Pith. sign in

REVIEW 3 major objections 5 minor 57 references

GenFAR: A generalized representation of brain structure, derived from 49,246 multi-cohort MRIs via deep learning

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read A modular network trained on 17 tasks and 49,246 brain MRIs learns a representation whose features match or beat directly trained models on 16 of 17 held-out tasks.

desk verdict A useful, openly released neuroimaging feature extractor with real transfer gains, but the headline leave-task-out numbers are inflated because the same tasks' labels were used to select the donor sequence. read the letter →

arxiv 2608.12185 v1 pith:VF2TK5GK submitted 2026-08-12 cs.CV

classification cs.CV
keywords brainMRIfoundationmodeltransferlearningmulti-tasksequentialsampleefficiencyleave-task-outevaluationneuroimaging
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 a single modular network trained on many clinical prediction tasks at once can learn a general representation of brain structure that transfers to new tasks. The authors train GenFAR on 49,246 T1-weighted MRIs from 11 cohorts across 17 classification and regression tasks, then freeze the learned features and attach lightweight predictors for tasks the network never trained on directly. In leave-task-out evaluation, the sequentially learned features matched or beat models trained directly on the full 3D images in 16 of 17 tasks, with the clearest gains in classification and in low-data settings. The claim matters because most neuroimaging studies are small; if features learned from large, diverse cohorts really transfer this well, small studies could build accurate predictors on frozen features instead of training from scratch.

What carries the argument

The load-bearing mechanism is sequential feature chaining. Each task channel is a 3D Squeeze-and-Excitation ResNet (a 3D convolutional network whose feature maps are recalibrated channel-wise), and in the sequential framework later channels take the concatenated 512- or 64-dimensional outputs of earlier channels as extra inputs, so knowledge flows forward without retraining earlier channels. Two design choices make the chain work: sequence length, set to six because analysis of 5,000 random sequences showed longer chains accumulate noise and overfit, and task order, set by the Donor Score, which compares the average performance improvement a receiver task gets from sequences that contain a candidate donor with sequences that do not. The five strong donors (age, AD/MCI, MMSE, hypertension, hyperlipidemia) form the base sequence, with each remaining task slotted into the sixth position, and this structure is what lets a frozen 3,200-feature representation serve as the foundation for secondary predictors.

What would settle it

Repeat the leave-task-out evaluation with the architecture chosen without any use of the held-out tasks: fix the five base tasks and the six-task sequence length by Donor Score analysis on a development set of tasks, then evaluate on a disjoint set of tasks whose labels never entered any design decision. If sequential features then beat direct training on only a minority of the untouched tasks, the 16-of-17 result depends on the selection step rather than on transfer; the released model and pipeline make this experiment runnable.

Watch

Extended reading notes

Core claim

GenFAR's central claim is that clinically supervised multi-task learning over many endpoints yields a general brain representation. The network is modular: each of 17 prediction channels is a 3D Squeeze-and-Excitation ResNet — a 3D convolutional network with channel-wise recalibration — followed by a task-specific head, and the channels' outputs are concatenated into a shared feature vector (8,704 features in the independent framework, 3,200 in the sequential framework). In the sequential framework, every task after the first receives both the brain scan and the concatenated features of all preceding tasks, so later tasks bootstrap from stronger earlier ones. The authors analysed 5,000 random task sequences, found that six-task chains transfer best while longer chains overfit, and introduced a Donor Score, $\operatorname{DonorScore}(d,r)=\mathrm{avg}[P(r\mid s):d\in s]-\mathrm{avg}[P(r\mid s):d\notin s]$, measuring how much a donor task's presence in a sequence raises a receiver task's performance; five tasks (age, AD/MCI, MMSE, hypertension, hyperlipidemia) emerged as consistently strong donors and form the fixed base of the final model. Freezing those channels and training a small head on a held-out task matched or beat full-image direct training in 16 of 17 tasks (the Boston Naming Test was the sole exception), and beat direct training on all four tasks tested on a separate external cohort.

Load-bearing premise

The leave-task-out comparison is only a fair test of pure transfer if the held-out tasks' labels had no influence on the design, but those same labels were used to pick the five base tasks and the six-task sequence length, so the held-out tasks were not fully unseen when the architecture was fixed.

