasets. If the central claim is valid, the approach would provide a practical feature extractor for low-data neurological disorder classification. The paper has clear strengths: it evaluates on four independent clinical datasets, includes comparisons with two recent fMRI foundation models, provides an ablation study isolating the contributions of self-supervised learning and meta-learning, and makes code publicly available. The domain-similarity analysis is a useful addition. However, the evaluation protocol has several unresolved threats—potential subject-level leakage in the Neurocon dataset, unclear handling of PCA in the feature pipeline, inconsistent support for the headline 'best on all four' claim across classifiers, and absent significance testing against foundation models—so the current evidence does not fully establish the central claim.

major comments (5)

[§4.6, §5.2.2] The Neurocon dataset is described in §4.6 as containing 27 PD patients and 16 controls with two replicate scans per subject, but the 5-fold cross-validation in §5.2.2 is not stated to be at the subject level. If the folds are split by scan rather than by subject, replicate scans from the same subject can appear in both the training and test sets of the linear probe, inflating the reported SVM AUC of 0.7529. The authors must clarify the splitting scheme and, if it is scan-level, rerun the evaluation with subject-level splits. Because Neurocon is one of four datasets supporting the 'best on all four' claim, this issue is load-bearing.
[§5.3, Evaluation] The feature evaluation pipeline applies PCA to reduce dimensionality 'before feeding the features into classifiers,' but the manuscript does not state that PCA is fit on the training folds only. If PCA is fit on the entire dataset before cross-validation, test information leaks into the probe training, which would bias the reported AUCs upward. The authors should specify the PCA fitting procedure and, if it was global, repeat the evaluation with PCA fit inside each training fold.
[§5.2.2, Table 2] The statement in §5.2.2 that 'MeTSK achieved the best performance on all four datasets, outperforming the two latest fMRI foundation models' is not uniformly supported by Table 2. On PTE, BrainLM's RF AUC is 0.5589 versus MeTSK's 0.5392; on Neurocon, BrainLM's RF AUC is 0.6774 versus MeTSK's 0.6230, and Thomas et al.'s RF AUC is 0.5813 versus MeTSK's 0.6230 as well. The claim appears to depend on choosing the SVM probe or on averaging across probes, but no aggregation rule is stated. The authors should either refine the claim to the specific probe and metric used or provide a principled aggregation across the three classifiers and confirm that the conclusion holds.
[§5.2.2, Table 2] No significance tests are reported for the differences between MeTSK and the foundation models in Table 2. Many of the reported standard deviations overlap substantially (e.g., PTE SVM: MeTSK 0.6415±0.0312 vs BrainLM 0.6011±0.0465; TaoWu SVM: 0.6831±0.1431 vs BrainLM 0.5725±0.1502), so the observed differences may be within noise. The authors should provide paired significance tests across folds (e.g., Wilcoxon signed-rank or permutation tests) for the MeTSK versus foundation-model comparisons on each dataset and classifier, and interpret the results accordingly.
[§5.3, §6] The key hyperparameters λ=30 and τ=30 are selected 'empirically by observing the convergence of the training loss,' but no sensitivity analysis is provided. Because λ scales the meta-learning target loss relative to the self-supervised loss and τ controls the sharpness of the contrastive loss, the reported performance could be sensitive to these choices. The authors should include a sensitivity analysis over reasonable ranges of λ and τ (e.g., on the ADHD-Peking or ABIDE-UM validation tasks) to demonstrate that the benefit of MeTSK is not confined to a single tuned setting.

minor comments (4)

[§5.3, Evaluation] The word 'Pinciple' in 'applied Pinciple Component Analysis' should be 'Principal.'
[Appendix A.2] The heading 'P-values Computed from Wilconxon Signed-Rank Tests' misspells 'Wilcoxon.'
[§6.1] The text refers to 'the same steps as described in Section 4.1.2,' but there is no Section 4.1.2; the intended reference appears to be Section 4.2 (ADHD-Peking) or the general preprocessing description in Section 4.
[Figure 3] The feature importance map in Figure 3 would benefit from a colorbar or an explicit statement of the color scale, since the reader cannot otherwise gauge the magnitude of the coefficients.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; MeTSK's held-out generalization claims rest on independent datasets and frozen-feature linear probing.

