OSCR

SelExNet: A Self-Supervised Physics-Informed Framework for Multi-Channel Joint RF and Gradient Waveform Optimization in 2D Spatially Selective Excitation.

Code ↔ Paper

4 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 4 matches
  1. [1] § Methods › Multi‐Channel Joint RF‐Gradient Optimization Network › Gradient Module: Parameterized Gradient Waveform Generation Network ↔ src/selexnet/model.py, lines 59–121 · score 0.75 · variable density, spiral trajectory, space trajectory, angular, radial, Gradient
  2. [2] § Methods › Loss Functions › Gradient Constraint Loss ↔ src/selexnet/train.py, lines 473–497 · score 0.61 · slew rate, Gradient amplitude, penalty, smooth, loss
  3. [3] § Methods › Loss Functions ↔ finetune.py, lines 73–139 · score 0.59 · weight decay, AdamW, trainable, training, optimizes
  4. [4] § Methods › Loss Functions ↔ main.py, lines 42–82 · score 0.54 · weight decay, AdamW, training, optimizes

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 · 562 lines · 16 KB · no license · 1 match

  1. #!/usr/bin/env python
  2. # coding=utf-8
  3. """
  4. Author : Chris Xiao [email hidden]
  5. Date : 2024-09-21 01:52:10
  6. LastEditors : Chris Xiao [email hidden]
  7. LastEditTime : 2025-03-12 00:14:32
  8. FilePath : /Documents/sTx_B0_1/src/model.py
  9. Description : DeepControlV2 network architecture
  10. I Love IU
  11. Copyright (c) 2024 by Chris Xiao [email hidden], All Rights Reserved.
  12. """
  14. import torch
  15. import torch.nn as nn
  16. import torch.nn.functional as F
  17. from typing import Any
  19. DEFAULT_NUM_GROUP = 32
  22. # Include the get_num_groups function as defined earlier
  23. def get_num_groups(num_channels, default_num_groups=32):
  24. num_groups = min(default_num_groups, num_channels)
  25. while num_groups > 0:
  26. if num_channels % num_groups == 0:
  27. return num_groups
  28. num_groups -= 1
  29. return 1 # Fallback to 1 if no divisor is found
  32. def _get_norm_layer(num_channels, norm_type, eps=1e-8, is_fc=False):
  33. if is_fc:
  34. return nn.RMSNorm(num_channels, eps=eps)
  35. if norm_type == "group":
  36. num_groups = get_num_groups(num_channels, DEFAULT_NUM_GROUP)
  37. return nn.GroupNorm(num_groups, num_channels, eps=eps, affine=True)
  38. elif norm_type == "instance":
  39. return nn.InstanceNorm2d(num_channels, eps=eps, affine=True)
  40. elif norm_type == "batch":
  41. return nn.BatchNorm2d(num_channels, eps=eps, affine=True)
  42. else:
  43. raise ValueError(f"Unsupported normalization type: {norm_type}")
  46. def _get_act_layer(act_type):
  47. if act_type == "swish":
  48. return MemoryEfficientSwish()
  49. elif act_type == "relu":
  50. return nn.ReLU()
  51. elif act_type == "leaky_relu":
  52. return nn.LeakyReLU(negative_slope=0.1)
  53. elif act_type == "gelu":
  54. return nn.GELU()
  55. else:
  56. raise ValueError(f"Unsupported activation type: {act_type}")
  59. def generate_spiral_trajectory_gradients(params, cfg):
  60. """
  61. Generates a spiral trajectory based on parameters.
  63. Args:
  64. num_points (int): Number of time points.
  65. k_max (float): Maximum k-space radius.
  66. params (dict): Contains 'n_turns', 'alpha', 'beta'.
  68. Returns:
  69. kx, ky (torch.Tensor): K-space trajectories.
  70. """
