Pith. sign in

REVIEW 3 major objections 6 minor 45 references

Unified HT-CNNs Architecture: Transfer Learning for Segmenting Diverse Brain Tumors in MRI from Gliomas to Pediatric Tumors

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read A unified ensemble of hybrid transformers and CNNs, pretrained on adult gliomas and fine-tuned on small pediatric and Sub-Saharan datasets, substantially improves brain tumor segmentation.

desk verdict Fine-tuning gains are real and useful, but the superiority claim over previous BraTS winners is not supported by the uncontrolled Table 2 comparison. read the letter →

arxiv 2412.08240 v1 pith:LDATE2LW submitted 2024-12-11 eess.IV cs.CVcs.LG

Ramy A. Zeineldin , Franziska Mathis-Ullrich This is my paper · ORCID
classification eess.IVcs.CVcs.LG
keywords braintumorsegmentationtransferlearningensemblehybridtransformerconvolutionalneuralnetworkMRIpediatrictumorsSub-SaharanAfrica
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

HT-CNNs is a single segmentation pipeline that combines three neural networks—a convolutional U-Net, a transformer-based segmenter, and an attention-enhanced decoder—and fuses their predictions with the STAPLE label-fusion method. The paper's central claim is that transfer learning from 1,251 adult glioma MRI cases to small pediatric (99 cases) and Sub-Saharan African (60 cases) datasets dramatically improves tumor boundary accuracy: pediatric Dice rises from 0.4097 to 0.6248 and HD95 falls from 163.37 to 37.46. On the adult glioma validation set, the ensemble reports an average Dice of 0.8842 and HD95 of 10.199, matching or beating the listed winners of previous years. The importance, if the claim holds, is that one architecture can serve many tumor types without large in-domain training sets.

What carries the argument

The load-bearing mechanism is the transfer-learning chain: pretraining on the full adult glioma dataset of 1,251 cases, fine-tuning the same weights on the small pediatric and African datasets, and then fusing the three network outputs with the STAPLE algorithm, which estimates a consensus label map from the individual predictions while weighting each network by its estimated reliability. The three components are a 3D U-Net backbone for local spatial features, a transformer encoder (tokenized volume patches) for global context, and an axial-attention decoder for refining boundaries. A two-stage post-processing rule replaces small or low-confidence enhancing-tumor predictions with necrosis and removes tiny connected components, which the authors say improves lesion-wise scores.

What would settle it

Re-run the previous winner models (2020–2023) on the same adult glioma validation cases used for HT-CNNs and compare the reported DSC and HD95; if the winner rows in Table 2 do not match the re-measured values, the claim of superiority over those winners is disproven. A quicker check is to see whether the row labeled 'Winner 2020' reproduces exactly the authors' own nnU-Net baseline scores, since that would suggest those numbers came from a baseline run rather than from an independent winner submission.

Watch

Extended reading notes

Core claim

The discovery is that an ensemble of a CNN, a transformer, and an attention-based network, when pretrained on adult gliomas and then fine-tuned on small specialized datasets, transfers learned representations well enough to lift segmentation accuracy on pediatric and Sub-Saharan African tumors to levels far above a from-scratch baseline. The reported evidence is the quantitative jump: for pediatric cases, Dice rises from 0.4097 to 0.6248 with fine-tuning and to 0.7621 when ensembling is added, while HD95 drops from 163.37 to 37.46 then to 25.72; for Sub-Saharan cases, Dice rises from 0.7832 to 0.8647 to 0.8872 and HD95 from 18.38 to 10.98 to 4.29. The same pipeline achieves an average Dice of 0.8842 on adult glioma validation, which the authors state is on par with the 2023 winner and higher than the 2020–2022 winners. This is presented as evidence that hybrid architectures plus transfer learning offer a unified solution across brain tumor types and demographics.

Load-bearing premise

The sweeping claim of beating previous challenge winners assumes that the winner scores listed for 2020–2023 were measured on the same validation set as the authors' models and are correctly attributed; if those numbers are not apples-to-apples, the superiority claim loses its support even though the fine-tuning gains may still hold.

