OSCR

Self-Supervised Text-Vision Alignment for Automated Brain MRI Abnormality Detection: A Multicenter Study (ALIGN Study).

Code ↔ Paper

The paper beside its authors' code: matches between them have not been computed for this paper yet.

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

Markdown · 80 lines · 2.6 KB · apache

  1. ---
  2. pipeline_tag: sentence-similarity
  3. tags:
  4. - sentence-transformers
  5. - feature-extraction
  6. - sentence-similarity
  7. - transformers
  8. - neuroradiology
  9. - medical
  10. license: apache-2.0
  11. ---
  13. # NeuroBERT
  15. A sentence-transformers model optimized for neuroradiology reports. Maps sentences to 768-dimensional embeddings for semantic similarity tasks.
  17. ## Overview
  19. NeuroBERT is a RoBERTa-based model with a **custom 10,000-word neuroradiology vocabulary** trained from scratch. Standard BERT tokenization fragments medical terms (e.g., "hemorrhage" → "he", "morr", "hage"), so we trained a domain-specific WordPiece vocabulary to preserve neuroradiologic terminology.
  21. **Training:**
  22. 1. **Masked language modeling** on neuroradiology reports (next sentence prediction omitted as adjacent sentences are often unrelated)
  23. 2. **Radiology section matching** using a SentenceBERT twin-network architecture to align Findings and Summary sections from the same report
  25. ## Usage
  27. ```bash
  28. pip install -U sentence-transformers
  29. ```
  31. ```python
  32. from sentence_transformers import SentenceTransformer, util
  34. model = SentenceTransformer('davvwood/NeuroBERT')
  36. # Reference templates for normal findings
  37. templates = [
  38. 'normal study',
  39. 'normal appearances of the brain',
  40. 'no intracranial abnormality identified'
  41. ]
  42. template_embeddings = model.encode(templates)
  44. # Example reports
  45. reports = [
  46. "mri head: there is restricted diffusion in the left paramedian ventral pons at the level of the middle cerebellar peduncle in keeping with an acute infarct.",
  47. "mri head: the ventricles and extra cerebral csf spaces are of normal size. no focal intracranial abnormality has been identified. conclusion: normal intracranial appearances"
  48. ]
  50. for report in reports:
  51. report_embedding = model.encode(report)
  52. similarities = [util.cos_sim(t_emb, report_embedding).item() for t_emb in template_embeddings]
  53. print(f"Max similarity to normal templates: {max(similarities):.3f}")
  54. ```
  56. ## Model Architecture
  58. ```
  59. SentenceTransformer(
  60. (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with RobertaModel
  61. (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_mean_tokens': True})
  62. )
  63. ```
  65. ## Citation
  67. If you use NeuroBERT, please cite:
  69. ```bibtex
  70. @article{wood2025neurobert,
  71. title={Self-supervised Text-vision Alignment for Automated Brain MRI Abnormality Detection: A Multicenter Study (ALIGN Study)},
  72. author={Wood, D. A. and Guilhem, E. and Kafiabadi, S. and Al Busaidi, A. and Dissanayake, K. and Hammam, A. and others},
  73. journal={Radiology: Artificial Intelligence},
  74. pages={e240619},
  75. year={2025},
  76. doi={10.1148/ryai.240619}
  77. }
  78. ```
  80. Paper: https://doi.org/10.1148/ryai.240619

