OSCR

Generating synthetic tau-PET scans in Alzheimer’s disease from MRI, blood biomarkers and demographics with deep learning

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. [1] § Methods › Model development and training › U-Net blocks ↔ src/models/unet3d.py, lines 158–183 · score 0.73 · transposed convolutions, decoder blocks, deeper, stride, upsampling, connection
  2. [2] § Methods › Participants ↔ notebooks/tau_load_plots.ipynb, lines 19–52 · score 0.70 · PREVENT AD, BACS, LLCF, LZAX, OASIS, UCSF
  3. [3] § Methods › Model development and training › U-Net architecture ↔ src/models/unet3d.py, lines 158–183 · score 0.64 · skip connections, decoder block, upsamples, layers, convolutional, encoder
  4. [4] § Methods › Model development and training › Loss function ↔ src/models/loss.py, lines 7–38 · score 0.61 · masked L1 loss, absolute, error, scalar, voxels, brain
  5. [5] § Methods › Participants ↔ notebooks/tau_spat_sim_plots.ipynb, lines 20–41 · score 0.60 · Berkeley, LLCF, LZAX, OASIS, UCSF, subset
  6. [6] § Methods › Training ↔ src/models/train.py, lines 116–221 · score 0.53 · validation loss, model weights, optimization, batch, training
  7. [7] § Methods › Model development and training › Loss function ↔ src/models/train.py, lines 12–68 · score 0.50 · masked L1 loss, scalar, brain, weight, predicted, model

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 · 318 lines · 9 KB · CC-BY-NC-4.0 · 2 matches