Editorial extensions

If this is right

  • If the transfer effect is real, a model pretrained on a large adult glioma collection can be adapted to a rare tumor type with only tens of annotated scans, cutting the annotation burden for new challenges.
  • The fine-tuning gains imply that the domain gap between adult and pediatric or African populations is substantially reducible in MRI segmentation, at least when preprocessing and label conventions are shared.
  • The reported parity with prior winners suggests that ensembling heterogeneous architectures—CNN, transformer, and attention decoder—remains a competitive strategy for benchmark segmentation.
  • The post-processing rules (voxel-count and probability thresholds) indicate that simple heuristic corrections can have a large effect on lesion-wise metrics, which the community may incorporate into their own pipelines.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural extension would be to test whether the same fine-tuning recipe transfers to meningioma or metastasis sub-challenges, which use different label schemes; the method's current results do not establish that.
  • The paper's Table 2 comparison may overstate the advantage over prior winners if the cited winner scores were not computed on the identical validation set; one way to check is to rerun those exact winner models on the authors' validation split.
  • Because the reported pediatric boost comes from only 99 training cases, the method suggests a steep learning curve for transfer; measuring DSC at 20, 40, and 60 fine-tuning cases would quantify how much data is actually needed.
  • If the approach is adopted in clinical practice, the ensemble's runtime and GPU demands (stated as a 4090 GPU) would need to be weighed against the accuracy gains, something the paper does not quantify.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The manuscript proposes HT-CNNs, an ensemble combining nnU-Net, TransBTS, and DeepSCAN with STAPLE label fusion, for brain tumor segmentation in the BraTS 2023 cluster of challenges, specifically the adult glioma (GLA), pediatric (PED), and Sub-Saharan African (SSA) datasets. The authors evaluate the ensemble on the GLA validation set, compare it with previous BraTS winners in Table 2, and report fine-tuning experiments in which models pre-trained on adult glioma data are adapted to the PED and SSA datasets, with large improvements in DSC and reductions in HD95. The paper also describes a two-stage post-processing strategy and reports lesion-wise metrics on the validation sets.

Significance. If the fine-tuning and ensemble gains are valid, the work is a useful contribution to cross-domain brain tumor segmentation, particularly for small pediatric and SSA datasets, and the publicly available Docker image is a practical reproducibility asset. The internal comparisons in Tables 3 and 4 are plausible and show large improvements from fine-tuning and ensembling. However, the headline claim of superiority over previous BraTS winners is not yet established: the comparison in Table 2 is not shown to be a controlled same-validation-set benchmark, and the manuscript reports no test-set results or statistical significance testing. The reported validation-set improvements may also be partly driven by post-processing thresholds tuned on the same data. The core transfer-learning idea and the reproducibility artifacts are strengths, but the central comparison claim needs substantial revision.

