Pith. sign in

REVIEW 5 major objections 6 minor 60 references

Interactive 3D medical segmentation can sustain refinement for dozens of clicks at 0.06 seconds per interaction and support prompt undo, by separating image encoding from prompt-driven refinement.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 05:04 UTC pith:XZ633Z66

load-bearing objection Well-engineered interactive segmentation with genuine low-latency and undo, but the SOTA accuracy claim rests on an oracle-click simulation the real-user study doesn't back up. the 5 major comments →

arxiv 2607.22332 v1 pith:XZ633Z66 submitted 2026-07-24 cs.CV

SLIP: Segmentation with Low-latency Interactive Prompting for 3D Medical Images

Baptiste Podvin , Alexandre Ancel , Flavio Milana , Chiara Innocenzi , Davide Arrigo , Federico Espinola Schulze , Guido Torzilli , Jacques Marescaux
show 3 more authors
Daniel George Alexandre Hostettler Toby Collins
This is my paper
classification cs.CV
keywords interactive segmentation3D medical imageslow latencyreversible promptingpatch memory bankuser studypoint promptsannotation efficiency
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to establish that annotation efficiency in interactive 3D medical segmentation is governed as much by interaction latency and workflow flexibility as by raw accuracy. SLIP separates image encoding from prompt-guided refinement: patch features are computed once and cached, and a fixed-capacity memory bank propagates interaction context across patches. On 13 public datasets, simulated users reach state-of-the-art final accuracy, with an average 0.06 s per interaction versus 0.31 s for the strongest published baseline, and continue improving past 30 clicks where the baseline plateaus. A prospective user study with six expert participants shows comparable or faster annotation for liver tasks and higher preference for low latency and undo, though the MRI uterus task still favors the baseline. The paper concludes that interactive segmentation systems should be evaluated on workflow measures, not segmentation scores alone.

Core claim

SLIP's central claim is that decoupling the image encoder from the prompt-conditioned decoder, plus an inference-time patch memory bank, lets a single 3D model respond to each click in about 0.06 s while still reaching state-of-the-art Dice scores after sustained interaction. The memory bank (six entries: three high-confidence and three recent patches) carries segmentation, uncertainty, and object-presence signals from previously processed patches, so a click in one patch can refine nearby patches without re-encoding the volume. Because image features are cached and only affected patches are revisited, the model also supports reversible prompting: users can undo any number of prompts, with i

What carries the argument

The load-bearing mechanism is the separation of image encoding from interactive reasoning, realized as (1) a shared 3D CNN encoder whose patch features are computed once and cached, (2) a fixed-capacity Patch Memory Bank (B = 6) that stores high-confidence and recently inferred patches and is updated breadth-first, (3) a Patch Contextualization Network that uses cross-attention from the current patch to the memory bank embedding to propagate prompt context across patches, and (4) a SAM-style 3D mask decoder plus a prompt-undo mechanism that restores cached compressed deltas of patch states. The memory bank is what allows sparse user guidance to spread beyond the clicked patch; the decoder se

Load-bearing premise

The simulated-user protocol places each corrective click in the largest false-positive or false-negative region, assuming users click near-optimally; the sustained late-stage advantage and the user-study preference results are the evidence that depends on this assumption, and the real-user study is too small (six participants) to confirm it.

What would settle it

Record the actual click sequences of more than six real annotators (ideally 20+) on the same MRI-uterus and CT-liver-lesion cases, replay those exact sequences through SLIP and the baseline, and check whether SLIP still achieves higher DSC at 50 clicks and comparable annotation time; if real clicks deviate from largest-error-region placement, the crossover at ~30 clicks may vanish.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If correct, interactive annotation of a typical CT or MR volume could be performed with sub-100-ms feedback per click, letting radiographers and surgeons refine challenging structures through many more correction rounds than current tools permit.
  • Reversible prompting could eliminate the restart workflow that occurs when a misplaced click derails an interactive session, as observed in the CT liver-lesion task where the baseline required up to three restarts per case.
  • The decoupled architecture implies image encoders can be swapped or upgraded without retraining the whole interactive system, since only the lightweight refinement modules depend on the prompt loop.
  • The reported CPU-only latency (~0.12 s per interaction) suggests interactive segmentation could move to radiology workstations and low-resource settings that lack GPUs during annotation.
  • Annotation efficiency comparisons should include cumulative interaction latency and undo support, not only DSC after a fixed number of clicks.

Where Pith is reading between the lines

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

  • Beyond the paper, the crossover at ~30 clicks suggests that studies using only 5-click budgets (e.g., challenge-style benchmarks) may systematically underestimate the value of low-latency methods; longer-horizon protocols would give a fairer picture.
  • If real users behave like the simulation when clicks are cheap and undoable, low latency might change optimal interaction strategy: users could afford to experiment and correct, rather than making a few cautious high-value clicks.
  • The mixed MRI-uterus result hints that low latency does not compensate for a weak initial model on a particular anatomy; a testable extension is to fine-tune the decoder on task-specific data (as the ultrasound task did) and re-run the user study to see whether the latency advantage then dominates.
  • The compressed-delta undo mechanism could be extended to multi-user collaborative annotation, where each editor needs to roll back their own prompts without recomputing the volume.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes SLIP, an interactive 3D medical image segmentation method that decouples image encoding from prompt-based refinement. Image patches are encoded once and reused across interactions; a patch memory bank and contextualization network propagate interaction context across patches, and reversible prompting is implemented by caching compressed patch states. The method is evaluated with simulated clicks on 13 public datasets against nnInteractive, SegVol, and SAM-Med3D, and in a prospective user study with six participants on three annotation tasks. The authors claim state-of-the-art accuracy, lower interaction latency, reversible prompting, and higher user preference.