  71. n_turns = params[:, 0].unsqueeze(1) * float(cfg.magnet.ktraj.n_turns[0]) + float(
  72. cfg.magnet.ktraj.n_turns[1]
  73. )
  74. alpha = params[:, 1].unsqueeze(1) * float(cfg.magnet.ktraj.alpha[0]) + float(
  75. cfg.magnet.ktraj.alpha[1]
  76. )
  77. beta = params[:, 2].unsqueeze(1) * float(cfg.magnet.ktraj.beta[0]) + float(
  78. cfg.magnet.ktraj.beta[1]
  79. )
  80. kmax_factor = params[:, 3].unsqueeze(1) * float(
  81. cfg.magnet.ktraj.kmax_factor[0]
  82. ) + float(cfg.magnet.ktraj.kmax_factor[1])
  83. T = cfg.magnet.tp
  84. num_points = cfg.model.rf_output_dim // 2
  85. gamma = cfg.magnet.gamma / (2 * torch.pi) # rad⋅s^-1⋅T^-1
  86. fov = cfg.image.fov # m
  87. N = cfg.image.N # pixels
  88. dt = T / num_points
  89. alpha = alpha.expand(-1, num_points)
  90. beta = beta.expand(-1, num_points)
  92. t = torch.linspace(0, T, num_points, device=params.device).unsqueeze(0)
  93. t_normalized = t / T # Normalize to [0, 1]
  95. # Radial component with variable density
  96. k_max = (
  97. torch.sqrt(
  98. torch.tensor(cfg.magnet.tx, dtype=params.dtype, device=params.device)
  99. )
  100. * N[0]
  101. ) / (kmax_factor * 2.0 * fov[0]) # 1/m
  102. r = k_max * (1.0 - torch.pow(t_normalized, alpha))
  104. # Angular component with variable speed
  105. theta = 2.0 * torch.pi * n_turns * torch.pow(t_normalized, beta)
  107. # Cartesian coordinates
  108. kx = r * torch.cos(theta) # 1/m
  109. ky = r * torch.sin(theta) # 1/m
  111. dkx = torch.gradient(kx, spacing=dt, dim=-1)[0] # rad/m/s
  112. dky = torch.gradient(ky, spacing=dt, dim=-1)[0] # rad/m/s
  114. # Compute gradients
  115. Gx = dkx / gamma # T/m
  116. Gy = dky / gamma # T/m
  117. Gx = Gx.repeat_interleave(repeats=2, dim=-1) # [B, T*2]
  118. Gy = Gy.repeat_interleave(repeats=2, dim=-1) # [B, T*2]
  119. Gz = torch.zeros_like(Gx).requires_grad_(False) # T/m
  121. return Gx, Gy, Gz
  124. # Memory-efficient Swish activation function
  125. class SwishImplementation(torch.autograd.Function):
  126. @staticmethod
  127. def forward(ctx: Any, x: torch.Tensor) -> torch.Tensor:
  128. result = x * torch.sigmoid(x)
  129. ctx.save_for_backward(x)
  130. return result
  132. @staticmethod
  133. def backward(ctx: Any, grad_outputs: torch.Tensor) -> torch.Tensor: # type: ignore
  134. x = ctx.saved_tensors[0]
  135. sigmoid_x = torch.sigmoid(x)
  136. grad_input = grad_outputs * (sigmoid_x * (1 + x * (1 - sigmoid_x)))
  137. return grad_input
  140. class MemoryEfficientSwish(nn.Module):
  141. def forward(self, x):
  142. return SwishImplementation.apply(x)
  145. # Bottleneck block with GroupNorm and optional dropout
  146. class Bottleneck(nn.Module):
  147. expansion = 4
  149. def __init__(
  150. self,
  151. in_channels,
  152. out_channels,
  153. stride=1,
  154. downsample=None,
  155. norm="group",
  156. act="swish",
  157. dropout_rate=0.1,
  158. ):
  159. super(Bottleneck, self).__init__()
  160. width = out_channels // self.expansion
  162. self.conv1 = nn.Conv2d(in_channels, width, kernel_size=1, bias=False)
  163. self.dropout1 = nn.Dropout2d(p=dropout_rate)
  164. self.norm1 = _get_norm_layer(width, norm)
  165. self.act1 = _get_act_layer(act)
  167. self.conv2 = nn.Conv2d(
  168. width, width, kernel_size=3, stride=stride, padding=1, bias=False
  169. )
  170. self.dropout2 = nn.Dropout2d(p=dropout_rate)