  1. """
  2. 3D U-Net architecture for synthetic tau-PET generation.
  3. """
  5. import numpy as np
  6. import tensorflow as tf
  8. from keras import Model
  9. from keras.layers import (
  10. Activation,
  11. BatchNormalization,
  12. Conv3D,
  13. Conv3DTranspose,
  14. Cropping3D,
  15. Dense,
  16. Input,
  17. LeakyReLU,
  18. MaxPooling3D,
  19. Reshape,
  20. ZeroPadding3D,
  21. add,
  22. concatenate,
  23. )
  26. def conv_block(inputs, num_filters, kernel_size=(3, 3, 3), activation=None, padding="same"):
  27. """
  28. Apply 3D convolution, batch normalization, and activation.
  30. Parameters
  31. ----------
  32. inputs : tf.Tensor
  33. Input tensor.
  34. num_filters : int
  35. Number of convolutional filters.
  36. kernel_size : tuple of int, optional
  37. Convolution kernel size. Default is (3, 3, 3).
  38. activation : keras.layers.Layer or None, optional
  39. Activation layer. Default is LeakyReLU().
  40. padding : str, optional
  41. Padding mode for convolution. Default is "same".
  43. Returns
  44. -------
  45. x : tf.Tensor
  46. Output tensor after convolution, batch norm, and activation.
  47. """
  48. if activation is None:
  49. activation = LeakyReLU()
  51. x = Conv3D(num_filters, kernel_size, padding=padding)(inputs)
  52. x = BatchNormalization(fused=False)(x)
  53. x = Activation(activation)(x)
  54. return x
  57. def residual_block(inputs, num_filters):
  58. """
  59. Residual 3D convolutional block with skip connection.
  61. Two convolutions with batch normalization and LeakyReLU activations,
  62. with a skip connection adding the input to the output.
  64. Parameters
  65. ----------
  66. inputs : tf.Tensor
  67. Input tensor with number of channels matching num_filters.
  68. num_filters : int
  69. Number of convolutional filters.
  71. Returns
  72. -------
  73. x : tf.Tensor
  74. Output tensor after residual connection.
  75. """
  76. x = Conv3D(num_filters, (3, 3, 3), padding="same")(inputs)
  77. x = BatchNormalization(fused=False)(x)
  78. x = Activation(LeakyReLU())(x)
  80. x = Conv3D(num_filters, (3, 3, 3), padding="same")(x)
  81. x = BatchNormalization(fused=False)(x)
  83. x = add([x, inputs])
  84. x = Activation(LeakyReLU())(x)
  85. return x
  88. def attention_block(inputs):
  89. """
  90. Self-attention block for spatial features.
  92. Computes scaled dot-product attention over spatial voxels, designed for
  93. bottleneck feature maps with small spatial dimensions.
  95. Parameters
  96. ----------
  97. inputs : tf.Tensor
  98. Input feature map.
  100. Returns
  101. -------
  102. output : tf.Tensor
  103. Output tensor after attention mechanism and skip connection.
  104. """
  105. query = Conv3D(filters=inputs.shape[-1] // 8, kernel_size=1)(inputs)
  106. key = Conv3D(filters=inputs.shape[-1] // 8, kernel_size=1)(inputs)
  107. value = Conv3D(filters=inputs.shape[-1], kernel_size=1)(inputs)
  108. attention_weights = tf.nn.softmax(tf.matmul(query, key, transpose_b=True) / tf.sqrt(tf.cast(inputs.shape[-1] // 8, dtype=tf.float32)))
  109. output = tf.matmul(attention_weights, value)
  110. output = add([output, inputs])
  112. return output
  115. def down_block(inputs, num_filters):
  116. """
  117. Encoder block combining convolution and residual connection.
  119. Parameters
  120. ----------
  121. inputs : tf.Tensor
  122. Input tensor.
  123. num_filters : int
  124. Number of convolutional filters.
  126. Returns
  127. -------
  128. x : tf.Tensor
  129. Output tensor after conv and residual blocks.
  130. """
  131. x = conv_block(inputs, num_filters)
  132. x = residual_block(x, num_filters)
  133. return x
  136. def bottleneck_block(inputs, num_filters):
  137. """
  138. Bottleneck block with attention.
  140. Parameters
  141. ----------
  142. inputs : tf.Tensor
  143. Input tensor.
  144. num_filters : int
  145. Number of convolutional filters.
  147. Returns
  148. -------
  149. x : tf.Tensor
  150. Output tensor after conv, attention, and conv blocks.
  151. """
  152. x = conv_block(inputs, num_filters)
  153. x = attention_block(x)
  154. x = conv_block(x, num_filters)
  155. return x
  158. def up_block(inputs, skip_connection, num_filters):
  159. """
  160. Decoder block with transposed convolution and skip connection.
  162. Applies transposed convolution to upsample, concatenates with skip connection,
  163. and applies conv and residual blocks.
  165. Parameters
  166. ----------
  167. inputs : tf.Tensor
  168. Input tensor (features from deeper layer).
  169. skip_connection : tf.Tensor
  170. Encoder feature map from corresponding down_block.
  171. num_filters : int
  172. Number of convolutional filters.
  174. Returns
  175. -------
  176. x : tf.Tensor
  177. Output tensor after upsampling, concatenation, and residual blocks.
  178. """