full rationale

The paper's derivation chain is self-contained with respect to its central claims. MeTSK is trained only on HCP (source) and ADHD-Peking/ABIDE-UM (target) data; the four clinical datasets used for generalization evaluation (PTE, OASIS-3, TaoWu, Neurocon) are held out during MeTSK training. Evaluation uses linear probing on frozen features, so the downstream classifiers cannot back-propagate into the representation or adapt it to the test folds. The reported superiority is an empirical comparison against external foundation models (Thomas et al. 2022; BrainLM) and traditional connectivity baselines, not a quantity fitted into the model. The self-citations to Akrami et al. (2021, 2024) are used for motivation and for post-hoc interpretability alignment only; the PTE classification result in Table 2 is computed from MeTSK features and an SVM probe, not imported from those prior papers. No equation in the paper equates a predicted quantity with an input by construction, and no load-bearing argument reduces to a self-citation. The possible Neurocon subject-level split leakage is an evaluation-validity concern, not a circularity of the derivation. Therefore the circularity score is 0.

Assumptions & free parameters 7 free parameters · 7 assumptions · 0 invented entities

The paper introduces no new physical entities or constants; the free parameters are training hyperparameters chosen empirically. The central load-bearing assumptions are the transferability of healthy-control self-supervised features and the adequacy of meta-learning to generalize beyond the single target domain.

free parameters (7)

lambda (meta-loss scaling) = 30
Scales the target-domain classification loss in the outer loop; chosen empirically by observing convergence (Sec 5.3).
tau (contrastive temperature) = 30
Temperature in the InfoNCE contrastive loss; value taken from You et al. 2020a but is not derived, and it strongly affects the loss scale.
inner loop learning rate alpha = 0.01
SGD learning rate for target head updates in the inner loop (Sec 5.3).
outer loop learning rate beta = 0.001
Adam learning rate for feature extractor and source head updates (Sec 5.3).
inner loop steps = 25
Number of fast adaptation steps for the target head (Sec 5.3).
sub-sequence length L = 128
Number of time points in sampled input sub-sequences for ST-GCN (Sec 5.3).
PCA dimension after feature extraction
The paper applies PCA to reduce graph-level feature dimensionality but does not report the number of components (Sec 5.3).

assumptions (7)

domain assumption The AAL atlas with 116 ROIs provides a valid node definition for fMRI graph construction across all datasets.
Sec 4 states all fMRI data were parcellated with the AAL atlas; results may depend on this choice (the appendix acknowledges atlas differences across prior works).
domain assumption Pearson correlation between ROI time series is a meaningful measure of functional connectivity for disorder classification.
Sec 3.1 defines graph edges via Pearson correlation; alternative connectivity measures are not tested.
domain assumption Contrastive self-supervised learning on the HCP healthy-control dataset yields features that are invariant across domains and useful for clinical classification.
Sec 3.3 hypothesizes that intrinsic features learned on controls are 'likely to be invariant across domains'; the entire transfer claim rests on this.
domain assumption Meta-learning with a single target domain improves generalization to unseen clinical datasets.
Sec 3.2 adopts the bi-level optimization from Liu et al. 2020a and assumes it transfers across domains beyond the single target used in training.
domain assumption Linear probing adequately evaluates representation quality for downstream disorder classification.
Sec 3.4 argues that good representations should be linearly separable; this is a standard but still an assumption.
domain assumption The HCP S1200 cohort is representative of healthy brain dynamics.
Sec 4.1 uses HCP as the source domain; if HCP is not representative of the general healthy population, the learned features may be biased.
standard math Graph convolution and InfoNCE contrastive loss equations (Eq. 1 and Eq. 4) are correct as stated.
These are standard formulations from Kipf and Welling and You et al., used without derivation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Generalizable Representation Learning for fMRI-based Neurological Disorder Identification." pith.science (2026). https://pith.science/paper/MPBNESPB

