Pith. sign in

REVIEW 3 major objections 5 minor 20 references

Mamba Based Feature Extraction And Adaptive Multilevel Feature Fusion For 3D Tumor Segmentation From Multi-modal Medical Image

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Pith's one-line read The paper claims that feeding each 3D modality through its own Mamba encoder and fusing the results with two-level attention beats seven CNN, Transformer, and Mamba baselines on BraTS2023 and improves PET/CT fusion over PET-only on…

desk verdict A plausible Mamba-fusion architecture whose BraTS2023 SOTA claim is not established because the baseline numbers are borrowed and the Hausdorff metric may be inconsistent. read the letter →

arxiv 2504.21281 v1 pith:3HHKY5NZ submitted 2025-04-30 cs.CV

classification cs.CV
keywords 3DmedicalimagesegmentationtumorMambamulti-modalfeaturefusionbi-levelattentionBraTS2023PET/CTstatespacemodel
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

The paper argues that 3D multimodal tumor segmentation improves when each imaging modality is processed by its own Mamba encoder before fusion, rather than stacking modalities as input channels. It introduces a bi-level synergistic integration block that learns modality-level and channel-level attention weights to combine features adaptively. On BraTS2023 MRI, the method reports a mean Dice score of 92.15 and a mean Hausdorff distance of 2.62 mm, beating the seven CNN, Transformer, and Mamba baselines listed in the comparison. On Hecktor2022 PET/CT, fusing both modalities raises the Dice score from 33.07 with PET-only input to 37.12. If the comparisons hold, the practical lesson is that explicit per-modality encoding plus learned attention fusion is a better use of Mamba than a single shared encoder.

What carries the argument

The central object is the bi-level synergistic integration block, a fusion module that pools the concatenated modality features, computes softmax-normalized modality weights through two linear layers with ReLU, computes sigmoid-normalized channel weights in parallel, and then recalibrates each modality's feature map by multiplying the modality weight with the channel-weighted features. The other load-bearing components are the specific modality Mamba encoder, which runs each 3D modality through its own Mamba block, defined as a state-space sequence model with linear-time long-range modeling and forward/backward selective scans, followed by a residual block, and the decoder, which adds a Mamba block before upsampling and merges multi-scale encoder features through skip connections. The mechanism that carries the argument is the adaptive weighting itself: instead of fixed early fusion or concatenation, pooled descriptors drive dynamic emphasis of informative modalities and channels, which is what makes the fusion claim concrete.

What would settle it

Re-run the seven compared methods under the exact same BraTS2023 split and preprocessing used in this paper, then test the mean Dice difference between this method and SegMamba, 92.15 versus 91.32, with a paired test across test cases; if the difference is not significant or reverses, the central outperformance claim fails.

Watch

Extended reading notes

Core claim

The central claim is that the proposed network, built from modality-specific 3D Mamba encoders, a bi-level synergistic integration block, and a decoder that restores spatial detail through skip connections, achieves the strongest reported tumor segmentation among the compared methods. The authors state that their approach outperforms all segmentation approaches in the BraTS2023 comparison, and that PET/CT fusion outperforms PET-only input on Hecktor2022. The design rationale is that Mamba gives linear-complexity long-range modeling, each modality keeps its distinct appearance through a dedicated encoder, and the two-level attention weights let the fusion adapt to which modality and which channels matter at each feature level. The paper identifies the fusion mechanism as the key contributor: adding it to a Mamba encoder raises the mean Dice score from 90.13 to 92.15 and cuts the mean Hausdorff distance from 3.78 to 2.62 mm.

Load-bearing premise

The baseline results on BraTS2023 are taken from the SegMamba paper rather than being re-run under this paper's own 70/10/20 split and preprocessing, so the reported advantage rests on the assumption that those published numbers are directly comparable.

Editorial extensions

If this is right

  • If the claim holds, per-modality Mamba encoders plus two-level attention becomes a reusable template for other aligned multimodal segmentation tasks, such as multi-contrast MRI or PET/CT in other body regions.
  • The reported gain of 2.02 Dice points from adding bi-level integration to the Mamba encoder suggests that the adaptive fusion mechanism contributes more than the choice of encoder backbone.
  • The strongest reported improvement is on enhancing tumor (ET), with a Dice score of 87.92 and a Hausdorff distance of 3.30 mm, which indicates the method particularly helps small, heterogeneous, boundary-ambiguous regions.
  • The Hecktor2022 comparison, showing 37.12 Dice for PET/CT fusion versus 33.07 for PET-only, supports the claim that combining metabolic and anatomical information is genuinely useful, not just adding more input channels.

