Pith. sign in

REVIEW 5 major objections 6 minor 50 references

Beyond Labels: Zero-Shot Diabetic Foot Ulcer Wound Segmentation with Self-attention Diffusion Models and the Potential for Text-Guided Customization

T0 review · 5 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read The paper claims that zero-shot, label-free segmentation of diabetic foot ulcers can beat supervised state-of-the-art models by reading self-attention maps from a pretrained diffusion model.

desk verdict A clinically motivated zero-shot segmentation pipeline built on existing diffusion-attention methods, but internally inconsistent metrics and an unspecified mask-selection step make the headline claim unverifiable. read the letter →

arxiv 2504.17628 v1 pith:I3XL2FOV submitted 2025-04-24 eess.IV cs.CV

Abderrachid Hamrani , Daniela Leizaola , Renato Sousa , Jose P. Ponce , Stanley Mathis , David G. Armstrong , Anuradha Godavarty This is my paper · ORCID
classification eess.IVcs.CV
keywords diabeticfootulcersegmentationzero-shotlearningself-attentiondiffusionmodelstext-guidedmedicalimageunsupervisedwoundassessment
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

This paper proposes that the internal self-attention maps of a pretrained image-generation model, Stable Diffusion, already contain enough object-grouping information to outline diabetic foot ulcers in real photographs without any labeled training data. The authors build ADZUS, a pipeline that extracts those maps, combines them across resolutions, merges similar regions, and turns the result into a segmentation mask guided by a text prompt. On the public chronic-wound test set, ADZUS reports the highest intersection-over-union (86.68%) and precision (94.69%) among compared models, narrowly ahead of the supervised FUSegNet; on a small clinician-curated white-light DFU set, it reports a median dice score of 75% versus about 45% for FUSegNet. If these numbers hold, zero-shot segmentation could remove the annotation bottleneck for wound assessment and let clinicians steer segmentation by describing what they are looking for.

What carries the argument

The load-bearing mechanism is the self-attention tensor of the diffusion U-Net: for each spatial location, a 2D attention map records how strongly every other location responds to it, and the paper assumes locations belonging to the same object have similar maps. ADZUS aggregates the 16 tensors from four resolutions ($8\times8$, $16\times16$, $32\times32$, $64\times64$) by bilinear upsampling and weighted summation, then merges nearby maps whose KL divergence is below a threshold, and applies non-maximum suppression to produce a binary mask. The aggregation weights proportional to resolution, the chosen threshold, and the time-step $t=300$ are hand-set hyperparameters; text guidance enters through cross-attention to condition which groups are emphasized.

What would settle it

Sweep the diffusion time-step from $t=100$ to $t=500$ and vary the resolution weights and merge threshold on the same 200-image chronic-wound set, then compare each resulting mask to the ground-truth tracings. If IoU and precision drop steeply or the masks shift off the wound under plausible settings, the zero-shot attention maps are not robustly encoding wound anatomy and the reported leaderboard numbers would not generalize.

Watch

Extended reading notes

Core claim

The paper's central claim is that the self-attention layers inside a pretrained Stable Diffusion U-Net encode spatial object groupings that are good enough to segment medical wounds without any training. ADZUS treats a real wound photograph as if it were a near-final denoised generated image, runs one unconditioned diffusion pass at a large time-step ($t=300$), and collects the 16 self-attention tensors produced at resolutions of 8, 16, 32, and 64. These tensors are upsampled, aggregated with resolution-proportional weights, merged iteratively using KL divergence between attention maps, and collapsed by non-maximum suppression into a final mask. A text prompt supplied through the model's cross-attention path shapes which regions the merged attention emphasizes. The paper reports that this label-free procedure reaches an IoU of 86.68% and precision of 94.69% on the chronic-wound benchmark, surpassing FUSegNet, and a median DSC of about 75% on its own 40-image DFU dataset, far above FUSegNet's 45%.

Load-bearing premise

The argument stands on the premise that the hand-picked diffusion settings, especially the time-step $t=300$ and the resolution weights, place real wound photographs in the attention regime where the self-attention maps mark actual wound boundaries, and the paper does not report sensitivity to these settings.

Editorial extensions

If this is right

  • If the reported results are correct, automated wound segmentation no longer needs a task-specific annotated training set; a general-purpose diffusion model can be pointed at a wound photograph and a descriptive prompt.
  • Text-guided customization becomes practical: changing the prompt to emphasize infection, inflammation, or granulation changes the output mask, so a clinician could request the region of interest without retraining.
  • The same zero-shot attention pipeline should transfer to other anatomical targets such as retinal vessels or tumors with prompt engineering alone, a claim the paper explicitly lists as future work.
  • On the benchmark, a label-free method matching a supervised specialist model implies the main cost of applying AI to new wound-imaging setups shifts from annotation to choosing prompts and hyperparameters.