@misc{pith2026241216197,
  author       = {Pith},
  title        = {Pith review of: Generalizable Representation Learning for fMRI-based Neurological Disorder Identification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MPBNESPB}},
  note         = {Machine review of arXiv:2412.16197}
}

read the original abstract

Despite the impressive advances achieved using deep learning for functional brain activity analysis, the heterogeneity of functional patterns and the scarcity of imaging data still pose challenges in tasks such as identifying neurological disorders. For functional Magnetic Resonance Imaging (fMRI), while data may be abundantly available from healthy controls, clinical data is often scarce, especially for rare diseases, limiting the ability of models to identify clinically-relevant features. We overcome this limitation by introducing a novel representation learning strategy integrating meta-learning with self-supervised learning to improve the generalization from normal to clinical features. This approach enables generalization to challenging clinical tasks featuring scarce training data. We achieve this by leveraging self-supervised learning on the control dataset to focus on inherent features that are not limited to a particular supervised task and incorporating meta-learning to improve the generalization across domains. To explore the generalizability of the learned representations to unseen clinical applications, we apply the model to four distinct clinical datasets featuring scarce and heterogeneous data for neurological disorder classification. Results demonstrate the superiority of our representation learning strategy on diverse clinically-relevant tasks. Code is publicly available at https://github.com/wenhui0206/MeTSK/tree/main

Figures

Figures reproduced from arXiv: 2412.16197 by the authors.

**Figure 2.** An illustration of the ST-GCN model architecture. Spatial graph convolution is first applied to the [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 3.** Feature importance map of PTE features generated from MeTSK shown as color-coded ROIs [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: A comparison of the linear probing performance between MeTSK model trained with a single [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]

**Figure 5.** Figure 5: During training, we observe that the self-supervised loss decreases well, while the meta-validation [PITH_FULL_IMAGE:figures/full_fig_p019_5.png]

**Figure 5.** Figure 5: Loss evolution during MeTSK training over 90 epochs. The first 20 epochs is a warm-up phase [PITH_FULL_IMAGE:figures/full_fig_p020_5.png]

**Figure 6.** Figure 6: A comparison of the domain similarity between HCP self-supervised features (HCP-SSL, from [PITH_FULL_IMAGE:figures/full_fig_p023_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 16 canonical work pages

[2]

Prediction of post traumatic epilepsy using mri-based imaging markers

Haleh Akrami, Wenhui Cui, Paul E Kim, Christianne N Heck, Andrei Irimia, Karim Jebri, Dileep Nair, Richard M Leahy, and Anand Joshi. Prediction of post traumatic epilepsy using mri-based imaging markers. bioRxiv, pp. 2024–01,

work page 2024
[5]

Zhi-An Huang, Rui Liu, and Kay Chen Tan

URLhttps://arxiv.org/abs/1911.05722. Zhi-An Huang, Rui Liu, and Kay Chen Tan. Multi-task learning for efficient diagnosis of asd and adhd using resting-state fmri data. In 2020 International Joint Conference on Neural Networks (IJCNN), pp. 1–7. IEEE,

arXiv 1911
[6]

Adam: a method for stochastic optimization (2014).arXiv preprint arXiv:1412.6980, 22,

15 Published in Transactions on Machine Learning Research (05/2025) Diederik P Kingma and Jimmy Ba. Adam: a method for stochastic optimization (2014).arXiv preprint arXiv:1412.6980, 22,

arXiv 2014
[10]

Meta-learning Transferable Representations with a Single Target Domain

Hong Liu, Jeff Z HaoChen, Colin Wei, and Tengyu Ma. Meta-learning transferable representations with a single target domain.arXiv preprint arXiv:2011.01418, 2020a. Hong Liu, Jeff Z. HaoChen, Colin Wei, and Tengyu Ma. Meta-learning transferable representations with a single target domain, 2020b. Xingdan Liu, Jiacheng Wu, Wenqi Li, Qian Liu, Lixia Tian, and ...

work page Pith review arXiv 2011
[11]