README.md at commit 69c974a, under apache · at the source

Overview

Authors: David A Wood1, Emily Guilhem2, Sina Kafiabadi2, Ayisha Al Busaidi2, Kishan Dissanayake2, Ahmed Hammam2, Nina Mansoor2, Matthew Townend1, Siddharth Agarwal2, Yiran Wei1, Asif Mazumder3, Gareth J Barker4, Peter Sasieni5, Sébastien Ourselin1, James H Cole6, Nikhil Nair7, Anil Geetha7, Chike Onyekwuluje7, Rob Dineen8, Permesh Dhillon8
and 10 other authorsCarolyn Costigan9, Kavi Fatania10, Mark Igra10, Rebecca Nichols11, Janak Saada12, Arne Juette12, Ramona-Rita Barbara12, Hilmar Spohr12, Thomas C Booth1,2, for the MIDI Consortium Group13
ORCID iDs: David A Wood, Emily Guilhem, Ayisha Al Busaidi, Kishan Dissanayake, Siddharth Agarwal, Asif Mazumder, Gareth J Barker, Peter Sasieni, James H Cole, Nikhil Nair, Anil Geetha, Chike Onyekwuluje, Rob Dineen, Permesh Dhillon, Carolyn Costigan, Kavi Fatania, Rebecca Nichols, Janak Saada, Arne Juette, Thomas C Booth
13 affiliations
  1. School of Biomedical Engineering and Imaging Sciences, King’s College London, Rayne Institute, 4th Floor, Lambeth Wing, London SE17 7EH, UK
  2. King’s College Hospital NHS Foundation Trust, SE5 9RS, London, United Kingdom
  3. Guy’s and St Thomas’ NHS Foundation Trust, SE1 9RT, London, United Kingdom
  4. Department of Neuroimaging, Institute of Psychiatry, Psychology, & Neuroscience, King’s College London, SE5 9NU, United Kingdom
  5. Wolfson Institute of Population Health, Queen Mary University of London, Charterhouse Square London, EC1M 6BQ, United Kingdom
  6. Centre for Medical Image Computing, Department of Computer Science, University College London, W1W 7T, United Kingdom
  7. Bedfordshire Hospitals NHS Foundation Trust, Bedford Hospital, South Wing, Kempston Road, Bedford, MK42 9DJ, United Kingdom
  8. Radiological Sciences, School of Medicine, University of Nottingham, Nottingham, United Kingdom
  9. University Hospitals NHS Trust, Nottingham, United Kingdom
  10. Department of Neuroradiology, Floor B, Clarendon Wing, Leeds General Infirmary, Leeds, LS1 3EX, United Kingdom
  11. Yeovil Hospital, Somerset NHS Foundation Trust, Yeovil, United Kingdom
  12. Department of Radiology, Norfolk and Norwich University Hospital, Colney Lane, Norwich, Norfolk, NR4 7UY, United Kingdom
  13. Members of the MIDI Consortium Group are listed at the end of the article
Institutions: King's College London (United Kingdom); King's College Hospital NHS Foundation Trust (United Kingdom); Guy's and St Thomas' NHS Foundation Trust (United Kingdom); Queen Mary University of London (United Kingdom); University College London (United Kingdom); Bedford Hospital NHS Trust (United Kingdom); Bedford Hospital (United Kingdom); University of Nottingham (United Kingdom); Nottingham University Hospitals NHS Trust (United Kingdom); Leeds General Infirmary (United Kingdom); Somerset NHS Foundation Trust; Norfolk and Norwich University Hospital (United Kingdom)
Journal: Radiology. Artificial intelligence, volume 8, issue 2, article e240619
Dates: received 19 September 2024; accepted 23 October 2025; published online 26 November 2025; in print March 2026
Type: Research article · Language: English
License: CC BY
Identifiers: DOI 10.1148/ryai.240619 · PMID 41295086 · PMCID PMC13019336 · OpenAlex W4416718099
Open access: hybrid, a free copy (OpenAlex)
Status: empty repository
Categories: structural MRI / diffusion (modality), human (organism), clinical / translational (subfield)
Methods: Connectivity, Statistics, Machine learning
Keywords: Head and Neck, Unsupervised Learning, Convolutional Neural Network (CNN), Neuroradiology
MeSH: Brain*, Brain Diseases*, Image Interpretation, Computer-Assisted*, Magnetic Resonance Imaging*, Neuroimaging*, Adult, Aged, Female, Humans, Male, Middle Aged, Neural Networks, Computer, Prospective Studies, Retrospective Studies (* major topic)
Topic: Intracerebral and Subarachnoid Hemorrhage Research (Neurology, Medicine), according to OpenAlex
Funding: Nvidia; Royal College of Radiologists; King's College London (WT203148/Z/16/Z); King's Health Partners; Medical Research Council (203148/Z/16/Z, MR/W021684/1); Engineering and Physical Sciences Research Council ([WT203148/Z/16/Z], WT 203148/Z/16/, 203148/ Z/16/Z); Centre For Medical Engineering, King’s College London (203148/Z/16/Z, WT 203148/Z/16/Z)
Citations: not cited yet (Europe PMC); 37 references in the paper
Notices: A comment on this paper has been published (41636622, from Europe PMC)

