Efficient Deep Learning Models for Predicting Individualized Task Activation From Resting-State Functional Connectivity.

Code ↔ Paper

5 matches between paragraphs of the paper and lines of its authors' code, computed by the harvester (lexical-v1). Click a colored paragraph or line to see its counterpart.

The 5 matches

[1] § Method › Models › BrainSurfGCN ↔ model/brain_surf_gnn.py, lines 19–61 · score 0.61 · BDLayer, Geometric, BrainSurfGCN, Graph, node, channels
[2] § Results › Mechanisms of Variability in Predictability › Resting‐State Signal Quality Shapes Prediction Error ↔ utils/utilities.py, lines 1–61 · score 0.56 · Math Story, Face Avg, Language, Motor
[3] § Method › Models › BrainSurfCNN ↔ model/brain_surf_cnn.py, lines 21–35 · score 0.52 · ResPoolBlock, Pooling Block, resolutions, convolutional, mesh
[4] § Method › Training Details ↔ train/train.py, lines 71–152 · score 0.51 · MSE loss, Adam, PyTorch, optimization, validation, trained
[5] § Method › Training Details ↔ train/train_gnn.py, lines 77–155 · score 0.51 · MSE loss, Adam, PyTorch, optimization, validation, trained

Paper

Loaded from Europe PMC by your browser, not stored by OSCR: doi.org · Europe PMC

The paper is loaded when this pane is shown.

The paper is loaded by your browser, with JavaScript: read it at doi.org or on Europe PMC.

The authors' code

Python · 61 lines · 2.3 KB · MIT · 1 match

import torch.nn as nn
from torch.nn import Linear
import torch.nn.functional as F
from torch_geometric.nn import GCNConv
class BDLayer(nn.Module):
def __init__(self, input_features, output_features, aggr='mean', activation=nn.LeakyReLU):
super(BDLayer, self).__init__()
self.conv = GCNConv(input_features, output_features, aggr=aggr)
self.activation = activation()
self.bn = nn.BatchNorm1d(output_features)
def forward(self, x, edge_index):
x = self.conv(x, edge_index)
x = self.activation(x)
x = self.bn(x)
return x
class BrainSurfGCN(nn.Module):
def __init__(self, in_ch, out_ch, hidden_channels=[32, 32, 64, 64]):
super(BrainSurfGCN, self).__init__()
assert(len(hidden_channels) > 0)
# self.in_conv = GCNConv(input_features, hidden_channels[0], aggr='mean')
down_layers = []
for i in range(len(hidden_channels)):
if i == 0:
down_layers.append(BDLayer(in_ch, hidden_channels[i], aggr='mean'))
else:
down_layers.append(BDLayer(hidden_channels[i-1], hidden_channels[i], aggr='mean'))
self.down_layers = nn.Sequential(*down_layers)
init_up = hidden_channels[-1]
up_layers = []
up_layers.append(BDLayer(init_up, hidden_channels[-1], aggr='mean'))
for i in range(1, len(hidden_channels)):
up_layers.append(BDLayer(hidden_channels[-i], hidden_channels[-i-1], aggr='mean'))
self.up_layers = nn.Sequential(*up_layers)
self.lin = Linear(hidden_channels[0], out_ch, bias=True)
self.in_ch = in_ch
self.out_ch = out_ch
def forward(self, data):
# data must be a torch_geometric Data() object
batch = len(data.ptr) - 1
num_nodes = data.num_nodes // batch # there are batch * num_nodes in the data structure, makes one big bipartite graph for batched data
x = data.x
edge_index = data.edge_index
res = []
for dl in self.down_layers:
x = dl(x, edge_index)
res.append(x)
for i, ul in enumerate(self.up_layers):
x = res[-(i+1)] + ul(x, edge_index)
x = self.lin(x)
x = x.reshape(batch, num_nodes, self.out_ch)
return x.transpose(1,2)

brain_surf_gnn.py at commit d0132ec, under MIT · at the source

Overview

Authors: Soren J. Madsen¹, Young‐Eun Lee¹, Shaun K. L. Quah¹, Lucina Q. Uddin², Jeanette A. Mumford³, Deanna M. Barch⁴, Damien A. Fair⁵, Ian H. Gotlib³, Russell A. Poldrack³, Amy Kuceyeski⁶, Manish Saggar¹

ORCID iDs: Young‐Eun Lee, Ian H. Gotlib, Manish Saggar

Institutions: Stanford Medicine (United States); Stanford University (United States); University of California, Los Angeles (United States); Washington University in St. Louis (United States); University of Minnesota (United States); Weill Cornell Medicine (United States)