Brainlm: A foundation model for brain activity recordings.bioRxiv, pp

Josue Ortega Caro, Antonio Henrique Oliveira Fonseca, Christopher Averill, Syed A Rizvi, Matteo Rosati, James L Cross, Prateek Mittal, Emanuele Zappala, Daniel Levine, Rahul M Dhodapkar, et al. Brainlm: A foundation model for brain activity recordings.bioRxiv, pp. 2023–09,

work page 2023
[12]

Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity mri.Cerebral cortex, 28(9):3095–3114,

16 Published in Transactions on Machine Learning Research (05/2025) Alexander Schaefer, Ru Kong, Evan M Gordon, Timothy O Laumann, Xi-Nian Zuo, Avram J Holmes, Simon B Eickhoff, and BT Thomas Yeo. Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity mri.Cerebral cortex, 28(9):3095–3114,

work page 2025
[14]

Contrastive functional connectivity graph learning for population-based fmri classification

Xuesong Wang, Lina Yao, Islem Rekik, and Yu Zhang. Contrastive functional connectivity graph learning for population-based fmri classification. InMedical Image Computing and Computer Assisted Intervention– MICCAI 2022: 25th International Conference, Singapore, September 18–22, 2022, Proceedings, Part I, pp. 221–230. Springer,

work page 2022
[15]

doi: 10.1109/tmi.2024.3392988

ISSN 1558-254X. doi: 10.1109/tmi.2024.3392988. URL http://dx.doi.org/10.1109/TMI.2024.3392988. Yuning You, Tianlong Chen, Yongduo Sui, Ting Chen, Zhangyang Wang, and Yang Shen. Graph contrastive learning with augmentations.Advances in Neural Information Processing Systems, 33:5812–5823, 2020a. 17 Published in Transactions on Machine Learning Research (05/...

arXiv 2024

Show all 25 references

[16]

When does self-supervision help graph convolutional networks? In international conference on machine learning, pp

0.4517± 0.0806 0 .4891± 0.0904 0 .4332± 0.1859 0 .4470± 0.1822 Yuning You, Tianlong Chen, Zhangyang Wang, and Yang Shen. When does self-supervision help graph convolutional networks? In international conference on machine learning, pp. 10871–10880. PMLR, 2020b. Z. Yu and G. He...

2005
[19]

This allows us to assess the robustness and generalization capability of MeTSK under data-scarce conditions—settings that are common in clinical practice

Unlike previous works that often leverage larger datasets, our experimental design simulates real-world clinical scenarios with limited training data. This allows us to assess the robustness and generalization capability of MeTSK under data-scarce conditions—settings that are ...

2024
[21]

(2024) 0.5 F1 Score 40% for testing 42 Harvard-Oxford 69Li et al

246Vigneshwaran et al. (2024) 0.5 F1 Score 40% for testing 42 Harvard-Oxford 69Li et al. (2021) 0.66 Accuracy 10-fold CV 42 Schaefer 100MeTSK (Ours) 0.7529 AUC 5-fold CV 42 AAL 116 Taowu Xu et al. (2024) 0.775 Accuracy 10-fold CV 40 Schaefer 100Vigneshwaran et al. (2024) 0.6 F...

2024
[22]

(2021) 0.675 Accuracy 10-fold CV 40 Schaefer 100MeTSK (Ours) 0.6831 AUC 5-fold CV 40 AAL 116 OASIS-3 Han et al

69Li et al. (2021) 0.675 Accuracy 10-fold CV 40 Schaefer 100MeTSK (Ours) 0.6831 AUC 5-fold CV 40 AAL 116 OASIS-3 Han et al. (2024) (Proposed GCN) 0.800 Accuracy 10-fold CV 1006 Schaefer 100Han et al. (2024) (Baseline GCN) 0.686 Accuracy 10-fold CV 1006 Schaefer 100Han et al. (...

2021
[23]

16Akrami et al. (2024) (GCN) 0.5900 AUC Leave-one-out CV 72 USCLobes 16MeTSK (Ours) 0.6415 AUC 5-fold CV 72 AAL 116 Table 6: Wilcoxon signed-rank test p-values between MeTSK and two other methods in Figure 4 across all datasets and classifiers. Columns represent comparisons ag...