  171. self.norm2 = _get_norm_layer(width, norm)
  172. self.act2 = _get_act_layer(act)
  174. self.conv3 = nn.Conv2d(width, out_channels, kernel_size=1, bias=False)
  175. self.norm3 = _get_norm_layer(out_channels, norm)
  177. self.act3 = _get_act_layer(act)
  178. self.downsample = downsample
  180. def forward(self, x):
  181. identity = x
  183. out = self.conv1(x)
  184. out = self.norm1(out)
  185. out = self.act1(out)
  186. out = self.dropout1(out)
  188. out = self.conv2(out)
  189. out = self.norm2(out)
  190. out = self.act2(out)
  191. out = self.dropout2(out)
  193. out = self.conv3(out)
  194. out = self.norm3(out)
  196. if self.downsample is not None:
  197. identity = self.downsample(x)
  199. out = out + identity
  200. out = self.act3(out)
  202. return out
  205. # Modified DCNV2 model with GroupNorm and dropout
  206. class RFNet(nn.Module):
  207. def __init__(
  208. self,
  209. input_dim=1,
  210. output_dim=2000,
  211. tx=8,
  212. dropout=0.5,
  213. norm="group",
  214. act="swish",
  215. block=[3, 4, 6, 3],
  216. ):
  217. super(RFNet, self).__init__()
  218. self.in_channels = 64
  219. self.tx = tx
  221. # Initial convolutional layer
  222. self.conv1 = nn.Conv2d(
  223. input_dim, self.in_channels, kernel_size=7, stride=2, padding=3, bias=False
  224. )
  225. self.norm1 = _get_norm_layer(self.in_channels, norm)
  226. self.act1 = _get_act_layer(act)
  227. self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
  229. # Define layers using the Bottleneck block
  230. self.layer1 = self._make_layer(
  231. Bottleneck,
  232. 64,
  233. blocks=block[0],
  234. dropout=dropout,
  235. stride=1,
  236. norm=norm,
  237. act=act,
  238. )
  239. self.layer2 = self._make_layer(
  240. Bottleneck,
  241. 128,
  242. blocks=block[1],
  243. dropout=dropout,
  244. stride=2,
  245. norm=norm,
  246. act=act,
  247. )
  248. self.layer3 = self._make_layer(
  249. Bottleneck,
  250. 256,
  251. blocks=block[2],
  252. dropout=dropout,
  253. stride=2,
  254. norm=norm,
  255. act=act,
  256. )
  257. self.layer4 = self._make_layer(
  258. Bottleneck,
  259. 512,
  260. blocks=block[3],
  261. dropout=dropout,
  262. stride=1,
  263. norm=norm,
  264. act=act,
  265. )
  267. # Global average pooling
  268. self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
  270. # Fully connected layers for amplitude and phase
  271. self.dropout = nn.Dropout(dropout)
  272. self.fc_amp = nn.Linear(512 * Bottleneck.expansion, output_dim * tx)
  273. self.fc_phase = nn.Linear(512 * Bottleneck.expansion, output_dim * tx)
  275. # Initialize weights
  276. self._initialize_weights()
  278. def _make_layer(
  279. self, block, planes, blocks, dropout=0.5, stride=1, norm="group", act="swish"
  280. ):
  281. downsample = None
  282. out_channels = planes * block.expansion
  284. if stride != 1 or self.in_channels != out_channels:
  285. downsample = nn.Sequential(
  286. nn.Conv2d(
  287. self.in_channels,
  288. out_channels,
  289. kernel_size=1,
  290. stride=stride,
  291. bias=False,
  292. ),
  293. _get_norm_layer(out_channels, norm),
  294. )
  296. layers = []
  297. layers.append(
  298. block(
  299. self.in_channels,
  300. out_channels,
  301. stride,
  302. downsample,
  303. norm=norm,
  304. act=act,
  305. dropout_rate=dropout,