Editorial extensions

If this is right

  • Frozen GenFAR features can stand in for full 3D image training on a new task: secondary predictors built on them matched or beat direct training in 16 of 17 leave-task-out experiments and on all four external-cohort tasks.
  • The advantage is largest where data are scarce: sequential features overtook direct training below roughly 2,000 training samples even for tasks where direct training won at full sample size (age, the CSF biomarkers, and the Boston Naming Test).
  • Which tasks are learned together matters: excluding negative-donor tasks (smoking, total-tau CSF) and fixing the chain at six tasks avoided the overfitting seen in longer sequences.
  • The released model and inference pipeline let outside researchers extract the same features from their own scans, so the transfer results can be checked on any cohort without retraining the foundation.

Reading between the lines

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

  • The paper's leave-task-out protocol withholds a held-out task's labels from the prediction channels, but those same labels entered the Donor Score and sequence-length analyses that fixed the base sequence; a stricter test would exclude held-out tasks from every design decision, and the 16-of-17 result could shrink under it.
  • The finding that age and AD/MCI are the strongest donors suggests that tasks tied to broad, whole-brain structural change are the best sources of general features; the same donor-score machinery could be used to pick anchor tasks for other modalities, such as diffusion or functional MRI.
  • Because the paper's closest self-supervised counterparts learn robust, acquisition-invariant features without labels, combining a contrastively pre-trained encoder with GenFAR's clinically anchored supervision is a natural untested next step that might improve both transfer and robustness.
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.

Referee Report

3 major / 5 minor

Summary. GenFAR trains modular 3D SE-ResNet channels on 49,246 T1-weighted MRIs from 11 cohorts using 17 classification and regression tasks (age, AD/MCI, MMSE, hypertension, hyperlipidemia, smoking, BMI, CSF biomarkers, and others). Two frameworks are compared: an independent set of 17 parallel channels and a sequential framework in which each task concatenates the frozen feature outputs of preceding tasks. From 5,000 random task sequences, the authors select sequence length six and a fixed base of five donor tasks via a proposed Donor Score, exclude two negative-donor tasks, and slot the remaining tasks as a 64-dimensional sixth task; the frozen channels then feed lightweight secondary predictors. The paper reports that sequential features beat or match direct training on 16 of 17 held-out tasks in leave-task-out (LTO) evaluation, large sample-efficiency gains as secondary-training data shrink to 100 subjects, and consistent gains on four tasks in the external MESA cohort. Model weights and CPU/GPU inference pipelines are publicly released.

Significance. If the claims hold, GenFAR is a useful community resource: a publicly released, clinically anchored brain-MRI feature extractor trained at unusually large scale (49,246 subjects, 11 cohorts, 17 diverse endpoints), with concrete evidence of low-data transfer gains and a genuinely external cross-cohort validation in MESA. The machine-checkable reproducibility elements — open weights, inference pipelines, and the web portal — are real strengths, as is the honesty of reporting channels that do not transfer well (e.g., BNT) rather than selecting only favorable tasks. The consistency of the transfer advantage across most tasks and its persistence in the external cohort make the central qualitative finding credible. However, the quantitative headline (16 of 17 tasks; 'unseen' tasks) is inflated by the model-selection protocol, because the donor sequence and sequence length were chosen using the same tasks that are later held out; the magnitude of the transfer benefit therefore needs re-estimation under a nested or pre-registered selection scheme.