2024
[24]

Fluctuations are expected, as the loss is computed on a small (39-subject) randomly sampled meta-validation set

During training, we observe that the self-supervised loss decreases well, while the meta-validation cross-entropy (classification) loss also shows a decreasing trend overall, although with some fluctuations. Fluctuations are expected, as the loss is computed on a small (39-sub...

2025
[25]

Notably, the pre-training datasets for these foundation models already included the HCP data, so aligning the training data required only fine- tuning on the ADHD-Peking dataset

A.4 Linear Probing Results of Fine-tuned Foundation Models To ensure a fair comparison, we fine-tuned the baseline foundation models on the same dataset—HCP combined with ADHD-Peking—that was used for training MeTSK. Notably, the pre-training datasets for these foundation mode...

2022
[2005]

2005.1521516

doi: 10.1109/ICME. 2005.1521516. Hao Zhang, Ran Song, Liping Wang, Lin Zhang, Dawei Wang, Cong Wang, and Wei Zhang. Classification of brain disorders in rs-fmri via local-to-global graph neural networks.IEEE Transactions on Medical Imaging, 42(2):444–455,

2005
[2013]

Large scale fine-grained categorization and domain-specific transfer learning,

14 Published in Transactions on Machine Learning Research (05/2025) Yin Cui, Yang Song, Chen Sun, Andrew Howard, and Serge Belongie. Large scale fine-grained categorization and domain-specific transfer learning,

2025
[2014]

Semi-supervised classification with graph convolutional networks.arXiv preprint arXiv:1609.02907,

Thomas N Kipf and Max Welling. Semi-supervised classification with graph convolutional networks.arXiv preprint arXiv:1609.02907,

arXiv
[2016]

Fine-tuning can distort pretrained features and underperform out-of-distribution.arXiv preprint arXiv:2202.10054,

Ananya Kumar, Aditi Raghunathan, Robbie Jones, Tengyu Ma, and Percy Liang. Fine-tuning can distort pretrained features and underperform out-of-distribution.arXiv preprint arXiv:2202.10054,

arXiv
[2018]

(2022) 0.667 AUC 10-fold CV 42 Brainnetome (Fan et al.,

100Shi et al. (2022) 0.667 AUC 10-fold CV 42 Brainnetome (Fan et al.,

2022
[2020]

Rao P Gullapalli

URLhttps://arxiv.org/abs/2006.07733. Rao P Gullapalli. Investigation of prognostic ability of novel imaging markers for traumatic brain injury (tbi). Technical report, BALTIMORE UNIV MD,

2006 arXiv
[2021]

Prediction of posttrau- matic epilepsy using machine learning

Haleh Akrami, Andrei Irimia, Wenhui Cui, Anand A Joshi, and Richard M Leahy. Prediction of posttrau- matic epilepsy using machine learning. InMedical Imaging 2021: Biomedical Applications in Molecular, Structural, and Functional Imaging, volume 11600, pp. 424–430. SPIE,

2021
[2022]

Oasis-3: longitudinal neuroimaging, clinical, and cognitive dataset for normal aging and alzheimer disease.MedRxiv, pp

Pamela J LaMontagne, Tammie LS Benzinger, John C Morris, Sarah Keefe, Russ Hornbeck, Chengjie Xiong, Elizabeth Grant, Jason Hassenstab, Krista Moulder, Andrei G Vlassenko, et al. Oasis-3: longitudinal neuroimaging, clinical, and cognitive dataset for normal aging and alzheimer...

2019
[2023]

Xi Sheryl Zhang, Fengyi Tang, Hiroko H Dodge, Jiayu Zhou, and Fei Wang

doi: 10.1109/TMI.2022.3219260. Xi Sheryl Zhang, Fengyi Tang, Hiroko H Dodge, Jiayu Zhou, and Fei Wang. Metapred: Meta-learning for clinical risk prediction with limited patient electronic health records. InProceedings of the 25th ACM SIGKDD International Conference on Knowledg...