  306. )
  307. )
  308. self.in_channels = out_channels
  309. for _ in range(1, blocks):
  310. layers.append(
  311. block(
  312. self.in_channels,
  313. out_channels,
  314. norm=norm,
  315. act=act,
  316. dropout_rate=dropout,
  317. )
  318. )
  320. return nn.Sequential(*layers)
  322. def forward(self, x):
  323. B = x.size(0)
  324. # Initial layers
  325. x = self.conv1(x)
  326. x = self.norm1(x)
  327. x = self.act1(x)
  328. x = self.maxpool(x)
  330. # Residual layers
  331. x = self.layer1(x)
  332. x = self.layer2(x)
  333. x = self.layer3(x)
  334. x = self.layer4(x)
  336. # Pooling and fully connected layers
  337. x = self.avgpool(x)
  338. x = torch.flatten(x, 1)
  339. x = self.dropout(x)
  341. # Real and Imaginary outputs
  342. re = self.fc_amp(x)
  343. im = self.fc_phase(x)
  345. re_out = re.view(B, self.tx, -1)
  346. im_out = im.view(B, self.tx, -1)
  348. return re_out, im_out
  350. def _initialize_weights(self):
  351. for m in self.modules():
  352. if isinstance(m, (nn.Conv2d, nn.Linear)):
  353. nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
  354. if m.bias is not None:
  355. nn.init.constant_(m.bias, 0)
  356. elif isinstance(m, (nn.GroupNorm, nn.InstanceNorm2d, nn.BatchNorm2d)):
  357. nn.init.constant_(m.weight, 1)
  358. nn.init.constant_(m.bias, 0)
  361. class GradNet(nn.Module):
  362. def __init__(
  363. self,
  364. cfg,
  365. input_dim=1,
  366. output_dim=4,
  367. dropout=0.5,
  368. norm="group",
  369. act="swish",
  370. block=[3, 4, 6, 3],
  371. ):
  372. super(GradNet, self).__init__()
  373. self.cfg = cfg
  374. self.in_channels = 64
  376. # Initial convolutional layer
  377. self.conv1 = nn.Conv2d(
  378. input_dim, self.in_channels, kernel_size=7, stride=2, padding=3, bias=False
  379. )
  380. self.norm1 = _get_norm_layer(self.in_channels, norm)
  381. self.act1 = _get_act_layer(act)
  382. self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
  384. # Define layers using the Bottleneck block
  385. self.layer1 = self._make_layer(
  386. Bottleneck,
  387. 64,
  388. blocks=block[0],
  389. dropout=dropout,
  390. stride=1,
  391. norm=norm,
  392. act=act,
  393. )
  394. self.layer2 = self._make_layer(
  395. Bottleneck,
  396. 128,
  397. blocks=block[1],
  398. dropout=dropout,
  399. stride=2,
  400. norm=norm,
  401. act=act,
  402. )
  403. self.layer3 = self._make_layer(
  404. Bottleneck,
  405. 256,
  406. blocks=block[2],
  407. dropout=dropout,
  408. stride=2,
  409. norm=norm,
  410. act=act,
  411. )
  412. self.layer4 = self._make_layer(
  413. Bottleneck,
  414. 512,
  415. blocks=block[3],
  416. dropout=dropout,
  417. stride=1,
  418. norm=norm,
  419. act=act,
  420. )
  422. # Global average pooling
  423. self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
  425. # Fully connected layers for amplitude and phase
  426. self.dropout = nn.Dropout(dropout)
  427. self.fc_1 = nn.Linear(512 * Bottleneck.expansion, 512)
  428. self.fc_norm1 = _get_norm_layer(512, norm, is_fc=True)
  429. self.fc_act1 = _get_act_layer(act)
  430. self.fc_2 = nn.Linear(512, 128)
  431. self.fc_norm2 = _get_norm_layer(128, norm, is_fc=True)
  432. self.fc_act2 = _get_act_layer(act)
  433. self.fc_3 = nn.Linear(128, 32)
  434. self.fc_norm3 = _get_norm_layer(32, norm, is_fc=True)