Reading between the lines

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

  • Inference: because the BraTS2023 baseline numbers are copied from the SegMamba paper rather than re-run under this paper's own split and preprocessing, the true ranking under identical conditions is untested; a fair re-run could either confirm or narrow the reported gap.
  • Inference: the bi-level attention module is not tied to the specific Mamba block used here, so it could likely be attached to other state-space or Transformer encoders and still provide adaptive multimodal fusion.
  • Inference: the authors' stated future goal of quantifying per-modality contributions could be approximated immediately by ablating individual modality streams and recording Dice changes, which would indicate how much each MRI sequence or PET/CT stream contributes to the final segmentation.
  • Inference: with fixed hyperparameters of SGD, learning rate 1e-3, and 1000 epochs, the reported gains may depend on optimization settings; testing other optimizers and losses would clarify whether the advantage comes from the architecture or from favorable training conditions.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper proposes a multi-modal 3D tumor segmentation architecture built on Mamba-based encoders. Each input modality is processed by a dedicated Mamba encoder to extract long-range features, and a 'bi-level synergistic integration block' fuses modalities using modality-level and channel-level attention. A decoder with skip connections produces the segmentation map. The method is evaluated on BraTS2023 (MRI, four modalities) and Hecktor2022 (PET/CT). The authors report ablation results showing incremental gains from each component, and a comparison table on BraTS2023 in which their method outperforms seven published CNN, Transformer, and Mamba baselines. On Hecktor2022, they compare their PET/CT fusion against a PET-only variant of the same network.

Significance. If the reported performance is reproducible, the architecture is a sensible and potentially useful contribution to multi-modal 3D segmentation: the modality-specific Mamba encoders and the adaptive fusion block are well motivated, and the internal ablation in Table 1 is self-consistent and shows a clear benefit from each added component. The paper also demonstrates the value of PET/CT fusion over PET-only input on Hecktor2022. However, the central claim of state-of-the-art performance on BraTS2023 is currently not supported with sufficient rigor, because the baseline numbers are imported from a single prior paper under a different experimental protocol, and the reported margins are small. The manuscript's significance therefore depends on whether the external comparison can be placed on a common footing.

major comments (3)
  1. [Section 3.3, Table 2, Section 3.4] The claim 'our approach outperforms all segmentation approaches' in Section 3.4 rests on Table 2, but six of the seven baseline rows are quoted from SegMamba [19] rather than re-run under the authors' own protocol. The authors state in Section 3.3: 'For the BraTS2023 dataset, the evaluation metrics for comparison methods are sourced from [19].' The authors train on their own random 70/10/20 split of BraTS2023, whereas the SegMamba numbers were obtained under a different split and preprocessing pipeline. The reported margins are small (0.83% mean Dice and 0.94 mm HD over SegMamba) and could plausibly be explained by split, preprocessing, or patch/crop differences. No error bars or significance tests are given. To support the state-of-the-art claim, the authors should re-run all baselines under the same split, preprocessing, and evaluation metric, or use the official BraTS2023 validation/test protocol, and report variance across multiple seeds.
  2. [Section 2 (Eq. 5) and Table 2] The Hausdorff distance in Eq. (5) is defined as the maximum Euclidean distance between boundary points, but most BraTS papers, including SegMamba [19], report the 95th percentile Hausdorff distance (HD95). If the baseline numbers in Table 2 are HD95 and the authors' numbers are maximum HD, then the HD column is comparing two different quantities, and the reported improvements are not meaningful. The authors must state explicitly which variant of HD is used for their own results and for each baseline, and ensure that all numbers in Table 2 are computed with the same metric (ideally by re-running the baselines).
  3. [Section 3.4, Table 3] The Hecktor2022 experiment in Table 3 compares only the authors' PET/CT model against a PET-only variant of the same network. This demonstrates that adding CT information improves segmentation within their architecture, but it does not support the abstract's claim of 'competitive performance compared to the state-of-the-art CNN, Transformer, and Mamba-based approaches' on this dataset. The authors should either add comparisons with published Hecktor2022 results under a common evaluation protocol, or explicitly restrict the claim on Hecktor2022 to demonstrating the benefit of multi-modal fusion rather than state-of-the-art performance.
minor comments (5)
  1. [Section 1, final paragraph] The sentence 'The structure of the paper is as follows: Section?? reviews the state-of-the-art segmentation methods' contains an unresolved placeholder 'Section??' and must be completed.
  2. [Throughout] There are grammatical and typographical errors, including 'an bi-level' in the abstract and Section 2.2, and 'quantifies the the maximum boundary deviation' in Section 3.2. These should be corrected.
  3. [Section 2.2] The description of the concatenation in the bi-level integration block is ambiguous: it is unclear whether the M modality feature maps are concatenated into a new modality dimension or along the channel dimension. The equations for the attention weights would benefit from explicit tensor shapes.
  4. [Section 3.3 and Figure 1] Implementation details are incomplete: the paper does not state the input patch size, cropping strategy, data augmentation, or the exact composition of the loss (cross-entropy only). Figure 1 is too small to read the individual block labels, which are crucial for understanding the architecture.
  5. [Table 2 caption] The caption of Table 2 should explicitly state that the baseline numbers are quoted from [19] and were not re-run by the authors under their own protocol, to avoid misleading readers.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the architecture is self-contained; the BraTS baseline sourcing from SegMamba is a benchmark comparability issue, not a circular derivation.

