OSCR

MedGAN-SSM: Multimodal brain image synthesis network integration using SSM empowered GAN.

Code ↔ Paper

7 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 7 matches · 1 of them tie a paragraph to a whole file, not to given lines: a weak match, whose lines are not tinted
  1. [1] § STAR★Methods › Quantification and statistical analysis › Evaluation metrics ↔ trainer.py, lines 69–136 · score 0.73 · Peak Signal, Absolute Error, Noise Ratio, metrics, SSIM, MAE
  2. [2] § STAR★Methods › Quantification and statistical analysis › Evaluation metrics ↔ tester.py, the whole file · a weak match · score 0.63 · Peak Signal, Noise Ratio, Error, metrics, SSIM, MAE
  3. [3] § STAR★Methods › Method details › State space construction and representation › Global information fusion ↔ models/generator.py, lines 4–20 · score 0.61 · state space construction, global contextual, GAP, linear, map, fusion
  4. [4] § STAR★Methods › Method details › Multi-dimensional S6 module design ↔ models/generator.py, lines 49–69 · score 0.57 · S6 module, residual blocks, model
  5. [5] § STAR★Methods › Method details › Multi-dimensional S6 module design › Cross-dimensional feature fusion ↔ models/generator.py, lines 105–181 · score 0.55 · MRI slices, available modality, decoder, S6
  6. [6] § STAR★Methods › Method details ↔ models/generator.py, lines 4–20 · score 0.53 · global context, state space, adaptive, maps, fusion, channel
  7. [7] § Results › Comparison with SOTA methods ↔ trainer.py, lines 69–136 · score 0.50 · absolute error, target modality, predicted, metrics, SSIM, MAE