Significance. If the technical results hold, SLIP is a practically relevant contribution: decoupling image encoding from prompt refinement is a sound way to reduce interaction latency, and the memory-bank design plus reversible prompting addresses real workflow needs. The paper also makes a genuine effort to complement simulated evaluation with a prospective human study, and it makes code publicly available. However, the strength of the evidence is not yet commensurate with the claims: the benchmark contains an apparent training/test overlap, the central accuracy advantage is produced by an oracle click simulation, and the user study is small and does not confirm the headline advantages. The paper is worth a major revision rather than rejection because the architecture and evaluation framework are promising, but the claims need substantial qualification and the evaluation needs additional analysis.

major comments (5)
  1. [Sec 3.1.2 / Sec 2.4.5 / Supp. Tables 1-2] The paper states in Sec 3.1.2 that none of the 13 test datasets are included in the training data, yet Sec 2.4.5 lists TRUSTED [49] among the ultrasound training datasets, and Supplementary Table 1 lists TRUSTED as a training set while Supplementary Table 2 lists TRUSTED as a test set. This is a direct internal contradiction. If any portion of TRUSTED was used for training, results on TRUSTED and the cross-dataset averages in Table 1 are contaminated. Please clarify the exact overlap (e.g., training on US volumes and testing on CT volumes from the same collection could still leak information), and if necessary rerun the benchmark without TRUSTED or report results separately with and without it.
  2. [Sec 3.1.1, Sec 3.2] The late-stage accuracy advantage of SLIP is produced by an oracle click simulation: corrective clicks are always placed in the largest false-positive/false-negative region (Sec 3.1.1). The paper itself acknowledges that simulated interactions cannot capture human decision-making (Sec 3.2), but the abstract's SOTA claim relies on this simulation. The prospective user study does not confirm the advantage: on MRI uterus nnInteractive is significantly faster (56% reduction, Table 3) and has significantly higher DSC (0.770 vs 0.730); on the two liver tasks the differences are not statistically significant. No analysis compares simulated click placement with the clicks actually made by the six participants. Please add such an analysis (e.g., distance between real clicks and the largest error region, or a simulated-user protocol using the observed click statistics) and qualify the simulation-b
  3. [Sec 3.1.1 / Table 1] Section 3.1.1 states that the mean and standard deviation of each metric are reported across all test cases, but Table 1 reports only point estimates. There are no standard deviations, confidence intervals, or per-dataset significance tests. Without these, the cross-dataset differences (final DSC 0.90 vs 0.88; DC-AUC@50 0.85 vs 0.86) cannot be distinguished from noise. Please add uncertainty measures and appropriate paired statistical tests (or effect sizes) for the simulated evaluation.
  4. [Abstract / Sec 3.1.3, Table 1] The abstract's unqualified claim of 'state-of-the-art interactive segmentation performance' is not fully supported by Table 1. SLIP has a lower DC-AUC@50 than nnInteractive (0.85 vs 0.86), and nnInteractive achieves the best AUC on 9 of 13 datasets. SLIP's advantage is in final DSC after 50 clicks and in interaction latency. The claim should be qualified, e.g., 'higher final DSC after 50 clicks with substantially lower latency, while early-interaction AUC remains lower'.
  5. [Sec 2.3.7 / Introduction, Limitation 2] The introduction states that reversible prompting is achieved 'without storing intermediate segmentation states,' but Sec 2.3.7 describes caching the prediction state of each affected patch and storing compressed deltas. These statements are contradictory. The actual contribution is a memory-efficient state cache, not state-free undo. Please correct the description in the introduction and abstract if it appears there.
minor comments (6)
  1. [Table 1 caption] The column header 'Final AUC Time' is ambiguous. Please define 'Final' as DSC after 50 clicks and 'AUC' as DC-AUC@50 in the caption or main text.
  2. [Sec 3.2.4] The first paragraph contains a duplicated phrase: 'demonstrating the efficiency advantage of interactive segmentation over fully manual annotation' appears twice. Please remove one occurrence.
  3. [Sec 3.2.4] The confidence interval for the MRI Uterus nnInteractive comparison reads '50.1.0%–61.3%' and appears to have a typo. It should presumably be '50.1%–61.3%'.
  4. [Sec 2.3.3] The rule 'After predicting six negative patches, the queue is no longer updated' is a design choice with no justification or ablation. Please add a brief explanation or a sensitivity analysis, since this rule could affect the late-stage propagation behavior central to the paper.
  5. [Fig. 4] The ablation in Fig. 4 shows a single liver experiment with no error bars or confidence intervals. Please clarify how representative this is and, if possible, provide results across more structures or cases.
  6. [Sec 3.2.1 / Sec 2.4.5] The US Liver Lesion task is task-supervised for SLIP because the model was trained on FLL-US-Train, whereas nnInteractive is used without task-specific fine-tuning. Even though the test split is held out, this is a confound when comparing interactive methods on the US task. Please acknowledge this limitation explicitly in the discussion of the user-study results.