  435. self.fc_act3 = _get_act_layer(act)
  436. self.fc_grad = nn.Linear(32, output_dim)
  437. self.fc_grad_act = nn.Sigmoid()
  439. # Initialize weights
  440. self._initialize_weights()
  442. def _make_layer(
  443. self, block, planes, blocks, dropout=0.5, stride=1, norm="group", act="swish"
  444. ):
  445. downsample = None
  446. out_channels = planes * block.expansion
  448. if stride != 1 or self.in_channels != out_channels:
  449. downsample = nn.Sequential(
  450. nn.Conv2d(
  451. self.in_channels,
  452. out_channels,
  453. kernel_size=1,
  454. stride=stride,
  455. bias=False,
  456. ),
  457. _get_norm_layer(out_channels, norm),
  458. )
  460. layers = []
  461. layers.append(
  462. block(
  463. self.in_channels,
  464. out_channels,
  465. stride,
  466. downsample,
  467. norm=norm,
  468. act=act,
  469. dropout_rate=dropout,
  470. )
  471. )
  472. self.in_channels = out_channels
  473. for _ in range(1, blocks):
  474. layers.append(
  475. block(
  476. self.in_channels,
  477. out_channels,
  478. norm=norm,
  479. act=act,
  480. dropout_rate=dropout,
  481. )
  482. )
  484. return nn.Sequential(*layers)
  486. def forward(self, x):
  487. # Initial layers
  488. x = self.conv1(x)
  489. x = self.norm1(x)
  490. x = self.act1(x)
  491. x = self.maxpool(x)
  493. # Residual layers
  494. x = self.layer1(x)
  495. x = self.layer2(x)
  496. x = self.layer3(x)
  497. x = self.layer4(x)
  499. # Pooling and fully connected layers
  500. x = self.avgpool(x)
  501. x = torch.flatten(x, 1)
  502. x = self.dropout(x)
  504. # Real and Imaginary outputs
  505. x = self.fc_1(x)
  506. x = self.fc_norm1(x)
  507. x = self.fc_act1(x)
  508. x = self.fc_2(x)
  509. x = self.fc_norm2(x)
  510. x = self.fc_act2(x)
  511. x = self.fc_3(x)
  512. x = self.fc_norm3(x)
  513. x = self.fc_act3(x)
  514. x = self.fc_grad(x)
  515. ktraj = self.fc_grad_act(x)
  517. gx, gy, gz = generate_spiral_trajectory_gradients(ktraj, self.cfg)
  519. return gx, gy, gz
  521. def _initialize_weights(self):
  522. for m in self.modules():
  523. if isinstance(m, (nn.Conv2d, nn.Linear)):
  524. nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity="relu")
  525. if m.bias is not None:
  526. nn.init.constant_(m.bias, 0)
  527. elif isinstance(
  528. m, (nn.GroupNorm, nn.InstanceNorm2d, nn.BatchNorm2d, nn.LayerNorm)
  529. ):
  530. nn.init.constant_(m.weight, 1)
  531. nn.init.constant_(m.bias, 0)
  534. class SelExNet(nn.Module):
  535. def __init__(self, cfg):
  536. super(SelExNet, self).__init__()
  537. self.rfnet = RFNet(
  538. input_dim=cfg.model.in_dim,
  539. output_dim=cfg.model.rf_output_dim,
  540. tx=cfg.magnet.tx,
  541. dropout=cfg.train.dropout,
  542. norm=cfg.model.norm,
  543. act=cfg.model.act,
  544. block=cfg.model.block,
  545. )
  546. if cfg.train.joint:
  547. self.gnet = GradNet(
  548. cfg,
  549. input_dim=cfg.model.in_dim,
  550. output_dim=cfg.model.grad_output_dim,
  551. dropout=cfg.train.dropout,
  552. norm=cfg.model.norm,
  553. act=cfg.model.act,
  554. block=cfg.model.block,
  555. )
  557. def forward(self, x):
  558. rf_re, rf_im = self.rfnet(x)
  559. if not hasattr(self, "gnet"):
  560. return rf_re, rf_im, None, None, None
  561. gx, gy, gz = self.gnet(x)
  562. return rf_re, rf_im, gx, gy, gz