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 authors' code

Python · 181 lines · 6.3 KB · MIT · 4 matches

  1. import torch
  2. import torch.nn as nn
  4. class StateSpaceModule(nn.Module):
  5. """State Space Construction Module"""
  6. def __init__(self, in_channels, state_dim):
  7. super().__init__()
  8. self.conv = nn.Conv2d(in_channels, state_dim, 1)
  9. self.gap = nn.AdaptiveAvgPool2d(1)
  10. self.fc = nn.Linear(in_channels, state_dim)
  12. def forward(self, x):
  13. # Basic state mapping
  14. s = torch.relu(self.conv(x))
  16. # Global context fusion
  17. g = self.gap(x).squeeze(-1).squeeze(-1)
  18. g = torch.relu(self.fc(g))[:, :, None, None]
  20. return s + g # Modulated state Ŝ
  22. class DynamicGate(nn.Module):
  23. """Dynamic Gating Module"""
  24. def __init__(self, in_channels, ratio=16):
  25. super().__init__()
  26. # Spatial gating branch
  27. self.spatial = nn.Sequential(
  28. nn.Conv2d(in_channels, in_channels//ratio, 3, padding=1),
  29. nn.ReLU(),
  30. NonLocalBlock(in_channels//ratio),
  31. nn.Conv2d(in_channels//ratio, 1, 3, padding=1),
  32. nn.Sigmoid()
  33. )
  35. # Channel gating branch
  36. self.channel = nn.Sequential(
  37. nn.AdaptiveAvgPool2d(1),
  38. nn.Conv2d(in_channels, in_channels//ratio, 1),
  39. nn.ReLU(),
  40. nn.Conv2d(in_channels//ratio, in_channels, 1),
  41. nn.Sigmoid()
  42. )
  44. def forward(self, x):
  45. g_spatial = self.spatial(x)
  46. g_channel = self.channel(x)
  47. return g_spatial * 0.6 + g_channel * 0.4 # Weighted fusion
  49. class MultiScaleS6(nn.Module):
  50. """Multi-scale S6 Module"""
  51. def __init__(self, channels):
  52. super().__init__()
  53. # Local branch
  54. self.local = nn.Sequential(
  55. ResidualBlock(channels, dilation=1),
  56. ResidualBlock(channels, dilation=2),
  57. ResidualBlock(channels, dilation=4)
  58. )
  60. # Global branch
  61. self.global_branch = nn.Sequential(
  62. nn.Conv2d(channels, channels, 7, padding=3),
  63. NonLocalBlock(channels)
  64. )
  66. def forward(self, x, gate):
  67. local_feat = self.local(x)
  68. global_feat = self.global_branch(x)
  69. return x + gate*local_feat + (1-gate)*global_feat
  71. class ResidualBlock(nn.Module):
  72. """Residual Block"""
  73. def __init__(self, channels, dilation=1):
  74. super().__init__()
  75. self.conv = nn.Sequential(
  76. nn.Conv2d(channels, channels, 3, padding=dilation, dilation=dilation),
  77. nn.InstanceNorm2d(channels),
  78. nn.ReLU(),
  79. nn.Conv2d(channels, channels, 3, padding=dilation, dilation=dilation),
  80. nn.InstanceNorm2d(channels))
  82. def forward(self, x):
  83. return x + self.conv(x)
  85. class NonLocalBlock(nn.Module):
  86. """Non-local Attention Module"""
  87. def __init__(self, channels):
  88. super().__init__()
  89. self.theta = nn.Conv2d(channels, channels//8, 1)
  90. self.phi = nn.Conv2d(channels, channels//8, 1)
  91. self.g = nn.Conv2d(channels, channels//8, 1)
  92. self.out = nn.Conv2d(channels//8, channels, 1)
  94. def forward(self, x):
  95. batch_size = x.size(0)
  96. theta = self.theta(x).view(batch_size, -1, x.size(2)*x.size(3))
  97. phi = self.phi(x).view(batch_size, -1, x.size(2)*x.size(3)).permute(0,2,1)
  98. g = self.g(x).view(batch_size, -1, x.size(2)*x.size(3))
  100. attn = torch.softmax(torch.bmm(theta, phi), dim=-1)
  101. out = torch.bmm(attn, g)
  102. out = out.view(batch_size, -1, x.size(2), x.size(3))
  103. return self.out(out) + x
  105. class Generator(nn.Module):
  106. """
  107. 2D Image Generator - Processing Multi-modal MRI Slices
  108. Input: 2D multi-modal MRI slices (variable channels, depending on available modalities)
  109. Output: Single modality 2D MRI slice
  110. """
  111. def __init__(self, in_channels=None, out_channels=1):
  112. super().__init__()
  113. if in_channels is None:
  114. in_channels = 3
  116. # Encoder
  117. self.encoder1 = nn.Sequential(
  118. nn.Conv2d(in_channels, 64, 4, stride=2, padding=1),
  119. nn.LeakyReLU(0.2),
  120. StateSpaceModule(64, 64)
  121. )
  122. self.encoder2 = nn.Sequential(
  123. nn.Conv2d(64, 128, 4, stride=2, padding=1),
  124. nn.InstanceNorm2d(128),
  125. nn.LeakyReLU(0.2)
  126. )
  127. self.encoder3 = nn.Sequential(
  128. nn.Conv2d(128, 256, 4, stride=2, padding=1),
  129. nn.InstanceNorm2d(256),
  130. nn.LeakyReLU(0.2)
  131. )
  132. self.encoder4 = nn.Sequential(
  133. nn.Conv2d(256, 512, 4, stride=2, padding=1),
  134. nn.InstanceNorm2d(512),
  135. nn.LeakyReLU(0.2)
  136. )
  138. # Core processing modules
  139. self.process = nn.Sequential(
  140. MultiScaleS6(512),
  141. DynamicGate(512),
  142. MultiScaleS6(512)
  143. )
  145. # Decoder
  146. self.decoder1 = nn.Sequential(
  147. nn.ConvTranspose2d(512, 256, 4, stride=2, padding=1),
  148. nn.InstanceNorm2d(256),
  149. nn.ReLU()
  150. )
  151. self.decoder2 = nn.Sequential(
  152. nn.ConvTranspose2d(512, 128, 4, stride=2, padding=1), # 512 = 256 + 256 (skip connection)
  153. nn.InstanceNorm2d(128),
  154. nn.ReLU()
  155. )
  156. self.decoder3 = nn.Sequential(
  157. nn.ConvTranspose2d(256, 64, 4, stride=2, padding=1), # 256 = 128 + 128 (skip connection)
  158. nn.InstanceNorm2d(64),
  159. nn.ReLU()
  160. )
  161. self.decoder4 = nn.Sequential(
  162. nn.ConvTranspose2d(128, out_channels, 4, stride=2, padding=1), # 128 = 64 + 64 (skip connection)
  163. nn.Tanh()
  164. )
  166. def forward(self, x):
  167. enc1 = self.encoder1(x)
  168. enc2 = self.encoder2(enc1)
  169. enc3 = self.encoder3(enc2)
  170. enc4 = self.encoder4(enc3)
  172. gate = self.process[1](enc4)
  173. proc = self.process[0](enc4, gate)
  174. proc = self.process[2](proc, gate)
  176. dec1 = self.decoder1(proc)
  177. dec2 = self.decoder2(torch.cat([dec1, enc3], dim=1)) # Skip connection
  178. dec3 = self.decoder3(torch.cat([dec2, enc2], dim=1)) # Skip connection
  179. dec4 = self.decoder4(torch.cat([dec3, enc1], dim=1)) # Skip connection
  181. return dec4

