Efficient Transformer-Based Localized Patch Sampling for Choroid Plexus Segmentation in Multiple Sclerosis

Alessandro Cagol; Andreea-Alexandra Aldea; Cristina Granziera; Federico Spagnolo; Jannis M\"uller; Jens Kuhle; Lester Melie-Garcia; Ludwig Kappos; Maria Pia Sormani; Marina Mastantuono

arxiv: 2606.03566 · v1 · pith:B4TSTKYSnew · submitted 2026-06-02 · 💻 cs.CV · cs.AI

Efficient Transformer-Based Localized Patch Sampling for Choroid Plexus Segmentation in Multiple Sclerosis

Po-Jui Lu , Alessandro Cagol , Mario Ocampo-Pineda , Federico Spagnolo , Marina Mastantuono , Andreea-Alexandra Aldea , Jannis M\"uller , \"Ozg\"ur Yaldizli

show 7 more authors

Matthias Weigel Lester Melie-Garcia Roberta Magliozzi Maria Pia Sormani Ludwig Kappos Jens Kuhle Cristina Granziera

This is my paper

Pith reviewed 2026-06-28 10:55 UTC · model grok-4.3

classification 💻 cs.CV cs.AI

keywords choroid plexus segmentationmultiple sclerosisMRISwinUNETRpatch samplingtransformerLVCPmedical image segmentation

0 comments

The pith

Localized small-patch sampling lets SwinUNETR segment the choroid plexus at 99 percent lower computational cost than full-volume models.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper builds an automatic segmentation pipeline for the lateral ventricle choroid plexus in multiple sclerosis using MRI. It trains a SwinUNETR transformer on 32x32x32 voxel patches restricted to intra- and peri-ventricular regions from both single and multi-modal scans. On a combined test set of 388 cases the model reaches a mean Dice score of 0.868, statistically higher than the UXNET baseline, and drops computational demand from 22,080 to 91.8 GFLOPs. Accuracy holds at 0.863 even when only FLAIR images are supplied. The efficiency gain removes a major barrier to using LVCP volume as a routine biomarker in large trials and follow-up studies.

Core claim

Integrating targeted intra- and peri-ventricular patch sampling with a SwinUNETR architecture produces a mean DSC of 0.868 (95 percent CI 0.863-0.872) on the extended test set when MPRAGE and FLAIR are combined, a statistically significant improvement over UXNET while lowering computational load by 99 percent.

What carries the argument

Targeted sampling of 32x32x32 voxel patches centered on the ventricles, processed by a SwinUNETR transformer, to segment the full lateral ventricle choroid plexus.

If this is right

The method maintains DSC 0.863 and low Hausdorff distance when only FLAIR is available.
Computational cost falls to 91.8 GFLOPs, enabling deployment on standard hardware.
Statistically superior performance holds across two independent MS cohorts totaling over 700 scans.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same localized sampling could be adapted to other small periventricular structures that are currently segmented at high cost.
If the patch strategy generalizes to lower-field or motion-corrupted scans, it could support routine clinical monitoring outside research centers.

Load-bearing premise

The assumption that small patches focused near the ventricles contain enough context to segment the entire choroid plexus without missing distant parts or creating edge errors.

What would settle it

Running the model on cases where the choroid plexus visibly extends outside the sampled patches and measuring whether Dice scores drop below the reported range.

read the original abstract

Background: The lateral ventricle choroid plexus (LVCP) is gaining recognition as a key imaging biomarker for multiple sclerosis (MS) related to physical disability and neuroinflammation. Yet, manual segmentation of the LVCP is highly tedious, restricting its use in broad clinical trials and longitudinal assessments. This research aims to develop a SwinUNETR-driven pipeline that leverages targeted intra- and peri-ventricular small patch sampling to automatically segment the LVCP in MS from both standalone and multi-modal MRI inputs. Methods: We retrospectively assessed 3T MRI scans across three sets of data stemming from two separate MS-dominant cohorts (Dataset 1: n=177; Dataset 2: n=177; expanded test set: n=388). Our method employed a SwinUNETR architecture trained on 32x32x32 voxel patches, benchmarking it against the 3D UXNET model. The primary metric for evaluation was the Dice Similarity Coefficient (DSC), supplemented by computational demand (GFLOPs) and the 95th percentile Hausdorff Distance (HD95). Results: On the extended test set, the SwinUNETR model secured a mean DSC of 0.868 (95% CI: 0.863-0.872) with MPRAGE and FLAIR combined, showing a statistically significant gain over UXNET (DSC: 0.858 [95% CI: 0.853-0.862], p<0.0001). When restricted to standalone FLAIR inputs, the transformer-based approach sustained a high DSC of 0.863, while the spatial localization of UXNET worsened considerably (HD95: 1.86 vs. 3.00 mm). Importantly, the proposed framework lowered computational load by 99% (91.8 vs. 22,080 GFLOPs). By integrating localized patch sampling with a SwinUNETR architecture, this methodology offers an accurate, robust, and statistically superior alternative to current leading models for LVCP segmentation. Its vast reduction in computational cost makes it ideal for widespread implementation in clinical and research environments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

SwinUNETR on 32^3 localized patches beats UXNET on DSC while cutting GFLOPs by 99%, but the coverage and stitching assumptions need explicit checks.

read the letter

The thing to know is that this work shows a SwinUNETR pipeline using small intra- and peri-ventricular patches can reach DSC 0.868 on the large held-out set with combined MPRAGE+FLAIR, edging out UXNET at 0.858 while dropping compute from 22k to 92 GFLOPs. The standalone FLAIR case also holds reasonably well. That efficiency angle is the practical takeaway for anyone running this on bigger cohorts.