  179. x = Conv3DTranspose(num_filters, (2, 2, 2), strides=(2, 2, 2), padding="same")(inputs)
  180. x = concatenate([x, skip_connection], axis=4)
  181. x = conv_block(x, num_filters)
  182. x = residual_block(x, num_filters)
  183. return x
  186. def covariate_bottleneck_layer(covariate_input, spatial_shape=(4, 4, 4), hidden_units=1024, name="covariate"):
  187. """
  188. Project scalar covariates into a 3D spatial feature map.
  190. Passes the input through dense layers and reshapes the output to match
  191. the specified spatial dimensions for concatenation with CNN features.
  193. Parameters
  194. ----------
  195. covariate_input : tf.Tensor, shape (batch, 1)
  196. Scalar covariate value (e.g., plasma, age).
  197. spatial_shape : tuple of int, optional
  198. Spatial dimensions of the output feature map. Default is (4, 4, 4).
  199. hidden_units : int, optional
  200. Number of units in the hidden dense layer. Default is 1024.
  201. name : str, optional
  202. Name prefix for the layers. Default is "covariate".
  204. Returns
  205. -------
  206. x : tf.Tensor, shape (batch, *spatial_shape, 1)
  207. Reshaped feature map ready for concatenation.
  208. """
  209. num_voxels = int(np.prod(spatial_shape))
  211. if name == "plasma":
  212. dense1_name, dense2_name = "data1", "data2"
  213. elif name == "age":
  214. dense1_name, dense2_name = "data3", "data4"
  215. else:
  216. dense1_name, dense2_name = f"{name}_dense1", f"{name}_dense2"
  218. x = Dense(hidden_units, activation="relu", name=f"{name}_dense1")(covariate_input)
  219. x = Dense(num_voxels, activation="relu", name=f"{name}_dense2")(x)
  220. x = Reshape((*spatial_shape, 1), name=f"{name}_reshape")(x)
  221. return x
  224. def build_tau_pet_unet(
  225. image_shape=(72, 90, 76, 1),
  226. padding=(28, 19, 26),
  227. encoder_filters=(32, 64, 128, 256, 512),
  228. bottleneck_filters=1024,
  229. covariate_shape=(4, 4, 4),
  230. output_activation=None,
  231. verbose=False,
  232. ):
  233. """
  234. Build a multimodal 3D U-Net for synthetic tau-PET generation.
  236. Constructs a U-Net encoder-decoder architecture with residual and attention
  237. blocks. Takes MRI and scalar covariates (plasma, age) as inputs and outputs
  238. synthetic tau-PET images. Covariates are projected into 3D bottleneck features
  239. and concatenated with CNN features.
  241. Parameters
  242. ----------
  243. image_shape : tuple of int, optional
  244. Input MRI image shape (H, W, D, channels). Default is (72, 90, 76, 1).
  245. padding : tuple of int, optional
  246. Zero padding applied to MRI input (H_pad, W_pad, D_pad). Default is (28, 19, 26).
  247. encoder_filters : tuple of int, optional
  248. Number of filters at each encoder level. Default is (32, 64, 128, 256, 512).
  249. bottleneck_filters : int, optional
  250. Number of filters in the bottleneck. Default is 1024.
  251. covariate_shape : tuple of int, optional
  252. Spatial shape of covariate bottleneck features. Default is (4, 4, 4).
  253. output_activation : keras.layers.Layer or None, optional
  254. Output activation layer. Default is LeakyReLU().
  255. verbose : bool, optional
  256. If True, print model summary. Default is False.
  258. Returns
  259. -------
  260. model : keras.Model
  261. Compiled U-Net model with inputs [mri, plasma, age] and output tau-PET.
  262. """
  263. if output_activation is None:
  264. output_activation = LeakyReLU()
  266. mri_input = Input(shape=image_shape, name="mri_input")
  267. plasma_input = Input(shape=(1,), name="plasma_input")
  268. age_input = Input(shape=(1,), name="age_input")
  270. x = ZeroPadding3D(padding=padding)(mri_input)
  272. skips = []
  273. for num_filters in encoder_filters:
  274. x = down_block(x, num_filters)
  275. skips.append(x)
  276. x = MaxPooling3D((2, 2, 2))(x)
  278. x = bottleneck_block(x, bottleneck_filters)
  280. plasma_features = covariate_bottleneck_layer(
  281. plasma_input,
  282. spatial_shape=covariate_shape,
  283. hidden_units=bottleneck_filters,
  284. name="plasma",
  285. )
  287. age_features = covariate_bottleneck_layer(
  288. age_input,
  289. spatial_shape=covariate_shape,
  290. hidden_units=bottleneck_filters,
  291. name="age",
  292. )
  294. x = concatenate([x, plasma_features], axis=4)
  295. x = concatenate([x, age_features], axis=4)
  297. for skip_connection, num_filters in zip(reversed(skips), reversed(encoder_filters)):
  298. x = up_block(x, skip_connection, num_filters)
  300. x = Conv3D(
  301. 1,
  302. (1, 1, 1),
  303. activation=output_activation,
  304. bias_initializer="zeros",
  305. )(x)
  307. output = Cropping3D(cropping=padding)(x)
  309. model = Model(
  310. inputs=[mri_input, plasma_input, age_input],
  311. outputs=output,
  312. name="tau_pet_unet",
  313. )
  315. if verbose:
  316. model.summary()
  318. return model