Reading between the lines

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

  • A sensible stress test would sweep $t$ and the merge threshold; if performance is stable across a wide range, the attention maps are genuinely carrying wound structure, and if it is not, the reported numbers may be tuned to the two datasets.
  • The text-guided examples show different prompts producing different masks, which suggests an untapped evaluation axis: measuring whether prompt semantics align with the selected tissue zones, rather than only measuring boundary overlap with a single ground-truth tracing.
  • Because ADZUS uses a frozen pretrained model, it could be combined with active learning: a handful of labeled wound images could select the best prompt and hyperparameters per clinic, keeping the zero-shot core while adapting to local imaging 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

5 major / 6 minor

Summary. The paper proposes ADZUS, a zero-shot unsupervised segmentation method for diabetic foot ulcers built on self-attention maps from a pretrained Stable Diffusion model. The method aggregates attention tensors from multiple resolutions, iteratively merges attention maps using KL divergence, applies non-maximum suppression to obtain object proposals, and optionally incorporates text prompts to guide the segmentation. On the chronic wound dataset (200 test images) the authors report IoU 86.68%, precision 94.69%, recall 92.46%, and DSC 91.98%, claiming to outperform supervised models including FUSegNet. On a custom-curated dataset of 40 white-light images, they report a median DSC of about 75% and IoU of about 68% for ADZUS, versus about 45% and 50% for FUSegNet. The paper also demonstrates text-guided segmentation qualitatively.

Significance. The underlying idea of exploiting pretrained diffusion self-attention for label-free medical image segmentation is timely and potentially valuable, and the authors include a data-contamination check using a public search interface, which is a useful precaution. If the quantitative claims could be verified, the result that a zero-shot, label-free method is competitive with supervised state-of-the-art on diabetic foot ulcer segmentation would be of clear interest to the biomedical imaging community. However, the central quantitative claims are not currently verifiable: the binary mask extraction protocol is unspecified, the reported metrics are internally inconsistent, and the key hyperparameters are not documented. The paper also provides no code, no sensitivity analysis, and no uncertainty quantification, so the headline claim of outperforming supervised models is not supported at the current level of reporting.

major comments (5)
  1. [Section II.B.3, Section III.A, Table II] The conversion from the multi-proposal ADZUS output to the single binary wound mask scored in Tables II and Figure 6 is never specified. Section II.B.3 states that ADZUS 'generates a comprehensive segmentation mask that delineates multiple regions' and that 'clinicians or users ... interactively select the relevant region of interest.' If the proposal used for evaluation was selected with knowledge of the ground truth, or by an unspecified oracle, then the reported IoU/DSC values are not zero-shot results. The authors must provide the exact automatic selection rule, or a fully specified human protocol that is blind to the ground truth, and ideally release code; without this, the central superiority claim cannot be verified from the paper alone.
  2. [Table II; Section III.A] The metrics reported for ADZUS are internally inconsistent. Given precision 94.69% and recall 92.46%, the pooled IoU must be 87.90% and DSC 93.56%, yet Table II lists IoU 86.68% and DSC 91.98%, while the text in Section III.A cites a DSC of 93.56. Every other row in Table II satisfies the identities IoU = PR/(P+R-PR) and DSC = 2IoU/(1+IoU) to rounding. If the metrics are per-image averages rather than pooled values, the aggregation protocol must be stated explicitly and applied uniformly; as written, the table is internally contradictory and the comparison to the baselines cannot be interpreted.
  3. [Section III.B, Figure 6] The custom-curated dataset results are also numerically impossible as stated. Because DSC = 2IoU/(1+IoU) is monotone, the median DSC must equal 2 times the median IoU divided by (1 plus the median IoU); a median IoU of about 68% would imply a median DSC of about 81%, not 75%, and FUSegNet's median IoU of about 50% would imply a median DSC of about 67%, not 45%. The authors need to report the paired per-image statistics or clarify which quantity is the median; the current numbers cannot both be medians of the same segmentations.
  4. [Section II.A, Section II.B] The method relies on several hand-set hyperparameters, including the diffusion timestep t about 300, the resolution weights R_k, the KL threshold tau, the anchor grid size M, and the iteration count N, but the paper does not report their values or provide a sensitivity analysis. The assertion that t=300 places clean images in the reverse-diffusion regime where attention maps reflect semantic grouping is central to the method and is stated without evidence. Without this information, the results are not reproducible and there is no check that the authors did not tune these parameters on the test set.
  5. [Section III.A, Table II] The claimed superiority over FUSegNet rests on a 0.28 percentage-point IoU difference (86.68% versus 86.40%) with no error bars, confidence intervals, or significance test. Given the metric inconsistencies described above, the statement that ADZUS 'surpasses' supervised state-of-the-art models is not supported at the current level of reporting; the authors should provide uncertainty quantification or qualify the claim as competitive rather than superior.
minor comments (6)
  1. [Appendix A, Eqs. (A1)-(A4)] The definitions should state whether TP/FP/FN are pooled over the dataset or computed per image, and how per-image metrics are aggregated.
  2. [Section II.B.1, Eq. (2)] The rendered equation is incomplete; the aggregation weights R_k and the upsampling notation should be defined explicitly so that the method is reproducible.
  3. [Section III.B, Figure 6] The figure should report the underlying distribution, for example as box plots with n=40, and the text should state whether the reported IoU and DSC values are medians of the same per-image scores.
  4. [Section I.B, Reference [37]] Reference [37] is described as the 'LAION-5B search tool,' but 'Have I Been Trained?' is a retrieval interface; the authors should describe its search modality and the limitations of the contamination check.
  5. [Abstract] The abstract contains the typo 'text-gui ded' and should read 'text-guided.'
  6. [Section II.B.4] The text-guided workflow in Section II.B.4 repeats the three-step description already given in Sections II.B.1-II.B.3; consider consolidating the two descriptions to avoid redundancy.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ADZUS's masks are obtained by direct post-processing of pretrained diffusion attention maps, and the reported comparisons are against external benchmarks.