Abstract

Purpose: To develop a self-supervised text-vision framework to detect abnormalities on brain MRI scans by leveraging free-text neuroradiology reports, eliminating the need for expert-labeled training datasets.

Materials and Methods: This retrospective and prospective multicenter study included 81 936 brain MRI examinations and corresponding radiology reports for adult patients at two UK National Health Service hospitals from January 2008 to December 2019 for training and internal testing and 1369 prospectively collected examinations between March 2022 and March 2024 from four separate National Health Service hospitals for external testing (ClinicalTrials.gov no. NCT04368481). A neuroradiology language model (NeuroBERT) was trained using self-supervised tasks to generate report embeddings. Convolutional neural networks (one per MRI sequence) were trained to map scans to embeddings by minimizing mean squared error loss. The framework then detected abnormalities in new examinations by scoring scans against query sentences using text-image similarity. Model diagnostic performance was assessed using the area under the receiver operating characteristic curve (AUC).

Results: The framework achieved an AUC of 0.95 (95% CI: 0.94, 0.97) for normal versus abnormal classification and generalized to external sites with examination-level AUCs of 0.90 (95% CI: 0.86, 0.93) in Bedford, 0.87 (95% CI: 0.83, 0.90) in Nottingham, 0.86 (95% CI: 0.83, 0.90) in Norwich, and 0.85 (95% CI: 0.81, 0.89) in Yeovil. In five zero-shot classification tasks—acute stroke, multiple sclerosis, intracranial hemorrhage, meningioma, and hydrocephalus—the framework achieved a mean AUC of 0.89 (range, 0.77–0.93). For visual-semantic image retrieval, mean precision was 0.84 among the top 15 images across seven pathologies.

Conclusion: The self-supervised text-vision framework accurately detected brain MRI abnormalities without expert-labeled datasets.

Clinical trial registration no. NCT04368481

Keywords: Head and Neck, Unsupervised Learning, Convolutional Neural Network (CNN), Neuroradiology

© The Author(s) 2025. Published by the Radiological Society of North America under a CC BY 4.0 license.

Supplemental material is available for this article.

See also commentary by Ghodasara in this issue.

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

Repository

Its files are read in the Code ↔ Paper reader above.

huggingface.co/davvwood/neurobert

License: apache
State: the link answers, verified on 30 September 2026
Evidence: files inventoried
Commit: 69c974a5a846953179f2737f5ef8d402f4a9e0d5, 7 January 2026
Size: 15 files, 0 scripts
Software Heritage: not archived
Found in: the text, “Self-Supervised Text-Vision Framework”
Holds: README
Not found: license file, CITATION.cff, environment file, tests, continuous integration, documentation
Availability: 1 check, the latest on 30 September 2026: the link answers
  • 30 September 2026: the link answers
1 file
At the source: huggingface.co/davvwood/NeuroBERT

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;
  • 0 scripts, each with its path and the digest of its content;
  • no match between paragraphs and code yet;
  • 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.

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, 30 authors, 4 keywords, 14 MeSH terms, 7 funders, 30 references, 1 integrity notice.