major comments (3)
  1. [Methods (Donor Score, Sequence length selection, LTO-CV); Results (Task ordering); Table 2] The leave-task-out evaluation does not produce genuinely unseen tasks, because the architecture evaluated on each held-out task was selected using that task's labels. The final configuration — the five-task base (Age, AD/MCI, MMSE, Hypertension, Hyperlipidemia), the length-6 design with a 64-dimensional slotted sixth task, the exclusion of Total-Tau and Smoking, and the choice of which tasks provide features — was derived from the same 5,000-sequence transfer experiments whose per-task results constitute the LTO outcome (Results 'Task ordering'; Methods 'Donor Score'). The sequence-length analysis (Methods 'Sequence length selection'; Figure 2) likewise compared model test performance across all tasks before any task was 'held out.' When task r is later held out, the model evaluated on r was chosen in part by optimizing the average transfer performance across all receivers, including r. This is not equivalence-by-definition, and the large, consistent gains in Table 2 indicate that genuine transfer exists, but the reported numbers measure transfer plus favorable selection, so the 16-of-17 claim is optimistic and the held-out tasks are not truly unseen. The manuscript itself notes that optimization-based sequence search overfit the validation set, yet the hand-picked donor and length selection from the same experiments is subject to the same bias in milder form. The MESA evaluation (Table 3) is a valid cross-cohort test because the data are external, but its four tasks also participated in the selection analysis, so it does not validate task-unseen transfer. Please restructure the evaluation so that donor set, order, and length are selected on a development subset disjoint from the transfer-test tasks (or pre-registered), or disclose the selection-dependence and provide a sensitivity analysis with architectures chosen without each held-out task.
  2. [Methods 'Donor Score'; Figure 3] The statistical significance analysis underlying the donor ranking is not valid as presented. The Welch t-tests compare mean performance across thousands of random task sequences that are far from independent: any two sequences of the same length share most of their tasks, so the effective number of comparisons is far smaller than the number of sequences, and the FDR-corrected p-values in Figure 3 are anti-conservative through pseudo-replication. In addition, the Donor Score pools sequences of lengths 3-9 even though Figure 2 shows strong length-dependent variation in test performance; since a donor's inclusion probability increases with sequence length, sequences containing a given donor are enriched for longer, generally worse-performing sequences, shifting the absolute scores and making the -0.1 threshold for 'negative donors' (Total-Tau, Smoking) hard to interpret without conditioning on length. A permutation test over task orderings, or a mixed-effects model with sequence length as a covariate and sequence as a random effect, would be more appropriate; at minimum the analysis should show that the donor ranking is stable when conditioning on sequence length.
  3. [Results (sample-efficiency subsection); Figure 4] The sample-efficiency evaluation, one of the paper's central claims, is under-specified. The text says the analysis began with 4,000 subjects and decreased to 100, but the Methods do not state which tasks contribute to the aggregate curves in Figure 4, how many random subsample replicates are performed at each size, whether the overlap-avoidance sampling of the LTO protocol is applied to the secondary-training subjects, or whether confidence intervals or significance tests support the claim that sequential features are superior 'across all sample size ranges.' The exact crossover thresholds (e.g., Age below 2,000 samples; Abeta at 1,000; BNT and Total-Tau at 500) should be backed by per-task curves with variability estimates. Please add the full protocol and per-task or error-barred results.
minor comments (5)
  1. [Results; Methods 'Sequence length selection'; Figure 2 caption] The reported sequence lengths are inconsistent: the Results describe 5,000 sequences of varying lengths 3-16 and claim superiority over 'Length 1,' while the Methods state that models were trained on lengths 3-9 with additional samples for 12 and 16, and Figure 2's caption lists lengths 1, 3, 6, 9, 12, and 16. Please reconcile these descriptions.
  2. [Results; Table 2] The '16 of 17 comparable or superior' count treats correlation differences below 0.03 as 'comparable,' but several such cases (e.g., Digit Span Forward: 0.413 vs 0.406 with MAE 1.31 vs 1.27) are statistical ties; the headline should be reported together with per-task effect sizes rather than as a binary count.
  3. [Abstract; Discussion] The abstract's '49,246 individuals across 11 cohorts' and the Discussion's '50,302 T1 brain MRIs across 12 studies' refer to different sets (primary training set vs. including the 1,056 MESA evaluation subjects); this should be clarified to avoid an apparent inconsistency.
  4. [Introduction; Discussion] The paper positions GenFAR against BrainIAC, Triad, SimCLR-based models, and BrainAge but provides no empirical comparison with any of them; adding at least one such comparison (e.g., on the MESA tasks) would considerably strengthen the transfer claims.
  5. [Abstract] The phrase 'various tasks beyond those included in the training set' is misleading because all 17 tasks participate in the sequence-length and Donor Score analyses; recommend wording such as 'tasks not used in feature training.'

Circularity Check

1 steps flagged · score 5.0 of 10

LTO transfer numbers are inflated by model selection: donor-score and sequence-length analyses used the very tasks later held out, so the held-out tasks are not truly unseen.

  1. fitted input called prediction [Methods 'Donor Score' and 'Leave-Task-Out Cross-Validation'; Results 'Task ordering' and Table 2]
    "Based on the Donor Score analysis (Figure 3B-D), we established a fixed base sequence using the five strongest positive donor tasks: Age (mean Donor Score: 0.12), AD/MCI (0.10), MMSE (0.09), Hypertension (0.07), and Hyperlipidemia (0.06). ... Each of the remaining 10 tasks was then inserted as the sixth task in this sequence, allowing it to benefit from the strong base features while preventing harmful tasks from disrupting the feature hierarchy."