generator.py at commit 624db77, under MIT · at the source

Overview

Authors: Tianming Song1, Mingzhi Wang2, Zhe Ren1, Wensi Li3, Jian Zhang1, Kexin Li1
ORCID iDs: Mingzhi Wang
  1. School of Integrated Circuit, Wuxi Vocational College of Science and Technology, Wuxi 214028, China
  2. School of Software, Dalian University of Technology, Dalian 116024, China
  3. Department of Emergency Internal Medicine, Heilongjiang Hospital of Beijing Children’s Hospital Affiliated to Capital Medical University, Harbin 150010, China
Institutions: Wuxi Vocational College of Science and Technology; Dalian University of Technology (China); Capital Medical University (China); Beijing Children’s Hospital (China)
Journal: iScience, volume 29, issue 6, article 115924
Dates: received 10 September 2025; accepted 24 April 2026; published online 30 April 2026
Type: Research article · Language: English
License: CC BY-NC
Identifiers: DOI 10.1016/j.isci.2026.115924 · PMID 42181291 · PMCID PMC13196568 · OpenAlex W7160127035
Open access: gold, a free copy (OpenAlex)
Status: code verified
Methods: Connectivity, Statistics, Machine learning
Keywords: Neuroscience, Computational bioinformatics
Topic: EEG and Brain-Computer Interfaces (Cognitive Neuroscience, Neuroscience), according to OpenAlex
Funding: General Project of Basic Science; Higher Education Institutions of Jiangsu Province (24KJB510047)
Citations: not cited yet (Europe PMC); 47 references in the paper

Abstract

Multimodal medical image synthesis is essential for addressing data scarcity and incomplete modality acquisition in clinical imaging. This study presents MedGAN-SSM, a framework that enhances multimodal brain MRI synthesis by integrating generative adversarial networks (GANs) with state space modeling (SSM). MedGAN-SSM uses a state space module to capture global semantic information via cross-layer transmission. A dynamic attention gate adjusts spatial and channel features, allowing the model to target relevant areas, while a multi-dimensional S6 module merges local and global features across scales to enhance synthesis quality and anatomical consistency. Experiments on the BraTS2020 and IXI datasets show that MedGAN-SSM outperforms existing methods in PSNR, SSIM, and MAE, demonstrating robustness in missing-modality scenarios while generating high-fidelity images that enhance segmentation performance. Overall, MedGAN-SSM reliably synthesizes high-quality multimodal brain MRI, preserving anatomical details while aiding automated analyses and clinical workflows in incomplete imaging conditions.

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

Repository

Its files are read in the Code ↔ Paper reader above, with 7 matches between paragraphs and lines of code.

Charles-wmz/MedGAN-SSM

License: MIT
State: the link answers, verified on 30 September 2026
Evidence: files inventoried
Commit: 624db770579c5807bdc10f6e83d1789b1a3504cb, 7 December 2025
Languages: Python (14)
Size: 27 files, 14 scripts
Software Heritage: not archived
Found in: “Data and code availability”
Holds: README, license file, environment (requirements.txt)
Not found: CITATION.cff, tests, continuous integration, documentation
Tools: NumPy (7 files), PyTorch (7 files), scikit-image (3 files), SimpleITK (3 files), NiBabel (2 files), Matplotlib (1 file)
Availability: 1 check, the latest on 30 September 2026: the link answers
  • 30 September 2026: the link answers
16 files
At the source: github.com/Charles-wmz/MedGAN-SSM

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:

  • 1 repository of the authors' code, each at its verified commit, with its license and how the link was found in the paper;
  • 14 scripts, each with its path and the digest of its content;
  • 7 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

No dataset and no data link were found in the paper.

Data and code availability

• The code for the proposed method is publicly available at https://github.com/Charles-wmz/MedGAN-SSM • The datasets used in this study (BraTS and IXI) are publicly available from their official repositories. • Any additional information required to reanalyze the data reported in this study is available from the lead contact upon request.

Reproduced under the paper's license (CC BY-NC), 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, 30 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 6 authors, 2 keywords, 2 funders, 37 references.