full rationale

The derivation chain is self-contained and non-circular. ADZUS takes a pretrained Stable Diffusion model and, without any labeled training, extracts self-attention tensors (Section II.A), aggregates them across resolutions with weights proportional to resolution (Eq. 2), iteratively merges attention maps using KL divergence (Eq. 4), and applies non-maximum suppression to obtain masks (Section II.B.3). None of these steps fits parameters to test labels or imports a conclusion from the authors' own prior work: the self-citations ([4], [8], [47], [48]) are contextual (review, skin-tone background, and device descriptions) and not load-bearing. The evaluation is benchmarked against external supervised models and the public chronic wound dataset [26], and the paper includes a LAION-5B contamination check to support the zero-shot claim. Two caveats affect verifiability rather than circularity: Section II.B.3 says users may 'interactively select the relevant region of interest,' leaving the exact binary mask scored in Table II unspecified, and Section III.C states that the accuracy of text-guided segmentation is 'part of our ongoing efforts.' These are reproducibility/correctness concerns, not definitional reductions, so the circularity score remains 0.

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

The method relies on several hand-chosen hyperparameters and the implicit validity of the attention-as-segmentation prior. No external benchmark confirms that the attention maps correspond to clinical ground truth beyond the reported metrics, and the search-tool verification is only as strong as the tool itself.

free parameters (6)
  • Diffusion timestep t = 300
    Chosen to treat real images as primarily denoised generated images; no sensitivity analysis is provided.
  • Attention aggregation weights R = proportional to resolution (exact values not reported)
    Weights for fusing attention tensors at different resolutions; hand-selected design choice.
  • KL merge threshold tau = not reported
    Threshold for merging attention maps; value likely tuned but not stated.
  • Anchor grid size M = not reported
    Number of anchor points for iterative merging; value not given.
  • Iteration count N = not reported
    Number of merging iterations; value not given.
  • Text prompt for segmentation = a detailed medical photograph of a diabetic foot ulcer with necrotic tissue, slough formation, granulation areas, and…
    Hand-crafted prompt used for all quantitative evaluations; the method's output depends on this input.
assumptions (3)
  • domain assumption Self-attention maps in Stable Diffusion encode object-level grouping that can be converted into segmentation masks.
    Inherited from DiffuMask and DAAS; invoked in Section II.B to justify the entire pipeline.
  • ad hoc to paper Setting t=300 places clean images in the reverse-diffusion regime where attention maps reflect semantic grouping rather than noise.
    Stated in Section II.A; a modeling assumption specific to this pipeline without a sensitivity study.
  • domain assumption The haveibeentrained.com LAION-5B search adequately confirms the test images are absent from Stable Diffusion's training data.
    Used in the Motivation section to rule out data leakage; the tool's completeness and coverage are not independently verified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Beyond Labels: Zero-Shot Diabetic Foot Ulcer Wound Segmentation with Self-attention Diffusion Models and the Potential for Text-Guided Customization." pith.science (2026). https://pith.science/paper/I3XL2FOV