Circularity Check

0 steps flagged

No circular derivation: SLIP's performance claims are empirically measured; only minor self-referential benchmark elements (TRUSTED, FLL-US-Train) slightly weaken external independence.

full rationale

The derivation chain is not circular. SLIP's contributions—decoupled image encoding, patch memory bank, contextualization network, reversible prompting—are implemented and trained end-to-end, and the headline numbers (DSC@50, DC-AUC@50, 0.06 s latency, user-study times) are measured against external baselines under a shared protocol. No predicted quantity is defined from a fitted parameter, and no architectural choice is justified by a self-cited uniqueness result. The closest methodological point is the simulated-user protocol (Sec 3.1.1), which places corrective clicks in the largest error region; the training procedure (Sec 2.4.2) uses a similar dominant-error-region simulation. This is a realism and potential overfitting concern, not a circular reduction, because the benchmark metric is not constructed from SLIP's own outputs. The paper itself acknowledges in Sec 3.2 that 'Automatically generated interactions cannot capture human decision-making, error-correction strategies, adaptation to intermediate predictions...' and the six-participant user study is independent evidence, though mixed (nnInteractive is significantly faster and more consistent on the MRI Uterus task). Two self-referential evaluation components should be noted: TRUSTED is a test dataset from the same group (Marescaux/George) and also appears in the training dataset table (Supp. Table 1: 'TRUSTED [37] Kidney 131 US'), which conflicts with Sec 3.1.2's statement that 'None of these datasets are included in the training data'; and the US user-study task uses FLL-US-Train, a private dataset that also contributed to training (held-out split). These affect benchmark independence, not the derivation itself, so they warrant a minor score adjustment rather than a circularity finding.

Axiom & Free-Parameter Ledger

8 free parameters · 5 axioms · 0 invented entities

The central accuracy/latency claims depend on several hand-chosen architectural constants (patch size, overlap, memory capacity, thresholds) and evaluation assumptions (simulated click model, dataset separation, STAPLE reference). No new physical or unobserved entities are introduced.

free parameters (8)
  • patch size = (32,192,192)
    Anisotropic patch size chosen for speed and typical medical volume aspect ratio (Sec 2.3.2).
  • patch overlap ratio s = 0.9
    Set to 90% (vs nnU-Net 50%) to trade patch density for interactive speed (Sec 2.3.2).
  • memory bank capacity B = 6 (3 high-confidence + 3 recent)
    Fixed-capacity memory; capacity chosen empirically (Sec 2.3.4).
  • confidence thresholds tau_u, tau_v, w_u = 0.5, 0.8, 1.5
    Empirical thresholds for memory bank allocation (Sec 2.3.6).
  • fusion sigma sfuse = 0.125
    Gaussian-weighted fusion smoothing set empirically (Sec 2.3.3).
  • queue negative-patch stop count = 6
    Heuristic: after six negative patches, queue stops updating (Sec 2.3.3).
  • mask decoder loss weights = lambda_focal=20, others 1
    Focal weight boosted to balance magnitudes (Sec 2.4.1).
  • training hyperparameters = lr 4e-5/8e-5/2e-5; batch 32; 300 epochs
    No formal HP search; set by experimentation (Sec 2.4.6).
axioms (5)
  • domain assumption Simulated clicks in the largest error region are a faithful proxy for real interactive refinement
    The simulated benchmark (Sec 3.1.1) assumes users click near-optimally; real behavior includes mistakes and suboptimal prompts, which the user study only partially covers.
  • domain assumption The 13 test datasets were not used in training
    The authors state this (Sec 3.1.2) but no data-version audit is provided.
  • domain assumption STAPLE consensus of six manual segmentations is an unbiased reference for comparing interactive methods
    Used to measure annotation reference consistency (Sec 3.2.3); consensus built only from manual masks to avoid bias toward either interactive method.
  • domain assumption Pseudo-labels generated by SAM/SAM2 provide valid supervision
    Training augmentation uses pseudo-labels (Sec 2.4.4); errors in pseudo-labels could bias learned representations.
  • ad hoc to paper Queue stops updating after six negative patches does not degrade propagation
    A designed heuristic in Sec 2.3.3 with no ablation.

pith-pipeline@v1.3.0-alltime-deepseek · 38008 in / 13220 out tokens · 120049 ms · 2026-08-01T05:04:26.234666+00:00 · methodology

0 comments
read the original abstract

Interactive deep image segmentation enables efficient medical image annotation by iteratively refining predictions from user prompts, such as positive and negative clicks. Recent patch-based methods, including nnInteractive, achieve strong segmentation performance but remain limited in annotation workflows by high interaction latency, limited responsiveness to successive interactions, and the lack of support for reversible prompting. Furthermore, evaluation relies predominantly on simulated rather than controlled real-user interaction studies. We present SLIP, an end-to-end trainable framework for interactive 3D medical image segmentation that decouples image encoding from prompt-guided refinement. Image features are computed once and reused, while a lightweight patch memory bank maintains an interaction-aware segmentation state shared across patches. This representation enables prediction updates by propagating interaction context throughout the image, supports reversible prompting without recomputing image features, and substantially reduces interaction latency. By separating image representation from interactive reasoning, SLIP remains compatible with a wide range of image encoders. We train a single SLIP model for general interactive segmentation across diverse anatomical structures and imaging modalities. Beyond standard simulated evaluation, we conduct a controlled prospective user study comparing manual segmentation, nnInteractive, and SLIP across three clinical annotation tasks, six expert participants, and subjective usability measures, addressing the limited human validation of interactive segmentation methods. SLIP achieves SOTA interactive segmentation performance across 13 public datasets while providing lower interaction latency, greater responsiveness, support for reversible prompting, and higher user preference than existing approaches.