Journal: Human brain mapping, volume 47, issue 9, article e70557

Dates: received 3 February 2026; accepted 15 May 2026; published online 18 June 2026; in print June 2026

Type: Research article · Language: English

License: CC BY

Identifiers: DOI 10.1002/hbm.70557 · PMID 42315989 · PMCID PMC13279125 · OpenAlex W7165170452

Open access: gold, a free copy (OpenAlex)

Status: code verified

Categories: fMRI (modality), human (organism), cognitive (subfield)

Methods: Connectivity, Statistics, Smoothing, state filtering, decompositions, Machine learning, Preprocessing

Keywords: deep learning, functional MRI, resting‐state, task contrast

MeSH: Brain*, Cerebral Cortex*, Connectome*, Deep Learning*, Magnetic Resonance Imaging*, Nerve Net*, Adult, Female, Humans, Male, Predictive Learning Models (* major topic)

Topic: Functional Brain Connectivity Studies (Cognitive Neuroscience, Neuroscience), according to OpenAlex

Funding: HHS | NIH | National Institute of Mental Health (NIMH) (MH127608); Stanford Maternal and Child Health Research Institute

Citations: not cited yet (Europe PMC); 42 references in the paper

Abstract

Deep learning models have demonstrated the potential to predict task‐evoked brain activation from resting‐state functional magnetic resonance imaging, offering a pathway toward individualized brain mapping without requiring task‐based data. In this study, we systematically evaluate architectural strategies for improving the efficiency and scalability of such models. Using data from the Human Connectome Project, we replicate the BrainSurfCNN framework and introduce two extensions: BrainSERF, which incorporates channel‐wise attention through squeeze‐and‐excitation modules, and BrainSurfGCN, a graph‐based model that leverages cortical mesh topology for efficient message passing. Across multiple evaluation metrics, including spatial correlation, Dice score, Dice AUC, and subject identification accuracy, all models achieve comparable predictive performance. Despite similar accuracy, the proposed models offer distinct advantages. BrainSERF provides modest improvements in capturing individual‐specific features, while BrainSurfGCN achieves substantial reductions in model size and training time, highlighting a favorable trade‐off between performance and computational efficiency. Beyond architectural comparisons, we investigate factors driving variability in prediction accuracy. We find that behavioral task performance, resting‐state data quality, and inter‐subject variability in task activation jointly constrain prediction fidelity. In particular, contrasts with lower signal reliability and higher variability exhibit reduced predictability across all models. Together, these findings demonstrate that incorporating topological and functional structural priors can improve the efficiency of deep learning models without sacrificing accuracy, while also emphasizing that prediction performance is fundamentally limited by the reliability of the underlying neural signals.

Reproduced under the paper's license (CC BY), from the paper cited above.

Repositories

Its files are read in the Code ↔ Paper reader above, with 5 matches between paragraphs and lines of code.

braindynamicslab/dl-task-contrast-prediction

License: MIT

State: the link answers, verified on 27 September 2026

Evidence: files inventoried

Commit: d0132ec125e8cb48875fac9827825d87def8b2bf, 30 January 2026

Languages: Python (19), Jupyter (4), Shell (2)

Size: 48 files, 25 scripts

Software Heritage: not archived

Found in: “Data Availability Statement”

Holds: README, license file, environment (requirements.txt), 4 notebooks

Not found: CITATION.cff, tests, continuous integration, documentation

Tools: NumPy (21 files), PyTorch (14 files), NiBabel (6 files), Matplotlib (5 files), PyTorch Geometric (5 files), neuromaps (4 files), seaborn (4 files), pandas (3 files), SciPy (3 files), Nilearn (1 file), scikit-learn (1 file)

Availability: 1 check, the latest on 27 September 2026: the link answers

27 September 2026: the link answers

27 files

model/brain_surf_cnn.py, Python, 379 lines, 1 match
model/brain_surf_gnn.py, Python, 61 lines, 1 match
model/utils.py, Python, 75 lines
posthoc_analysis/correlation_accuracy.ipynb, Jupyter, 802 lines
posthoc_analysis/cortical_plots.ipynb, Jupyter, 1,010 lines
posthoc_analysis/dice_plots.ipynb, Jupyter, 541 lines
posthoc_analysis/extract_tsnr.py, Python, 63 lines
posthoc_analysis/tsnr_plots.ipynb, Jupyter, 676 lines
predict/predict.py, Python, 73 lines
predict/predict_attn.py, Python, 72 lines
predict/predict_final.py, Python, 73 lines
predict/predict_gnn.py, Python, 85 lines
preprocess/1_separate_rs_cifti.sh, Shell, 35 lines
preprocess/2_compute_rs_fingerprint.py, Python, 116 lines
preprocess/3_separate_task_cifti.sh, Shell, 44 lines
preprocess/4_join_all_task_contrasts.py, Python, 52 lines
train/train.py, Python, 152 lines, 1 match
train/train_attn.py, Python, 153 lines
train/train_gnn.py, Python, 155 lines, 1 match
utils/compute_within_and_across_subj_loss.py, Python, 72 lines
utils/dataset.py, Python, 148 lines
utils/experiment.py, Python, 91 lines
utils/gifti_generation.py, Python, 95 lines
utils/parser.py, Python, 182 lines
utils/utilities.py, Python, 140 lines, 1 match
LICENSE, License, 21 lines
README.md, Text, 77 lines

