OSCR

Radiomics model integrating MRI and ECV enhances prediction accuracy for progression in high-grade glioma.

Code ↔ Paper

2 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 2 matches
  1. [1] § Methods › Model building ↔ Model.py, lines 173–313 · score 0.88 · cls token, LayerNorm, Transformer encoders, dropout, positional, blocks
  2. [2] § Methods › Model building ↔ Model.py, lines 173–313 · score 0.81 · MobileNetV2, ReLU, blocks, encoders, head, ECV images

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 · 313 lines · 10 KB · no license · 2 matches

  1. import nibabel as nib
  2. import numpy as np
  3. from skimage.transform import resize
  5. def resample_to_spacing(img_nii, target_spacing=(3.0, 3.0, 3.0)):
  6. """
  7. 将 NIfTI 图像重采样到指定体素间距 (3mm, 3mm, 3mm)。
  8. img_nii: nibabel Nifti1Image
  9. 返回: resampled_volume (numpy array, float32)
  10. """
  11. volume = img_nii.get_fdata().astype(np.float32)
  12. original_spacing = img_nii.header.get_zooms()[:3] # (sx, sy, sz)
  13. zoom = np.array(original_spacing) / np.array(target_spacing)
  14. # 目标体素数
  15. new_shape = np.round(np.array(volume.shape) * zoom).astype(int)
  17. # 使用 scikit-image resize 实现(三线性插值)
  18. resampled = resize(
  19. volume,
  20. new_shape,
  21. order=1, # 线性插值
  22. mode='edge',
  23. anti_aliasing=False
  24. ).astype(np.float32)
  26. return resampled
  28. def zscore_normalize(volume, eps=1e-8):
  29. """
  30. 对单模态 3D 体数据做 Z-Score 标准化: (x - mean) / std
  31. """
  32. mean = volume.mean()
  33. std = volume.std()
  34. volume_norm = (volume - mean) / (std + eps)
  35. return volume_norm
  37. import random
  38. from scipy.ndimage import rotate
  40. def random_flip_3d(volume):
  41. # 概率 0.5 沿三个维度随机翻转
  42. if random.random() < 0.5:
  43. axis = random.choice([0, 1, 2])
  44. volume = np.flip(volume, axis=axis).copy()
  45. return volume
  47. def random_rotate_3d(volume, max_angle=10):
  48. # 在 [-max_angle, max_angle] 范围随机旋转某一个轴平面
  49. angle = random.uniform(-max_angle, max_angle)
  50. axis = random.choice([(0, 1), (0, 2), (1, 2)]) # 选择旋转平面
  51. volume = rotate(volume, angle=angle, axes=axis, reshape=False, order=1, mode='nearest')
  52. return volume
  54. def random_crop_3d(volume, crop_size=(64, 64, 64)):
  55. """
  56. 简单随机裁剪，假定 volume 形状 >= crop_size
  57. """
  58. z, y, x = volume.shape
  59. cz, cy, cx = crop_size
  60. z1 = random.randint(0, max(0, z - cz))
  61. y1 = random.randint(0, max(0, y - cy))
  62. x1 = random.randint(0, max(0, x - cx))
  63. return volume[z1:z1+cz, y1:y1+cy, x1:x1+cx]
  66. import torch
  67. from torch.utils.data import Dataset
  69. class SingleModality3DDataset(Dataset):
  70. """
  71. 用于 ECV model 或 T2 model：
  72. - image_paths: list[str], 指向 NIfTI 文件路径
  73. - labels: list[int] 或 list[float]，二分类/风险标签
  74. - transform: 可选的 3D 数据增强函数
  75. """
  76. def __init__(self, image_paths, labels, transform=None, target_spacing=(3,3,3), crop_size=(64,64,64)):
  77. self.image_paths = image_paths
  78. self.labels = labels
  79. self.transform = transform
  80. self.target_spacing = target_spacing
  81. self.crop_size = crop_size
  83. def __len__(self):
  84. return len(self.image_paths)
  86. def __getitem__(self, idx):
  87. path = self.image_paths[idx]
