Pith. sign in

REVIEW 3 major objections 4 minor 31 references

Segment Anything for Cell Tracking

T0 review · 3 major / 4 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read A frozen, unmodified SAM2 model can link cell masks across time-lapse microscopy frames and detect mitotic divisions without any dataset-specific training, performing on par with supervised trackers across the Cell Tracking Challenge.

desk verdict A genuinely zero-shot 2D cell-linking method built on SAM2, wrapped in a 3D pipeline that fine-tunes on the target data and reports training-set scores; the 2D part deserves a referee, the 3D claims need revisions. read the letter →

arxiv 2509.09943 v1 pith:KPFFUB2E submitted 2025-09-12 cs.CV

Zhu Chen , Mert Edgü , Er Jin , Johannes Stegmaier This is my paper
classification cs.CV
keywords celltrackingSAM2zero-shotmitosisdetectionfoundationmodelmicroscopysegmentationChallenge
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 claims that a pre-trained video-segmentation foundation model, SAM2, can track cells in microscopy time-lapse sequences without any labeled training data. The authors feed consecutive image patches and prompts derived from an existing segmentation mask into the frozen model, and it predicts the cell's mask in the adjacent frame, enabling mask linking and mitosis detection. Across 13 diverse Cell Tracking Challenge datasets, this zero-shot linking approach reaches top-3 accuracy on average, competing with supervised and tuned methods. For large 3D volumes with thousands of cells, the same SAM2 linking is combined with SAM-Med3D, fine-tuned only on rough watershed masks, to produce volumetric segmentation and tracking with accuracy near leading supervised pipelines. The central promise is that foundation models can remove the need for expensive manual annotation in cell tracking.

What carries the argument

The load-bearing mechanism is SAM2's promptable video mask propagation: a short sequence of two image patches, a mask-derived bounding box, and positive/negative point prompts let the frozen model hallucinate the cell mask in the previous frame, while the model's memory encoder produces feature vectors whose cosine similarity links candidate cells in large-scale tracking. For 3D volumes, SAM-Med3D—a medical foundation model not trained on microscopy—is fine-tuned on rough watershed masks so a single click generates a 3D segmentation mask.

What would settle it

Run the authors' released code on the official CTC test set for Fluo-N3DL-TRIC and Fluo-N3DL-TRIF and compare the reported training-set SEG/TRA values against the test-set values; if the test-set scores drop materially, the central 3D competitive claim is not supported. A second check: apply the zero-shot linking pipeline to a held-out microscopy dataset with ground-truth lineages and see whether LNK remains top-tier without any threshold tuning.

Watch

Extended reading notes

Core claim

The paper's central claim is that a fixed, pre-trained SAM2 model—used without any fine-tuning or dataset-specific adaptation—can serve as the core of a cell-tracking pipeline. Given a known mask at time t, the model receives a crop of frames t and t−1 plus bounding-box and point prompts, and predicts the corresponding mask at t−1; comparing this prediction to pre-segmented masks links the cell across frames, and two cells linking back to the same parent signals mitosis. The authors report top-3 average LNK score over 13 blind test datasets, and for large-scale 3D data they add SAM-Med3D, fine-tuned on rough watershed-generated masks, to segment cells during forward tracking, achieving secon

Load-bearing premise

For the large-scale 3D results, the paper assumes that scores measured on the CTC training set after fine-tuning SAM-Med3D on rough masks from those same sequences are representative of what the method would achieve on the official blind test set, an equivalence asserted without tabulated test-set scores.

Editorial extensions

If this is right

  • Cell tracking on new microscopy datasets no longer requires collecting manual segmentation and tracking annotations; a frozen foundation model plus hand-set thresholds may suffice.
  • Tracking quality becomes directly tied to the quality of the initial segmentation masks, since SAM2 propagates prompts rather than learning dataset-specific appearance priors.
  • Because the pipeline is prompt-based, users can interactively correct detections in one frame and have corrections propagate backward or forward, enabling human-in-the-loop refinement.
  • Large-scale 3D time-lapse data can be processed patch-wise in parallel, making tracking of thousands of cells computationally feasible on a single GPU.
  • If the method holds on official test sets, it offers a strong baseline for evaluating new supervised trackers, since it uses no training data of any kind.