They handle the evaluation cleanly enough: three datasets, confidence intervals, p-values, and a direct GFLOPs comparison. The numbers are reported on an expanded test set of 388 scans, which adds some weight. The architecture itself is off-the-shelf, so the localized sampling strategy is what they are really testing.

The soft spot is the one flagged in the stress test. Everything rests on the patches enclosing every LVCP voxel and the model handling 32-voxel context without edge artifacts when masks are stitched. The abstract gives no coverage statistics, no ablation on stride or overlap, and no detail on how the localization step itself is performed or validated. If that step fails on some cases, the reported superiority does not fully carry over. That is the main assumption to pressure in review.

This is for MS imaging groups or people building efficient segmentation tools for small structures. A reader who needs to process hundreds of scans would see the value in the compute reduction. It has enough quantified results and a clear empirical setup to deserve peer review rather than a desk reject.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a SwinUNETR pipeline for LVCP segmentation in MS that uses targeted intra- and peri-ventricular 32^3 patch sampling from MPRAGE and/or FLAIR MRI. On an extended test set of 388 scans it reports mean DSC 0.868 (95% CI 0.863-0.872) for the combined-modality case, statistically superior to UXNET (DSC 0.858, p<0.0001), together with a reduction from 22,080 to 91.8 GFLOPs.

Significance. If the patch-sampling premise holds, the work supplies a concrete route to deploy LVCP segmentation at scale in MS trials by cutting computational cost by two orders of magnitude while preserving or improving Dice scores; the explicit confidence intervals and hypothesis tests are a positive feature of the empirical reporting.

major comments (2)

[Abstract] Abstract/Results: the reported DSC of 0.868 and the 99% GFLOPs reduction both presuppose that the upstream localization step places 32^3 patches so that every LVCP voxel is enclosed and that stitching produces no boundary artifacts. No coverage fraction, missed-voxel count, or ablation on patch stride/overlap is supplied, which directly undermines the claim that the localized method is reliably superior to full-volume UXNET.
[Methods] Methods: the localization procedure that generates the intra-/peri-ventricular patches is described only at the level of 'targeted ... small patch sampling'; without an explicit algorithm, ventricle mask source, or guarantee that all LVCP voxels fall inside the sampled region, the generalizability of the 0.868 DSC across the two cohorts cannot be assessed.

minor comments (2)

[Abstract] Abstract: the relationship between Dataset 1 (n=177), Dataset 2 (n=177) and the 'expanded test set (n=388)' is not stated; it is unclear whether the latter is an independent hold-out or contains the training/validation data.
[Results] Results: the HD95 comparison for standalone FLAIR is given only as '1.86 vs. 3.00 mm'; the first number should be explicitly attributed to SwinUNETR.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We address each point below and will revise the manuscript to provide the requested details on the localization procedure.

read point-by-point responses

Referee: [Abstract] Abstract/Results: the reported DSC of 0.868 and the 99% GFLOPs reduction both presuppose that the upstream localization step places 32^3 patches so that every LVCP voxel is enclosed and that stitching produces no boundary artifacts. No coverage fraction, missed-voxel count, or ablation on patch stride/overlap is supplied, which directly undermines the claim that the localized method is reliably superior to full-volume UXNET.

Authors: We agree that explicit quantification of patch coverage, missed voxels, and boundary effects is needed to fully support the claims. In the revised manuscript we will add: (1) the fraction of LVCP voxels captured by the 32^3 patches, (2) any observed missed-voxel statistics on the test set, and (3) an ablation on patch stride/overlap. These additions will directly address the concern that the reported superiority may rest on unverified localization assumptions. revision: yes
Referee: [Methods] Methods: the localization procedure that generates the intra-/peri-ventricular patches is described only at the level of 'targeted ... small patch sampling'; without an explicit algorithm, ventricle mask source, or guarantee that all LVCP voxels fall inside the sampled region, the generalizability of the 0.868 DSC across the two cohorts cannot be assessed.