Figures

Figures reproduced from arXiv: 2607.22332 by Alexandre Ancel, Alexandre Hostettler, Baptiste Podvin, Chiara Innocenzi, Daniel George, Davide Arrigo, Federico Espinola Schulze, Flavio Milana, Guido Torzilli, Jacques Marescaux, Toby Collins.

Figure 1
Figure 1. Figure 1: Overview of the SLIP inference pipeline and model architecture. Inference is divided into two stages. (1) Patch pre￾processing: the input 3D image is partitioned into overlapping patches, each of which is encoded once by a shared CNN encoder to produce patch feature embeddings that are cached for the remainder of the interactive session. (2) Real-time user interaction: after each user prompt, only affected… view at source ↗
Figure 2
Figure 2. Figure 2: Qualitative comparison across three representative test datasets in the simulated-user evaluation. Each row shows a representative case, with a 2D image slice (top) and corresponding 3D rendering (bottom). Columns show the ground truth (GT) and segmentations produced by SegVol, SAMMed3D, nnInteractive, and SLIP after 10 simulated iterative clicks. 3. Interactive segmentation with SLIP, implemented within t… view at source ↗
Figure 4
Figure 4. Figure 4: Ablation of patch feature contextualization for simulated interactive liver segmentation. Mean DSC is plotted against the number of interaction points for models with and without patch feature contextualization, illustrating its role in propagating sparse user guidance during the earliest stages of refinement. 5.6% longer annotation time on average (𝑝 = 0.520, 95% CI: −10.6% to +24.9%). Despite requiring m… view at source ↗
Figure 3
Figure 3. Figure 3: Quantitative comparison of segmentation perfor￾mance in the simulated-user evaluation over 50 iterative interactions. (a) Mean DSC versus the number of interaction points, illustrating prompt efficiency. (b) Mean DSC versus cumulative interaction latency, illustrating the cumulative waiting time required to achieve a given segmentation quality during an interactive annotation session. reversible prompting.… view at source ↗
Figure 1
Figure 1. Figure 1: Architecture of the SLIP model. CNN blocks follow a residual structure with two 3 [PITH_FULL_IMAGE:figures/full_fig_p022_1.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

60 extracted references · 20 canonical work pages

  1. [1]

    Prostate158-an expert-annotated 3t mri dataset and algorithm for prostate cancer detection

    Adams, L.C., Makowski, M.R., Engel, G., Rattunde, M., Busch, F., Asbach, P., Niehues, S.M., Vinaya- halingam, S., van Ginneken, B., Litjens, G., et al., 2022. Prostate158-an expert-annotated 3t mri dataset and algorithm for prostate cancer detection. Computers in biology and medicine 148, 105817. doi:10.1016/j.compbiomed.2022.105817

    arXiv 2022

  2. [2]

    Data from nsclc-radiomics-genomics

    Aerts, H.J., Rios Velazquez, E., Leijenaar, R.T., Parmar, C., Grossmann, P., Carvalho, S., Bussink, J., Monshouwer, R., Haibe-Kains, B., Rietveld, D., et al., 2015. Data from nsclc-radiomics-genomics. (No Title) doi:10.7937/K9/TCIA.2015.PF0M9REI

    doi:10.7937/k9/tcia.2015.pf0m9rei 2015

  3. [3]

    Covid-19 lung ct scans

    Aria, M., Ghaderzadeh, M., Asadi, F., 2021. Covid-19 lung ct scans. doi:10.34740/KAGGLE/DSV/1875670

    arXiv 2021

  4. [4]

    Data from lidc-idri

    Armato III, S.G., McLennan, G., Bidaut, L., McNitt-Gray, M.F., Meyer, C.R., Reeves, A.P., Zhao, B., Aberle, D.R., Henschke, C.I., Hoffman, E.A., et al., 2015. Data from lidc-idri. (No Title) doi:/10.7937/K9/TCIA.2 015.LO9QL9SX

    doi:10.7937/k9/tcia.2 2015

  5. [5]

    Spatial proteomics in three-dimensional intact specimens

    Bhatia, H.S., Brunner, A.D., ¨Ozt¨ urk, F., Kapoor, S., Rong, Z., Mai, H., Thielert, M., Ali, M., Al-Maskari, R., Paetzold, J.C., Kofler, F., Todorov, M.I., Molbay, M., Kolabas, Z.I., Negwer, M., Hoeher, L., Steinke, H., Dima, A., Gupta, B., Kaltenecker, D., ¨Oz¨ um Sehnaz Caliskan, Brandt, D., Krahmer, N., M¨ uller, S., Lichtenthaler, S.F., Hellal, F., B...