Reading between the lines

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

  • If confirmed on blind tests, this zero-shot approach could substantially lower the entry barrier for cell tracking in understudied organisms or unusual imaging modalities where labeled data are scarce.
  • The same prompt-based linking mechanism could transfer to non-microscopy object tracking tasks that supply mask sequences, since SAM2's video capabilities are not microscopy-specific.
  • The reliance on pre-segmented masks means the pipeline inherits their errors; combining the linking step with self-supervised segmentation refinement could remove the need for even the rough watershed masks in the 3D setting.
  • The method's weakness on large appearance changes and ID switches suggests that integrating whole-structure motion prediction, as the authors state they plan to do, might close the gap to global optimization approaches on BIO scores.
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 / 4 minor

Summary. The paper proposes a cell tracking pipeline built on the pre-trained SAM2 foundation model. For 2D and small 3D data, the method links pre-segmented masks by propagating a mask backward in time through SAM2 prompted with the previous frame's mask, recovering missed detections and detecting mitoses when two daughter masks link to one parent. For large-scale 3D+t data, the pipeline combines SAM2-based similarity linking with SAM-Med3D, which is fine-tuned on rough watershed masks generated from the target sequences, to produce volumetric segmentation and tracking. The authors report top-3 LNK ranks on several CTC blind test datasets and competitive SEG/TRA scores on two large 3D datasets, and claim the framework is zero-shot, fully unsupervised, and requires no fine-tuning or dataset-specific adaptation.

Significance. The 2D linking contribution is potentially valuable: it demonstrates that an unmodified pre-trained video segmentation model can be prompted to link cells across diverse microscopy modalities without any labeled training, and the evaluation on blind test datasets gives credible evidence. The public code is a concrete strength. However, the advertised scope is significantly broader than what is actually demonstrated. The large-3D pipeline fine-tunes SAM-Med3D on masks derived from the same sequences that are later evaluated, and the reported 3D scores are computed on the CTC training set rather than the official test set. These two issues undercut the central claims of 'zero-shot,' 'without fine-tuning,' and 'competitive accuracy in large-scale 3D' as stated in the abstract. The paper is therefore a solid candidate after substantial revision that aligns claims with evidence and supplies missing test-set measurements.

major comments (3)
  1. [Abstract; §2.2; §3.3] The central claim of zero-shot / no-fine-tuning operation is internally contradicted. Section 2.2 states 'We then fine-tune SAM-Med3D using these rough masks' where the rough masks are generated from the target sequences, while Section 3.3 claims 'we apply it directly to the training set without parameter fine-tuning'. Fine-tuning SAM-Med3D on target-data-derived masks is dataset-specific adaptation even if the masks are unannotated. This directly contradicts the abstract's 'without fine-tuning' and 'eliminating the need for dataset-specific adaptation.' Please either remove the fine-tuning step, report results without it, or revise the abstract/contributions to claim only 'no manual annotations' for the 2D path and clearly label the large-3D variant as using unsupervised fine-tuning on target data.
  2. [Table 2; §3.3] The large-3D comparison is not valid as presented. Table 2 reports SEG/TRA scores on the CTC training set, while the competing methods are evaluated on the official CTC test set. The statement that 'training and test set scores are highly similar in our case' is not supported by tabulated test-set numbers; supplementary videos are qualitative and cannot substitute for quantitative evaluation. Please provide actual test-set SEG/TRA scores under the official CTC protocol, or explicitly relabel Table 2 as a non-comparable training-set benchmark. As written, the 'third in SEG / second in TRA' claim is not established.
  3. [§3.2; Table 1] The 2D linking claim is 'top 3 in LNK score on average across all 13 datasets,' but Table 1 shows only six of the 13 datasets and reports ranks rather than the average LNK value. It is unclear whether 'top 3' refers to mean LNK score or mean rank. Please provide the full per-dataset table (or a supplementary table) with LNK/BIO values and state the averaging procedure so the 2D claim is verifiable. Without this, the headline 2D result is not fully reproducible from the manuscript.
minor comments (4)
  1. [§2.2] The search radius τ and patch side length d are introduced as predefined but no values are given in the experiments. Please report the exact values used for each dataset.
  2. [§2.2] The mitosis threshold (similarity difference below 0.1) and the linking threshold (0.8) are stated as fixed, but the paper also says the latter 'might need adjustments in other circumstances.' Please clarify whether these thresholds were fixed across all datasets or tuned per dataset, and how they relate to the claim of no dataset-specific adaptation.
  3. [Table 1/Table 2] In Table 1, the superscript numbers are method identifiers, but in Table 2 the bold row for the proposed method is not explicitly labeled with a superscript. Please add a note explaining the notation consistently.
  4. [General] The supplementary videos are referenced but not accessible in the manuscript; please ensure the supplementary material is available to reviewers and readers, and consider adding quantitative per-sequence results in the supplementary text.