    The Donor Score used to build the fixed base sequence averages performance over every receiver task in the 17-task set, and the sequence-length analysis compared 5,000 random sequences on those same tasks. A task later 'held out' in LTO therefore influenced which five donors, which sequence length, and which slot-in architecture are evaluated for it; LTO removes only the task's labels from direct channel training, not from architecture and model selection. Consequently, the sequential-LTO numbers in Table 2 combine true transfer with favorable selection on the held-out task's own labels.

full rationale

The paper's strongest transfer claim rests on leave-task-out evaluation, but that evaluation is not fully independent of the held-out tasks: the donor-score and sequence-length analyses used all 17 tasks to select the base sequence, sequence length, slot-in architecture, and exclusion of Total-Tau and Smoking. When a task is later 'held out', the architecture evaluated for it was chosen partly on the basis of that task's own labels, so the held-out task is not genuinely unseen. This is a real, but partial, circularity: subject-level leakage is controlled, and the external MESA evaluation (Table 3) uses a truly unseen cohort and provides independent support for the method's utility. The paper also contains no load-bearing self-citation or definitional equivalence beyond this model-selection leakage. Overall, the central LTO claim is inflated by construction to an unknown degree, but the method retains independent empirical content, giving a moderate circularity score of 5.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claims rest on two unverified selections: the optimal sequence length and the donor task set are chosen using the same evaluation framework where transfer performance is reported. The representation's value also presupposes that T1 MRI contains learnable signal for all 17 labels. No new physical or mathematical entities are introduced.

free parameters (3)
  • Optimal sequence length = 6
    Chosen as best-performing length among 5,000 random task sequences using validation and test performance; this model-selection step uses the same evaluation setup where transfer gains are later reported.
  • Donor task base set = Age, AD/MCI, MMSE, Hypertension, Hyperlipidemia; exclude Smoking and Total-Tau CSF
    Base sequence selected from the same Donor Score analysis used to evaluate transfer, creating selection dependence between architecture choice and reported outcomes.
  • Feature dimensionality for slotted sixth task = 64 vs 512 for first five tasks
    Architectural choice that reduces sequential feature dimension; the paper provides no independent justification beyond design preference.
assumptions (3)
  • domain assumption T1-weighted brain MRI contains sufficient signal to predict each of the 17 clinical labels.
    The entire framework presupposes these labels are learnable from T1 images; the paper tests this empirically, but the value of the representation depends on it.
  • ad hoc to paper Donor Score and sequence-length analyses using validation and test performance can be used to select the final model without biasing later leave-task-out evaluation.
    The LTO protocol assumes holding out a task removes all influence of that task's labels, but donor-score selection used those labels to choose the architecture.
  • domain assumption The LTO sampling strategy prevents all cross-task data leakage.
    The paper reserves held-out samples and drops in-sample tasks with fewer than 300 samples, but overlapping labels and donor-score selection still create indirect information flow.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GenFAR: A generalized representation of brain structure, derived from 49,246 multi-cohort MRIs via deep learning." pith.science (2026). https://pith.science/paper/VF2TK5GK

@misc{pith2026260812185,
  author       = {Pith},
  title        = {Pith review of: GenFAR: A generalized representation of brain structure, derived from 49,246 multi-cohort MRIs via deep learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VF2TK5GK}},
  note         = {Machine review of arXiv:2608.12185}
}
read the original abstract

Deep learning models for neuroimaging have largely been developed for individual tasks, limiting knowledge transfer across applications. Here we introduce GenFAR, a modular deep learning framework that learns general, clinically informed features from brain MRIs. We trained this modular architecture on 49,246 individuals across 11 cohorts, using 17 diverse classification and regression tasks spanning cognition, clinical, diagnosis, demographics, and biomarkers. This yields aggregated, focused feature sets that capture rich, clinically- and biologically-relevant brain representations. We developed a sequential learning approach where tasks progressively build on previously learned representations. Through an analysis of 5,000 task sequences, we identified an optimal sequence length of six tasks and introduced a Donor Score metric to quantify each task's contribution to downstream performance. This analysis revealed five consistently strong donor tasks (Age, AD/MCI, MMSE, Hypertension, Hyperlipidemia) that formed the base of our sequential model. We demonstrated the utility of our learned representation, in various tasks beyond those included in the training set, to serve as the foundation for specialized secondary predictors. We further showed that using the learned feature representation can substantially increase the sample efficiency of secondary deep learning training tasks and models, as well as improve their accuracy.