Authors: We acknowledge the description of the localization step is currently high-level. The revised Methods section will include: the explicit algorithm (including how intra- and peri-ventricular regions are defined), the source of the ventricle masks used for patch placement, and empirical verification that all annotated LVCP voxels lie within the sampled patches on both cohorts. This will allow readers to evaluate generalizability. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical ML segmentation study with held-out evaluation

full rationale

This paper presents an empirical machine-learning pipeline for LVCP segmentation using SwinUNETR trained on 32^3 patches, benchmarked against UXNET on independent test sets (n=388). All reported results (DSC 0.868, GFLOPs reduction, statistical tests) are computed directly from held-out data splits with no derivations, first-principles predictions, or fitted quantities that reduce to the inputs by construction. No self-citation load-bearing steps, uniqueness theorems, or ansatzes appear in the described method; the localized patch sampling is a design choice whose validity is assessed empirically rather than assumed tautologically. The central claims rest on external test-set performance, satisfying the criteria for a self-contained, non-circular empirical study.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The performance claims depend on the choice of patch size and the assumption that the sampled regions are representative of the full structure across the patient population.

free parameters (2)

patch size = 32x32x32
Chosen to enable localized sampling around ventricles
training hyperparameters
Various model and optimization parameters tuned on training data

axioms (1)

domain assumption Localized patches contain sufficient information for accurate LVCP segmentation
The method assumes that focusing on intra- and peri-ventricular regions captures the choroid plexus adequately.

pith-pipeline@v0.9.1-grok · 6003 in / 1449 out tokens · 44317 ms · 2026-06-28T10:55:29.905742+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

29 extracted references · 24 canonical work pages

[1]

Choroid plexus volume changes in mulƟple sclerosis: insights from a systemaƟc review and meta-analysis of magneƟc resonance imaging studies

Kolahi S, Zarei D, Issaiy M, Shakiba M, Azizi N, Firouznia K. Choroid plexus volume changes in mulƟple sclerosis: insights from a systemaƟc review and meta-analysis of magneƟc resonance imaging studies. Neuroradiology. 2024;c:1869–1886. doi: 10.1007/s00234-024-03439-3. Cited in PMID: 39105769

work page doi:10.1007/s00234-024-03439-3 2024
[2]

Choroid Plexus Volume Change—A Candidate for a New Radiological Marker of MS Progression

Jankowska A, Chwojnicki K, Grzywińska M, Trzonkowski P , Szurowska E. Choroid Plexus Volume Change—A Candidate for a New Radiological Marker of MS Progression. DiagnosƟcs. 2023;13. doi: 10.3390/diagnosƟcs13162668

work page doi:10.3390/diagnos 2023
[3]

Pathogenesis of Alzheimer’s disease: Involvement of the choroid plexus

Čarna M, Onyango IG, KaƟna S, Holub D, Novotny JS, Nezvedova M, Jha D, Nedelska Z, Lacovich V, Vyvere T Vande, et al. Pathogenesis of Alzheimer’s disease: Involvement of the choroid plexus. Alzheimer’s and DemenƟa. 2023;19:3537–3554. doi: 10.1002/alz.12970. Cited in PMID: 36825691

work page doi:10.1002/alz.12970 2023
[4]

AssociaƟon of choroid plexus volume with motor symptoms and dopaminergic degeneraƟon in Parkinson’s disease

Jeong SH, Park CJ, Jeong HJ, Sunwoo MK, Ahn SS, Lee SK, Lee PH, Kim YJ, Sohn YH, Chung SJ. AssociaƟon of choroid plexus volume with motor symptoms and dopaminergic degeneraƟon in Parkinson’s disease. Journal of Neurology, Neurosurgery and Psychiatry. 2023;94:1047–1055. doi: 10.1136/jnnp-2023-331170. Cited in PMID: 37399288

work page doi:10.1136/jnnp-2023-331170 2023
[5]

Neuropathological and cerebrospinal ﬂuid correlates of choroid plexus inﬂammaƟon in progressive mulƟple sclerosis

Magliozzi R, Hametner S, Mastantuono M, Mensi A, Karimian M, Griﬃths L, Watkins LM, Poli A, BerƟ GM, Barusolo E, et al. Neuropathological and cerebrospinal ﬂuid correlates of choroid plexus inﬂammaƟon in progressive mulƟple sclerosis. Brain Pathology. 2024;1–20. doi: 10.1111/bpa.13322

work page doi:10.1111/bpa.13322 2024
[6]

Chronic AcƟve Lesions and Larger Choroid Plexus Explain CogniƟon and FaƟgue in MulƟple Sclerosis

Preziosa P , Pagani E, Meani A, Storelli L, Margoni M, Yudin Y , Tedone N, Biondi D, Rubin M, Rocca MA, et al. Chronic AcƟve Lesions and Larger Choroid Plexus Explain CogniƟon and FaƟgue in MulƟple Sclerosis. Neurol Neuroimmunol Neuroinﬂamm. 2024;11:e200205. doi: 10.1212/NXI.0000000000200205

work page doi:10.1212/nxi.0000000000200205 2024
[7]

Choroid Plexus Enlargement in MulƟple Sclerosis Correlates with CorƟcal and Phase Rim Lesions on 7T MRI and Predicts Progression Independent of Relapse AcƟvity

BarbuƟ E, ConƟ A, Treaba CA, Miscioscia A, BarleƩa VT , Herranz E, Sloane JA, Klawiter EC, Toschi N, Mainero C. Choroid Plexus Enlargement in MulƟple Sclerosis Correlates with CorƟcal and Phase Rim Lesions on 7T MRI and Predicts Progression Independent of Relapse AcƟvity. AJNR Am J Neuroradiol. 2026;47:557–565. doi: 10.3174/ajnr.A8983

work page doi:10.3174/ajnr.a8983 2026
[8]