Circularity Check

1 steps flagged · score 6.0 of 10

Large-scale 3D results are produced by fine-tuning SAM-Med3D on rough masks from the same sequences that are later scored, contradicting the zero-shot/no-fine-tuning claim.

  1. fitted input called prediction [Section 2.2 (Large-scale 3D) and Section 3.3 / Table 2]
    "We then fine-tune SAM-Med3D using these rough masks, enabling it to generate 3D segmentation masks with a single click on the cell area. ... Since our approach is fully-unsupervised, meaning no ground truth information from the training set is used, we apply it directly to the training set without parameter fine-tuning and evaluate its performance using ground truth annotations."

    The fine-tuning step adapts SAM-Med3D to rough masks extracted from the exact sequences (Fluo-N3DL-TRIC/TRIF) whose ground-truth SEG/TRA are then reported in Table 2. The paper calls this 'without parameter fine-tuning' and 'zero-shot,' but the 3D segmentation weights are fitted to the evaluated data before scoring. The reported 3D scores are therefore training-set measurements after target-data adaptation, not zero-shot predictions; the subsequent assertion that 'training and test set scores are highly similar' is unsupported by any tabulated test-set numbers, so the competitive-accuracy claim rests on this fitted evaluation. The 'prediction' is statistically dependent on the input data it purports to generalize from.

full rationale

The 2D linking and small-3D results are genuinely independent: SAM2 is used as-is with hand-set prompts and no training on target data. The core circularity is confined to the large-scale 3D contribution, where the model is fine-tuned on rough masks from the same sequences scored in Table 2, and then the paper asserts without tabulated evidence that this transfers to test data. Since this is the basis for the abstract's 'large-scale 3D without dataset-specific adaptation' claim, the central claim partially reduces to a fit on evaluated data. No self-citation chain or uniqueness-theorem circularity is present.

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

The pipeline adds no new physical or conceptual entities; it introduces hand-set linking thresholds and relies on the assumed transferability of two foundation models (SAM2, SAM-Med3D) to microscopy data.

free parameters (4)
  • similarity threshold = 0.8
    Hand-fixed threshold for linking candidates in large-scale 3D tracking; the paper states it 'might need adjustments in other circumstances'.
  • mitosis difference threshold = 0.1
    Hand-fixed difference in similarity scores below which two equally similar candidates are classified as daughter cells.
  • search radius tau = not specified
    Neighborhood radius for candidate search in 3D tracking; no value or selection criterion is given.
  • patch side length d = not specified
    Square patch size for SAM2 input; said to be based on cell size but no formula or value is provided.
assumptions (4)
  • domain assumption SAM2's video mask propagation transfers to microscopy image patches despite being trained on natural video.
    Section 2.1 relies on SAM2 predicting same-cell masks across adjacent frames from a bounding box plus point prompts.
  • domain assumption Cosine similarity of SAM2 memory-encoded features is a reliable proxy for cell identity across frames.
    Section 2.2 uses this similarity to select links and to detect mitosis by comparing scores.
  • domain assumption SAM-Med3D, trained on medical images, can be fine-tuned on rough watershed masks to produce accurate 3D cell masks.
    Section 2.2 fine-tunes SAM-Med3D on rough masks and uses it to generate all 3D segmentations.
  • domain assumption CTC ground-truth lineage annotations are correct and the AOGM/TRA/SEG metrics are appropriate for comparing tracking methods.
    Used without discussion in Section 3.1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Segment Anything for Cell Tracking." pith.science (2026). https://pith.science/paper/KPFFUB2E