Cite

This paper

Wood, D. A., Guilhem, E., Kafiabadi, S., Al Busaidi, A., Dissanayake, K., Hammam, A., Mansoor, N., Townend, M., Agarwal, S., Wei, Y., Mazumder, A., Barker, G. J., Sasieni, P., Ourselin, S., Cole, J. H., Nair, N., Geetha, A., Onyekwuluje, C., Dineen, R., . . . for the MIDI Consortium Group. (2026). Self-Supervised Text-Vision Alignment for Automated Brain MRI Abnormality Detection: A Multicenter Study (ALIGN Study). Radiology. Artificial intelligence, 8(2), e240619. https://doi.org/10.1148/ryai.240619

BibTeX

@article{wood2026self,
author = {Wood, David A and Guilhem, Emily and Kafiabadi, Sina and Al Busaidi, Ayisha and Dissanayake, Kishan and Hammam, Ahmed and Mansoor, Nina and Townend, Matthew and Agarwal, Siddharth and Wei, Yiran and Mazumder, Asif and Barker, Gareth J and Sasieni, Peter and Ourselin, Sébastien and Cole, James H and Nair, Nikhil and Geetha, Anil and Onyekwuluje, Chike and Dineen, Rob and Dhillon, Permesh and Costigan, Carolyn and Fatania, Kavi and Igra, Mark and Nichols, Rebecca and Saada, Janak and Juette, Arne and Barbara, Ramona-Rita and Spohr, Hilmar and Booth, Thomas C and {for the MIDI Consortium Group}},
title = {{Self-Supervised Text-Vision Alignment for Automated Brain MRI Abnormality Detection: A Multicenter Study (ALIGN Study)}},
journal = {Radiology. Artificial intelligence},
year = {2026},
month = mar,
volume = {8},
number = {2},
pages = {e240619},
publisher = {Radiological Society of North America},
issn = {2638-6100},
doi = {10.1148/ryai.240619},
url = {https://doi.org/10.1148/ryai.240619},
pmid = {41295086},
pmcid = {PMC13019336}
}

RIS

TY - JOUR
AU - Wood, David A
AU - Guilhem, Emily
AU - Kafiabadi, Sina
AU - Al Busaidi, Ayisha
AU - Dissanayake, Kishan
AU - Hammam, Ahmed
AU - Mansoor, Nina
AU - Townend, Matthew
AU - Agarwal, Siddharth
AU - Wei, Yiran
AU - Mazumder, Asif
AU - Barker, Gareth J
AU - Sasieni, Peter
AU - Ourselin, Sébastien
AU - Cole, James H
AU - Nair, Nikhil
AU - Geetha, Anil
AU - Onyekwuluje, Chike
AU - Dineen, Rob
AU - Dhillon, Permesh
AU - Costigan, Carolyn
AU - Fatania, Kavi
AU - Igra, Mark
AU - Nichols, Rebecca
AU - Saada, Janak
AU - Juette, Arne
AU - Barbara, Ramona-Rita
AU - Spohr, Hilmar
AU - Booth, Thomas C
AU - for the MIDI Consortium Group
TI - Self-Supervised Text-Vision Alignment for Automated Brain MRI Abnormality Detection: A Multicenter Study (ALIGN Study)
T2 - Radiology. Artificial intelligence
J2 - Radiol Artif Intell
PY - 2026
DA - 2026/03/01
VL - 8
IS - 2
SP - e240619
SN - 2638-6100
PB - Radiological Society of North America
DO - 10.1148/ryai.240619
UR - https://doi.org/10.1148/ryai.240619
LA - en
ER -