model.py at commit ef28e27, no license · at the source

Overview

Authors: Yuliang Xiao1,2, Jason Rock1,2, Zhe Wu3, Jamie Near1,2, Mark Chiew1,2, Simon J. Graham1,2
ORCID iDs: Yuliang Xiao, Jason Rock, Zhe Wu, Jamie Near, Mark Chiew, Simon J. Graham
  1. Physical Sciences Platform Sunnybrook Research Institute Toronto Ontario Canada
  2. Department of Medical Biophysics, Temerty Faculty of Medicine University of Toronto Toronto Ontario Canada
  3. Siemens Healthcare Limited Oakville Ontario Canada
Institutions: University of Toronto (Canada); Sunnybrook Research Institute; Siemens (Canada) (Canada)
Journal: Magnetic resonance in medicine, volume 96, issue 3, pages 1219-1234
Dates: received 5 December 2025; accepted 2 May 2026; published online 14 May 2026; in print September 2026
Type: Research article · Language: English
License: CC BY-NC-ND
Identifiers: DOI 10.1002/mrm.70431 · PMID 42136067 · PMCID PMC13327443 · OpenAlex W7161247274
Open access: hybrid, a free copy (OpenAlex)
Status: code verified
Categories: structural MRI / diffusion (modality), human (organism), cellular / molecular (subfield)
Methods: Connectivity, Smoothing, state filtering, decompositions, Evoked potentials, Machine learning, fMRI & imaging
Keywords: Bloch equations, joint optimization, multi‐channel transmit, RF pulse, self‐supervised learning, variable‐density spiral trajectory
MeSH: Brain*, Image Processing, Computer-Assisted*, Magnetic Resonance Imaging*, Algorithms, Humans, Phantoms, Imaging, Radio Waves (* major topic)
Journal subjects: Imaging Methodology
Topic: Advanced MRI Techniques and Applications (Radiology, Nuclear Medicine and Imaging, Medicine), according to OpenAlex
Funding: Natural Sciences and Engineering Research Council of Canada (RGPIN‐2023‐03410, RGPIN‐2023‐04408); Canadian Institutes of Health Research (PJT‐186038)
Citations: not cited yet (Europe PMC); 57 references in the paper

Abstract

The abstract is not reproduced here: the paper's license (CC BY-NC-ND) does not allow it. Read it in the paper, at the publisher or on Europe PMC.

Repository

Its files are read in the Code ↔ Paper reader above, with 4 matches between paragraphs and lines of code.

chiew-group/SelExNet

License: none: the authors keep all their rights
State: the link answers, verified on 28 September 2026
Evidence: files inventoried
Commit: ef28e2706b41325f09c90d603f15dca2a2101dc9, 13 September 2026
Languages: Python (11)
Size: 21 files, 11 scripts
Software Heritage: not archived
Found in: “Data Availability Statement”
Holds: README, environment (pyproject.toml, uv.lock)
Not found: license file, CITATION.cff, tests, continuous integration, documentation
Tools: PyTorch (10 files), NumPy (4 files), SciPy (3 files), Matplotlib (2 files), Pillow (2 files), OpenCV (1 file)
Availability: 1 check, the latest on 28 September 2026: the link answers
  • 28 September 2026: the link answers
12 files
At the source: github.com/chiew-group/SelExNet.git

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;
  • 11 scripts, each with its path and the digest of its content;
  • 4 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.

Code and data availability statement

The paper has a code and data availability statement. Its license (CC BY-NC-ND) does not allow reproducing it here; in short, from what the harvester recognized in it:

Read it in the paper: doi.org/10.1002/mrm.70431.

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 2, 28 September 2026

  • Publisher: n/a → Wiley

Version 1, 28 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 6 authors, 6 keywords, 7 MeSH terms, 2 funders, 55 references.

Cite

This paper

Xiao, Y., Rock, J., Wu, Z., Near, J., Chiew, M., & Graham, S. J. (2026). SelExNet: A Self-Supervised Physics-Informed Framework for Multi-Channel Joint RF and Gradient Waveform Optimization in 2D Spatially Selective Excitation. Magnetic resonance in medicine, 96(3), 1219-1234. https://doi.org/10.1002/mrm.70431

BibTeX

@article{xiao2026selexnet,
author = {Xiao, Yuliang and Rock, Jason and Wu, Zhe and Near, Jamie and Chiew, Mark and Graham, Simon J.},
title = {{SelExNet: A Self-Supervised Physics-Informed Framework for Multi-Channel Joint RF and Gradient Waveform Optimization in 2D Spatially Selective Excitation}},
journal = {Magnetic resonance in medicine},
year = {2026},
month = may,
volume = {96},
number = {3},
pages = {1219--1234},
publisher = {Wiley},
issn = {0740-3194},
doi = {10.1002/mrm.70431},
url = {https://doi.org/10.1002/mrm.70431},
pmid = {42136067},
pmcid = {PMC13327443}
}