@misc{pith2026250909943,
  author       = {Pith},
  title        = {Pith review of: Segment Anything for Cell Tracking},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KPFFUB2E}},
  note         = {Machine review of arXiv:2509.09943}
}
read the original abstract

Tracking cells and detecting mitotic events in time-lapse microscopy image sequences is a crucial task in biomedical research. However, it remains highly challenging due to dividing objects, low signal-tonoise ratios, indistinct boundaries, dense clusters, and the visually similar appearance of individual cells. Existing deep learning-based methods rely on manually labeled datasets for training, which is both costly and time-consuming. Moreover, their generalizability to unseen datasets remains limited due to the vast diversity of microscopy data. To overcome these limitations, we propose a zero-shot cell tracking framework by integrating Segment Anything 2 (SAM2), a large foundation model designed for general image and video segmentation, into the tracking pipeline. As a fully-unsupervised approach, our method does not depend on or inherit biases from any specific training dataset, allowing it to generalize across diverse microscopy datasets without finetuning. Our approach achieves competitive accuracy in both 2D and large-scale 3D time-lapse microscopy videos while eliminating the need for dataset-specific adaptation.

Figures

Figures reproduced from arXiv: 2509.09943 by the authors.

Figure 1
Figure 1. Overview of the cell linking method. Patches are cropped from consecutive images based on the mask center at time point t. Bounding box and point prompts are then generated from the mask at time t. SAM2 predicts the segmentation masks of the same cell in both time frames, and the masks are linked to form tracklets. The depicted example is the Fluo-N2DL-HeLa dataset from Cell Tracking Challenge (CTC) [19]. R w×h , ea… view at source ↗
Figure 2
Figure 2. For a given pair of consecutive time frames, potential cells in time point t + 1 are searched within the neighborhood of the tracked cell in time point t. The bounding box of the tracked cell is used as a prompt for all local patches of potential cells. We use SAM2 to extract memory feature vectors for all patches, and the cell with the highest similarity score is linked. A mitotic event is detected when two potenti… view at source ↗
Figure 3
Figure 3. Visualization of several examples of SAM2 prediction results. Left: input image patch at time t with prompts. Right: output image patch at time t−1 with the predicted mask. The depicted examples are from CTC [19]. 3.2 Cell Linking Results We evaluate cell linking performance using 2D and 3D datasets from the Cell Tracking Challenge (CTC) [19,28]. Each dataset provides an incomplete or com￾plete set of segmentation m… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Visualization of several failed cases of SAM2 prediction results. Left: input image patch at time t with prompts. Right: output image patch at time t − 1 with the predicted mask and the ground truth mask. The depicted examples are from CTC [19]. (a) and (c) show cases …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

31 extracted references · 1 canonical work pages

  1. [1]

    Nature Methods (2025)

    Archit, A., Freckmann, L., Nair, S., Khalid, N., Hilt, P., Rajashekar, V., Freitag, M., Teuber, C., Buckley, G., von Haaren, S., et al.: Segment anything for mi- croscopy. Nature Methods (2025)

    2025

  2. [2]

    In: 17th European Conference on Computer Vision, ECCV

    Ben-Haim, T., Raviv, T.R.: Graph neural network for cell tracking in microscopy videos. In: 17th European Conference on Computer Vision, ECCV. pp. 610–626. Springer (2022)

    2022

  3. [3]

    bioRxiv:2024.09.02.610652v1 (2024)

    Bragantini, J., Theodoro, I., Zhao, X., Huijben, T.A., Hirata-Miyasaki, E., VijayKumar, S., Balasubramanian, A., Lao, T., Agrawal, R., Xiao, S., et al.: Ultrack: pushing the limits of cell tracking across biological scales. bioRxiv:2024.09.02.610652v1 (2024)

    2024

  4. [4]

    Nature Communications11(1), 6254 (2020)

    Cao, J., Guan, G., Ho, V.W.S., Wong, M.K., Chan, L.Y., Tang, C., Zhao, Z., Yan, H.: Establishment of a morphological atlas of the caenorhabditis elegans embryo using deep-learning-based 4d segmentation. Nature Communications11(1), 6254 (2020)

    2020

  5. [5]

    Cell147(5), 992–1009 (2011)

    Friedl, P., Alexander, S.: Cancer invasion and the microenvironment: plasticity and reciprocity. Cell147(5), 992–1009 (2011)