@misc{pith2026250417628,
  author       = {Pith},
  title        = {Pith review of: Beyond Labels: Zero-Shot Diabetic Foot Ulcer Wound Segmentation with Self-attention Diffusion Models and the Potential for Text-Guided Customization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I3XL2FOV}},
  note         = {Machine review of arXiv:2504.17628}
}
read the original abstract

Diabetic foot ulcers (DFUs) pose a significant challenge in healthcare, requiring precise and efficient wound assessment to enhance patient outcomes. This study introduces the Attention Diffusion Zero-shot Unsupervised System (ADZUS), a novel text-guided diffusion model that performs wound segmentation without relying on labeled training data. Unlike conventional deep learning models, which require extensive annotation, ADZUS leverages zero-shot learning to dynamically adapt segmentation based on descriptive prompts, offering enhanced flexibility and adaptability in clinical applications. Experimental evaluations demonstrate that ADZUS surpasses traditional and state-of-the-art segmentation models, achieving an IoU of 86.68\% and the highest precision of 94.69\% on the chronic wound dataset, outperforming supervised approaches such as FUSegNet. Further validation on a custom-curated DFU dataset reinforces its robustness, with ADZUS achieving a median DSC of 75\%, significantly surpassing FUSegNet's 45\%. The model's text-guided segmentation capability enables real-time customization of segmentation outputs, allowing targeted analysis of wound characteristics based on clinical descriptions. Despite its competitive performance, the computational cost of diffusion-based inference and the need for potential fine-tuning remain areas for future improvement. ADZUS represents a transformative step in wound segmentation, providing a scalable, efficient, and adaptable AI-driven solution for medical imaging.

Figures

Figures reproduced from arXiv: 2504.17628 by the authors.

Figure 1
Figure 1. Comparison of Stable Diffusion-generated (a, b, c) and real diabetic foot ulcer images (d, e, f). Given this capability, an important question arises: has Stable Diffusion been trained on medical images (e.g., wound-related data), in particular the datasets used in this study? To address this, we conducted a data similarity verification analysis using the publicly available LAION-5B search tool [37], which forms the… view at source ↗
Figure 3
Figure 3. Workflow of text-to-image diffusion integration in ADZUS for wound segmentation [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Sample qualitative segmentation on the chronic wound dataset [26]: original boundaries (red) and predicted boundaries (green) displayed on cropped images for enhanced visualization [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Sample comparison of segmentation results: ADZUS vs. benchmark models on chronic wound dataset [26], showing original boundaries (red) and predicted boundaries (green) on cropped images [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Comparison of ADZUS with FUSegNet model based on (a) IoU and (b) DSC scores [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Sample comparison of wound segmentation results using ADZUS and FUSegNet for one sample DFU case from our custom-curated dataset. Red outlines represent clinical ground truth tracings, while blue outlines denote segmentations produced by the respective methods [PITH_F…
Figure 8
Figure 8. Figure 8: Demonstration of text-guided segmentation in ADZUS: (a) original wound image, (b) segmentation result using Prompt 1, (c) confidence map corresponding to Prompt 1, (d) segmentation result using Prompt 2, and (e) confidence map corresponding to Prompt 2 [PITH_FULL_IMAG…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 32 canonical work pages

  1. [1]

    Diabetic Foot Ulcers: A Review,

    D. G. Armstrong, T. W. Tan, A. J. M. Boulton, and S. A. Bus, “Diabetic Foot Ulcers: A Review,” JAMA, vol. 330, no. 1, pp. 62–75, Jul. 2023, doi: 10.1001/JAMA.2023.10578

    arXiv 2023

  2. [2]

    Diabetic Foot Ulcers and Their Recurrence,

    D. G. Armstrong, A. J. M. Boulton, and S. A. Bus, “Diabetic Foot Ulcers and Their Recurrence,” N. Engl. J. Med., vol. 376, no. 24, pp. 2367–2375, Jun. 2017, doi: 10.1056/NEJMRA1615439/SUPPL_FILE/NEJMRA1 615439_DISCLOSURES.PDF

    work page doi:10.1056/nejmra1615439/suppl_file/nejmra1 2017

  3. [3]

    A. K. Madeshiya et al., Nutrition and diabetic wound healing. Elsevier, 2020. doi: 10.1016/B978-0-12- 816413-6.00020-4

    doi:10.1016/b978-0-12- 2020

  4. [4]