major comments (3)
  1. [Results, Table 2] Table 2 does not support the abstract's claim of 'superior segmentation results ... over the previous winning methods.' The Winner 2020 row cites the nnU-Net paper [10] yet its DSC values (0.8402, 0.8718, 0.9213, avg 0.8778) are identical to the authors' own nnU-Net baseline in Table 1, with only slightly different HD95 values; this strongly suggests that the row is a re-used baseline rather than an independent evaluation of the actual 2020 winner on the BraTS 2023 GLA validation set. The Winner 2022 row cites the authors' own BraTS 2022 solution paper [28] without establishing official-winner status. Prior winners were ranked on their own edition's validation and test sets, which differ in cohort composition, scanner protocols, and label distributions, so the numbers cannot be compared directly against HT-CNNs on the 2023 GLA validation set. The authors should either recompute all prior-winner methods on the same 2023 GLA validation cases or cite the official challenge leaderboard values with clear provenance, and then re-evaluate the superiority claim.
  2. [Post-processing Strategy] The post-processing thresholds (the enhancing-tumor replacement threshold, the 16-voxel connected-component cutoff, the 0.9 probability cutoff, and the 73-voxel count cutoff) were 'fine-tuned via cross-validation on the mean Dice and ranking scores,' and the results are reported on the same BraTS validation sets. This creates a selection-on-validation risk: the improvements in Tables 3-5, and the HT-CNNs ranking in Table 2, may be inflated by tuning to the same data used for evaluation. Please clarify whether the thresholds were selected only on training folds and provide at least one hold-out or official test-set evaluation to confirm that the gains are not an artifact of validation-set tuning.
  3. [Experiments and Results] No test-set evaluation is provided for any of the BraTS 2023 tasks, and the DSC and HD95 differences in Tables 3 and 4 are reported without confidence intervals or significance tests. Because the BraTS 2023 challenge provides official test labels, the authors should report the test-set results for GLA, PED, and SSA if available, or state explicitly that they are unavailable. Without independent test-set confirmation, the claims of 'superior' performance and 'substantial enhancement' remain limited to validation sets that were also used for post-processing selection.
minor comments (6)
  1. [Figure 3] The caption uses 'EC' for the enhancing tumor while the text and tables consistently use 'ET'; please make the notation uniform.
  2. [Tables 3 and 4] The 'Baseline' row is not defined in the main text; specify whether it is the model without fine-tuning and without ensembling, and clarify what 'Baseline + TR + EN' adds beyond 'Baseline + TR'.
  3. [Methods, Network Components] The Transformer Network and Attention Network subsections describe generic components but do not explicitly map them onto TransBTS and DeepSCAN as used in Figure 1; in particular, DeepSCAN is described as an attention-focused component even though the original DeepSCAN is a CNN. Please clarify the mapping and the role of axial attention in each component.
  4. [Table 5] The lesion-wise comparison to the BraTS-PED 2023 winner CNMCPMI2023 is mentioned only in the text; adding a row to Table 5 would make the comparison direct and transparent.
  5. [Introduction] The sentence stating that the upper layers of U-Net capture broad contextual information and the lower layers are rich in spatial detail is inverted relative to the standard low-level/high-level feature terminology; please correct it.
  6. [Introduction, Contributions] The contributions state that the framework is tailored for meningioma and brain metastasis segmentation, but no experiments or results are reported for those tasks; either add those results or revise the claim to match the evaluated scope.

Circularity Check

3 steps flagged · score 6.0 of 10

Validation-set performance is partially circular: the post-processing ET threshold was tuned on the same validation metrics used to report 'superior' results, and Table 2's 'previous winners' include the authors' own nnU-Net baseline and own BraTS 2022 paper.

  1. fitted input called prediction [Post-processing Strategy; Experiments and Results, BraTS 2023 validation sets (Tables 1-5)]
    "In cases where the reference segmentation lacked enhancing tumor regions, we removed enhancing tumor predictions below a certain threshold, replacing them with necrosis predictions. This approach optimized the model's ranking while managing potential true positive losses, with the threshold fine-tuned via cross-validation on the mean Dice and ranking scores."

    The reported validation-set DSC, HD95, and ranking scores are the very metrics used to tune the post-processing ET threshold. A free parameter was therefore selected to optimize the same outcome on the same validation cases that are later presented as evidence of the model's superiority. The validation results are thus not an independent prediction of the model's performance; one component of the reported score is fitted to the evaluation set by construction.

  2. renaming known result [Table 1 vs Table 2, BraTS 2023 GLA validation results]
    "Table 1 ... nnU-Net 0.8402 0.8718 0.9213 0.8778 ... Table 2 ... Winner 2020 [10] 0.8402 0.8718 0.9213 0.8778"

    The DSC values for 'Winner 2020' in Table 2 are numerically identical, to four decimal places in all three subregions and the average, to the paper's own nnU-Net baseline in Table 1. The claimed superiority over 'Winner 2020' is therefore just the ensemble's advantage over one of its own components relabeled as a prior champion. This is a self-referential comparison rather than an evaluation against an independently measured previous winning method.

1 more flagged steps
  1. self citation load bearing [Table 2, row 'Winner 2022 [28]' and Reference [28]]
    "Winner 2022 [28] ... Zeineldin RA, Karar ME, Burgert O, Mathis-Ullrich F (2023) Multimodal CNN Networks for Brain Tumor Segmentation in MRI: A BraTS 2022 Challenge Solution."