    2011

  6. [6]

    Nature reviews Molecular cell biology10(7), 445–457 (2009)

    Friedl, P., Gilmour, D.: Collective cell migration in morphogenesis, regeneration and cancer. Nature reviews Molecular cell biology10(7), 445–457 (2009)

    2009

  7. [7]

    In: 18th European Conference on Computer Vision, ECCV

    Gallusser, B., Weigert, M.: Trackastra: Transformer-based cell tracking for live- cell microscopy. In: 18th European Conference on Computer Vision, ECCV. pp. 467–484. Springer (2024)

    2024

  8. [8]

    In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 2019, Proceedings, Part I 22

    Hayashida, J., Bise, R.: Cell tracking with deep learning for cell detection and motion estimation in low-frame-rate. In: Medical Image Computing and Computer Assisted Intervention–MICCAI 2019: 22nd International Conference, Shenzhen, China, October 13–17, 2019, Proceedings, Part I 22. pp. 397–405. Springer (2019)

    2019

Show all 31 references
  1. [9]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Hayashida, J., Nishimura, K., Bise, R.: Mpm: Joint representation of motion and position map for cell tracking. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 3823–3832 (2020)

    2020

  2. [10]

    bioRxiv:2023.11.17.567630v5 (2024)

    Israel, U., Marks, M., Dilip, R., Li, Q., Yu, C., Laubscher, E., Li, S., Schwartz, M., Pradhan, E., Ates, A., et al.: A foundation model for cell segmentation. bioRxiv:2023.11.17.567630v5 (2024)

    2023

  3. [11]

    In: Proceedings of the IEEE/CVF International Conference on Computer Vision

    Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: Proceedings of the IEEE/CVF International Conference on Computer Vision. pp. 4015–4026 (2023)

    2023

  4. [12]

    arXiv preprint arXiv:2212.03465 (2022)

    Lee, G., Kim, S., Kim, J., Yun, S.: Mediar: Harmony of data-centric and model- centric for multi-modality microscopy.(2022). arXiv preprint arXiv:2212.03465 (2022)

    2022 arXiv

  5. [13]

    IEEE Access10, 77147–77157 (2022)

    Löffler, K., Mikut, R.: Embedtrack—simultaneous cell segmentation and tracking through learning offsets and clustering bandwidths. IEEE Access10, 77147–77157 (2022)

    2022

  6. [14]

    In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI)

    Lux, F., Matula, P.: Cell tracking based on integer linear programming and prob- ability scores. In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI). pp. 1–5. IEEE (2024)

    2024

  7. [15]

    Ma, J., Kim, S., Li, F., Baharoon, M., Asakereh, R., Lyu, H., Wang, B.: Segment anythinginmedicalimagesandvideos:Benchmarkanddeployment.arXivpreprint arXiv:2408.03322 (2024) Segment Anything for Cell Tracking 11

    2024 arXiv

  8. [16]

    Magnusson, K.E.: Segmentation and tracking of cells and particles in time-lapse microscopy. Ph.D. thesis, KTH Royal Institute of Technology (2016)

    2016

  9. [17]

    IEEE Transactions on Medical Imaging34(4), 911–929 (2014)

    Magnusson, K.E., Jaldén, J., Gilbert, P.M., Blau, H.M.: Global linking of cell tracks using the viterbi algorithm. IEEE Transactions on Medical Imaging34(4), 911–929 (2014)

    2014

  10. [18]

    Nature biotech- nology41(1), 44–49 (2023)

    Malin-Mayor, C., Hirsch, P., Guignard, L., McDole, K., Wan, Y., Lemon, W.C., Kainmueller, D., Keller, P.J., Preibisch, S., Funke, J.: Automated reconstruction of whole-embryo cell lineages by learning from sparse annotations. Nature biotech- nology41(1), 44–49 (2023)

    2023

  11. [19]

    Nature Methods 20(7), 1010–1020 (2023)

    Maška, M., Ulman, V., Delgado-Rodriguez, P., Gómez-de Mariscal, E., Nečasová, T., Guerrero Peña, F.A., Ren, T.I., Meyerowitz, E.M., Scherr, T., Löffler, K., et al.: The cell tracking challenge: 10 years of objective benchmarking. Nature Methods 20(7), 1010–1020 (2023)