Choroid Plexus Enlargement in Inﬂammatory MulƟple Sclerosis: 3.0-T MRI and Translocator Protein PET EvaluaƟon

Ricigliano VAG, Morena E, Colombi A, TonieƩo M, Hamzaoui M, Poirion E, BoƩlaender M, Gervais P , Louapre C, Bodini B, et al. Choroid Plexus Enlargement in Inﬂammatory MulƟple Sclerosis: 3.0-T MRI and Translocator Protein PET EvaluaƟon. Radiology. 2021;301:166–177. doi: 10.1148/radiol.2021204426

work page doi:10.1148/radiol.2021204426 2021
[9]

Understanding the Role of the Choroid Plexus in MulƟple Sclerosis as an MRI Biomarker of Disease AcƟvity

Müller J, Noteboom S, Granziera C, Schoonheim MM. Understanding the Role of the Choroid Plexus in MulƟple Sclerosis as an MRI Biomarker of Disease AcƟvity. Neurology. 2023;100:405–406. doi: 10.1212/WNL.0000000000206806. Cited in PMID: 36543568

work page doi:10.1212/wnl.0000000000206806 2023
[10]

Choroid Plexus Volume in MulƟple Sclerosis vs NeuromyeliƟs OpƟca Spectrum Disorder: A RetrospecƟve, Cross-secƟonal Analysis

Müller J, Sinnecker T, Wendebourg MJ, Schläger R, Kuhle J, Schädelin S, Benkert P , Derfuss T , Caƫ n P , Jud C, et al. Choroid Plexus Volume in MulƟple Sclerosis vs NeuromyeliƟs OpƟca Spectrum Disorder: A RetrospecƟve, Cross-secƟonal Analysis. Neurol Neuroimmunol Neuroinﬂamm. 2022;9:e1147. doi: 10.1212/NXI.0000000000001147. Cited in PMID: 35217580

work page doi:10.1212/nxi.0000000000001147 2022
[11]

Morphological changes of the choroid plexus in the lateral ventricle across the lifespan: 5551 subjects from fetus to elderly

Li J, Gao Y , Xu Y , Dai W, Hu Y , Feng X, Wu D, Zhao L. Morphological changes of the choroid plexus in the lateral ventricle across the lifespan: 5551 subjects from fetus to elderly. NeuroImage. 2025;318:121392. doi: 10.1016/j.neuroimage.2025.121392

work page doi:10.1016/j.neuroimage.2025.121392 2025
[12]

ASCHOPLEX : a generalizable approach for the automaƟc segmentaƟon of choroid plexus

Visani V, Veronese M, Pizzini FB, Colombi A, Marjin C, TamanƟ A, Schubert JJ, Althubaity N, Harrison NA, Bullmore ET , et al. ASCHOPLEX : a generalizable approach for the automaƟc segmentaƟon of choroid plexus. 2024;182. doi: 10.1016/j.compbiomed.2024.109164

work page doi:10.1016/j.compbiomed.2024.109164 2024
[13]

U-Net: ConvoluƟonal Networks for Biomedical Image SegmentaƟon [Internet]

Ronneberger O, Fischer P , Brox T. U-Net: ConvoluƟonal Networks for Biomedical Image SegmentaƟon [Internet]. arXiv; 2015 [cited 2025 Nov 24]. Available from: hƩps://arxiv.org/abs/1505.04597

Pith/arXiv arXiv 2015
[14]

doi:10.1016/j.neuroimage.2012.01.021

Fischl B. FreeSurfer. NeuroImage. 2012;62:774–781. doi: 10.1016/j.neuroimage.2012.01.021

work page doi:10.1016/j.neuroimage.2012.01.021 2012
[15]

Deep learning-based SegmentaƟon of the Choroid Plexus in MulƟple Sclerosis Using MPRAGE, FLAIR, and MP2RAGE

Lu P-J, Cagol A, Ocampo-Pineda M, Mastantuono M, Aldea A-A, Müller J, Yaldizli Ö, Weigel M, Melie-Garcia L, Magliozzi R, et al. Deep learning-based SegmentaƟon of the Choroid Plexus in MulƟple Sclerosis Using MPRAGE, FLAIR, and MP2RAGE. Mult Scler . 2025;31:1032. doi: 10.1177/13524585251358343

work page doi:10.1177/13524585251358343 2025
[16]

Swin UNETR: Swin Transformers for SemanƟc SegmentaƟon of Brain Tumors in MRI Images [Internet]

Hatamizadeh A, Nath V, Tang Y , Yang D, Roth H, Xu D. Swin UNETR: Swin Transformers for SemanƟc SegmentaƟon of Brain Tumors in MRI Images [Internet]. arXiv; 2022 [cited 2026 Feb 4]. Available from: hƩp://arxiv.org/abs/2201.01266

arXiv 2022
[17]

3D UX-Net: A Large Kernel Volumetric ConvNet Modernizing Hierarchical Transformer for Medical Image SegmentaƟon [Internet]

Lee HH, Bao S, Huo Y , Landman BA. 3D UX-Net: A Large Kernel Volumetric ConvNet Modernizing Hierarchical Transformer for Medical Image SegmentaƟon [Internet]. arXiv; 2022 [cited 2025 Nov 24]. Available from: hƩps://arxiv.org/abs/2209.15076

arXiv 2022
[18]