Cite

This paper

Song, T., Wang, M., Ren, Z., Li, W., Zhang, J., & Li, K. (2026). MedGAN-SSM: Multimodal brain image synthesis network integration using SSM empowered GAN. iScience, 29(6), 115924. https://doi.org/10.1016/j.isci.2026.115924

BibTeX

@article{song2026medgan,
author = {Song, Tianming and Wang, Mingzhi and Ren, Zhe and Li, Wensi and Zhang, Jian and Li, Kexin},
title = {{MedGAN-SSM: Multimodal brain image synthesis network integration using SSM empowered GAN}},
journal = {iScience},
year = {2026},
month = apr,
volume = {29},
number = {6},
pages = {115924},
publisher = {Elsevier},
issn = {2589-0042},
doi = {10.1016/j.isci.2026.115924},
url = {https://doi.org/10.1016/j.isci.2026.115924},
pmid = {42181291},
pmcid = {PMC13196568}
}

RIS

TY - JOUR
AU - Song, Tianming
AU - Wang, Mingzhi
AU - Ren, Zhe
AU - Li, Wensi
AU - Zhang, Jian
AU - Li, Kexin
TI - MedGAN-SSM: Multimodal brain image synthesis network integration using SSM empowered GAN
T2 - iScience
J2 - iScience
PY - 2026
DA - 2026/04/30
VL - 29
IS - 6
SP - 115924
SN - 2589-0042
PB - Elsevier
DO - 10.1016/j.isci.2026.115924
UR - https://doi.org/10.1016/j.isci.2026.115924
LA - en
ER -

CSL-JSON

{
"id": "10.1016/j.isci.2026.115924",
"type": "article-journal",
"title": "MedGAN-SSM: Multimodal brain image synthesis network integration using SSM empowered GAN",
"container-title": "iScience",
"author": [
{
"family": "Song",
"given": "Tianming"
},
{
"family": "Wang",
"given": "Mingzhi"
},
{
"family": "Ren",
"given": "Zhe"
},
{
"family": "Li",
"given": "Wensi"
},
{
"family": "Zhang",
"given": "Jian"
},
{
"family": "Li",
"given": "Kexin"
}
],
"container-title-short": "iScience",
"volume": "29",
"issue": "6",
"page": "115924",
"DOI": "10.1016/j.isci.2026.115924",
"PMID": "42181291",
"PMCID": "PMC13196568",
"ISSN": "2589-0042",
"publisher": "Elsevier",
"URL": "https://doi.org/10.1016/j.isci.2026.115924",
"language": "en",
"issued": {
"date-parts": [
[
2026,
4,
30
]
]
}
}

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:10.1038/s41467-026-69578-8 [code]
Contrast-free identification of glioma blood-brain barrier status via generative diffusion AI and non-contrast MRI.
Journal: Nature communications
In common: SimpleITK, scikit-image, NiBabel, 2 other tools, 1 reference
[2] doi:10.1007/s12021-026-09817-x [code]
Circle of Willis-Guided Localization for Simultaneous Detection and Classification of Large Vessel Occlusions in Brain CTA.
Journal: Neuroinformatics
In common: SimpleITK, scikit-image, NiBabel, 3 other tools
[3] doi:10.1016/j.isci.2026.117446 [code]
Volume-inflation registration (INFREG) for morphometric analysis of human focal cortical dysplasia type II.
Journal: iScience
In common: SimpleITK, scikit-image, NiBabel, 3 other tools
[4] doi:10.1002/hipo.70124 [code]
Association Between Anterior Hippocampal Gyrification and Episodic Memory Performance in Neurotypical Young Adults.
Journal: Hippocampus
In common: SimpleITK, scikit-image, NiBabel, 3 other tools
[5] doi:10.21037/qims-2026-0792 [code]
An nnU-Net-based framework with adaptive feature representation for 3D brain tumor segmentation.
Journal: Quantitative imaging in medicine and surgery
In common: SimpleITK, scikit-image, NiBabel, 3 other tools
[6] doi:10.1162/imag.a.1326 [code]
RAVEN: Robust, generalizable, multi-resolution structural MRI upsampling using autoencoders.
Journal: Imaging neuroscience (Cambridge, Mass.)
In common: SimpleITK, scikit-image, NiBabel, 3 other tools
[7] doi:10.3389/fmed.2026.1875760 [code]
Adaptive multi-stage domain unlearning for white-matter lesion segmentation.
Journal: Frontiers in medicine
In common: SimpleITK, scikit-image, NiBabel, 3 other tools
[8] doi:10.1038/s41467-026-76011-7 [code]
Human cortex organizes dynamic co-fluctuations along the sensorimotor-association axis.
Journal: Nature communications
In common: SimpleITK, scikit-image, NiBabel, 3 other tools
[9] doi:10.1371/journal.pcbi.1014555 [code]
Body surface potential driven personalisation of electrophysiological digital twins in hypertrophic cardiomyopathy.
Journal: PLoS computational biology
In common: SimpleITK, scikit-image, NiBabel, 3 other tools
[10] doi:10.64898/2026.07.15.26357954 [code]
Portable Ultra-Low Field MRI Deep-Learning Algorithms for White Matter Lesion Segmentation Improve Accuracy and Reflect Clinical Disability in Multiple Sclerosis
Journal: medRxiv (preprint)
In common: SimpleITK, scikit-image, NiBabel, 3 other tools

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.