  88. label = self.labels[idx]
  90. nii = nib.load(path)
  91. vol = resample_to_spacing(nii, target_spacing=self.target_spacing)
  92. vol = zscore_normalize(vol)
  94. # 数据增强
  95. if self.transform:
  96. vol = self.transform(vol)
  98. # 若需要裁剪
  99. vol = random_crop_3d(vol, crop_size=self.crop_size)
  101. # [D, H, W] -> [1, D, H, W]
  102. vol = np.expand_dims(vol, axis=0)
  103. vol_t = torch.from_numpy(vol).float()
  104. label_t = torch.tensor(label).float()
  106. return vol_t, label_t
  109. from PIL import Image
  110. import torchvision.transforms as T
  112. class Combined2DClinicalDataset(Dataset):
  113. """
  114. 用于 MobHy-Net:
  115. - 2D T2-FLAIR + 2D 虚拟 ECV + 临床特征
  116. """
  117. def __init__(self, t2_paths, ecv_paths, clinical_features, labels, img_size=224):
  118. self.t2_paths = t2_paths
  119. self.ecv_paths = ecv_paths
  120. self.clinical_features = clinical_features
  121. self.labels = labels
  122. self.img_transform = T.Compose([
  123. T.Resize((img_size, img_size)),
  124. T.RandomHorizontalFlip(),
  125. T.RandomRotation(10),
  126. T.ToTensor(),
  127. T.Normalize(mean=[0.5]*3, std=[0.5]*3),
  128. ])
  130. def __len__(self):
  131. return len(self.labels)
  133. def __getitem__(self, idx):
  134. t2_img = Image.open(self.t2_paths[idx]).convert('RGB')
  135. ecv_img = Image.open(self.ecv_paths[idx]).convert('RGB')
  137. t2_tensor = self.img_transform(t2_img) # [3, H, W]
  138. ecv_tensor = self.img_transform(ecv_img) # [3, H, W]
  140. clin = torch.tensor(self.clinical_features[idx], dtype=torch.float32)
  141. label = torch.tensor(self.labels[idx], dtype=torch.float32)
  143. return t2_tensor, ecv_tensor, clin, label
  146. import torch.nn as nn
  147. import math
  149. class PatchEmbedding3D(nn.Module):
  150. """
  151. 将 3D 体数据切分为 3D patches，并映射到 D 维嵌入空间。
  152. 输入: (B, C, D, H, W)
  153. 输出: (B, N_patches, embed_dim)
  154. """
  155. def __init__(self, in_channels=1, patch_size=(4,4,4), embed_dim=128):
  156. super().__init__()
  157. self.patch_size = patch_size
  158. self.proj = nn.Conv3d(
  159. in_channels,
  160. embed_dim,
  161. kernel_size=patch_size,
  162. stride=patch_size
  163. )
  165. def forward(self, x):
  166. # x: (B, C, D, H, W)
  167. x = self.proj(x) # (B, embed_dim, D', H', W')
  168. x = x.flatten(2) # (B, embed_dim, N_patches)
  169. x = x.transpose(1, 2) # (B, N_patches, embed_dim)
  170. return x
  173. class ViT3D(nn.Module):
  174. def __init__(
  175. self,
  176. in_channels=1,
  177. patch_size=(4,4,4),
  178. embed_dim=128,
  179. depth=6,
  180. num_heads=4,
  181. mlp_ratio=4.0,
  182. num_classes=1,
  183. use_cls_token=True,
  184. dropout=0.1
  185. ):
  186. super().__init__()
  187. self.use_cls_token = use_cls_token
  189. self.patch_embed = PatchEmbedding3D(in_channels, patch_size, embed_dim)
  191. # 这里先假设输入体裁剪为固定大小，例如 [1, 64, 64, 64]
  192. # 需要根据 patch_size 推出 N_patches
  193. dummy = torch.zeros(1, in_channels, 64, 64, 64)
  194. with torch.no_grad():
  195. n_patches = self.patch_embed(dummy).shape[1]
  197. # cls token
  198. if use_cls_token:
  199. self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim))
  200. n_positions = n_patches + 1
  201. else:
  202. self.cls_token = None
  203. n_positions = n_patches