    2022

  6. [6]

    Segmenting the Inferior Alveolar Canal in CBCTs Volumes: the ToothFairy Challenge

    Bolelli, F., et al., 2024. Segmenting the Inferior Alveolar Canal in CBCTs Volumes: the ToothFairy Challenge. IEEE Transactions on Medical Imaging , 1–17doi:10.1109/TMI.2024.3523096

    arXiv 2024

  7. [7]

    Automatic Segmentation and Alignment of Uterine Shapes from 3D Ultrasound Data

    Boneˇ s, E., Gergolet, M., Bohak, C., ˇZiga Lesar, Marolt, M., 2024. Automatic Segmentation and Alignment of Uterine Shapes from 3D Ultrasound Data. Computers in Biology and Medicine 178, 108794. doi:10.101 6/j.compbiomed.2024.108794

    arXiv 2024

  8. [8]

    Miccai challenge on circuit reconstruction from electron microscopy images.https://cremi

    CREMI, 2016. Miccai challenge on circuit reconstruction from electron microscopy images.https://cremi. org/. Accessed: 2020-10-27

    2016

  9. [9]

    Totalsegmentator mri: robust sequence-independent segmentation of multiple anatomic structures in mri

    D’Antonoli, T.A., Berger, L.K., Indrakanti, A.K., Vishwanathan, N., Weiß, J., Jung, M., Berkarda, Z., Rau, A., Reisert, M., K¨ ustner, T., et al., 2024. Totalsegmentator mri: robust sequence-independent segmentation of multiple anatomic structures in mri. arXiv preprint arXiv:2405.19492 doi:10.1148/radiol.241613

    Pith/arXiv arXiv 2024

  10. [10]

    Ctspine1k: A large-scale dataset for spinal vertebrae segmentation in computed tomography

    Deng, Y., Wang, C., Hui, Y., Li, Q., Li, J., Luo, S., Sun, M., Quan, Q., Yang, S., Hao, Y., et al., 2021. Ctspine1k: A large-scale dataset for spinal vertebrae segmentation in computed tomography. arXiv preprint arXiv:2105.14711 doi:10.59275/j.melba.2025-gf84

    Pith/arXiv arXiv 2021

  11. [11]

    Lnq 2023 challenge: Benchmark of weakly-supervised techniques for mediastinal lymph node quantification

    Dorent, R., Khajavi, R., Idris, T., Ziegler, E., Somarouthu, B., Jacene, H., LaCasce, A., Deissler, J., Ehrhardt, J., Engelson, S., et al., 2024. Lnq 2023 challenge: Benchmark of weakly-supervised techniques for mediastinal lymph node quantification. arXiv preprint arXiv:2408.10069 doi:10.59275/j.melba.2025-d482

    Pith/arXiv arXiv 2024

  12. [12]

    Ultrasound segmentation analysis via distinct and completed anatomical borders

    Duque, V.G., Marquardt, A., Velikova, Y., Lacourpaille, L., Nordez, A., Crouzier, M., Lee, H.J., Mateus, D., Navab, N., 2024. Ultrasound segmentation analysis via distinct and completed anatomical borders. International Journal of Computer Assisted Radiology and Surgery 19, 1419–1427. doi:10.1371/journal.po ne.0268550

    doi:10.1371/journal.po 2024

  13. [13]

    A large-scale multicenter breast cancer dce-mri benchmark dataset with expert segmentations

    Garrucho, L., Kushibar, K., Reidel, C.A., Joshi, S., Osuala, R., Tsirikoglou, A., Bobowicz, M., Del Riego, J., Catanese, A., Gwo´ zdziewicz, K., et al., 2025. A large-scale multicenter breast cancer dce-mri benchmark dataset with expert segmentations. Scientific data 12, 453. doi:10.1038/s41597-025-04707-4

    work page doi:10.1038/s41597-025-04707-4 2025

  14. [14]

    A whole-body fdg-pet/ct dataset with manually annotated tumor lesions

    Gatidis, S., Hepp, T., Fr¨ uh, M., La Foug` ere, C., Nikolaou, K., Pfannenberg, C., Sch¨ olkopf, B., K¨ ustner, T., Cyran, C., Rubin, D., 2022. A whole-body fdg-pet/ct dataset with manually annotated tumor lesions. Scientific Data 9, 601. doi:10.1038/s41597-022-01718-3

    doi:10.1038/s41597-022-01718-3 2022

  15. [15]

    Spider-lumbar spine segmentation in mr images: a dataset and a public benchmark

    van der Graaf, J., van Hooff, M., Buckens, C., et al., 2023. Spider-lumbar spine segmentation in mr images: a dataset and a public benchmark. Boston: Zendo 10. doi:10.1038/s41597-024-03090-w

    doi:10.1038/s41597-024-03090-w 2023

  16. [16]

    Deep learning enables automatic detection and segmentation of brain metastases on multisequence mri

    Grøvik, E., Yi, D., Iv, M., Tong, E., Rubin, D., Zaharchuk, G., 2020. Deep learning enables automatic detection and segmentation of brain metastases on multisequence mri. Journal of Magnetic Resonance Imaging 51, 175–182. doi:10.1002/jmri.26766. 6

    work page doi:10.1002/jmri.26766 2020

  17. [17]

    Multi-class segmentation of aortic branches and zones in computed tomography angiography: the aortaseg24 challenge

    Imran, M., Krebs, J.R., Sivaraman, V.B., Zhang, T., Kumar, A., Ueland, W.R., Fassler, M.J., Huang, J., Sun, X., Wang, L., et al., 2025. Multi-class segmentation of aortic branches and zones in computed tomography angiography: the aortaseg24 challenge. arXiv preprint arXiv:2502.05330 doi:10.1016/j.media.2026.104188

    Pith/arXiv arXiv 2025

  18. [18]

    Towards unifying anatomy segmentation: Automated generation of a full-body ct dataset, in: 2024 IEEE International Conference on Image Processing (ICIP), pp