    The baseline labeled 'Winner 2022' is the authors' own prior challenge paper, not an independently established external winner. The paper's headline claim of outperforming 'previous winning methods' depends on this row being the official 2022 champion, but the cited source is the authors' own solution. The comparison to 'previous winners' thus partly reduces to a self-citation that establishes the baseline identity, making the claimed superiority over prior winners depend on the authors' own prior work.

full rationale

The paper's central claim that HT-CNNs 'achieves superior segmentation results across the BraTS validation datasets over the previous winning methods' is supported mainly by Table 2 and by the validation-set metric improvements in Tables 1, 3, and 4. The fine-tuning improvements for PED and SSA are empirical measurements and are not inherently circular: they compare the same model with and without transfer learning. However, the headline 'superior over previous winners' claim has two self-referential components. First, the post-processing ET threshold was explicitly tuned via cross-validation on the mean Dice and ranking scores, and the results are reported on the same BraTS 2023 validation sets; this makes the validation performance at least partly a fitted outcome rather than a pure prediction. Second, Table 2's 'Winner 2020' row has DSC values exactly matching the paper's own nnU-Net baseline, and 'Winner 2022' cites the authors' own prior BraTS 2022 paper. These baselines are not independently verified external winners, so the claimed superiority reduces in part to comparing the ensemble against its own components and prior self-authored systems. The fine-tuning results may remain valid, but the central claim of superiority over previous winning methods is partially circular because it depends on self-referential baselines and on metrics that were optimized on the evaluation set. Score 6 reflects this partial circularity rather than a fully circular derivation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claims rest on standard deep learning assumptions: expert ground truth, transferability of pretrained weights to 3D MRI, and usefulness of adult glioma features for other tumor types. The post-processing rules introduce hand-chosen thresholds that are fitted to performance metrics. No new physical entities are introduced.

free parameters (5)
  • ET replacement threshold = not reported
    Threshold for replacing enhancing-tumor predictions with necrosis was fine-tuned via cross-validation on mean Dice and ranking scores (Post-processing Strategy); value not disclosed.
  • Connected component size and probability cutoffs = 16 voxels, 0.9
    Components smaller than 16 voxels with mean probability below 0.9 are relabeled as necrosis, a hand-chosen rule.
  • ET voxel count and probability cutoffs = 73 voxels, 0.9
    If predicted ET voxel count is below 73 with mean probability below 0.9, all ET voxels are replaced by necrosis.
  • Sliding window overlap = 0.6
    Hand-chosen inference parameter in Implementation Details.
  • Training epochs = 500
    Hand-chosen training duration in Implementation Details.
assumptions (4)
  • domain assumption The BraTS ground-truth annotations are expert-validated and correct for all subregions.
    The paper uses BraTS labels as ground truth without independent verification (Data section).
  • domain assumption ImageNet-1K pre-trained weights transfer to 3D MRI feature embeddings in TransBTS.
    Implementation Details state the Transformer was pre-trained on ImageNet-1K weights and fine-tuned on 3D MRI; no evidence is given that this transfer is valid for volumetric medical data.
  • domain assumption Adult glioma representations transfer to pediatric and Sub-Saharan African tumors.
    The transfer learning strategy assumes shared features across tumor types and demographics (Transfer Learning section); this is the hypothesis under test, not an independently established fact.
  • domain assumption STAPLE yields a better consensus segmentation than simple averaging or individual models.
    The paper relies on STAPLE [20] as its ensemble method without comparing it to other fusion strategies.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unified HT-CNNs Architecture: Transfer Learning for Segmenting Diverse Brain Tumors in MRI from Gliomas to Pediatric Tumors." pith.science (2026). https://pith.science/paper/LDATE2LW

@misc{pith2026241208240,
  author       = {Pith},
  title        = {Pith review of: Unified HT-CNNs Architecture: Transfer Learning for Segmenting Diverse Brain Tumors in MRI from Gliomas to Pediatric Tumors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LDATE2LW}},
  note         = {Machine review of arXiv:2412.08240}
}
read the original abstract