Figures

Figures reproduced from arXiv: 2608.12185 by the authors.

Figure 1
Figure 1. Overview of the GenFAR model architecture and training frameworks. (A) [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Sequential learning performance across varying sequence lengths. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Donor Score analysis reveals task relationships in sequential learning. (A) [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Sample efficiency using GenFAR features in a secondary predictor versus direct training. [PITH_FULL_IMAGE:figures/full_fig_p014_4.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 49 canonical work pages

  1. [1]

    Deep learning in neuroimaging data analysis: Applications, challenges, and solutions

    LK Avberšek and G Repovš. Deep learning in neuroimaging data analysis: Applications, challenges, and solutions. Frontiers in Neuroimaging, 1, 2022

  2. [2]

    AD Yao, DL Cheng, I Pan, and F Kitamura. Deep learning in neuroradiology: A systematic review of current algorithms and approaches for the new wave of imaging technology.Radiology: Artificial Intelligence, 2(2):e190026, 2020

  3. [3]

    On the opportunities and risks of foundation models.arXiv preprint arXiv:2108.07258, 2021

    R Bommasani, DA Hudson, E Adeli, et al. On the opportunities and risks of foundation models.arXiv preprint arXiv:2108.07258, 2021

  4. [4]

    Zero-shot text-to-image generation

    A Ramesh, M Pavlov, G Goh, et al. Zero-shot text-to-image generation. InPMLR, pages 8821–8831, 2021

  5. [5]

    Language models are few-shot learners

    T Brown, B Mann, N Ryder, et al. Language models are few-shot learners. InAdvances in Neural Information Processing Systems, volume 33, pages 1877–1901, 2020

  6. [6]

    Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

    OpenAI. Gpt-4 technical report.arXiv preprint arXiv:2303.08774, 2023

  7. [7]

    High-resolution image synthesis with latent diffusion models

    R Rombach, A Blattmann, D Lorenz, P Esser, and B Ommer. High-resolution image synthesis with latent diffusion models. pages 10684–10695, 2022

  8. [8]

    Bloom: A 176b-parameter open-access multilingual language model.arXiv preprint arXiv:2211.05100, 2022

    TL Scao, A Fan, C Akiki, et al. Bloom: A 176b-parameter open-access multilingual language model.arXiv preprint arXiv:2211.05100, 2022

Show all 57 references
  1. [9]

    Foundation models for generalist medical artificial intelligence.Nature, 616(7956):259–265, 2023

    M Moor, O Banerjee, ZSH Abad, et al. Foundation models for generalist medical artificial intelligence.Nature, 616(7956):259–265, 2023

  2. [10]

    Representation learning: A review and new perspectives.IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(8):1798–1828, 2013

    Y Bengio, A Courville, and P Vincent. Representation learning: A review and new perspectives.IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(8):1798–1828, 2013

  3. [11]

    Multi-task neural networks for joint hippocampus segmentation and clinical score regression.Multimedia Tools and Applications, 77(22):29669–29686, 2018

    L Cao, L Li, J Zheng, et al. Multi-task neural networks for joint hippocampus segmentation and clinical score regression.Multimedia Tools and Applications, 77(22):29669–29686, 2018. 28

  4. [12]

    A fully automated multimodal mri-based multi-task learning for glioma segmentation and idh genotyping.IEEE Transactions on Medical Imaging, 2022

    J Cheng, J Liu, H Kuang, and J Wang. A fully automated multimodal mri-based multi-task learning for glioma segmentation and idh genotyping.IEEE Transactions on Medical Imaging, 2022

  5. [13]

    A multi-task learning framework for automated segmen- tation and classification of breast tumors from ultrasound images.Ultrasonic Imaging, 44(1):3–12, 2022

    J Chowdary, P Yogarajah, P Chaurasia, and V Guruviah. A multi-task learning framework for automated segmen- tation and classification of breast tumors from ultrasound images.Ultrasonic Imaging, 44(1):3–12, 2022

  6. [14]

    A feature transfer enabled multi-task deep learning model on medical imaging

    F Gao, H Yoon, T Wu, and X Chu. A feature transfer enabled multi-task deep learning model on medical imaging. Expert Systems with Applications, 143:112957, 2020