RIS

TY - JOUR
AU - Xiao, Yuliang
AU - Rock, Jason
AU - Wu, Zhe
AU - Near, Jamie
AU - Chiew, Mark
AU - Graham, Simon J.
TI - SelExNet: A Self-Supervised Physics-Informed Framework for Multi-Channel Joint RF and Gradient Waveform Optimization in 2D Spatially Selective Excitation
T2 - Magnetic resonance in medicine
J2 - Magn Reson Med
PY - 2026
DA - 2026/05/14
VL - 96
IS - 3
SP - 1219
EP - 1234
SN - 0740-3194
PB - Wiley
DO - 10.1002/mrm.70431
UR - https://doi.org/10.1002/mrm.70431
LA - en
ER -

CSL-JSON

{
"id": "10.1002/mrm.70431",
"type": "article-journal",
"title": "SelExNet: A Self-Supervised Physics-Informed Framework for Multi-Channel Joint RF and Gradient Waveform Optimization in 2D Spatially Selective Excitation",
"container-title": "Magnetic resonance in medicine",
"author": [
{
"family": "Xiao",
"given": "Yuliang"
},
{
"family": "Rock",
"given": "Jason"
},
{
"family": "Wu",
"given": "Zhe"
},
{
"family": "Near",
"given": "Jamie"
},
{
"family": "Chiew",
"given": "Mark"
},
{
"family": "Graham",
"given": "Simon J."
}
],
"container-title-short": "Magn Reson Med",
"volume": "96",
"issue": "3",
"page": "1219-1234",
"DOI": "10.1002/mrm.70431",
"PMID": "42136067",
"PMCID": "PMC13327443",
"ISSN": "0740-3194",
"publisher": "Wiley",
"URL": "https://doi.org/10.1002/mrm.70431",
"language": "en",
"issued": {
"date-parts": [
[
2026,
5,
14
]
]
}
}

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:10.1002/nbm.70263 [code]
Cross-Site Generalization of CNN-Based $$ {B}_1^{+} $$ Mapping in UHF MRI.
Journal: NMR in biomedicine
In common: PyTorch, SciPy, Matplotlib, 1 other tool, structural MRI / diffusion, 5 references
[2] doi:10.3389/frai.2026.1771088 [code]
Few-shot deployment of pretrained MRI transformers in brain imaging tasks.
Journal: Frontiers in artificial intelligence
In common: OpenCV, Pillow, PyTorch, 3 other tools, structural MRI / diffusion, 1 reference
[3] doi:10.1038/s41598-026-53152-9 [code]
Layer-specific feature preference in a trained AlexNet model revealed by stylized image analysis.
Journal: Scientific reports
In common: OpenCV, Pillow, PyTorch, 2 other tools, 2 references
[4] doi:10.1007/s00429-026-03166-w [code]
Autoencoders for unsupervised analysis of rat myeloarchitecture.
Journal: Brain structure & function
In common: OpenCV, Pillow, PyTorch, 3 other tools, cellular / molecular, 1 reference
[5] doi:10.1167/jov.26.8.1 [code]
MAME: Multidimensional adaptive metamer exploration with human perceptual feedback.
Journal: Journal of vision
In common: OpenCV, Pillow, PyTorch, 3 other tools, 1 reference
[6] doi:10.1371/journal.pcbi.1014263 [code]
MIRAGE: Robust multi-modal architectures translate fMRI-to-image models from vision to mental imagery.
Journal: PLoS computational biology
In common: OpenCV, Pillow, PyTorch, 3 other tools, 1 reference
[7] doi:10.1371/journal.pone.0344600 [code]
Robust disease prognosis via diagnostic knowledge preservation: A sequential learning approach.
Journal: PloS one
In common: OpenCV, Pillow, PyTorch, 3 other tools, 1 reference
[8] doi:10.7554/elife.107933 [code]
Modality-agnostic decoding of vision and language from fMRI.
Journal: eLife
In common: OpenCV, Pillow, PyTorch, 3 other tools, 1 reference
[9] doi:10.1016/j.patter.2026.101538 [code]
A multi-modal foundation model for brain disease diagnosis and medical imaging.
Journal: Patterns (New York, N.Y.)
In common: OpenCV, Pillow, PyTorch, 3 other tools, 1 reference
[10] doi:10.1016/j.isci.2026.116168 [code]
See the small lesions: Frequency-guided spatial debiasing GAN for multimodal medical image fusion.
Journal: iScience
In common: OpenCV, Pillow, PyTorch, 3 other tools, 1 reference

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.