unet3d.py at commit 4226694, under CC-BY-NC-4.0 · at the source

Overview

Authors: Linda Karlsson1, Olof Strandberg1, Ruben Smith1,2, Weizhong Tang1, Ida Arvidsson3, Kalle Åström3, Kevin Oliveira Hauer1,2, Shorena Janelidze1, Erik Stomrud1,2, Sebastian Palmqvist1,2, Philip B. Verghese4, Joel B. Braunstein4, Alzheimer’s Disease Neuroimaging Initiative, PREVENT-AD Research Group, Gregory Klein5, Sergey Shcherbinin6, William J. Jagust7, Sylvia Villeneuve8, Renaud La Joie9, Gil D. Rabinovici9,10, Niklas Mattsson-Carlgren1,2, Jacob W. Vogel11, Oskar Hansson1
ORCID iDs: Linda Karlsson, Renaud La Joie, Oskar Hansson
  1. Clinical Memory Research Unit, Department of Clinical Sciences in Malmö, Lund University, Lund, Sweden
  2. Memory Clinic, Skåne University Hospital, Malmö, Sweden
  3. Centre for Mathematical Sciences, Lund University, Lund, Sweden
  4. C2N Diagnostics LLC, St Louis, MO, USA
  5. Pharma Research and Early Development, F Hoffmann-La Roche Ltd., Basel, Switzerland
  6. Eli Lilly and Company, Indianapolis, Indiana, USA
  7. Department of Neuroscience, University of California, Berkeley, California,USA
  8. Centre for Studies in the Prevention of Alzheimer’s Disease, Douglas Mental Health Institute, McGill University, Montreal, QC, Canada
  9. Department of Neurology, Edward and Pearl Fein Memory and Aging Center, Weill Institute for Neurosciences, University of California, San Francisco, California, USA
  10. Department of Radiology and Biomedical Imaging, Edward and Pearl Fein Memory and Aging Center, Weill Institute for Neurosciences, University of California, San Francisco, California, USA
  11. Clinical Memory Research Unit, SciLifeLab, Department of Clinical Sciences in Malmö, Lund University, Lund, Sweden
Institutions: Lund University (Sweden); Skåne University Hospital (Sweden); C2N Diagnostics (United States) (United States); Roche (Switzerland) (Switzerland); Eli Lilly and Company, Indianapolis, Indiana, USA; University of California, Berkeley (United States); McGill University (Canada); University of California, San Francisco (United States)
Journal: medRxiv (preprint)
Dates: published online 7 May 2026
Type: Preprint · Language: English
License: CC BY
Identifiers: DOI 10.64898/2026.05.06.26352540 · OpenAlex W7160551948
Open access: green, a free copy (OpenAlex)
Status: code verified
Categories: structural MRI / diffusion (modality), other (modality), PET / SPECT (modality), human (organism), Alzheimer's / dementia (population), clinical / translational (subfield)
Methods: Connectivity, Statistics, Preprocessing, Machine learning, fMRI & imaging
Topic: Dementia and Cognitive Impairment Research (Psychiatry and Mental health, Medicine), according to OpenAlex
Funding: Michael J. Fox Foundation for Parkinson's Research (MJFF‐025507, 2022-Projekt0107); Alzheimer's Association (ZEN24-1069572, SG-23-1061717); Eli Lilly and Company; Cure Alzheimer’s Fund; Australian Government (2022-Projekt0107, 2022-Projekt0080); GHR Foundation (2022-00775); Lunds Universitet (AF-980907); Knut och Alice Wallenbergs Stiftelse (KAW 2020.0239, (2022-0231), 2022-Projekt0080, 2022-00775, AF-980907, 2022-1259, FO2021-0293, 2020-O000028, 1412/22, 2020.0239); Vetenskapsrådet (ERAPERMED2021-184, 2024-03642, 2022-00775, 2022-Projekt0080, FO2021-0293, AF-980907, 2018, 1412/22, 2018-02052, 2021-02219, 2020-O000028, 2022-1259); Konung Gustaf V:s och Drottning Victorias Frimurarestiftelse (2020-O000028, WASP/DDLS22-066); Science for Life Laboratory (KAW 2020.0239); Skånes universitetssjukhus (2022-1259, 2020-O000028, 2022-Projekt0080); Stiftelsen Bundy Academy; University of California, San Francisco; Parkinsonfonden (1412/22)
Citations: not cited yet (Europe PMC); 82 references in the paper