  205. # 位置编码
  206. self.pos_embed = nn.Parameter(torch.zeros(1, n_positions, embed_dim))
  207. nn.init.trunc_normal_(self.pos_embed, std=0.02)
  209. self.pos_drop = nn.Dropout(dropout)
  211. self.blocks = nn.ModuleList([
  212. TransformerEncoderBlock(embed_dim, num_heads, mlp_ratio, dropout)
  213. for _ in range(depth)
  214. ])
  216. self.norm = nn.LayerNorm(embed_dim)
  217. self.head = nn.Sequential(
  218. nn.LayerNorm(embed_dim),
  219. nn.Linear(embed_dim, num_classes)
  220. )
  222. def forward(self, x):
  223. # x: (B, C, D, H, W)
  224. B = x.shape[0]
  225. x = self.patch_embed(x) # (B, N, C)
  227. if self.use_cls_token:
  228. cls_tokens = self.cls_token.expand(B, -1, -1) # (B, 1, C)
  229. x = torch.cat((cls_tokens, x), dim=1) # (B, 1+N, C)
  231. x = x + self.pos_embed
  232. x = self.pos_drop(x)
  234. for blk in self.blocks:
  235. x = blk(x)
  237. x = self.norm(x)
  239. if self.use_cls_token:
  240. feat = x[:, 0] # (B, C)
  241. else:
  242. feat = x.mean(dim=1) # mean pooling
  244. logits = self.head(feat).squeeze(-1) # (B,) for num_classes=1
  245. return logits
  248. class MobHyNet(nn.Module):
  249. def __init__(self, clinical_in_dim, num_classes=1, img_feat_dim=512,
  250. attn_heads=4, dropout=0.1):
  251. super().__init__()
  252. # 两个 MobileNetV2 提取器
  253. self.t2_extractor = MobileNetFeatureExtractor(pretrained=True)
  254. self.ecv_extractor = MobileNetFeatureExtractor(pretrained=True)
  256. # 将通道从 2560 -> 512
  257. self.conv_reduce = nn.Conv2d(
  258. in_channels=self.t2_extractor.out_channels * 2,
  259. out_channels=img_feat_dim,
  260. kernel_size=1
  261. )
  262. self.bn_reduce = nn.BatchNorm2d(img_feat_dim)
  263. self.act = nn.ReLU(inplace=True)
  265. # 临床特征 -> 100 维
  266. self.clin_fc = nn.Sequential(
  267. nn.Linear(clinical_in_dim, 100),
  268. nn.ReLU(inplace=True),
  269. nn.Dropout(dropout)
  270. )
  272. # 将图像特征展平为 [B, N, C] 形式
  273. # 注意 N = Hf * Wf, C = img_feat_dim
  274. self.attn_block = SimpleSelfAttnBlock(embed_dim=img_feat_dim + 100,
  275. num_heads=attn_heads,
  276. dropout=dropout)
  278. self.classifier = nn.Sequential(
  279. nn.LayerNorm(img_feat_dim + 100),
  280. nn.Linear(img_feat_dim + 100, 128),
  281. nn.ReLU(inplace=True),
  282. nn.Dropout(dropout),
  283. nn.Linear(128, num_classes)
  284. )
  286. def forward(self, t2_img, ecv_img, clin_feat):
  287. # 图像特征
  288. t2_f = self.t2_extractor(t2_img) # (B, C, Hf, Wf)
  289. ecv_f = self.ecv_extractor(ecv_img) # (B, C, Hf, Wf)
  291. x = torch.cat([t2_f, ecv_f], dim=1) # (B, 2C, Hf, Wf)
  292. x = self.conv_reduce(x)
  293. x = self.bn_reduce(x)
  294. x = self.act(x) # (B, img_feat_dim, Hf, Wf)
  296. B, C, Hf, Wf = x.shape
  297. x = x.view(B, C, -1).transpose(1, 2) # (B, N=Hf*Wf, C)
  299. # 临床特征
  300. clin = self.clin_fc(clin_feat) # (B, 100)
  301. # 扩展到与 N 对齐（简单广播，每个 patch 拼同一临床向量）
  302. clin_expand = clin.unsqueeze(1).expand(B, x.size(1), 100) # (B, N, 100)
  304. x_comb = torch.cat([x, clin_expand], dim=-1) # (B, N, C+100)
  306. # 多头自注意力
  307. x_attn = self.attn_block(x_comb) # (B, N, C+100)
  309. # 这里使用 mean pooling 汇聚
  310. feat = x_attn.mean(dim=1) # (B, C+100)
  312. logits = self.classifier(feat).squeeze(-1) # (B,)
  313. return logits