full rationale

The paper is an empirical architecture paper and contains no derivation chain that reduces to its own inputs. The Mamba equations (Eqs. 1-3) are standard state-space model background cited to prior external work, and the bi-level synergistic integration block is a concrete attention mechanism defined in Section 2.2 with explicit formulas. No parameter is fitted to a target metric and then renamed as a prediction; the ablation study (Table 1) is internal and self-consistent, and the reported gains are measured on the authors' own 70/10/20 split. The self-citations in the introduction ([10], [11]) concern MRI super-resolution and are not load-bearing for the segmentation claim. The benchmarking weakness flagged by the reader is that seven baseline rows in Table 2 are quoted from SegMamba [19], an external prior work, rather than re-run under the authors' split and preprocessing, and Eq. 5 defines Hausdorff distance as the symmetric maximum while many BraTS reports use the 95th percentile. That is a validity or comparability concern, not circularity: quoting an external baseline does not make the central claim equivalent to its inputs by construction, and [19] is not a self-citation. The paper also does not claim a first-principles derivation of its performance, so no self-definitional, fitted-input, uniqueness-imported, or ansatz-smuggling pattern is present. No circular step can be exhibited with a specific equation reduction, so the appropriate score is 0.

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

The central claim is an empirical architecture result. It relies on standard components from prior literature (Mamba, Res blocks, attention), on public datasets and metrics, and on hand-chosen hyperparameters. No new fundamental constants or independent measurements are supplied, so the contribution is an engineering combination rather than a parameter-free derivation.

free parameters (6)
  • learning rate = 1e-3
    Chosen by hand for SGD optimizer; affects optimization trajectory and final metrics.
  • weight decay = 1e-5
    Chosen by hand to regularize training.
  • number of epochs = 1000
    Chosen by hand; no early stopping or convergence check is reported.
  • batch size = 1
    Chosen by hand; affects gradient noise and memory usage.
  • BraTS2023 split ratios = 70/10/20
    Chosen by hand; may differ from the splits used to produce the baseline numbers in [19].
  • Hecktor2022 split ratios = 60/20/20
    Chosen by hand; no comparison to state-of-the-art is provided on this dataset.
assumptions (3)
  • domain assumption Mamba state space model equations and zero-order hold discretization (Eqs. 1-3) are valid for extracting long-range features from 3D medical images.
    The method builds entirely on the Mamba model from Gu & Dao (2023) and earlier linear state space layer work; the paper does not provide an independent validation or derivation of this modeling choice for medical volumes.
  • domain assumption The ground truth annotations in BraTS2023 and Hecktor2022 are accurate and the Dice and Hausdorff metrics are appropriate measures of segmentation quality.
    The paper accepts the public datasets and standard metrics as reliable without performing any independent review of the labels or metric biases.
  • domain assumption Baseline results for BraTS2023 reported in [19] (SegMamba) are comparable to the authors' own results despite potentially different train/validation/test splits.
    The paper states the comparison metrics are sourced from [19] without describing the exact splits and preprocessing used there, so comparability is assumed rather than verified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Mamba Based Feature Extraction And Adaptive Multilevel Feature Fusion For 3D Tumor Segmentation From Multi-modal Medical Image." pith.science (2026). https://pith.science/paper/3HHKY5NZ