  7. [15]

    Eis-net: Segmenting early infarct and scoring aspects simultaneously on non-contrast ct of patients with acute ischemic stroke.Medical Image Analysis, 70:101984, 2021

    H Kuang, BK Menon, SI Sohn, and W Qiu. Eis-net: Segmenting early infarct and scoring aspects simultaneously on non-contrast ct of patients with acute ischemic stroke.Medical Image Analysis, 70:101984, 2021

  8. [16]

    Multi-task knowledge distillation for eye disease prediction

    S Chelaramani, M Gupta, V Agarwal, P Gupta, and R Habash. Multi-task knowledge distillation for eye disease prediction. pages 3983–3993, 2021

  9. [17]

    Canet: cross-disease attention network for joint diabetic retinopathy and diabetic macular edema grading.IEEE Transactions on Medical Imaging, 39(5):1483–1493, 2019

    X Li, X Hu, L Yu, L Zhu, C-W Fu, and P-A Heng. Canet: cross-disease attention network for joint diabetic retinopathy and diabetic macular edema grading.IEEE Transactions on Medical Imaging, 39(5):1483–1493, 2019

  10. [18]

    Deep learning of imaging phenotype and genotype for predicting overall survival time of glioblastoma patients.IEEE Transactions on Medical Imaging, 39(6):2100–2109, 2020

    Z Tang, Y Xu, L Jin, et al. Deep learning of imaging phenotype and genotype for predicting overall survival time of glioblastoma patients.IEEE Transactions on Medical Imaging, 39(6):2100–2109, 2020

  11. [19]

    Knowledge-guided multi-task attention network for survival risk prediction using multi-center computed tomography images.Neural Networks, 152:394–406, 2022

    L Zhang, L Zhong, C Li, et al. Knowledge-guided multi-task attention network for survival risk prediction using multi-center computed tomography images.Neural Networks, 152:394–406, 2022

  12. [20]

    Meta-matching as a simple framework to translate phenotypic predictive models from big to small data.Nature Neuroscience, 25(6):795–804, 2022

    T He, L An, P Chen, et al. Meta-matching as a simple framework to translate phenotypic predictive models from big to small data.Nature Neuroscience, 25(6):795–804, 2022

  13. [21]

    Brainiac: A foundation model for generalized brain mri analysis.medRxiv,

    D Tak, BA Garomsa, TL Chaunzwa, et al. Brainiac: A foundation model for generalized brain mri analysis.medRxiv,

  14. [22]

    Building a general simclr self-supervised foundation model across neurological diseases to advance 3d brain mri diagnoses.arXiv preprint arXiv:2509.10620, 2025

    E Kaczmarek, J Szeto, B Nichyporuk, and T Arbel. Building a general simclr self-supervised foundation model across neurological diseases to advance 3d brain mri diagnoses.arXiv preprint arXiv:2509.10620, 2025

  15. [23]

    DA Wood, M Townend, E Guilhem, et al. Brainage: Optimising brain age estimation through transfer learning: A suite of pre-trained foundation models for improved performance and generalisability in a clinical setting.Human Brain Mapping, 2024

  16. [24]

    Triad: Vision foundation model for 3d magnetic resonance imaging.Research Square (Preprint), 03 2025

    S Wang, M Safari, Q Li, et al. Triad: Vision foundation model for 3d magnetic resonance imaging.Research Square (Preprint), 03 2025

  17. [25]

    Brain foundation models with hypergraph dynamic adapter for brain disease analysis.arXiv e-prints, 2025

    Z Deng, H Wang, Z Huang, et al. Brain foundation models with hypergraph dynamic adapter for brain disease analysis.arXiv e-prints, 2025. arXiv:2505.00627

  18. [26]

    Brainfounder: Towards brain foundation models for neuroimage analysis.arXiv preprint arXiv:2406.10395, 2024

    J Cox, P Liu, SE Stolte, et al. Brainfounder: Towards brain foundation models for neuroimage analysis.arXiv preprint arXiv:2406.10395, 2024

  19. [27]

    Lamim: Medical image foundation models in assisting diagnosis of brain tumors: a pilot study.European Radiology, 2024

    M Chen, M Zhang, L Yin, et al. Lamim: Medical image foundation models in assisting diagnosis of brain tumors: a pilot study.European Radiology, 2024

  20. [28]