Enhanced choroid plexus segmentaƟon with 3D UX-Net and its associaƟon with disease progression in mulƟple sclerosis

Wang X, Wang X, Yan Z, Yin F, Li Y , Liu X, Liu Y . Enhanced choroid plexus segmentaƟon with 3D UX-Net and its associaƟon with disease progression in mulƟple sclerosis. MulƟple Sclerosis and Related Disorders. 2024;88:105750. doi: 10.1016/j.msard.2024.105750

work page doi:10.1016/j.msard.2024.105750 2024
[19]

Deep learning segmentaƟon of the choroid plexus from structural magneƟc resonance imaging (MRI): validaƟon and normaƟve ranges across the adult lifespan

Eisma JJ, McKnight CD, HeƩ K, Elenberger J, Han CJ, Song AK, Considine C, Claassen DO, Donahue MJ. Deep learning segmentaƟon of the choroid plexus from structural magneƟc resonance imaging (MRI): validaƟon and normaƟve ranges across the adult lifespan. Fluids Barriers CNS. 2024;21:21. doi: 10.1186/s12987-024-00525-9

work page doi:10.1186/s12987-024-00525-9 2024
[20]

Isensee, M

Isensee F, Schell M, Pﬂueger I, Brugnara G, Bonekamp D, Neuberger U, Wick A, Schlemmer H, Heiland S, Wick W, et al. Automated brain extracƟon of mulƟsequence MRI using arƟﬁcial neural networks. Human Brain Mapping. 2019;40:4952–4964. doi: 10.1002/hbm.24750

work page doi:10.1002/hbm.24750 2019
[21]

Tustison, Philip A

TusƟson NJ, Cook PA, Holbrook AJ, Johnson HJ, Muschelli J, Devenyi GA, Duda JT, Das SR, Cullen NC, Gillen DL, et al. The ANTsX ecosystem for quanƟtaƟve biological and medical imaging. Sci Rep. 2021;11:9068. doi: 10.1038/s41598-021-87564-6

work page doi:10.1038/s41598-021-87564-6 2021
[22]

A contrast-adapƟve method for simultaneous whole-brain and lesion segmentaƟon in mulƟple sclerosis

Cerri S, PuonƟ O, Meier DS, Wuerfel J, Mühlau M, Siebner HR, Van Leemput K. A contrast-adapƟve method for simultaneous whole-brain and lesion segmentaƟon in mulƟple sclerosis. NeuroImage. 2021;225:117471. doi: 10.1016/j.neuroimage.2020.117471

work page doi:10.1016/j.neuroimage.2020.117471 2021
[23]

NeuroImage62(2), 782–790 (2012)

Jenkinson M, Beckmann CF, Behrens TEJ, Woolrich MW , Smith SM. Review FSL. NeuroImage. 2012; doi: 10.1016/j.neuroimage.2011.09.015

work page doi:10.1016/j.neuroimage.2011.09.015 2012
[24]

User-guided 3D acƟve contour segmentaƟon of anatomical structures: Signiﬁcantly improved eﬃciency and reliability

Yushkevich PA, Piven J, HazleƩ HC, Smith RG, Ho S, Gee JC, Gerig G. User-guided 3D acƟve contour segmentaƟon of anatomical structures: Signiﬁcantly improved eﬃciency and reliability. NeuroImage. 2006;31:1116–1128. doi: 10.1016/j.neuroimage.2006.01.015

work page doi:10.1016/j.neuroimage.2006.01.015 2006
[25]

Focal Loss for Dense Object DetecƟon

Lin T-Y , Goyal P , Girshick R, He K, Dollár P . Focal Loss for Dense Object DetecƟon. 2017

2017
[26]

Decoupled weight decay regularizaƟon

Loshchilov I, HuƩer F. Decoupled weight decay regularizaƟon. 7th InternaƟonal Conference on Learning RepresentaƟons, ICLR 2019. 2019

2019
[27]

Metrics for evaluaƟng 3D medical image segmentaƟon: analysis, selecƟon, and tool

Taha AA, Hanbury A. Metrics for evaluaƟng 3D medical image segmentaƟon: analysis, selecƟon, and tool. BMC Med Imaging. 2015;15:29. doi: 10.1186/s12880-015-0068-x

work page doi:10.1186/s12880-015-0068-x 2015
[28]

FastSurfer - A fast and accurate deep learning based neuroimaging pipeline

Henschel L, ConjeƟ S, Estrada S, Diers K, Fischl B, Reuter M. FastSurfer - A fast and accurate deep learning based neuroimaging pipeline. NeuroImage. 2020;219:117012. doi: 10.1016/j.neuroimage.2020.117012

work page doi:10.1016/j.neuroimage.2020.117012 2020
[29]

YOLO in Healthcare: A Comprehensive Review of DetecƟon Architectures, Domain ApplicaƟons, and Future InnovaƟons