    Jaus, A., Seibold, C., Hermann, K., Shahamiri, N., Walter, A., Giske, K., Haubold, J., Kleesiek, J., Stiefel- hagen, R., 2024. Towards unifying anatomy segmentation: Automated generation of a full-body ct dataset, in: 2024 IEEE International Conference on Image Processing (ICIP), pp. 41–47. doi:10.1109/ICIP51287.20 24.10647307

    work page doi:10.1109/icip51287.20 2024

  19. [19]

    3d whole body preclinical micro-ct database of subcutaneous tumors in mice with annotations from 3 annotators

    Jensen, M., Clemmensen, A., Hansen, J.G., van Krimpen Mortensen, J., Christensen, E.N., Kjaer, A., Ripa, R.S., 2024. 3d whole body preclinical micro-ct database of subcutaneous tumors in mice with annotations from 3 annotators. Scientific Data 11, 1021. doi:10.1038/s41597-024-03814-y

    work page doi:10.1038/s41597-024-03814-y 2024

  20. [20]

    Tracked 3d ultrasound and deep neural network-based thyroid segmentation reduce interobserver variability in thyroid volumetry

    Kr¨ onke, M., Eilers, C., Dimova, D., K¨ ohler, M., Buschner, G., Schweiger, L., Konstantinidou, L., Makowski, M., Nagarajah, J., Navab, N., et al., 2022. Tracked 3d ultrasound and deep neural network-based thyroid segmentation reduce interobserver variability in thyroid volumetry. Plos one 17, e0268550. doi:10.1371/jo urnal.pone.0268550

    doi:10.1371/jo 2022

  21. [21]

    Emidec: a database usable for the automatic evaluation of myocardial infarction from delayed-enhancement cardiac mri

    Lalande, A., Chen, Z., Decourselle, T., Qayyum, A., Pommier, T., Lorgis, L., de La Rosa, E., Cochet, A., Cottin, Y., Ginhac, D., et al., 2020. Emidec: a database usable for the automatic evaluation of myocardial infarction from delayed-enhancement cardiac mri. Data 5, 89. doi:10.3390/data5040089

    work page doi:10.3390/data5040089 2020

  22. [22]

    Deep learning for segmentation using an open large-scale dataset in 2d echocardiography

    Leclerc, S., Smistad, E., Pedrosa, J., Østvik, A., Cervenansky, F., Espinosa, F., Espeland, T., Berg, E.A.R., Jodoin, P.M., Grenier, T., et al., 2019. Deep learning for segmentation using an open large-scale dataset in 2d echocardiography. IEEE transactions on medical imaging 38, 2198–2210. doi:10.1109/TMI.2019.2900516

    arXiv 2019

  23. [23]

    Abdomenatlas: A large-scale, detailed-annotated, & multi-center dataset for efficient transfer learning and open algorithmic benchmarking

    Li, W., Qu, C., Chen, X., Bassi, P.R., Shi, Y., Lai, Y., Yu, Q., Xue, H., Chen, Y., Lin, X., et al., 2024. Abdomenatlas: A large-scale, detailed-annotated, & multi-center dataset for efficient transfer learning and open algorithmic benchmarking. Medical Image Analysis 97, 103285. doi:10.1016/j.media.2024.103285

    arXiv 2024

  24. [24]

    Lin, Z., Wei, D., Petkova, M.D., Wu, Y., Ahmed, Z., K, K.S., Zou, S., Wendt, N., Boulanger-Weill, J., Wang, X., et al., 2021. Nucmm dataset: 3d neuronal nuclei instance segmentation at sub-cubic millimeter scale, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 164–174. doi:10.1007/978-3-030-87193-2_16

    work page doi:10.1007/978-3-030-87193-2_16 2021

  25. [25]

    Liu, Y., Yibulayimu, S., Sang, Y., Zhu, G., Wang, Y., Zhao, C., Wu, X., 2023. Pelvic fracture segmentation using a multi-scale distance-weighted neural network, in: International Conference on Medical Image Com- puting and Computer-Assisted Intervention, Springer. pp. 312–321. doi:10.1007/978-3-031-43996-4_30

    work page doi:10.1007/978-3-031-43996-4_30 2023

  26. [26]

    Sdr-former: A siamese dual-resolution transformer for liver lesion classification using 3d multi-phase imaging

    Lou, M., Ying, H., Liu, X., Zhou, H.Y., Zhang, Y., Yu, Y., 2025. Sdr-former: A siamese dual-resolution transformer for liver lesion classification using 3d multi-phase imaging. Neural Networks , 107228doi:10.101 6/j.neunet.2025.107228

    arXiv 2025

  27. [27]

    Segrap2023: A benchmark of organs-at-risk and gross tumor volume segmentation for radiotherapy planning of nasopharyngeal carcinoma

    Luo, X., Fu, J., Zhong, Y., Liu, S., Han, B., Astaraki, M., Bendazzoli, S., Toma-Dasu, I., Ye, Y., Chen, Z., et al., 2025. Segrap2023: A benchmark of organs-at-risk and gross tumor volume segmentation for radiotherapy planning of nasopharyngeal carcinoma. Medical image analysis 101, 103447. doi:10.1016/j.me dia.2024.103447

    arXiv 2025

  28. [28]

    Towards data-efficient learning: A benchmark for covid-19 ct lung and infection segmentation