At the source: github.com/braindynamicslab/dl-task-contrast-prediction

ngohgia/brain-surf-cnn

License: none: the authors keep all their rights

State: the link answers, verified on 27 September 2026

Evidence: files inventoried

Commit: 4d2c9fc661e6ec50cbe6e699ea02c0e66d41b14f, 3 May 2022

Languages: Shell (16), Python (12), Jupyter (3)

Size: 58 files, 31 scripts

Software Heritage: archived

Found in: the end of the paper

Holds: README, environment (environment.yml), 3 notebooks

Not found: license file, CITATION.cff, tests, continuous integration, documentation

Tools: NumPy (14 files), PyTorch (8 files), NiBabel (6 files), SciPy (4 files), Matplotlib (3 files), pandas (2 files), seaborn (2 files), scikit-learn (1 file)

Availability: 1 check, the latest on 27 September 2026: the link answers

27 September 2026: the link answers

32 files

data/fs_LR_mesh_templates/scripts/generate_fslr_icospheres.ipynb, Jupyter, 309 lines
experiments/MICCAI2020/posthoc_analysis.ipynb, Jupyter, 210 lines
experiments/MICCAI2020/scripts/1_train.sh, Shell, 32 lines
experiments/MICCAI2020/scripts/2_predict.sh, Shell, 28 lines
experiments/MICCAI2020/scripts/3_predict_on_train_subj.sh, Shell, 28 lines
experiments/MICCAI2020/scripts/4_compute_within_and_across_subj_loss.sh, Shell, 23 lines
experiments/MICCAI2020/scripts/5_finetune_with_rc_loss.sh, Shell, 33 lines
experiments/MICCAI2020/scripts/run_preprocess_pipeline.sh, Shell, 43 lines
experiments/MICCAI2020/scripts/submit.sh, Shell, 3 lines
experiments/test_example/posthoc_analysis.ipynb, Jupyter, 117 lines
experiments/test_example/scripts/0_run_preprocess_pipeline.sh, Shell, 44 lines
experiments/test_example/scripts/1_train.sh, Shell, 32 lines
experiments/test_example/scripts/2_predict.sh, Shell, 29 lines
experiments/test_example/scripts/3_predict_on_train_subj.sh, Shell, 28 lines
experiments/test_example/scripts/4_compute_within_and_across_subj_loss.sh, Shell, 23 lines
experiments/test_example/scripts/5_finetune_with_rc_loss.sh, Shell, 35 lines
experiments/test_example/scripts/submit.sh, Shell, 3 lines
model/brain_surf_cnn.py, Python, 272 lines
model/utils.py, Python, 75 lines
predict.py, Python, 72 lines
preprocess/1_separate_rs_cifti.sh, Shell, 35 lines
preprocess/2_compute_rs_fingerprint.py, Python, 116 lines
preprocess/3_separate_task_cifti.sh, Shell, 42 lines
preprocess/4_join_all_task_contrasts.py, Python, 52 lines
preprocess/sample_train_subj.py, Python, 7 lines
train.py, Python, 140 lines
utils/compute_within_and_across_subj_loss.py, Python, 68 lines
utils/dataset.py, Python, 26 lines
utils/experiment.py, Python, 91 lines
utils/parser.py, Python, 148 lines
utils/utilities.py, Python, 140 lines
README.md, Text, 51 lines

At the source: github.com/ngohgia/brain-surf-cnn

The paper's code and data availability statement is in the Data section.

Tracing map

Proposed by the machine: these links were found in the paper and verified at the source, without human review. The map will receive a Zenodo DOI once one of the paper's authors has validated it with their ORCID.

What the map holds:

2 repositories of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
56 scripts, each with its path and the digest of its content;
5 matches between paragraphs of the paper and lines of the code (method lexical-v1);
neither the text of the paper nor the code itself.

Its JSON (tracing-map.json) is deposited on Zenodo with its DOI once the map is validated.