    Diabetic Foot Ulcer Imaging: An Overview and Future Directions,

    A. Godavarty, K. Leiva, N. Amadi, D. C. Klonoff, and D. G. Armstrong, “Diabetic Foot Ulcer Imaging: An Overview and Future Directions,” J. Diabetes Sci. Technol., vol. 17, no. 6, pp. 1662–1675, Nov. 2023, doi: 10.1177/19322968231187660/ASSET/IMAGES/LAR GE/10.1177_19322968231187660-FIG1.JPEG

    work page doi:10.1177/19322968231187660/asset/images/lar 2023

  5. [5]

    Recognition of ischaemia and infection in diabetic foot ulcers: Dataset and techniques,

    M. Goyal, N. D. Reeves, S. Rajbhandari, N. Ahmad, C. Wang, and M. H. Yap, “Recognition of ischaemia and infection in diabetic foot ulcers: Dataset and techniques,” Comput. Biol. Med., vol. 117, p. 103616, Feb. 2020, doi: 10.1016/J.COMPBIOMED.2020.103616

    arXiv 2020

  6. [6]

    Is my wound infected? A study on the use of hyperspectral imaging to assess wound infection,

    J. L. Ramirez-GarciaLuna et al., “Is my wound infected? A study on the use of hyperspectral imaging to assess wound infection,” Front. Med., vol. 10, p. 1165281, Aug. 2023, doi: 10.3389/FMED.2023.1165281/BIBTEX

    arXiv 2023

  7. [7]

    Prevention and treatment of diabetic foot ulcers,

    J. Z. M. Lim, N. S. L. Ng, and C. Thomas, “Prevention and treatment of diabetic foot ulcers,” https://doi.org/10.1177/0141076816688346, vol. 110, no. 3, pp. 104–109, Jan. 2017, doi: 10.1177/0141076816688346

    work page doi:10.1177/0141076816688346 2017

  8. [8]

    AI Dermatochroma Analytica (AIDA): Smart Technology for Robust Skin Color Classification and Segmentation,

    A. Hamrani et al., “AI Dermatochroma Analytica (AIDA): Smart Technology for Robust Skin Color Classification and Segmentation,” Cosmet. 2024, Vol. 11, Page 218, vol. 11, no. 6, p. 218, Dec. 2024, doi: 10.3390/COSMETICS11060218

    work page doi:10.3390/cosmetics11060218 2024

Show all 50 references
  1. [9]

    Diabetic foot disease: From the evaluation of the ‘foot at risk’ to the novel diabetic ulcer treatment modalities,

    N. Amin and J. Doupis, “Diabetic foot disease: From the evaluation of the ‘foot at risk’ to the novel diabetic ulcer treatment modalities,” World J. Diabetes, vol. 7, no. 7, p. 153, 2016, doi: 10.4239/WJD.V7.I7.153

    2016 doi

  2. [10]

    Current Challenges and Opportunities in the Prevention and Management of Diabetic Foot Ulcers,

    W. J. Jeffcoate, L. Vileikyte, E. J. Boyko, D. G. Armstrong, and A. J. M. Boulton, “Current Challenges and Opportunities in the Prevention and Management of Diabetic Foot Ulcers,” Diabetes Care, vol. 41, no. 4, pp. 645–652, Apr. 2018, doi: 10.2337/DC17-1836

    2018 doi

  3. [11]

    WHO IS AT RISK FOR DIABETIC FOOT ULCERATION?,

    D. G. Armstrong, L. A. Lavery, and L. B. Harkless, “WHO IS AT RISK FOR DIABETIC FOOT ULCERATION?,” Clin. Podiatr. Med. Surg., vol. 15, 2 1002 TPDSC TP FP FN   , (A1) 100TPIoU TP FP FN   , (A2) 100TPprecision TP FP   , (A3) 100TPrecall TP FN  ...

    1998 doi

  4. [12]

    Quantifying dermal microcirculatory changes of neuropathic and neuroischemic diabetic foot ulcers using spatial frequency domain imaging: a shade of things to come?,

    G. A. Murphy, R. P. Singh-Moon, A. Mazhar, D. J. Cuccia, V. L. Rowe, and D. G. Armstrong, “Quantifying dermal microcirculatory changes of neuropathic and neuroischemic diabetic foot ulcers using spatial frequency domain imaging: a shade of things to come?,” BMJ Open Diabetes R...