Abstract

Tau protein aggregation in the brain is a hallmark of Alzheimer’s disease (AD). Positron emission tomography (PET) is the only in vivo method to visualize tau pathology and estimate both its burden and regional distribution, but the use of tau-PET is constrained by high cost and limited accessibility. Here, we develop a deep learning model to synthesize tau-PET scans from more accessible data: structural magnetic resonance imaging (MRI), demographics, and when available, blood biomarkers. We included 5,191 participants across the AD continuum or with another neurological disorder from 13 cohorts (mean age 70 years, 51% female) and optimized a 3D U-Net neural network with residual and attention units for this task. In held-out test data, synthetic tau-PET reliably modeled tau burden, with correlations of R=0.77–0.86 with true tau-PET across individuals in common AD regions of interest. Spatial similarity between synthetic and true tau-PET was likewise high, with mean regional correlation of R=0.75. Synthetic scans also captured clinically meaningful prognostic information comparable to true tau-PET, including distinction between early (HR=12, p<0.001) and late (HR=45, p<0.001) stages of tau accumulation. These findings demonstrate that clinically informative synthetic tau-PET scans can be generated from widely available modalities using deep learning, potentially offering a scalable and cost-effective approach for estimating tau AD pathology in the brain.

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

DeMONLab-BioFINDER/karlsson_synthetic_taupet

License: CC-BY-NC-4.0
State: the link answers, verified on 28 September 2026
Evidence: files inventoried
Commit: 4226694b0d3ca9d15e3a427f1f0bc49d0259a05e, 29 June 2026
Languages: Python (15), Jupyter (4), Shell (1)
Size: 59 files, 20 scripts
Software Heritage: not archived
Found in: “Code availability”
Holds: README, license file, environment (environment.yml, pyproject.toml, requirements.txt), 4 notebooks
Not found: CITATION.cff, tests, continuous integration, documentation
Tools: NumPy (11 files), TensorFlow (10 files), NiBabel (7 files), pandas (7 files), Matplotlib (6 files), scikit-learn (4 files), Keras (3 files), SciPy (3 files), scikit-image (2 files), seaborn (2 files)
Availability: 1 check, the latest on 28 September 2026: the link answers
  • 28 September 2026: the link answers
22 files
At the source: github.com/DeMONLab-BioFINDER/karlsson_synthetic_taupet

Code availability

The code and model will be made available at https://github.com/DeMONLab-BioFINDER/karlsson_synthetic_taupet.

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

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;
  • 20 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 availability

Participant data from open access cohorts (ADNI, A4, OASIS and PREVENT-AD) can be obtained from http://adni.loni.usc.edu/, https://ida.loni.usc.edu/, https://sites.wustl.edu/oasisbrains/, and https://openpreventad.loris.ca/. Pseudonymized data from the other cohorts (BF1, BF2, Avid studies, BACS, UCSF) can be shared with qualified academic researcher upon request to respective principal investigator (for UCSF: submit a request form at https://memory.ucsf.edu/research-trials/professional/open-science) for the purpose of replicating procedures and results presented in the study. Data transfer must be in agreement with the data protection regulation at the institution and decisions by the local ethics review board.

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 2, 28 September 2026

  • Language: n/a → en
  • Authors: added Renaud La Joie (0000-0003-2581-8100); removed Renaud La Joie
  • Funding: added Michael J. Fox Foundation for Parkinson's Research: MJFF‐025507, 2022-Projekt0107; Alzheimer's Association: ZEN24-1069572, SG-23-1061717; Eli Lilly and Company; Cure Alzheimer's Fund; Australian Government: 2022-Projekt0107, 2022-Projekt0080; GHR Foundation: 2022-00775; Lunds Universitet: AF-980907; Knut och Alice Wallenbergs Stiftelse: KAW 2020.0239, (2022-0231), 2022-Projekt0080, 2022-00775, AF-980907, 2022-1259, FO2021-0293, 2020-O000028, 1412/22, 2020.0239; Vetenskapsrådet: ERAPERMED2021-184, 2024-03642, 2022-00775, 2022-Projekt0080, FO2021-0293, AF-980907, 2018, 1412/22, 2018-02052, 2021-02219, 2020-O000028, 2022-1259; Konung Gustaf V:s och Drottning Victorias Frimurarestiftelse: 2020-O000028, WASP/DDLS22-066; Science for Life Laboratory: KAW 2020.0239; Skånes universitetssjukhus: 2022-1259, 2020-O000028, 2022-Projekt0080; Stiftelsen Bundy Academy; University of California, San Francisco; Parkinsonfonden: 1412/22

Version 1, 28 September 2026: the first record

Recorded: type, journal, dates, 23 authors, 72 references.