    2023

  12. [20]

    PloS one10(12), e0144959 (2015)

    Matula, P., Maška, M., Sorokin, D.V., Matula, P., Ortiz-de Solórzano, C., Kozubek, M.: Cell tracking accuracy measurement based on comparison of acyclic oriented graphs. PloS one10(12), e0144959 (2015)

    2015

  13. [21]

    Biology Open7(7), bio027730 (2018)

    May, M., Denecke, B., Schroeder, T., Götz, M., Faissner, A.: Cell tracking in vitro reveals that the extracellular matrix glycoprotein tenascin-c modulates cell cycle length and differentiation in neural stem/progenitor cells of the developing mouse spinal cord. Biology Open7(...

    2018

  14. [22]

    https://doi.org/10.5281/zenodo.14914189

    Miles, A., jakirkham, Hamman, J., Orfanos, D.P., Stansby, D., Bussonnier, M., Moore, J., Bennett, D., Augspurger, T., Rzepka, N., et al.: zarr-developers/zarr- python: v3.0.4 (2025). https://doi.org/10.5281/zenodo.14914189

    2025 doi

  15. [23]

    bioRxiv:803205v2 (2019)

    Moen, E., Borba, E., Miller, G., Schwartz, M., Bannon, D., Koe, N., Camplisson, I., Kyme, D., Pavelchek, C., Price, T., et al.: Accurate cell tracking and lineage construction in live-cell imaging experiments with deep learning. bioRxiv:803205v2 (2019)

    2019

  16. [24]

    arXiv preprint arXiv:2401.13220 (2024)

    Na, S., Guo, Y., Jiang, F., Ma, H., Huang, J.: Segment any cell: A sam-based auto-prompting fine-tuning framework for nuclei segmentation. arXiv preprint arXiv:2401.13220 (2024)

    2024 arXiv

  17. [25]

    arXiv preprint arXiv:2408.00714 (2024)

    Ravi, N., Gabeur, V., Hu, Y.T., Hu, R., Ryali, C., Ma, T., Khedr, H., Rädle, R., Rolland, C., Gustafson, L., et al.: Sam 2: Segment anything in images and videos. arXiv preprint arXiv:2408.00714 (2024)

    2024 arXiv

  18. [26]

    PloS one9(2), e90036 (2014)

    Stegmaier, J., Otte, J.C., Kobitski, A., Bartschat, A., Garcia, A., Nienhaus, G.U., Strähle, U., Mikut, R.: Fast segmentation of stained nuclei in terabyte-scale, time resolved 3d microscopy image stacks. PloS one9(2), e90036 (2014)

    2014

  19. [27]

    Nature Methods (2025)

    Stringer, C., Pachitariu, M.: Cellpose3: one-click image restoration for improved cellular segmentation. Nature Methods (2025)

    2025

  20. [28]

    Nature Methods14(12), 1141–1152 (2017)

    Ulman, V., Maška, M., Magnusson, K.E., Ronneberger, O., Haubold, C., Harder, N., Matula, P., Matula, P., Svoboda, D., Radojevic, M., et al.: An objective com- parison of cell-tracking algorithms. Nature Methods14(12), 1141–1152 (2017)

    2017

  21. [29]

    arXiv preprint arXiv:2310.15161 (2023)

    Wang, H., Guo, S., Ye, J., Deng, Z., Cheng, J., Li, T., Chen, J., Su, Y., Huang, Z., Shen, Y., et al.: Sam-med3d: towards general-purpose segmentation models for volumetric medical images. arXiv preprint arXiv:2310.15161 (2023)

    2023 arXiv

  22. [30]

    In: 2022 IEEE International Symposium on Biomedical Imaging Challenges (ISBIC)

    Weigert, M., Schmidt, U.: Nuclei instance segmentation and classification in histopathology images with stardist. In: 2022 IEEE International Symposium on Biomedical Imaging Challenges (ISBIC). pp. 1–4. IEEE (2022)

    2022

  23. [31]

    CoRR (2024)

    Zhu, J., Qi, Y., Wu, J.: Medical sam 2: Segment medical images as video via segment anything model 2. CoRR (2024)

    2024

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.