    2020 doi

  5. [13]

    Deep learning in diabetic foot ulcers detection: A comprehensive evaluation,

    M. H. Yap et al., “Deep learning in diabetic foot ulcers detection: A comprehensive evaluation,” Comput. Biol. Med., vol. 135, p. 104596, Aug. 2021, doi: 10.1016/J.COMPBIOMED.2021.104596

    2021

  6. [14]

    The impact of machine learning on the prediction of diabetic foot ulcers – A systematic review,

    T. Weatherall et al., “The impact of machine learning on the prediction of diabetic foot ulcers – A systematic review,” J. Tissue Viability, vol. 33, no. 4, pp. 853– 863, Nov. 2024, doi: 10.1016/J.JTV.2024.07.004

    2024 doi

  7. [15]

    Automated Detection of Infection in Diabetic Foot Ulcer Using Pre-trained Fast Convolutional Neural Network with U++net,

    S. V. N. Murthy, K. N. Bhargavi, S. Isaac, and E. N.Ganesh, “Automated Detection of Infection in Diabetic Foot Ulcer Using Pre-trained Fast Convolutional Neural Network with U++net,” SN Comput. Sci., vol. 5, no. 6, pp. 1–15, Aug. 2024, doi: 10.1007/S42979-024-02981-4/FIGURES/12

    2024 doi

  8. [16]

    Can deep learning wound segmentation algorithms developed for a dataset be effective for another dataset? A specific focus on diabetic foot ulcers,

    S. Lucho, R. Naemi, B. Castaneda, and S. Treuillet, “Can deep learning wound segmentation algorithms developed for a dataset be effective for another dataset? A specific focus on diabetic foot ulcers,” IEEE Access, 2024, doi: 10.1109/ACCESS.2024.3502467

    2024

  9. [17]

    State of the Art on Diffusion Models for Visual Computing,

    R. Po et al., “State of the Art on Diffusion Models for Visual Computing,” Comput. Graph. Forum, vol. 43, no. 2, p. e15063, May 2024, doi: 10.1111/CGF.15063

    2024 doi

  10. [18]

    A comprehensive review of generative adversarial networks: Fundamentals, applications, and challenges,

    M. Megahed and A. Mohammed, “A comprehensive review of generative adversarial networks: Fundamentals, applications, and challenges,” Wiley Interdiscip. Rev. Comput. Stat., vol. 16, no. 1, p. e1629, Jan. 2024, doi: 10.1002/WICS.1629

    2024 doi

  11. [19]

    Recent Advances in Variational Autoencoders with Representation Learning for Biomedical Informatics: A Survey,

    R. Wei and A. Mahm ood, “Recent Advances in Variational Autoencoders with Representation Learning for Biomedical Informatics: A Survey,” IEEE Access, vol. 9, pp. 4939–4956, 2021, doi: 10.1109/ACCESS.2020.3048309

    2021

  12. [20]

    MedSegDiff: Medical Image Segmentation with Diffusion Probabilistic Model

    J. Wu et al., “MedSegDiff: Medical Image Segmentation with Diffusion Probabilistic Model.” PMLR, pp. 1623–1639, Jan. 23, 2024. Accessed: Jan. 05, 2025. [Online]. Available: https://proceedings.mlr.press/v227/wu24a.html

    2024

  13. [21]

    Diffusion model-based text-guided enhancement network for medical image segmentation,

    Z. Dong, G. Yuan, Z. Hua, and J. Li, “Diffusion model-based text-guided enhancement network for medical image segmentation,” Expert Syst. Appl., vol. 249, p. 123549, Sep. 2024, doi: 10.1016/J.ESWA.2024.123549

    2024

  14. [22]

    Semantic segmentation of diabetic foot ulcer images: Dealing with small dataset in dl approaches,

    N. Rania, H. Douzi, L. Yves, and T. Sylvie, “Semantic segmentation of diabetic foot ulcer images: Dealing with small dataset in dl approaches,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 12119 LNCS, pp. 162–169, 202...

    2020 doi

  15. [23]

    A few-shot diabetes foot ulcer image classification method based on deep ResNet and transfer learning,

    C. Wang, Z. Yu, Z. Long, H. Zhao, and Z. Wang, “A few-shot diabetes foot ulcer image classification method based on deep ResNet and transfer learning,” Sci. Reports 2024 141, vol. 14, no. 1, pp. 1–9, Dec. 2024, doi: 10.1038/s41598-024-80691-w

    2024 doi

  16. [24]

    Improving Limited Supervised Foot Ulcer Segmentation Using Cross-Domain Augmentation,

    S.-J. Kuo, P.-H. Huang, C.-C. Lin, J.-L. Li, and M.-C. Chang, “Improving Limited Supervised Foot Ulcer Segmentation Using Cross-Domain Augmentation,” Jan. 2024, Accessed: Jan. 05, 2025. [Online]. Available: https://arxiv.org/abs/2401.08422v1

    2024 arXiv

  17. [25]

    A framework of wound segmentation based on deep convolutional networks,

    X. Liu, C. Wang, F. Li, X. Zhao, E. Zhu, and Y. Peng, “A framework of wound segmentation based on deep convolutional networks,” Proc. - 2017 10th Int. Congr. Image Signal Process. Biomed. Eng. Informatics, CISP-BMEI 2017, vol. 2018-January, pp. 1–7, Jul. 2017, doi: 10.1109/CIS...