Cite

This paper

Karlsson, L., Strandberg, O., Smith, R., Tang, W., Arvidsson, I., Åström, K., Hauer, K. O., Janelidze, S., Stomrud, E., Palmqvist, S., Verghese, P. B., Braunstein, J. B., Alzheimer’s Disease Neuroimaging Initiative, PREVENT-AD Research Group, Klein, G., Shcherbinin, S., Jagust, W. J., Villeneuve, S., La Joie, R., . . . Hansson, O. (2026). Generating synthetic tau-PET scans in Alzheimer’s disease from MRI, blood biomarkers and demographics with deep learning. medRxiv (preprint). https://doi.org/10.64898/2026.05.06.26352540

BibTeX

@article{karlsson2026generating,
author = {Karlsson, Linda and Strandberg, Olof and Smith, Ruben and Tang, Weizhong and Arvidsson, Ida and Åström, Kalle and Hauer, Kevin Oliveira and Janelidze, Shorena and Stomrud, Erik and Palmqvist, Sebastian and Verghese, Philip B. and Braunstein, Joel B. and {Alzheimer’s Disease Neuroimaging Initiative} and {PREVENT-AD Research Group} and Klein, Gregory and Shcherbinin, Sergey and Jagust, William J. and Villeneuve, Sylvia and La Joie, Renaud and Rabinovici, Gil D. and Mattsson-Carlgren, Niklas and Vogel, Jacob W. and Hansson, Oskar},
title = {{Generating synthetic tau-PET scans in Alzheimer’s disease from MRI, blood biomarkers and demographics with deep learning}},
journal = {medRxiv (preprint)},
year = {2026},
month = may,
publisher = {medRxiv},
doi = {10.64898/2026.05.06.26352540},
url = {https://doi.org/10.64898/2026.05.06.26352540}
}

RIS

TY - JOUR
AU - Karlsson, Linda
AU - Strandberg, Olof
AU - Smith, Ruben
AU - Tang, Weizhong
AU - Arvidsson, Ida
AU - Åström, Kalle
AU - Hauer, Kevin Oliveira
AU - Janelidze, Shorena
AU - Stomrud, Erik
AU - Palmqvist, Sebastian
AU - Verghese, Philip B.
AU - Braunstein, Joel B.
AU - Alzheimer’s Disease Neuroimaging Initiative
AU - PREVENT-AD Research Group
AU - Klein, Gregory
AU - Shcherbinin, Sergey
AU - Jagust, William J.
AU - Villeneuve, Sylvia
AU - La Joie, Renaud
AU - Rabinovici, Gil D.
AU - Mattsson-Carlgren, Niklas
AU - Vogel, Jacob W.
AU - Hansson, Oskar
TI - Generating synthetic tau-PET scans in Alzheimer’s disease from MRI, blood biomarkers and demographics with deep learning
T2 - medRxiv (preprint)
J2 - medRxiv
PY - 2026
DA - 2026/05/07
PB - medRxiv
DO - 10.64898/2026.05.06.26352540
UR - https://doi.org/10.64898/2026.05.06.26352540
LA - en
ER -

CSL-JSON

{
"id": "10.64898/2026.05.06.26352540",
"type": "article",
"title": "Generating synthetic tau-PET scans in Alzheimer’s disease from MRI, blood biomarkers and demographics with deep learning",
"container-title": "medRxiv (preprint)",
"author": [
{
"family": "Karlsson",
"given": "Linda"
},
{
"family": "Strandberg",
"given": "Olof"
},
{
"family": "Smith",
"given": "Ruben"
},
{
"family": "Tang",
"given": "Weizhong"
},
{
"family": "Arvidsson",
"given": "Ida"
},
{
"family": "Åström",
"given": "Kalle"
},
{
"family": "Hauer",
"given": "Kevin Oliveira"
},
{
"family": "Janelidze",
"given": "Shorena"
},
{
"family": "Stomrud",
"given": "Erik"
},
{
"family": "Palmqvist",
"given": "Sebastian"
},
{
"family": "Verghese",
"given": "Philip B."
},
{
"family": "Braunstein",
"given": "Joel B."
},
{
"literal": "Alzheimer’s Disease Neuroimaging Initiative"
},
{
"literal": "PREVENT-AD Research Group"
},
{
"family": "Klein",
"given": "Gregory"
},
{
"family": "Shcherbinin",
"given": "Sergey"
},
{
"family": "Jagust",
"given": "William J."
},
{
"family": "Villeneuve",
"given": "Sylvia"
},
{
"family": "La Joie",
"given": "Renaud"
},
{
"family": "Rabinovici",
"given": "Gil D."
},
{
"family": "Mattsson-Carlgren",
"given": "Niklas"
},
{
"family": "Vogel",
"given": "Jacob W."
},
{
"family": "Hansson",
"given": "Oskar"
}
],
"container-title-short": "medRxiv",
"DOI": "10.64898/2026.05.06.26352540",
"publisher": "medRxiv",
"URL": "https://doi.org/10.64898/2026.05.06.26352540",
"language": "en",
"issued": {
"date-parts": [
[
2026,
5,
7
]
]
}
}