Data

Datasets cited

humanconnectome.org/study/hcp-young-adult, at Human Connectome Project; found in “Data Availability Statement”

Data Availability Statement

All code used for this article is made publicly available on GitHub at https://github.com/braindynamicslab/dl‐task‐contrast‐prediction (https://github.com/braindynamicslab/dl-task-contrast-prediction). The data that support the findings of this study are available in Human Connectome Project at https://www.humanconnectome.org/study/hcp‐young‐adult/document/hcp‐young‐adult‐2025‐release (https://www.humanconnectome.org/study/hcp-young-adult/document/hcp-young-adult-2025-release). These data were derived from the following resources available in the public domain: https://www.humanconnectome.org/study/hcp‐young‐adult/docume (https://www.humanconnectome.org/study/hcp-young-adult/docume), https://www.humanconnectome.org/study/hcp‐young‐adult/document/hcp‐young‐adult‐2025‐release (https://www.humanconnectome.org/study/hcp-young-adult/document/hcp-young-adult-2025-release).

Reproduced under the paper's license (CC BY), from the paper cited above.

Versions

The history of this record: each version stored by the harvester or made by a correction of its authors or of the maintainers of its code, and what changed in its facts. The texts of the paper (its abstract, its availability statements) are not part of it; versions that changed only those are not listed.

Version 1, 27 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 11 authors, 4 keywords, 11 MeSH terms, 2 funders, 31 references.

Cite

This paper

Madsen, S. J., Lee, Y., Quah, S. K. L., Uddin, L. Q., Mumford, J. A., Barch, D. M., Fair, D. A., Gotlib, I. H., Poldrack, R. A., Kuceyeski, A., & Saggar, M. (2026). Efficient Deep Learning Models for Predicting Individualized Task Activation From Resting-State Functional Connectivity. Human brain mapping, 47(9), e70557. https://doi.org/10.1002/hbm.70557

BibTeX

@article{madsen2026efficient, author = {Madsen, Soren J. and Lee, Young‐Eun and Quah, Shaun K. L. and Uddin, Lucina Q. and Mumford, Jeanette A. and Barch, Deanna M. and Fair, Damien A. and Gotlib, Ian H. and Poldrack, Russell A. and Kuceyeski, Amy and Saggar, Manish}, title = {{Efficient Deep Learning Models for Predicting Individualized Task Activation From Resting-State Functional Connectivity}}, journal = {Human brain mapping}, year = {2026}, month = jun, volume = {47}, number = {9}, pages = {e70557}, publisher = {Wiley}, issn = {1065-9471}, doi = {10.1002/hbm.70557}, url = {https://doi.org/10.1002/hbm.70557}, pmid = {42315989}, pmcid = {PMC13279125} }

RIS

TY - JOUR AU - Madsen, Soren J. AU - Lee, Young‐Eun AU - Quah, Shaun K. L. AU - Uddin, Lucina Q. AU - Mumford, Jeanette A. AU - Barch, Deanna M. AU - Fair, Damien A. AU - Gotlib, Ian H. AU - Poldrack, Russell A. AU - Kuceyeski, Amy AU - Saggar, Manish TI - Efficient Deep Learning Models for Predicting Individualized Task Activation From Resting-State Functional Connectivity T2 - Human brain mapping J2 - Hum Brain Mapp PY - 2026 DA - 2026/06/01 VL - 47 IS - 9 SP - e70557 SN - 1065-9471 PB - Wiley DO - 10.1002/hbm.70557 UR - https://doi.org/10.1002/hbm.70557 LA - en ER -

CSL-JSON

{ "id": "10.1002/hbm.70557", "type": "article-journal", "title": "Efficient Deep Learning Models for Predicting Individualized Task Activation From Resting-State Functional Connectivity", "container-title": "Human brain mapping", "author": [ { "family": "Madsen", "given": "Soren J." }, { "family": "Lee", "given": "Young‐Eun" }, { "family": "Quah", "given": "Shaun K. L." }, { "family": "Uddin", "given": "Lucina Q." }, { "family": "Mumford", "given": "Jeanette A." }, { "family": "Barch", "given": "Deanna M." }, { "family": "Fair", "given": "Damien A." }, { "family": "Gotlib", "given": "Ian H." }, { "family": "Poldrack", "given": "Russell A." }, { "family": "Kuceyeski", "given": "Amy" }, { "family": "Saggar", "given": "Manish" } ], "container-title-short": "Hum Brain Mapp", "volume": "47", "issue": "9", "page": "e70557", "DOI": "10.1002/hbm.70557", "PMID": "42315989", "PMCID": "PMC13279125", "ISSN": "1065-9471", "publisher": "Wiley", "URL": "https://doi.org/10.1002/hbm.70557", "language": "en", "issued": { "date-parts": [ [ 2026, 6, 1 ] ] } }

The tracing map gets a citation of its own once an author has validated it and it has a DOI.