    2017

  18. [26]

    Fully automatic wound segmentation with deep convolutional neural networks,

    C. Wang et al., “Fully automatic wound segmentation with deep convolutional neural networks,” Sci. Reports 2020 101, vol. 10, no. 1, pp. 1–9, Dec. 2020, doi: 10.1038/s41598-020-78799-w

    2020 doi

  19. [27]

    Automatic Foot Ulcer Segmentation Using an Ensemble of Convolutional Neural Networks,

    A. Mahbod, G. Schaefer, R. Ecker, and I. Ellinger, “Automatic Foot Ulcer Segmentation Using an Ensemble of Convolutional Neural Networks,” 2022 26th Int. Conf. Pattern Recognit., vol. 2022-August, pp. 4358–4364, Aug. 2022, doi: 10.1109/ICPR56361.2022.9956253

    2022

  20. [28]

    Translating Clinical Delineation of Diabetic Foot Ulcers into Machine Interpretable Segmentation,

    C. Kendrick et al., “Translating Clinical Delineation of Diabetic Foot Ulcers into Machine Interpretable Segmentation,” arXiv, Apr. 2022, Accessed: Jan. 07,

    2022

  21. [29]

    OCRNet for Diabetic Foot Ulcer Segmentation Combined with Edge Loss,

    H. Yi et al., “OCRNet for Diabetic Foot Ulcer Segmentation Combined with Edge Loss,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 13797 LNCS, pp. 31–39, 2023, doi: 10.1007/978-3-031-26354- 5_3/TABLES/3

    2023 doi

  22. [30]

    Diabetic Foot Ulcer Segmentation Using Convolutional and Transformer-Based Models,

    M. Hassib, M. Ali, A. Mohamed, M. Torki, and M. Hussein, “Diabetic Foot Ulcer Segmentation Using Convolutional and Transformer-Based Models,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 13797 LNCS, pp. 83–91, 2023, d...

    2023 doi

  23. [31]

    Automated Tissue Classification Framework for Reproducible Chronic Wound Assessment,

    R. Mukherjee, D. D. Manohar, D. K. Das, A. Achar, A. Mitra, and C. Chakraborty, “Automated Tissue Classification Framework for Reproducible Chronic Wound Assessment,” Biomed Res. Int., vol. 2014, no. 1, p. 851582, Jan. 2014, doi: 10.1155/2014/851582

    2014 doi

  24. [32]

    Wound tissue segmentation by computerised image analysis of clinical pressure injury photographs: a pilot study,

    D. Li, C. Mathews, C. Zamarripa, F. Zhang, and Q. Xiao, “Wound tissue segmentation by computerised image analysis of clinical pressure injury photographs: a pilot study,” https://doi.org/10.12968/jowc.2022.31.8.710, vol. 31, no. 8, pp. 710–719, Aug. 2022, doi: 10.12968/JOWC.20...

    2022 doi

  25. [33]

    Fully Automated Wound Tissue Segmentation Using Deep Learning on Mobile Devices: Cohort Study,

    D. Ramachandram, J. L. Ramirez-GarciaLuna, R. D. J. Fraser, M. A. Martínez-Jiménez, J. E. Arriaga- Caballero, and J. Allport, “Fully Automated Wound Tissue Segmentation Using Deep Learning on Mobile Devices: Cohort Study,” JMIR mHealth uHealth, vol. 12 10, no. 4, p. e36977, Ap...

    2022 doi

  26. [34]

    Simultaneous wound border segmentation and tissue classification using a conditional generative adversarial network,

    S. Sarp, M. Kuzlu, M. Pipattanasomporn, and O. Guler, “Simultaneous wound border segmentation and tissue classification using a conditional generative adversarial network,” J. Eng., vol. 2021, no. 3, pp. 125–134, Mar. 2021, doi: 10.1049/TJE2.12016

    2021 doi

  27. [35]

    Wound Tissue Segmentation in Diabetic Foot Ulcer Images Using Deep Learning: A Pilot Study,

    M. K. Dhar et al., “Wound Tissue Segmentation in Diabetic Foot Ulcer Images Using Deep Learning: A Pilot Study,” Jun. 2024, Accessed: Jan. 07, 2025. [Online]. Available: https://arxiv.org/abs/2406.16012v1

    2024 arXiv

  28. [36]

    Integrated image and location analysis for wound classification: a deep learning approach,

    Y. Patel et al., “Integrated image and location analysis for wound classification: a deep learning approach,” Sci. Reports 2024 141, vol. 14, no. 1, pp. 1–20, Mar. 2024, doi: 10.1038/s41598-024-56626-w

    2024 doi

  29. [37]

    Have I been Trained ?