2022
[2024]

Chunlei Shi, Xianwei Xin, and Jiacai Zhang

URL https://arxiv.org/abs/2408.04154. Chunlei Shi, Xianwei Xin, and Jiacai Zhang. Domain adaptation using a three-way decision improves the identification of autism patients from multisite fmri data.Brain Sciences, 11(5):603,

arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [2]

Prediction of post traumatic epilepsy using mri-based imaging markers

Haleh Akrami, Wenhui Cui, Paul E Kim, Christianne N Heck, Andrei Irimia, Karim Jebri, Dileep Nair, Richard M Leahy, and Anand Joshi. Prediction of post traumatic epilepsy using mri-based imaging markers. bioRxiv, pp. 2024–01,

work page 2024

[2] [5]

Zhi-An Huang, Rui Liu, and Kay Chen Tan

URLhttps://arxiv.org/abs/1911.05722. Zhi-An Huang, Rui Liu, and Kay Chen Tan. Multi-task learning for efficient diagnosis of asd and adhd using resting-state fmri data. In 2020 International Joint Conference on Neural Networks (IJCNN), pp. 1–7. IEEE,

arXiv 1911

[3] [6]

Adam: a method for stochastic optimization (2014).arXiv preprint arXiv:1412.6980, 22,

15 Published in Transactions on Machine Learning Research (05/2025) Diederik P Kingma and Jimmy Ba. Adam: a method for stochastic optimization (2014).arXiv preprint arXiv:1412.6980, 22,

arXiv 2014

[4] [10]

Meta-learning Transferable Representations with a Single Target Domain

Hong Liu, Jeff Z HaoChen, Colin Wei, and Tengyu Ma. Meta-learning transferable representations with a single target domain.arXiv preprint arXiv:2011.01418, 2020a. Hong Liu, Jeff Z. HaoChen, Colin Wei, and Tengyu Ma. Meta-learning transferable representations with a single target domain, 2020b. Xingdan Liu, Jiacheng Wu, Wenqi Li, Qian Liu, Lixia Tian, and ...

work page Pith review arXiv 2011

[5] [11]

Brainlm: A foundation model for brain activity recordings.bioRxiv, pp

Josue Ortega Caro, Antonio Henrique Oliveira Fonseca, Christopher Averill, Syed A Rizvi, Matteo Rosati, James L Cross, Prateek Mittal, Emanuele Zappala, Daniel Levine, Rahul M Dhodapkar, et al. Brainlm: A foundation model for brain activity recordings.bioRxiv, pp. 2023–09,

work page 2023

[6] [12]

Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity mri.Cerebral cortex, 28(9):3095–3114,

16 Published in Transactions on Machine Learning Research (05/2025) Alexander Schaefer, Ru Kong, Evan M Gordon, Timothy O Laumann, Xi-Nian Zuo, Avram J Holmes, Simon B Eickhoff, and BT Thomas Yeo. Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity mri.Cerebral cortex, 28(9):3095–3114,

work page 2025

[7] [14]

Contrastive functional connectivity graph learning for population-based fmri classification

Xuesong Wang, Lina Yao, Islem Rekik, and Yu Zhang. Contrastive functional connectivity graph learning for population-based fmri classification. InMedical Image Computing and Computer Assisted Intervention– MICCAI 2022: 25th International Conference, Singapore, September 18–22, 2022, Proceedings, Part I, pp. 221–230. Springer,

work page 2022

[8] [15]

doi: 10.1109/tmi.2024.3392988

ISSN 1558-254X. doi: 10.1109/tmi.2024.3392988. URL http://dx.doi.org/10.1109/TMI.2024.3392988. Yuning You, Tianlong Chen, Yongduo Sui, Ting Chen, Zhangyang Wang, and Yang Shen. Graph contrastive learning with augmentations.Advances in Neural Information Processing Systems, 33:5812–5823, 2020a. 17 Published in Transactions on Machine Learning Research (05/...

arXiv 2024