Accurate segmentation of brain tumors from 3D multimodal MRI is vital for diagnosis and treatment planning across diverse brain tumors. This paper addresses the challenges posed by the BraTS 2023, presenting a unified transfer learning approach that applies to a broader spectrum of brain tumors. We introduce HT-CNNs, an ensemble of Hybrid Transformers and Convolutional Neural Networks optimized through transfer learning for varied brain tumor segmentation. This method captures spatial and contextual details from MRI data, fine-tuned on diverse datasets representing common tumor types. Through transfer learning, HT-CNNs utilize the learned representations from one task to improve generalization in another, harnessing the power of pre-trained models on large datasets and fine-tuning them on specific tumor types. We preprocess diverse datasets from multiple international distributions, ensuring representativeness for the most common brain tumors. Our rigorous evaluation employs standardized quantitative metrics across all tumor types, ensuring robustness and generalizability. The proposed ensemble model achieves superior segmentation results across the BraTS validation datasets over the previous winning methods. Comprehensive quantitative evaluations using the DSC and HD95 demonstrate the effectiveness of our approach. Qualitative segmentation predictions further validate the high-quality outputs produced by our model. Our findings underscore the potential of transfer learning and ensemble approaches in medical image segmentation, indicating a substantial enhancement in clinical decision-making and patient care. Despite facing challenges related to post-processing and domain gaps, our study sets a new precedent for future research for brain tumor segmentation. The docker image for the code and models has been made publicly available, https://hub.docker.com/r/razeineldin/ht-cnns.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 25 canonical work pages

  1. [10]

    0.8402 0.8718 0.9213 0.8778 16.028 8.953 3.823 9.601 Winner 2021

    work page 2021

  2. [28]

    0.8438 0.8753 0.9271 0.8821 17.504 7.533 3.595 9.544 Winner 2023

    work page 2023

  3. [1]

    International Journal of Computer Assisted Radiology and Surgery (2024) 1 Unified HT-CNNs Architecture: Transfer Learning for Segmenting Diverse Brain Tumors in MRI from Gliomas to Pediatric Tumors Ramy A. Zeineldin1,2* · Franziska Mathis-Ullrich1 1Department of Artificial Intelligence in Biomedical Engineering (AIBE), Friedrich-Alexander-University Erlan...

    work page 2024

  4. [2]

    Transfer Learning In our HT-CNNs framework, transfer learning played a pivotal role in enhancing model performance across diverse brain tumor segmentation tasks

    Axial attention is employed individually for each axis in the output obtained from transpose convolution [25]. Transfer Learning In our HT-CNNs framework, transfer learning played a pivotal role in enhancing model performance across diverse brain tumor segmentation tasks. Transfer learning is a machine learning technique where a model developed for a 6 pa...

    work page 2023

  5. [3]

    The predictions displayed in the figure are generated using our proposed model, and the rows correspond to samples from the GLA, SSA, and PED datasets

    These results showcase the effectiveness of our approach in accurately segmenting brain tumors across different tasks. The predictions displayed in the figure are generated using our proposed model, and the rows correspond to samples from the GLA, SSA, and PED datasets. From Figure 3, it is evident that HT-CNNs model produces segmentation results of high ...

    work page 2023

  6. [4]

    Sci Data 4 (1):170117

    Bakas S, Akbari H, Sotiras A, Bilello M, Rozycki M, Kirby JS, Freymann JB, Farahani K, Davatzikos C (2017) Advancing The Cancer Genome Atlas glioma MRI collections with expert segmentation labels and radiomic features. Sci Data 4 (1):170117. doi:10.1038/sdata.2017.117

    doi:10.1038/sdata.2017.117 2017

  7. [5]

    It achieved an average DSC of 0.8842, which is on par with the winner of 2023 and higher than the winners of 2020, 2021, and

    work page 2023

  8. [6]

    The cancer imaging archive 286

    Bakas S, Akbari H, Sotiras A, Bilello M, Rozycki M, Kirby J, Freymann J, Farahani K, Davatzikos C (2017) Segmentation labels and radiomic features for the pre-operative scans of the TCGA-LGG collection. The cancer imaging archive 286

    work page 2017

Show all 45 references
  1. [7]

    doi:10.48550/arXiv.2305.19369

    Adewole M, Rudie JD, Gbadamosi A, Toyobo O, Raymond C, Zhang D, Omidiji O, Akinola R, Abba Suwaid M, Emegoakor A, Ojo N, Aguh K, Kalaiwo C, Babatunde G, Ogunleye A, Gbadamosi Y, Iorpagher K, Calabrese E, Aboian M, Linguraru M, Albrecht J, Wiestler B, Kofler F, Janas A, LaBella...