    “Have I been Trained ?” Accessed: Feb. 20, 2025. [Online]. Available: https://haveibeentrained.com/

    2025

  30. [38]

    High-Resolution Image Synthesis With Latent Diffusion Models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-Resolution Image Synthesis With Latent Diffusion Models,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 10684–10695. Accessed: Jun. 04, 2024. [Online]. Available: https://g...

    2022

  31. [39]

    Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding,

    C. Saharia et al., “Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding,” Adv. Neural Inf. Process. Syst., vol. 35, pp. 36479–36494, Dec. 2022

    2022

  32. [40]

    Denoising Diffusion Probabilistic Models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising Diffusion Probabilistic Models,” Adv. Neural Inf. Process. Syst., vol. 33, pp. 6840–6851, 2020

    2020

  33. [41]

    DiffuMask: Synthesizing Images with Pixel-level Annotations for Semantic Segmentation Using Diffusion Models,

    W. Wu, Y. Zhao, M. Z. Shou, H. Zhou, and C. Shen, “DiffuMask: Synthesizing Images with Pixel-level Annotations for Semantic Segmentation Using Diffusion Models,” pp. 1206–1217, 2023, doi: 10.1109/ICCV51070.2023.00117

    2023

  34. [42]

    Diffuse Attend and Segment: Unsupervised Zero-Shot Segmentation using Stable Diffusion,

    J. Tian, L. Aggarwal, A. Colaco, Z. Kira, and M. Gonzalez-Franco, “Diffuse Attend and Segment: Unsupervised Zero-Shot Segmentation using Stable Diffusion,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024, pp. 3554–3563

    2024

  35. [43]

    FUSegNet: A deep convolutional neural network for foot ulcer segmentation,

    M. K. Dhar, T. Zhang, Y. Patel, S. Gopalakrishnan, and Z. Yu, “FUSegNet: A deep convolutional neural network for foot ulcer segmentation,” Biomed. Signal Process. Control, vol. 92, p. 106057, Jun. 2024, doi: 10.1016/J.BSPC.2024.106057

    2024

  36. [44]

    DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs,

    L. C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 40, no. 4, pp. 834–848, Apr. 2018, doi: 10.1109...

    2018

  37. [45]

    Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation,

    L. C. Chen, Y. Zhu, G. Papandreou, F. Schroff, and H. Adam, “Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 11211 LNCS, pp. 833–851, Fe...

    2018 doi

  38. [46]

    SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers,

    E. Xie, W. Wang, Z. Yu, A. Anandkumar, J. M. Alvarez, and P. Luo, “SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers,” Adv. Neural Inf. Process. Syst., vol. 15, pp. 12077–12090, May 2021, Accessed: Jan. 05,

    2021

  39. [47]

    Development of a Smartphone-Based Optical Device to Measure Hemoglobin Concentration Changes for Remote Monitoring of Wounds,

    K. Kaile, C. Fernandez, and A. Godavarty, “Development of a Smartphone-Based Optical Device to Measure Hemoglobin Concentration Changes for Remote Monitoring of Wounds,” Biosens. 2021, Vol. 11, Page 165, vol. 11, no. 6, p. 165, May 2021, doi: 10.3390/BIOS11060165

    2021 doi

  40. [48]

    Available: https://arxiv.org/abs/2105.15203v3

    [Online]. Available: https://arxiv.org/abs/2105.15203v3

    arXiv

  41. [50]

    Development and Validation of a Smartphone-Based Near-Infrared Optical Imaging Device to Measure Physiological Changes In-Vivo,

    K. Kaile and A. Godavarty, “Development and Validation of a Smartphone-Based Near-Infrared Optical Imaging Device to Measure Physiological Changes In-Vivo,” Micromachines 2019, Vol. 10, Page 180, vol. 10, no. 3, p. 180, Mar. 2019, doi: 10.3390/MI10030180

    2019 doi

  42. [2025]

    Available: https://arxiv.org/abs/2204.11618v2

    [Online]. Available: https://arxiv.org/abs/2204.11618v2

    arXiv

Pith tools

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