    Ma, J., Wang, Y., An, X., Ge, C., Yu, Z., Chen, J., Zhu, Q., Dong, G., He, J., He, Z., Cao, T., Zhu, Y., Nie, Z., Yang, X., 2021. Towards data-efficient learning: A benchmark for covid-19 ct lung and infection segmentation. Medical Physics 48, 1197–1210. doi:10.1002/mp.14676

    doi:10.1002/mp.14676 2021

  29. [29]

    Automatic organ and pan-cancer segmentation in abdomen ct: the flare 2023 challenge

    Ma, J., Zhang, Y., Gu, S., Ge, C., Wang, E., Zhou, Q., Huang, Z., Lyu, P., He, J., Wang, B., 2024. Automatic organ and pan-cancer segmentation in abdomen ct: the flare 2023 challenge. arXiv preprint arXiv:2408.12534 doi:10.48550/arXiv.2408.12534

    Pith/arXiv arXiv doi:10.48550/arxiv.2408.12534 2024

  30. [30]

    Duke liver dataset: A publicly available liver mri dataset with liver segmentation masks and series labels

    Macdonald, J.A., Zhu, Z., Konkel, B., Mazurowski, M.A., Wiggins, W.F., Bashir, M.R., 2023. Duke liver dataset: A publicly available liver mri dataset with liver segmentation masks and series labels. Radiology: Artificial Intelligence 5, e220275. doi:10.1148/ryai.220275

    doi:10.1148/ryai.220275 2023

  31. [31]

    Open access series of imaging studies (oasis): cross-sectional mri data in young, middle aged, nondemented, and demented older adults

    Marcus, D.S., Wang, T.H., Parker, J., Csernansky, J.G., Morris, J.C., Buckner, R.L., 2007. Open access series of imaging studies (oasis): cross-sectional mri data in young, middle aged, nondemented, and demented older adults. Journal of cognitive neuroscience 19, 1498–1507. doi:10.1162/jocn.2007.19.9.1498

    doi:10.1162/jocn.2007.19.9.1498 2007

  32. [32]

    Deep learning segmentation of the right ventricle in cardiac mri: the m&ms challenge

    Mart ´ ın-Isla, C., Campello, V.M., Izquierdo, C., Kushibar, K., Sendra-Balcells, C., Gkontra, P., Sojoudi, A., Fulton, M.J., Arega, T.W., Punithakumar, K., et al., 2023. Deep learning segmentation of the right ventricle in cardiac mri: the m&ms challenge. IEEE Journal of Biomedical and Health Informatics 27, 3302–3313. doi:10.1109/JBHI.2023.3267857. 7

    arXiv 2023

  33. [33]

    type b aortic dissection cta collection with true and false lumen expert annotations for the development of ai-based algorithms

    Mayer, C., Pepe, A., Hossain, S., Karner, B., Arnreiter, M., Kleesiek, J., Schmid, J., Janisch, M., Hannes, D., Fuchsj¨”ager, M., et al., 2024. type b aortic dissection cta collection with true and false lumen expert annotations for the development of ai-based algorithms. Scientific Data 11, 596. doi:10.1038/s41597-024-0 3284-2

    work page doi:10.1038/s41597-024-0 2024

  34. [34]

    Cervical cancer—tumor heterogeneity: Serial functional and molecular imaging across the radiation therapy course in advanced cervical cancer (version 1)

    Mayr, N., Yuh, W., Bowen, S., Harkenrider, M., Knopp, M., Lee, E., Leung, E., Lo, S., Small Jr, W., Wolfson, H., 2023. Cervical cancer—tumor heterogeneity: Serial functional and molecular imaging across the radiation therapy course in advanced cervical cancer (version 1). The Cancer Imaging Archive doi:/10.7937/ERZ5-QZ59

    work page doi:10.7937/erz5-qz59 2023

  35. [35]

    Voxel-level segmentation of pathologically-proven adrenocortical carcinoma with ki-67 expression (adrenal-acc-ki67-seg)[data set]

    Moawad, A.W., Ahmed, A.A., ElMohr, M., Eltaher, M., Habra, M.A., Fisher, S., Perrier, N., Zhang, M., Fuentes, D., Elsayes, K., 2023a. Voxel-level segmentation of pathologically-proven adrenocortical carcinoma with ki-67 expression (adrenal-acc-ki67-seg)[data set]. The Cancer Imaging Archive 8. doi:10.7937/1FPG-V M46

    work page doi:10.7937/1fpg-v

  36. [36]

    Multimodality annotated hepatocellular carcinoma data set including pre-and post-tace with imaging segmentation

    Moawad, A.W., Morshid, A., Khalaf, A.M., Elmohr, M.M., Hazle, J.D., Fuentes, D., Badawy, M., Kaseb, A.O., Hassan, M., Mahvash, A., et al., 2023b. Multimodality annotated hepatocellular carcinoma data set including pre-and post-tace with imaging segmentation. Scientific data 10, 33. doi:10.7937/TCIA.5FNA-0924

    doi:10.7937/tcia.5fna-0924

  37. [37]

    Trusted: The paired 3d transabdominal ultrasound and ct human data for kidney segmentation and registration research

    Ndzimbong, W., Fourniol, C., Themyr, L., Thome, N., Keeza, Y., Sauer, B., Pi´ echaud, P.T., M´ ejean, A., Marescaux, J., George, D., et al., 2025. Trusted: The paired 3d transabdominal ultrasound and ct human data for kidney segmentation and registration research. Scientific Data 12, 615. doi:10.1038/s41597-025-04467-1

    doi:10.1038/s41597-025-04467-1 2025

  38. [38]