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:10.1093/brain/awaf375 [code]
Reference proteins to improve Core 1 and Core 2 Alzheimer's disease CSF and plasma biomarkers.
Journal: Brain : a journal of neurology
In common: scikit-learn, pandas, Matplotlib, 1 other tool, PET / SPECT, other, Alzheimer's / dementia, 1 other category, 10 references, 2 authors
[2] doi:10.1002/hbm.70508 [code]
Cyclic 2.5D Perceptual Loss for Cross-Modal 3D Medical Image Synthesis: T1w MRI to Tau PET.
Journal: Human brain mapping
In common: NiBabel, NumPy, PET / SPECT, Alzheimer's / dementia, structural MRI / diffusion, 11 references
[3] doi:10.1093/brain/awaf413 [code]
Estimating the time course of biomarker changes in Alzheimer's disease.
Journal: Brain : a journal of neurology
In common: PET / SPECT, Alzheimer's / dementia, structural MRI / diffusion, 1 other category, 9 references, author O Hansson
[4] doi:10.1038/s41467-026-71732-1 [code]
Trajectories of plasma and CSF MTBR-tau243 and phosphorylated-tau species across the Alzheimer's disease continuum.
Journal: Nature communications
In common: PET / SPECT, Alzheimer's / dementia, 9 references, author O Hansson
[5] doi:10.1093/braincomms/fcag176 [code]
Tau topography subtypes account for clinical heterogeneity and longitudinal trajectories in early-onset Alzheimer's disease.
Journal: Brain communications
In common: seaborn, scikit-learn, pandas, 3 other tools, PET / SPECT, Alzheimer's / dementia, clinical / translational, 3 references, author Renaud La Joie
[6] doi:10.1002/alz.71711 [code]
Exploring longitudinal relationships among Alzheimer's disease biomarkers.
Journal: Alzheimer's & dementia : the journal of the Alzheimer's Association
In common: PET / SPECT, other, Alzheimer's / dementia, 1 other category, 7 references
[7] doi:10.1002/alz.71530 [code]
Differential associations of plasma biomarkers with Alzheimer's disease and small vessel disease: A multimodal imaging study.
Journal: Alzheimer's & dementia : the journal of the Alzheimer's Association
In common: NiBabel, pandas, SciPy, 2 other tools, PET / SPECT, other, Alzheimer's / dementia, 2 other categories, 4 references
[8] doi:10.1038/s41398-026-04081-8 [code]
Functional system-specific brain aging across the Alzheimer's disease continuum.
Journal: Translational psychiatry
In common: Keras, TensorFlow, NiBabel, 6 other tools, Alzheimer's / dementia, structural MRI / diffusion, clinical / translational, 1 reference
[9] doi:10.1186/s12880-026-02481-2 [code]
Deep learning-based neuroanatomical profiling reveals population-specific brain changes in multiple sclerosis: a large-scale Middle Eastern study.
Journal: BMC medical imaging
In common: Keras, TensorFlow, scikit-image, 7 other tools, structural MRI / diffusion, clinical / translational
[10] doi:10.1080/07853890.2026.2685416 [code]
Pulmonary and cerebral damage in COVID-19 survivors: is there any association?
Journal: Annals of medicine
In common: Keras, TensorFlow, scikit-image, 6 other tools, other, structural MRI / diffusion, clinical / translational

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.