Model.py at commit 1718af6, no license · at the source

Overview

Authors: Gefei Jiang1,2, Xingjian Sun1,2, Yuchen Zhu1,2, Yinjiao Fei3, Weilin Xu1, Zhichao Jiang1,2, Tianchi Shao1,2, Yuandong Cao1, Liting Li4, Shu Zhou1
  1. Department of Radiation Oncology, The First Affiliated Hospital with Nanjing Medical University,Nanjing, China
  2. The First School of Clinical Medicine, Nanjing Medical University,Nanjing, China
  3. Department of Oncology, Sir Run Run Hospital, Nanjing Medical University,Nanjing, China
  4. Department of Radiation Oncology, Sun Yat-Sen Memorial Hospital, Sun Yat-Sen University,Guangzhou, China
Institutions: Nanjing Medical University (China); Sun Yat-sen University (China)
Journal: NPJ precision oncology, volume 10, issue 1, article 289
Dates: received 24 September 2025; accepted 29 April 2026; published online 13 May 2026
Type: Research article · Language: English
License: CC BY-NC-ND
Identifiers: DOI 10.1038/s41698-026-01475-1 · PMID 42129328 · PMCID PMC13402302 · OpenAlex W7161064856
Open access: gold, a free copy (OpenAlex)
Status: code verified
Categories: structural MRI / diffusion (modality), other condition (population), clinical / translational (subfield)
Methods: Connectivity, Statistics, Machine learning, Preprocessing, fMRI & imaging
Keywords: Biomarkers, Cancer, Computational biology and bioinformatics, Diseases, Oncology
Topic: Radiomics and Machine Learning in Medical Imaging (Radiology, Nuclear Medicine and Imaging, Medicine), according to OpenAlex
Citations: not cited yet (Europe PMC); 53 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 2 matches between paragraphs and lines of code.

410312774/MobHyNet

License: none: the authors keep all their rights
State: the link answers, verified on 28 September 2026
Evidence: files inventoried
Commit: 1718af6e8285a1080bcaa8c87e0779ee3eb380c3, 28 November 2025
Languages: Python (1)
Size: 4 files, 1 script
Software Heritage: not archived
Found in: “Code availability”
Holds: README
Not found: license file, CITATION.cff, environment file, tests, continuous integration, documentation
Tools: NiBabel (1 file), NumPy (1 file), Pillow (1 file), PyTorch (1 file), scikit-image (1 file), SciPy (1 file)
Availability: 1 check, the latest on 28 September 2026: the link answers
  • 28 September 2026: the link answers
2 files
At the source: github.com/410312774/MobHyNet

Code availability statement

The paper has a code 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/10.1038/s41698-026-01475-1.

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;
  • 1 script, each with its path and the digest of its content;
  • 2 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 availability statement

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

  • no repository, dataset or request procedure was recognized in it

Read it in the paper: doi.org/10.1038/s41698-026-01475-1.

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, 28 September 2026: the first record

Recorded: type, language, journal, volume, issue, pages, dates, 10 authors, 5 keywords, 49 references.

Cite

This paper

Jiang, G., Sun, X., Zhu, Y., Fei, Y., Xu, W., Jiang, Z., Shao, T., Cao, Y., Li, L., & Zhou, S. (2026). Radiomics model integrating MRI and ECV enhances prediction accuracy for progression in high-grade glioma. NPJ precision oncology, 10(1), 289. https://doi.org/10.1038/s41698-026-01475-1

BibTeX

@article{jiang2026radiomics,
author = {Jiang, Gefei and Sun, Xingjian and Zhu, Yuchen and Fei, Yinjiao and Xu, Weilin and Jiang, Zhichao and Shao, Tianchi and Cao, Yuandong and Li, Liting and Zhou, Shu},
title = {{Radiomics model integrating MRI and ECV enhances prediction accuracy for progression in high-grade glioma}},
journal = {NPJ precision oncology},
year = {2026},
month = may,
volume = {10},
number = {1},
pages = {289},
publisher = {Nature Publishing Group},
issn = {2397-768X},
doi = {10.1038/s41698-026-01475-1},
url = {https://doi.org/10.1038/s41698-026-01475-1},
pmid = {42129328},
pmcid = {PMC13402302}
}

RIS

TY - JOUR
AU - Jiang, Gefei
AU - Sun, Xingjian
AU - Zhu, Yuchen
AU - Fei, Yinjiao
AU - Xu, Weilin
AU - Jiang, Zhichao
AU - Shao, Tianchi
AU - Cao, Yuandong
AU - Li, Liting
AU - Zhou, Shu
TI - Radiomics model integrating MRI and ECV enhances prediction accuracy for progression in high-grade glioma
T2 - NPJ precision oncology
J2 - NPJ Precis Oncol
PY - 2026
DA - 2026/05/13
VL - 10
IS - 1
SP - 289
SN - 2397-768X
PB - Nature Publishing Group
DO - 10.1038/s41698-026-01475-1
UR - https://doi.org/10.1038/s41698-026-01475-1
LA - en
ER -