    2023 doi arXiv

  2. [8]

    doi:10.48550/arXiv.2305.17033

    Fathi Kazerooni A, Khalili N, Liu X, Haldar D, Jiang Z, Muhammed Anwar S, Albrecht J, Adewole M, Anazodo U, Anderson H, Bagheri S, Baid U, Bergquist T, Borja AJ, Calabrese E, Chung V, Conte G-M, Dako F, Eddy J, Ezhov I, Familiar A, Farahani K, Haldar S, Eugenio Iglesias J, Jan...

    2023 doi arXiv

  3. [9]

    In: Crimi A, Bakas S (eds) Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries, Cham, 2022//

    Zeineldin RA, Karar ME, Mathis-Ullrich F, Burgert O Ensemble CNN Networks for GBM Tumors Segmentation Using Multi-parametric MRI. In: Crimi A, Bakas S (eds) Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries, Cham, 2022//

    2022

  4. [11]

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 9351:234-241

    Ronneberger O, Fischer P, Brox T (2015) U-net: Convolutional networks for biomedical image segmentation. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 9351:234-241. doi:10.1007/978-3-319-24574-4_28

    2015 doi

  5. [12]

    Brain Pathol 30 (4):844-856

    Louis DN, Wesseling P, Aldape K, Brat DJ, Capper D, Cree IA, Eberhart C, Figarella-Branger D, Fouladi M, Fuller GN, Giannini C, Haberler C, Hawkins C, Komori T, Kros JM, Ng HK, Orr BA, Park SH, Paulus W, Perry A, Pietsch 12 T, Reifenberger G, Rosenblum M, Rous B, Sahm F, Sarka...

    2020

  6. [13]

    Baid U, Ghodasara S, Bilello M, Mohan S, Calabrese E, Colak E, Farahani K, Kalpathy-Cramer J, Kitamura FC, Pati S, Prevedello LM, Rudie JD, Sako C, Shinohara RT, Bergquist T, Chai R, Eddy J, Elliott J, Reade W, Schaffter T, Yu T, Zheng J, Annotators B, Davatzikos C, Mongan J, ...

    2021 arXiv

  7. [14]

    IEEE Trans Med Imaging 34 (10):1993-2024

    Menze BH, Jakab A, Bauer S, Kalpathy-Cramer J, Farahani K, Kirby J, Burren Y, Porz N, Slotboom J, Wiest R, Lanczi L, Gerstner E, Weber MA, Arbel T, Avants BB, Ayache N, Buendia P, Collins DL, Cordier N, Corso JJ, Criminisi A, Das T, Delingette H, Demiralp C, Durst CR, Dojat M,...

    2015

  8. [15]

    Advances in neural information processing systems 30

    Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, Kaiser Ł, Polosukhin I (2017) Attention is all you need. Advances in neural information processing systems 30

    2017

  9. [16]

    The Cancer Imaging Archive

    Bakas S, Akbari H, Sotiras A (2017) Segmentation labels for the pre-operative scans of the TCGA-GBM collection. The Cancer Imaging Archive. ed,

    2017

  10. [17]

    Sci Rep 14 (1):3713

    Zeineldin RA, Karar ME, Elshaer Z, Coburger J, Wirtz CR, Burgert O, Mathis-Ullrich F (2024) Explainable hybrid vision transformers and convolutional network for multimodal glioma segmentation in brain MRI. Sci Rep 14 (1):3713. doi:10.1038/s41598-024-54186-7

    2024 doi

  11. [18]

    Comput Methods Programs Biomed 229:107307

    Zheng J, Liu H, Feng Y, Xu J, Zhao L (2023) CASF-Net: Cross-attention and cross-scale fusion network for medical image segmentation. Comput Methods Programs Biomed 229:107307. doi:10.1016/j.cmpb.2022.107307

    2023

  12. [19]

    Despite achieving impressive performance, there remains the risk of neglecting small yet clinically significant features due to these reductions in dimensionality and complexity

    are prime examples, with the former utilizing a factorized attention mechanism and the latter projecting features into a lower-dimensional space to alleviate computational demands. Despite achieving impressive performance, there remains the risk of neglecting small yet clinica...