    Synthseg: Segmentation of brain mri scans of any contrast and resolution without retraining.Medical Image Analysis, 86:102789, 2023

    B Billot, DN Greve, O Puonti, et al. Synthseg: Segmentation of brain mri scans of any contrast and resolution without retraining.Medical Image Analysis, 86:102789, 2023

  21. [29]

    Model zoo: A growing “brain” that learns continually

    Rahul Ramesh and Pratik Chaudhari. Model zoo: A growing “brain” that learns continually. InInternational Conference on Learning Representations (ICLR), 2022. arXiv preprint arXiv:2106.03027

  22. [30]

    TR Austin, IM Nasrallah, G Erus, et al. Association of brain volumes and white matter injury with race, ethnicity, and cardiovascular risk factors: the multi-ethnic study of atherosclerosis.Journal of the American Heart Association, 11(7):e023159, 2022

  23. [31]

    Assessing brain and biological aging trajectories associated with alzheimer’s disease.Frontiers in Neuroscience, 16, 2022

    WS Liang, LH Goetz, and NJ Schork. Assessing brain and biological aging trajectories associated with alzheimer’s disease.Frontiers in Neuroscience, 16, 2022

  24. [32]

    Cardiometabolic risk factors associated with brain age and accelerate brain ageing.Human Brain Mapping, 43(2):700–720, 2022

    D Beck, AMG de Lange, ML Pedersen, et al. Cardiometabolic risk factors associated with brain age and accelerate brain ageing.Human Brain Mapping, 43(2):700–720, 2022

  25. [33]

    Prediction of brain age suggests accelerated atrophy after traumatic brain injury.Annals of Neurology, 77(4):571–581, 2015

    JH Cole, R Leech, DJ Sharp, and AsDN Initiative. Prediction of brain age suggests accelerated atrophy after traumatic brain injury.Annals of Neurology, 77(4):571–581, 2015. 29

  26. [34]

    Mri signatures of brain age and disease over the lifespan based on a deep brain network and 14,468 individuals worldwide.Brain, 143(7):2312–2324, 2020

    VM Bashyam, G Erus, J Doshi, et al. Mri signatures of brain age and disease over the lifespan based on a deep brain network and 14,468 individuals worldwide.Brain, 143(7):2312–2324, 2020

  27. [35]

    M Habes, R Pomponio, H Shou, et al. The brain chart of aging: Machine-learning analytics reveals links between brain aging, white matter disease, amyloid burden, and cognition in the istaging consortium of 10,216 harmonized mr scans.Alzheimer’s & Dementia, 17(1):89–102, 2021

  28. [36]

    Grad-cam++: Gen- eralized gradient-based visual explanations for deep convolutional networks

    Aditya Chattopadhay, Anirban Sarkar, Prantik Howlader, and Vineeth N Balasubramanian. Grad-cam++: Gen- eralized gradient-based visual explanations for deep convolutional networks. In2018 IEEE Winter Conference on Applications of Computer Vision (WACV), pages 839–847, 2018

  29. [37]

    The wisconsin registry for alzheimer’s prevention: a review of findings and current directions.Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring, 10:130–142, 2018

    SC Johnson, RL Koscik, EM Jonaitis, et al. The wisconsin registry for alzheimer’s prevention: a review of findings and current directions.Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring, 10:130–142, 2018

  30. [38]

    Cognitive changes preceding clinical symptom onset of mild cognitive impairment and relationship to apoe genotype.Current Alzheimer Research, 11(8):773–784, 2014

    M Albert, A Soldan, R Gottesman, et al. Cognitive changes preceding clinical symptom onset of mild cognitive impairment and relationship to apoe genotype.Current Alzheimer Research, 11(8):773–784, 2014

  31. [39]

    AL Gross, JJ Hassenstab, SC Johnson, et al. A classification algorithm for predicting progression from normal cognition to mild cognitive impairment across five cohorts: the preclinical ad consortium.Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring, 8:147–155, 2017

  32. [40]

    MA Espeland, KM Hayden, SN Lockhart, et al. Sex-related differences in brain volumes and cerebral blood flow among overweight and obese adults with type 2 diabetes: Exploratory analyses from the action for health in diabetes brain magnetic resonance imaging study.The Journals ...