@misc{pith2026250421281,
  author       = {Pith},
  title        = {Pith review of: Mamba Based Feature Extraction And Adaptive Multilevel Feature Fusion For 3D Tumor Segmentation From Multi-modal Medical Image},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3HHKY5NZ}},
  note         = {Machine review of arXiv:2504.21281}
}
read the original abstract

Multi-modal 3D medical image segmentation aims to accurately identify tumor regions across different modalities, facing challenges from variations in image intensity and tumor morphology. Traditional convolutional neural network (CNN)-based methods struggle with capturing global features, while Transformers-based methods, despite effectively capturing global context, encounter high computational costs in 3D medical image segmentation. The Mamba model combines linear scalability with long-distance modeling, making it a promising approach for visual representation learning. However, Mamba-based 3D multi-modal segmentation still struggles to leverage modality-specific features and fuse complementary information effectively. In this paper, we propose a Mamba based feature extraction and adaptive multilevel feature fusion for 3D tumor segmentation using multi-modal medical image. We first develop the specific modality Mamba encoder to efficiently extract long-range relevant features that represent anatomical and pathological structures present in each modality. Moreover, we design an bi-level synergistic integration block that dynamically merges multi-modal and multi-level complementary features by the modality attention and channel attention learning. Lastly, the decoder combines deep semantic information with fine-grained details to generate the tumor segmentation map. Experimental results on medical image datasets (PET/CT and MRI multi-sequence) show that our approach achieve competitive performance compared to the state-of-the-art CNN, Transformer, and Mamba-based approaches.

Figures

Figures reproduced from arXiv: 2504.21281 by the authors.

Figure 1
Figure 1. Our architecture mainly includes Mamba block, bi-level synergistic integration block, and Res block, shown through a brain tumor segmentation with 4-sequence MRI. characteristics of each modality. Therefore, we use specific modality Mamba en￾coder to realize specialized feature extraction for each modality, capturing the most relevant tumor-related information and reducing the impact of modality￾specific artifacts o… view at source ↗
Figure 2
Figure 2. Qualitative results on BraTS2023 dataset. Quantitative results Our approach is also compared with the state-of-the-art CNN-based (SegresNet [14], UX-Net [12], MedNeXt [16]), Transformer-based (UNETR [6], SwinUNETR [5], SwinUNETRV2 [7]), and Mamba-based (Seg￾Mamba [19]) approaches validated on BraTS2023 dataset. The quantitative ex￾perimental results on BraTS2023 dataset are summarized in Tab. 2. It can be seen that … view at source ↗
Figure 3
Figure 3. Qualitative results on Hecktor2022 dataset. were compared: network using PET-only input and with multi-modality (Ours). Tab. 3 indicates that the fusion of PET and CT outperforms the PET-only seg￾mentation. PET imaging excels at capturing metabolic activity, which is crucial for identifying regions with abnormal cellular function, such as tumors. However, PET images often suffer from low spatial resolution and lack … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 18 canonical work pages

  1. [19]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI

    Xing, Z., Ye, T., Yang, Y., Liu, G., Zhu, L.: Segmamba: Long-range sequential modeling mamba for 3d medical image segmentation. In: International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI. pp. 578–588 (2024)

  2. [1]

    In: International Conference on Learning Representations, ICLR (2021)

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., Gelly, S., Uszkoreit, J., Houlsby, N.: An image is worth 16x16 words: Transformers for image recognition at scale. In: International Conference on Learning Representations, ICLR (2021)

  3. [2]

    IEEE Transactions on Circuits and Systems II: Ex- press Briefs 55(12), 1269–1273 (2008)

    Galias, Z., Yu, X.: Analysis of zero-order holder discretization of two-dimensional sliding-mode control systems. IEEE Transactions on Circuits and Systems II: Ex- press Briefs 55(12), 1269–1273 (2008)

  4. [3]

    arXiv preprint arXiv:2312.00752 (2023)

    Gu, A., Dao, T.: Mamba: Linear-time sequence modeling with selective state spaces. arXiv preprint arXiv:2312.00752 (2023)

  5. [4]

    Advances in neural information processing systems34, 572–585 (2021)

    Gu, A., Johnson, I., Goel, K., Saab, K., Dao, T., Rudra, A., Ré, C.: Combining recurrent, convolutional, and continuous-time models with linear state space layers. Advances in neural information processing systems34, 572–585 (2021)

  6. [5]

    In: International MICCAI brainlesion workshop

    Hatamizadeh, A., Nath, V., Tang, Y., Yang, D., Roth, H.R., Xu, D.: Swin unetr: Swin transformers for semantic segmentation of brain tumors in mri images. In: International MICCAI brainlesion workshop. pp. 272–284 (2021)