    2023

  13. [20]

    This approach integrated the outputs of multiple models to create a consensus segmentation

    for transfer learning. This approach integrated the outputs of multiple models to create a consensus segmentation. By leveraging the strengths of various models, the STAPLE technique helped mitigate individual model biases and improved overall segmentation robustness. This ens...

    2023

  14. [21]

    Nat Methods 18 (2):203-211

    Isensee F, Jaeger PF, Kohl SAA, Petersen J, Maier-Hein KH (2021) nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nat Methods 18 (2):203-211. doi:10.1038/s41592-020-01008-z

    2021 doi

  15. [22]

    Dosovitskiy A, Beyer L, Kolesnikov A, Weissenborn D, Zhai X, Unterthiner T, Dehghani M, Minderer M, Heigold G, Gelly S, Uszkoreit J, Houlsby N (2021) An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    2021

  16. [23]

    In: Medical Image Computing and Computer-Assisted Intervention – MICCAI

    Çiçek Ö, Abdulkadir A, Lienkamp SS, Brox T, Ronneberger O (2016) 3D U-Net: Learning Dense Volumetric International Journal of Computer Assisted Radiology and Surgery 13 Segmentation from Sparse Annotation. In: Medical Image Computing and Computer-Assisted Intervention – MICCAI

    2016

  17. [24]

    In: Computer Vision – ECCV

    Wang H, Zhu Y, Green B, Adam H, Yuille A, Chen L-C (2020) Axial-DeepLab: Stand-Alone Axial-Attention for Panoptic Segmentation. In: Computer Vision – ECCV

    2020

  18. [25]

    0.8451 0.8781 0.9275 0.8836 20.73 7.623 3.47 10.608 Winner 2022

    2022

  19. [26]

    Comput Biol Med 156:106718

    Huang Q, Zhao L, Ren G, Wang X, Liu C, Wang W (2023) NAG-Net: Nested attention-guided learning for segmentation of carotid lumen-intima interface and media-adventitia interface. Comput Biol Med 156:106718. doi:10.1016/j.compbiomed.2023.106718

    2023

  20. [27]

    Pattern Recognition

    Luo Y, Huang Q, Li X (2022) Segmentation information with attention integration for classification of breast tumor in ultrasound image. Pattern Recognition

    2022

  21. [29]

    These studies compared the performance of models before and after applying transfer learning and ensemble techniques

    0.8464 0.8769 0.9294 0.8842 17.81 11.12 4.26 11.06 HT-CNNs 0.8449 0.8786 0.9291 0.8842 16.189 7.88 4.21 10.199 • Bold and underlined values correspond to best and second-best scores Fine-tuning Results Studies To validate the effectiveness of our transfer learning techniques, ...

  22. [30]

    Comput Methods Programs Biomed 233:107493

    Liu Y, Zhu Y, Xin Y, Zhang Y, Yang D, Xu T (2023) MESTrans: Multi-scale embedding spatial transformer for medical image segmentation. Comput Methods Programs Biomed 233:107493. doi:10.1016/j.cmpb.2023.107493

    2023

  23. [32]

    In: Medical Image Computing and Computer Assisted Intervention – MICCAI

    Gao Y, Zhou M, Metaxas DN (2021) UTNet: A Hybrid Transformer Architecture for Medical Image Segmentation. In: Medical Image Computing and Computer Assisted Intervention – MICCAI

    2021

  24. [34]

    IEEE Transactions on Medical Imaging 23 (7):903-921

    Warfield SK, Zou KH, Wells WM (2004) Simultaneous Truth and Performance Level Estimation (STAPLE): An Algorithm for the Validation of Image Segmentation. IEEE Transactions on Medical Imaging 23 (7):903-921. doi:10.1109/tmi.2004.828354

    2004

  25. [35]