  33. [41]

    P Cermakova, M Muller, AC Armstrong, et al. Subclinical cardiac dysfunction and brain health in midlife: Cardia (coronary artery risk development in young adults) brain magnetic resonance imaging substudy.Journal of the American Heart Association, 6(12):e006750, 2017

  34. [42]

    RH Pietrzak, YY Lim, D Ames, et al. Trajectories of memory decline in preclinical alzheimer’s disease: Results from the australian imaging, biomarkers and lifestyle flagship study of ageing.Neurobiology of Aging, 36(3):1231–1238, 2015

  35. [43]

    DS Marcus, TH Wang, J Parker, JG Csernansky, JC Morris, and RL Buckner. Open access series of imaging studies (oasis): Cross-sectional mri data in young, middle-aged, nondemented, and demented older adults.Journal of Cognitive Neuroscience, 19(9):1498–1507, 2007

  36. [44]

    The baltimore longitudinal study of aging (blsa): A 50-year-long journey and plans for the future

    L Ferrucci. The baltimore longitudinal study of aging (blsa): A 50-year-long journey and plans for the future. pages 1416–1419. Oxford University Press, 2008

  37. [45]

    DA Wolk, Z Zhang, S Boudhar, CM Clark, MJ Pontecorvo, and SE Arnold. Amyloid imaging in alzheimer’s disease: Comparison of florbetapir and pittsburgh compound-b positron emission tomography.Journal of Neurology, Neurosurgery & Psychiatry, 83(9):923–926, 2012

  38. [46]

    Postmenopausal hormone therapy and regional brain volumes: The whims-mri study.Neurology, 72(2):135–142, 2009

    SM Resnick, MA Espeland, SA Jaramillo, et al. Postmenopausal hormone therapy and regional brain volumes: The whims-mri study.Neurology, 72(2):135–142, 2009

  39. [47]

    The alzheimer’s disease neuroimaging initiative: Progress report and future plans.Alzheimer’s & Dementia, 6(3):202–211.e7, 2010

    MW Weiner, PS Aisen, CR Jack Jr, et al. The alzheimer’s disease neuroimaging initiative: Progress report and future plans.Alzheimer’s & Dementia, 6(3):202–211.e7, 2010

  40. [48]

    The alzheimer’s disease neuroimaging initiative (adni): Mri methods

    CR Jack Jr, MA Bernstein, NC Fox, et al. The alzheimer’s disease neuroimaging initiative (adni): Mri methods. Journal of Magnetic Resonance Imaging, 27(4):685–691, 2008

  41. [49]

    Image processing and quality control for the first 10,000 brain imaging datasets from uk biobank.NeuroImage, 166:400–424, 2018

    F Alfaro-Almagro, M Jenkinson, NK Bangerter, et al. Image processing and quality control for the first 10,000 brain imaging datasets from uk biobank.NeuroImage, 166:400–424, 2018

  42. [50]

    Lea & Febiger, Philadelphia, 1983

    E Kaplan, H Goodglass, and S Weintraub.Boston Naming Test. Lea & Febiger, Philadelphia, 1983

  43. [51]

    A new clinical scale for the staging of dementia.The British Journal of Psychiatry, 140(6):566–572, 1982

    CP Hughes, L Berg, W Danziger, LA Coben, and RL Martin. A new clinical scale for the staging of dementia.The British Journal of Psychiatry, 140(6):566–572, 1982

  44. [52]

    mini-mental state

    MF Folstein, SE Folstein, and PR McHugh. "mini-mental state": A practical method for grading the cognitive state of patients for the clinician.Journal of Psychiatric Research, 12(3):189–198, 1975

  45. [53]

    Squeeze-and-excitation networks

    Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 7132–7141, 2018

  46. [54]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    DP Kingma and J Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014. 30

  47. [55]

    Pytorch: An imperative style, high-performance deep learning library

    A Paszke, S Gross, F Massa, et al. Pytorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems, volume 32, 2019

  48. [56]

    Monai: An open-source framework for deep learning in healthcare.arXiv preprint arXiv:2211.02701, 2022

    MJ Cardoso, W Li, R Brown, et al. Monai: An open-source framework for deep learning in healthcare.arXiv preprint arXiv:2211.02701, 2022

  49. [57]

    Pytorch lightning

    WA Falcon. Pytorch lightning. GitHub, 2019. 31 Supplemental Materials Supplemental Text 1: Task-specific prediction channels learn representations that probe the brain across various regions, contributing to predictive performance on new tasks We analyzed the relationship betw...

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.