  7. [6]

    In: Proceedings of the IEEE/CVF winter conference on applications of computer vi- sion, WACV

    Hatamizadeh, A., Tang, Y., Nath, V., Yang, D., Myronenko, A., Landman, B., Roth, H.R., Xu, D.: Unetr: Transformers for 3d medical image segmentation. In: Proceedings of the IEEE/CVF winter conference on applications of computer vi- sion, WACV. pp. 574–584 (2022)

  8. [7]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI

    He, Y., Nath, V., Yang, D., Tang, Y., Myronenko, A., Xu, D.: Swinunetr-v2: Stronger swin transformers with stagewise convolutions for 3d medical image segmentation. In: International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI. pp. 416–426 (2023)

Show all 20 references
  1. [8]

    Medical Image Analysis99, 103334 (2025) 12 F

    Huang, J., Yang, L., Wang, F., Wu, Y., Nan, Y., Wu, W., Wang, C., Shi, K., Aviles- Rivero, A.I., Schönlieb, C.B., et al.: Enhancing global sensitivity and uncertainty quantification in medical image reconstruction with monte carlo arbitrary-masked mamba. Medical Image Analysis...

  2. [9]

    Information fusion91, 737–756 (2023)

    Huang, L., Ruan, S., Denœux, T.: Application of belief functions to medical image segmentation: A review. Information fusion91, 737–756 (2023)

  3. [10]

    Pattern Recognition Letters187, 93–99 (2025)

    Ji, Z., Zou, B., Kui, X., Li, H., Vera, P., Ruan, S.: Generation of super-resolution for medical image via a self-prior guided mamba network with edge-aware constraint. Pattern Recognition Letters187, 93–99 (2025)

  4. [11]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI

    Ji, Z., Zou, B., Kui, X., Vera, P., Ruan, S.: Deform-mamba network for mri super-resolution. In: International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI. pp. 242–252 (2024)

  5. [12]

    In: The Eleventh International Conference on Learning Representations, ICLR (2023)

    Lee, H.H., Bao, S., Huo, Y., Landman, B.A.: 3d ux-net: A large kernel volumetric convnet modernizing hierarchical transformer for medical image segmentation. In: The Eleventh International Conference on Learning Representations, ICLR (2023)

  6. [13]

    IEEE Transactions on Geoscience and Remote Sensing (2024)

    Li,Y.,Luo,Y.,Zhang,L.,Wang,Z.,Du,B.:Mambahsi:Spatial-spectralmambafor hyperspectral image classification. IEEE Transactions on Geoscience and Remote Sensing (2024)

  7. [14]

    In: International MICCAI brainlesion workshop

    Myronenko,A.:3dmribraintumorsegmentationusingautoencoderregularization. In: International MICCAI brainlesion workshop. pp. 311–320 (2019)

  8. [15]

    The Computer Journal10(4), 383–384 (1968)

    Osborne, M.R., Watson, G.A.: Note on two methods of solving ordinary linear differential equations. The Computer Journal10(4), 383–384 (1968)

  9. [16]

    In: International Conference on Medical Image Computing and Computer-Assisted Intervention, MICCAI

    Roy, S., Koehler, G., Ulrich, C., Baumgartner, M., Petersen, J., Isensee, F., Jaeger, P.F., Maier-Hein, K.H.: Mednext: transformer-driven scaling of convnets for medi- cal image segmentation. In: International Conference on Medical Image Computing and Computer-Assisted Interve...

  10. [17]

    Biomedical Signal Processing and Control103, 107376 (2025)

    Wang, W., Jin, Z., Chen, X.: Cdrmamba: A framework for automated craniomax- illofacial defect reconstruction using mamba-based modeling. Biomedical Signal Processing and Control103, 107376 (2025)

  11. [18]

    Biomedical Signal Processing and Control84, 104791 (2023)

    Xiao, H., Li, L., Liu, Q., Zhu, X., Zhang, Q.: Transformers in medical image seg- mentation: A review. Biomedical Signal Processing and Control84, 104791 (2023)

  12. [20]

    In: IEEE 17th international symposium on biomedical imaging, (ISBI)

    Zhou, T., Ruan, S., Guo, Y., Canu, S.: A multi-modality fusion network based on attention mechanism for brain tumor segmentation. In: IEEE 17th international symposium on biomedical imaging, (ISBI). pp. 377–380 (2020)

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.