    Int J Comput Assist Radiol Surg 15 (6):909-920

    Zeineldin RA, Karar ME, Coburger J, Wirtz CR, Burgert O (2020) DeepSeg: deep neural network framework for automatic brain tumor segmentation using magnetic resonance FLAIR images. Int J Comput Assist Radiol Surg 15 (6):909-920. doi:10.1007/s11548-020-02186-z

    2020 doi

  26. [37]

    IEEE Transactions on Medical Imaging 42 (8):2451-2461

    Lin J, Lin J, Lu C, Chen H, Lin H, Zhao B, Shi Z, Qiu B, Pan X, Xu Z, Huang B, Liang C, Han G, Liu Z, Han C (2023) CKD-TransBTS: Clinical Knowledge-Driven Hybrid Transformer With Modality-Correlated Cross-Attention for Brain Tumor Segmentation. IEEE Transactions on Medical Ima...

    2023

  27. [40]

    In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries

    Luu HM, Park S-H (2022) Extending nn-UNet for Brain Tumor Segmentation. In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Lecture Notes in Computer Science. pp 173-186. doi:10.1007/978-3-031-09002-8_16

    2022 doi

  28. [41]

    Med Image Anal 54:280-296

    Cheplygina V, de Bruijne M, Pluim JPW (2019) Not-so-supervised: A survey of semi-supervised, multi-instance, and transfer learning in medical image analysis. Med Image Anal 54:280-296. doi:10.1016/j.media.2019.03.009

    2019 doi

  29. [42]

    Comput Biol Med 154:106555

    Zhang S, Miao Y, Chen J, Zhang X, Han L, Ran D, Huang Z, Pei N, Liu H, An C (2023) Twist-Net: A multi-modality transfer learning network with the hybrid bilateral encoder for hypopharyngeal cancer segmentation. Comput Biol Med 154:106555. doi:10.1016/j.compbiomed.2023.106555

    2023

  30. [43]

    In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries

    Zeineldin RA, Karar ME, Burgert O, Mathis-Ullrich F (2023) Multimodal CNN Networks for Brain Tumor Segmentation in MRI: A BraTS 2022 Challenge Solution. In: Brainlesion: Glioma, Multiple Sclerosis, Stroke and Traumatic Brain Injuries. Lecture Notes in Computer Science. pp 127-...

    2023 doi

  31. [44]

    doi:10.48550/arXiv.2402.17317

    Ferreira A, Solak N, Li J, Dammann P, Kleesiek J, Alves V, Egger J (2024) How we won BraTS 2023 Adult Glioma challenge? Just faking it! Enhanced Synthetic Data Augmentation and Model Ensemble for brain tumour segmentation.arXiv:2402.17317. doi:10.48550/arXiv.2402.17317

    2024 doi arXiv

  32. [45]

    doi:10.48550/arXiv.2407.08855

    Fathi Kazerooni A, Khalili N, Liu X, Haldar D, Jiang Z, Zapaishchykova A, Pavaine J, Shah LM, Jones BV, Sheth N, Prabhu SP, McAllister AS, Tu W, Nandolia KK, Rodriguez AF, Salman Shaikh I, Sanchez Montano M, Lai HA, Adewole M, Albrecht J, Anazodo U, Anderson H, Muhammed Anwar ...

    2024 doi arXiv

  33. [124]

    doi:10.1016/j.patcog.2021.108427

    2021

  34. [2016]

    pp 424-432

    Lecture Notes in Computer Science. pp 424-432. doi:10.1007/978-3-319-46723-8_49

    doi

  35. [2020]

    pp 108-126

    Lecture Notes in Computer Science. pp 108-126. doi:10.1007/978-3-030-58548-8_7

    doi

  36. [2021]

    pp 61-71

    Lecture Notes in Computer Science. pp 61-71. doi:10.1007/978-3-030-87199-4_6

    doi

  37. [2022]

    Table 1 Performance of five-fold cross-validation models on BraTS 2023 GLA Validation Cases

    Additionally, the HT-CNNs model demonstrates an improvement in the HD95 metric over the 2021, 2022, and 2023 winners, indicating its efficacy in precise tumor delineation. Table 1 Performance of five-fold cross-validation models on BraTS 2023 GLA Validation Cases. Model DSC HD...

    2021

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.