Palaniappan D, Jain R, Premavathi T, Parmar K, Ghribi W, Ahmed AM, Ahmad N. YOLO in Healthcare: A Comprehensive Review of DetecƟon Architectures, Domain ApplicaƟons, and Future InnovaƟons. IEEE Access. 2025;13:145714–145735. doi: 10.1109/ACCESS.2025.3599358. Figures Figure 1. Lateral ventricle choroid plexus segmentaƟon of a person with mulƟple sclerosis....

work page doi:10.1109/access.2025.3599358 2025

[1] [1]

Choroid plexus volume changes in mulƟple sclerosis: insights from a systemaƟc review and meta-analysis of magneƟc resonance imaging studies

Kolahi S, Zarei D, Issaiy M, Shakiba M, Azizi N, Firouznia K. Choroid plexus volume changes in mulƟple sclerosis: insights from a systemaƟc review and meta-analysis of magneƟc resonance imaging studies. Neuroradiology. 2024;c:1869–1886. doi: 10.1007/s00234-024-03439-3. Cited in PMID: 39105769

work page doi:10.1007/s00234-024-03439-3 2024

[2] [2]

Choroid Plexus Volume Change—A Candidate for a New Radiological Marker of MS Progression

Jankowska A, Chwojnicki K, Grzywińska M, Trzonkowski P , Szurowska E. Choroid Plexus Volume Change—A Candidate for a New Radiological Marker of MS Progression. DiagnosƟcs. 2023;13. doi: 10.3390/diagnosƟcs13162668

work page doi:10.3390/diagnos 2023

[3] [3]

Pathogenesis of Alzheimer’s disease: Involvement of the choroid plexus

Čarna M, Onyango IG, KaƟna S, Holub D, Novotny JS, Nezvedova M, Jha D, Nedelska Z, Lacovich V, Vyvere T Vande, et al. Pathogenesis of Alzheimer’s disease: Involvement of the choroid plexus. Alzheimer’s and DemenƟa. 2023;19:3537–3554. doi: 10.1002/alz.12970. Cited in PMID: 36825691

work page doi:10.1002/alz.12970 2023

[4] [4]

AssociaƟon of choroid plexus volume with motor symptoms and dopaminergic degeneraƟon in Parkinson’s disease

Jeong SH, Park CJ, Jeong HJ, Sunwoo MK, Ahn SS, Lee SK, Lee PH, Kim YJ, Sohn YH, Chung SJ. AssociaƟon of choroid plexus volume with motor symptoms and dopaminergic degeneraƟon in Parkinson’s disease. Journal of Neurology, Neurosurgery and Psychiatry. 2023;94:1047–1055. doi: 10.1136/jnnp-2023-331170. Cited in PMID: 37399288

work page doi:10.1136/jnnp-2023-331170 2023

[5] [5]

Neuropathological and cerebrospinal ﬂuid correlates of choroid plexus inﬂammaƟon in progressive mulƟple sclerosis

Magliozzi R, Hametner S, Mastantuono M, Mensi A, Karimian M, Griﬃths L, Watkins LM, Poli A, BerƟ GM, Barusolo E, et al. Neuropathological and cerebrospinal ﬂuid correlates of choroid plexus inﬂammaƟon in progressive mulƟple sclerosis. Brain Pathology. 2024;1–20. doi: 10.1111/bpa.13322

work page doi:10.1111/bpa.13322 2024

[6] [6]

Chronic AcƟve Lesions and Larger Choroid Plexus Explain CogniƟon and FaƟgue in MulƟple Sclerosis

Preziosa P , Pagani E, Meani A, Storelli L, Margoni M, Yudin Y , Tedone N, Biondi D, Rubin M, Rocca MA, et al. Chronic AcƟve Lesions and Larger Choroid Plexus Explain CogniƟon and FaƟgue in MulƟple Sclerosis. Neurol Neuroimmunol Neuroinﬂamm. 2024;11:e200205. doi: 10.1212/NXI.0000000000200205

work page doi:10.1212/nxi.0000000000200205 2024

[7] [7]

Choroid Plexus Enlargement in MulƟple Sclerosis Correlates with CorƟcal and Phase Rim Lesions on 7T MRI and Predicts Progression Independent of Relapse AcƟvity

BarbuƟ E, ConƟ A, Treaba CA, Miscioscia A, BarleƩa VT , Herranz E, Sloane JA, Klawiter EC, Toschi N, Mainero C. Choroid Plexus Enlargement in MulƟple Sclerosis Correlates with CorƟcal and Phase Rim Lesions on 7T MRI and Predicts Progression Independent of Relapse AcƟvity. AJNR Am J Neuroradiol. 2026;47:557–565. doi: 10.3174/ajnr.A8983

work page doi:10.3174/ajnr.a8983 2026

[8] [8]

Choroid Plexus Enlargement in Inﬂammatory MulƟple Sclerosis: 3.0-T MRI and Translocator Protein PET EvaluaƟon

Ricigliano VAG, Morena E, Colombi A, TonieƩo M, Hamzaoui M, Poirion E, BoƩlaender M, Gervais P , Louapre C, Bodini B, et al. Choroid Plexus Enlargement in Inﬂammatory MulƟple Sclerosis: 3.0-T MRI and Translocator Protein PET EvaluaƟon. Radiology. 2021;301:166–177. doi: 10.1148/radiol.2021204426

work page doi:10.1148/radiol.2021204426 2021

[9] [9]