CSL-JSON

{
"id": "10.1038/s41698-026-01475-1",
"type": "article-journal",
"title": "Radiomics model integrating MRI and ECV enhances prediction accuracy for progression in high-grade glioma",
"container-title": "NPJ precision oncology",
"author": [
{
"family": "Jiang",
"given": "Gefei"
},
{
"family": "Sun",
"given": "Xingjian"
},
{
"family": "Zhu",
"given": "Yuchen"
},
{
"family": "Fei",
"given": "Yinjiao"
},
{
"family": "Xu",
"given": "Weilin"
},
{
"family": "Jiang",
"given": "Zhichao"
},
{
"family": "Shao",
"given": "Tianchi"
},
{
"family": "Cao",
"given": "Yuandong"
},
{
"family": "Li",
"given": "Liting"
},
{
"family": "Zhou",
"given": "Shu"
}
],
"container-title-short": "NPJ Precis Oncol",
"volume": "10",
"issue": "1",
"page": "289",
"DOI": "10.1038/s41698-026-01475-1",
"PMID": "42129328",
"PMCID": "PMC13402302",
"ISSN": "2397-768X",
"publisher": "Nature Publishing Group",
"URL": "https://doi.org/10.1038/s41698-026-01475-1",
"language": "en",
"issued": {
"date-parts": [
[
2026,
5,
13
]
]
}
}

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:10.2463/mrms.mp.2024-0149 [code]
Image Distortion Correction for Diffusion MR Imaging Using a Transformer-based U-Net.
Journal: Magnetic resonance in medical sciences : MRMS : an official journal of Japan Society of Magnetic Resonance in Medicine
In common: scikit-image, Pillow, NiBabel, 3 other tools, structural MRI / diffusion, clinical / translational
[2] doi:10.1002/alz.71649 [code]
Postmortem brain MRI reveals differential associations of subcortical and limbic volumes with cortical thinning and neurodegenerative pathologies.
Journal: Alzheimer's & dementia : the journal of the Alzheimer's Association
In common: scikit-image, Pillow, NiBabel, 3 other tools, structural MRI / diffusion, other condition
[3] doi:10.1371/journal.pone.0344600 [code]
Robust disease prognosis via diagnostic knowledge preservation: A sequential learning approach.
Journal: PloS one
In common: scikit-image, Pillow, NiBabel, 3 other tools, clinical / translational, other condition
[4] doi:10.1038/s41598-026-48496-1 [code]
A unified FLAIR hyperintensity segmentation model for various CNS tumor types and acquisition time points.
Journal: Scientific reports
In common: scikit-image, Pillow, NiBabel, 2 other tools, structural MRI / diffusion, other condition, 1 reference
[5] doi:10.1126/sciadv.aed4944 [code]
Focused ultrasound blood-brain barrier opening reveals a paradoxical remote metabolic response in the primate brain.
Journal: Science advances
In common: scikit-image, NiBabel, PyTorch, 2 other tools, structural MRI / diffusion, clinical / translational, 1 reference
[6] doi:10.1002/hbm.70542 [code]
The In Vivo Microstructural Profile of Human Hippocampal Subfield CA1 and Its Relation to Memory Performance.
Journal: Human brain mapping
In common: scikit-image, Pillow, NiBabel, 2 other tools, structural MRI / diffusion, 1 reference
[7] 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: scikit-image, NiBabel, PyTorch, 2 other tools, structural MRI / diffusion, other condition, 1 reference
[8] doi:10.1162/imag.a.1326 [code]
RAVEN: Robust, generalizable, multi-resolution structural MRI upsampling using autoencoders.
Journal: Imaging neuroscience (Cambridge, Mass.)
In common: scikit-image, Pillow, NiBabel, 3 other tools, structural MRI / diffusion
[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: scikit-image, Pillow, NiBabel, 3 other tools, structural MRI / diffusion
[10] doi:10.3390/jimaging12070276 [code]
Hyperelastic Regularization for Near-Diffeomorphic Transformer-Based Brain MRI Registration.
Journal: Journal of imaging
In common: scikit-image, Pillow, NiBabel, 3 other tools, structural MRI / diffusion

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.