CSL-JSON

{
"id": "10.1148/ryai.240619",
"type": "article-journal",
"title": "Self-Supervised Text-Vision Alignment for Automated Brain MRI Abnormality Detection: A Multicenter Study (ALIGN Study)",
"container-title": "Radiology. Artificial intelligence",
"author": [
{
"family": "Wood",
"given": "David A"
},
{
"family": "Guilhem",
"given": "Emily"
},
{
"family": "Kafiabadi",
"given": "Sina"
},
{
"family": "Al Busaidi",
"given": "Ayisha"
},
{
"family": "Dissanayake",
"given": "Kishan"
},
{
"family": "Hammam",
"given": "Ahmed"
},
{
"family": "Mansoor",
"given": "Nina"
},
{
"family": "Townend",
"given": "Matthew"
},
{
"family": "Agarwal",
"given": "Siddharth"
},
{
"family": "Wei",
"given": "Yiran"
},
{
"family": "Mazumder",
"given": "Asif"
},
{
"family": "Barker",
"given": "Gareth J"
},
{
"family": "Sasieni",
"given": "Peter"
},
{
"family": "Ourselin",
"given": "Sébastien"
},
{
"family": "Cole",
"given": "James H"
},
{
"family": "Nair",
"given": "Nikhil"
},
{
"family": "Geetha",
"given": "Anil"
},
{
"family": "Onyekwuluje",
"given": "Chike"
},
{
"family": "Dineen",
"given": "Rob"
},
{
"family": "Dhillon",
"given": "Permesh"
},
{
"family": "Costigan",
"given": "Carolyn"
},
{
"family": "Fatania",
"given": "Kavi"
},
{
"family": "Igra",
"given": "Mark"
},
{
"family": "Nichols",
"given": "Rebecca"
},
{
"family": "Saada",
"given": "Janak"
},
{
"family": "Juette",
"given": "Arne"
},
{
"family": "Barbara",
"given": "Ramona-Rita"
},
{
"family": "Spohr",
"given": "Hilmar"
},
{
"family": "Booth",
"given": "Thomas C"
},
{
"literal": "for the MIDI Consortium Group"
}
],
"container-title-short": "Radiol Artif Intell",
"volume": "8",
"issue": "2",
"page": "e240619",
"DOI": "10.1148/ryai.240619",
"PMID": "41295086",
"PMCID": "PMC13019336",
"ISSN": "2638-6100",
"publisher": "Radiological Society of North America",
"URL": "https://doi.org/10.1148/ryai.240619",
"language": "en",
"issued": {
"date-parts": [
[
2026,
3,
1
]
]
}
}

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

Similar papers

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

[1] doi:10.1162/imag.a.1352 [code]
Brain-age in ultra-low-field MRI: How well does it work?
Journal: Imaging neuroscience (Cambridge, Mass.)
In common: structural MRI / diffusion, 3 authors
[2] doi:10.1093/braincomms/fcag162 [code]
Disability profiles in progressive multiple sclerosis reflect pathology distribution, independent of clinical phenotype.
Journal: Brain communications
In common: structural MRI / diffusion, clinical / translational, author James H Cole
[3] doi:10.1002/hbm.70425 [code]
Brain Age Estimation on T2-FLAIR Scans for Application to Multiple Sclerosis.
Journal: Human brain mapping
In common: structural MRI / diffusion, clinical / translational, author James H Cole
[4] doi:10.1038/s41398-026-04010-9 [code]
Bullying victimization and brain development: a longitudinal structural magnetic resonance imaging study from adolescence to early adulthood.
Journal: Translational psychiatry
In common: structural MRI / diffusion, author Gareth J Barker
[5] doi:10.1038/s43856-026-01707-2 [code]
Decreased amyloid-related structure-function coupling in preclinical Alzheimer's disease.
Journal: Communications medicine
In common: clinical / translational, author James H Cole
[6] doi:10.1126/sciadv.aed0772 [code]
Deep learning reveals a neurocomputational mechanism predicting depression risk in adolescents.
Journal: Science advances
In common: author Gareth J Barker
[7] doi:10.1038/s41398-026-04114-2 [code]
Plasma Glial Fibrillary Acidic Protein (GFAP) shows age-dependent associations with externalizing psychopathology and atypical brain connectivity.
Journal: Translational psychiatry
In common: author Gareth J Barker
[8] 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: structural MRI / diffusion, 2 references
[9] doi:10.1038/s41591-026-04497-1 [code]
Health system learning enables generalist neuroimaging models.
Journal: Nature medicine
In common: structural MRI / diffusion, clinical / translational, 1 reference
[10] doi:10.3389/fonc.2026.1928589
LiteFreqMamba:lightweight frequency-enhanced Mamba for efficient 3D brain tumor segmentation.
Journal: Frontiers in oncology
In common: 2 references

Contribute

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

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

Request its removal

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

Discussion, reproductions, activity

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

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

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