Understanding the Role of the Choroid Plexus in MulƟple Sclerosis as an MRI Biomarker of Disease AcƟvity

Müller J, Noteboom S, Granziera C, Schoonheim MM. Understanding the Role of the Choroid Plexus in MulƟple Sclerosis as an MRI Biomarker of Disease AcƟvity. Neurology. 2023;100:405–406. doi: 10.1212/WNL.0000000000206806. Cited in PMID: 36543568

work page doi:10.1212/wnl.0000000000206806 2023

[10] [10]

Choroid Plexus Volume in MulƟple Sclerosis vs NeuromyeliƟs OpƟca Spectrum Disorder: A RetrospecƟve, Cross-secƟonal Analysis

Müller J, Sinnecker T, Wendebourg MJ, Schläger R, Kuhle J, Schädelin S, Benkert P , Derfuss T , Caƫ n P , Jud C, et al. Choroid Plexus Volume in MulƟple Sclerosis vs NeuromyeliƟs OpƟca Spectrum Disorder: A RetrospecƟve, Cross-secƟonal Analysis. Neurol Neuroimmunol Neuroinﬂamm. 2022;9:e1147. doi: 10.1212/NXI.0000000000001147. Cited in PMID: 35217580

work page doi:10.1212/nxi.0000000000001147 2022

[11] [11]

Morphological changes of the choroid plexus in the lateral ventricle across the lifespan: 5551 subjects from fetus to elderly

Li J, Gao Y , Xu Y , Dai W, Hu Y , Feng X, Wu D, Zhao L. Morphological changes of the choroid plexus in the lateral ventricle across the lifespan: 5551 subjects from fetus to elderly. NeuroImage. 2025;318:121392. doi: 10.1016/j.neuroimage.2025.121392

work page doi:10.1016/j.neuroimage.2025.121392 2025

[12] [12]

ASCHOPLEX : a generalizable approach for the automaƟc segmentaƟon of choroid plexus

Visani V, Veronese M, Pizzini FB, Colombi A, Marjin C, TamanƟ A, Schubert JJ, Althubaity N, Harrison NA, Bullmore ET , et al. ASCHOPLEX : a generalizable approach for the automaƟc segmentaƟon of choroid plexus. 2024;182. doi: 10.1016/j.compbiomed.2024.109164

work page doi:10.1016/j.compbiomed.2024.109164 2024

[13] [13]

U-Net: ConvoluƟonal Networks for Biomedical Image SegmentaƟon [Internet]

Ronneberger O, Fischer P , Brox T. U-Net: ConvoluƟonal Networks for Biomedical Image SegmentaƟon [Internet]. arXiv; 2015 [cited 2025 Nov 24]. Available from: hƩps://arxiv.org/abs/1505.04597

Pith/arXiv arXiv 2015

[14] [14]

doi:10.1016/j.neuroimage.2012.01.021

Fischl B. FreeSurfer. NeuroImage. 2012;62:774–781. doi: 10.1016/j.neuroimage.2012.01.021

work page doi:10.1016/j.neuroimage.2012.01.021 2012

[15] [15]

Deep learning-based SegmentaƟon of the Choroid Plexus in MulƟple Sclerosis Using MPRAGE, FLAIR, and MP2RAGE

Lu P-J, Cagol A, Ocampo-Pineda M, Mastantuono M, Aldea A-A, Müller J, Yaldizli Ö, Weigel M, Melie-Garcia L, Magliozzi R, et al. Deep learning-based SegmentaƟon of the Choroid Plexus in MulƟple Sclerosis Using MPRAGE, FLAIR, and MP2RAGE. Mult Scler . 2025;31:1032. doi: 10.1177/13524585251358343

work page doi:10.1177/13524585251358343 2025

[16] [16]

Swin UNETR: Swin Transformers for SemanƟc SegmentaƟon of Brain Tumors in MRI Images [Internet]

Hatamizadeh A, Nath V, Tang Y , Yang D, Roth H, Xu D. Swin UNETR: Swin Transformers for SemanƟc SegmentaƟon of Brain Tumors in MRI Images [Internet]. arXiv; 2022 [cited 2026 Feb 4]. Available from: hƩp://arxiv.org/abs/2201.01266

arXiv 2022

[17] [17]

3D UX-Net: A Large Kernel Volumetric ConvNet Modernizing Hierarchical Transformer for Medical Image SegmentaƟon [Internet]

Lee HH, Bao S, Huo Y , Landman BA. 3D UX-Net: A Large Kernel Volumetric ConvNet Modernizing Hierarchical Transformer for Medical Image SegmentaƟon [Internet]. arXiv; 2022 [cited 2025 Nov 24]. Available from: hƩps://arxiv.org/abs/2209.15076

arXiv 2022

[18] [18]

Enhanced choroid plexus segmentaƟon with 3D UX-Net and its associaƟon with disease progression in mulƟple sclerosis