Similar papers

The papers with a page that share the most with this one: the tools found in their code, their categories, datasets, cited references and authors, the rarest counting most.

[1] doi:10.21203/rs.3.rs-9326213/v1 [code]: Multi-task fMRI outperforms resting-state fMRI for revealing task-invariant organization of the human brain
Journal: Research Square (preprint)
In common: neuromaps, Nilearn, NiBabel, 7 other tools, humanconnectome.org/study/hcp-young-adult, fMRI, cognitive, 7 references
[2] doi:10.1016/j.isci.2026.116903 [code]: Neurobiological and behavioral relevance of intrinsic functional connectome constraints on task-evoked neural activation.
Journal: iScience
In common: Nilearn, NiBabel, PyTorch, 4 other tools, humanconnectome.org/study/hcp-young-adult, fMRI, cognitive, 5 references
[3] doi:10.1002/hbm.70533 [code]: Trait-Relevant Tasks Improve Personality Prediction From Structural-Functional Brain Network Coupling.
Journal: Human brain mapping
In common: Nilearn, NiBabel, pandas, 1 other tool, humanconnectome.org/study/hcp-young-adult, cognitive, 5 references
[4] doi:10.1038/s41467-026-72931-6 [code]: Three parsimonious spatiotemporal patterns in cerebellum reveal individual traits in function and behavior.
Journal: Nature communications
In common: Nilearn, NiBabel, PyTorch, 6 other tools, humanconnectome.org/study/hcp-young-adult, fMRI, cognitive, 1 reference
[5] doi:10.1186/s12916-026-04903-y [code]: Structural connectome architecture and biological vulnerability shape cortical atrophy in cocaine use disorder.
Journal: BMC medicine
In common: neuromaps, Nilearn, seaborn, 5 other tools, humanconnectome.org/study/hcp-young-adult, 2 references
[6] doi:10.1016/j.isci.2026.116671 [code]: A high-resolution functional network-organized atlas of human superficial white matter from ultra-high-field diffusion MRI.
Journal: iScience
In common: Nilearn, NiBabel, PyTorch, 5 other tools, humanconnectome.org/study/hcp-young-adult, 2 references
[7] doi:10.1162/netn.a.549 [code]: Distributed cortical network dynamics of binocular convergent eye movements in humans.
Journal: Network neuroscience (Cambridge, Mass.)
In common: NiBabel, Matplotlib, NumPy, fMRI, 7 references
[8] doi:10.64898/2026.08.13.26360304 [code]: Lifespan brain structural variation reveals shared organization across mental health conditions
Journal: medRxiv (preprint)
In common: Nilearn, NiBabel, seaborn, 5 other tools, 1 reference, author Ian H. Gotlib
[9] doi:10.1038/s41467-026-75959-w [code]: Charting higher-order models of brain function beyond pairwise interactions.
Journal: Nature communications
In common: neuromaps, Nilearn, NiBabel, 6 other tools, 3 references
[10] doi:10.1016/j.patter.2026.101619 [code]: Sampling bias corrections for discrete and Gaussian partial information decompositions.
Journal: Patterns (New York, N.Y.)
In common: Nilearn, NiBabel, seaborn, 5 other tools, humanconnectome.org/study/hcp-young-adult, 2 references

Contribute

The authors of this paper can claim it, correct its record and validate its tracing map, and the maintainers of its code (its owner, or a public member of its organization) correct what it says of their repository; anyone signed in can ask for its removal. Every request goes to OSCR's own machine, which answers it; your account page follows them.

Sign in with ORCID to claim this paper as one of its authors, correct its record or validate its tracing map: when the paper's metadata lists your ORCID iD, you are recognized at once. Maintainers of its code: sign in with GitHub, then claim the repository on your account page.

Request its removal

To ask OSCR to remove this record, the copies of its authors' scripts or its tracing map, use the removal request page: signed in, you say who you are, what to remove and why, then review and confirm the request. Published rules decide every request (how).

Discussion, reproductions, activity

Discussion: questions and error reports about this paper and its code, from signed-in readers and its authors. It opens with sign-in.

Reproductions: reports from readers who ran the authors' code: what they reproduced, with which environment, commit and data. It opens with sign-in.

Activity: what happens around this paper: new versions of its record, its map's validation, discussions and reproductions. It opens with sign-in.