Wang X, Wang X, Yan Z, Yin F, Li Y , Liu X, Liu Y . Enhanced choroid plexus segmentaƟon with 3D UX-Net and its associaƟon with disease progression in mulƟple sclerosis. MulƟple Sclerosis and Related Disorders. 2024;88:105750. doi: 10.1016/j.msard.2024.105750

work page doi:10.1016/j.msard.2024.105750 2024

[19] [19]

Deep learning segmentaƟon of the choroid plexus from structural magneƟc resonance imaging (MRI): validaƟon and normaƟve ranges across the adult lifespan

Eisma JJ, McKnight CD, HeƩ K, Elenberger J, Han CJ, Song AK, Considine C, Claassen DO, Donahue MJ. Deep learning segmentaƟon of the choroid plexus from structural magneƟc resonance imaging (MRI): validaƟon and normaƟve ranges across the adult lifespan. Fluids Barriers CNS. 2024;21:21. doi: 10.1186/s12987-024-00525-9

work page doi:10.1186/s12987-024-00525-9 2024

[20] [20]

Isensee, M

Isensee F, Schell M, Pﬂueger I, Brugnara G, Bonekamp D, Neuberger U, Wick A, Schlemmer H, Heiland S, Wick W, et al. Automated brain extracƟon of mulƟsequence MRI using arƟﬁcial neural networks. Human Brain Mapping. 2019;40:4952–4964. doi: 10.1002/hbm.24750

work page doi:10.1002/hbm.24750 2019

[21] [21]

Tustison, Philip A

TusƟson NJ, Cook PA, Holbrook AJ, Johnson HJ, Muschelli J, Devenyi GA, Duda JT, Das SR, Cullen NC, Gillen DL, et al. The ANTsX ecosystem for quanƟtaƟve biological and medical imaging. Sci Rep. 2021;11:9068. doi: 10.1038/s41598-021-87564-6

work page doi:10.1038/s41598-021-87564-6 2021

[22] [22]

A contrast-adapƟve method for simultaneous whole-brain and lesion segmentaƟon in mulƟple sclerosis

Cerri S, PuonƟ O, Meier DS, Wuerfel J, Mühlau M, Siebner HR, Van Leemput K. A contrast-adapƟve method for simultaneous whole-brain and lesion segmentaƟon in mulƟple sclerosis. NeuroImage. 2021;225:117471. doi: 10.1016/j.neuroimage.2020.117471

work page doi:10.1016/j.neuroimage.2020.117471 2021

[23] [23]

NeuroImage62(2), 782–790 (2012)

Jenkinson M, Beckmann CF, Behrens TEJ, Woolrich MW , Smith SM. Review FSL. NeuroImage. 2012; doi: 10.1016/j.neuroimage.2011.09.015

work page doi:10.1016/j.neuroimage.2011.09.015 2012

[24] [24]

User-guided 3D acƟve contour segmentaƟon of anatomical structures: Signiﬁcantly improved eﬃciency and reliability

Yushkevich PA, Piven J, HazleƩ HC, Smith RG, Ho S, Gee JC, Gerig G. User-guided 3D acƟve contour segmentaƟon of anatomical structures: Signiﬁcantly improved eﬃciency and reliability. NeuroImage. 2006;31:1116–1128. doi: 10.1016/j.neuroimage.2006.01.015

work page doi:10.1016/j.neuroimage.2006.01.015 2006

[25] [25]

Focal Loss for Dense Object DetecƟon

Lin T-Y , Goyal P , Girshick R, He K, Dollár P . Focal Loss for Dense Object DetecƟon. 2017

2017

[26] [26]

Decoupled weight decay regularizaƟon

Loshchilov I, HuƩer F. Decoupled weight decay regularizaƟon. 7th InternaƟonal Conference on Learning RepresentaƟons, ICLR 2019. 2019

2019

[27] [27]

Metrics for evaluaƟng 3D medical image segmentaƟon: analysis, selecƟon, and tool

Taha AA, Hanbury A. Metrics for evaluaƟng 3D medical image segmentaƟon: analysis, selecƟon, and tool. BMC Med Imaging. 2015;15:29. doi: 10.1186/s12880-015-0068-x

work page doi:10.1186/s12880-015-0068-x 2015

[28] [28]

FastSurfer - A fast and accurate deep learning based neuroimaging pipeline

Henschel L, ConjeƟ S, Estrada S, Diers K, Fischl B, Reuter M. FastSurfer - A fast and accurate deep learning based neuroimaging pipeline. NeuroImage. 2020;219:117012. doi: 10.1016/j.neuroimage.2020.117012

work page doi:10.1016/j.neuroimage.2020.117012 2020

[29] [29]

YOLO in Healthcare: A Comprehensive Review of DetecƟon Architectures, Domain ApplicaƟons, and Future InnovaƟons

Palaniappan D, Jain R, Premavathi T, Parmar K, Ghribi W, Ahmed AM, Ahmad N. YOLO in Healthcare: A Comprehensive Review of DetecƟon Architectures, Domain ApplicaƟons, and Future InnovaƟons. IEEE Access. 2025;13:145714–145735. doi: 10.1109/ACCESS.2025.3599358. Figures Figure 1. Lateral ventricle choroid plexus segmentaƟon of a person with mulƟple sclerosis....

work page doi:10.1109/access.2025.3599358 2025