    Hvsmr-2.0: A 3d cardiovascular mr dataset for whole-heart segmentation in congenital heart disease

    Pace, D.F., Contreras, H.T., Romanowicz, J., Ghelani, S., Rahaman, I., Zhang, Y., Gao, P., Jubair, M.I., Yeh, T., Golland, P., et al., 2024. Hvsmr-2.0: A 3d cardiovascular mr dataset for whole-heart segmentation in congenital heart disease. Scientific Data 11, 721. doi:10.1038/s41597-024-03469-9

    doi:10.1038/s41597-024-03469-9 2024

  39. [39]

    Han-seg: The head and neck organ-at-risk ct and mr segmentation dataset

    Podobnik, G., Strojan, P., Peterlin, P., Ibragimov, B., Vrtovec, T., 2023. Han-seg: The head and neck organ-at-risk ct and mr segmentation dataset. Medical physics 50, 1917–1927. doi:10.1002/mp.16197

    doi:10.1002/mp.16197 2023

  40. [40]

    Quinton, F., Popoff, R., Presles, B., Leclerc, S., Meriaudeau, F., Nodari, G., Lopez, O., Pellegrinelli, J., Chevallier, O., Ginhac, D., et al., 2023. A tumor and liver automatic segmentation (atlas) dataset: A hepatocellular carcinoma dataset for automatic segmentation of liver and liver tumours on contrast-enhanced magnetic resonance imaging. Data 8, 79...

    work page doi:10.3390/data8050079 2023

  41. [41]

    Calibration and uncertainty for multirater volume assessment in multiorgan segmentation (curvas) challenge results

    Riera-Mar ´ ın, M., Sikha, O., Rodr ´ ıguez-Comas, J., May, M.S., Pan, Z., Zhou, X., Liang, X., Erick, F.X., Prenner, A., H´ emon, C., et al., 2025. Calibration and uncertainty for multirater volume assessment in multiorgan segmentation (curvas) challenge results. Computers in Biology and Medicine 197, 111024. doi:10 .1016/j.compbiomed.2025.111024

    arXiv 2025

  42. [42]

    A new 2.5 d representation for lymph node detection in ct

    Roth, H., Le, L., Ari, S., Cherry, K., Hoffman, J., Wang, S., Summers, R., 2018. A new 2.5 d representation for lymph node detection in ct. The Cancer Imaging Archive doi:10.7937/K9/TCIA.2015.AQIIDCNM

    work page doi:10.7937/k9/tcia.2015.aqiidcnm 2018

  43. [43]

    Preoperative ct and survival data for patients undergoing resection of colorectal liver metastases

    Simpson, A.L., Peoples, J., Creasy, J.M., Fichtinger, G., Gangai, N., Keshavamurthy, K.N., Lasso, A., Shia, J., D’Angelica, M.I., Do, R.K., 2024. Preoperative ct and survival data for patients undergoing resection of colorectal liver metastases. Scientific data 11, 172. doi:10.7937/QXK2-QG03

    work page doi:10.7937/qxk2-qg03 2024

  44. [44]

    Thyroid ultrasound cine-clip

    StanfordAIMI, 2026. Thyroid ultrasound cine-clip. URL:https://aimi.stanford.edu/, doi:10.71718/7m5 n-rh16

    work page doi:10.71718/7m5 2026

  45. [45]

    Aeropath: An airway segmentation benchmark dataset with challenging pathology and baseline method

    Støverud, K.H., Bouget, D., Pedersen, A., Leira, H.O., Amundsen, T., Langø, T., Hofstad, E.F., 2024. Aeropath: An airway segmentation benchmark dataset with challenging pathology and baseline method. Plos one 19, e0311416. doi:10.1371/journal.pone.0311416

    doi:10.1371/journal.pone.0311416 2024

  46. [46]

    Generation of digital phantoms of cell nuclei and simulation of image formation in 3d image cytometry

    Svoboda, D., Kozubek, M., Stejskal, S., 2009. Generation of digital phantoms of cell nuclei and simulation of image formation in 3d image cytometry. Cytometry Part A: The Journal of the International Society for Advancement of Cytometry 75, 494–509. doi:10.1002/cyto.a.20714

    work page doi:10.1002/cyto.a.20714 2009

  47. [47]

    Wahid, K.A., Dede, C., El-Habashy, D.M., Kamel, S., Rooney, M.K., Khamis, Y., Abdelaal, M.R., Ahmed, S., Corrigan, K.L., Chang, E., et al., 2024. Overview of the head and neck tumor segmentation for magnetic resonance guided applications (hnts-mrg) 2024 challenge, in: Challenge on Head and neck tumor segmentation for MRI-guided applications. Springer, pp....

    work page doi:10.1007/978-3-031-83274-1_1 2024

  48. [48]

    Sam-med3d: Towards general-purpose segmentation models for volumetric medical images

    Wang, H., Guo, S., Ye, J., Deng, Z., Cheng, J., Li, T., Chen, J., Su, Y., Huang, Z., Shen, Y., Fu, B., Zhang, S., He, J., Qiao, Y., 2024. Sam-med3d: Towards general-purpose segmentation models for volumetric medical images. doi:10.1007/978-3-031-91721-9_4,arXiv:2310.15161. 8

    Pith/arXiv arXiv 2024

  49. [49]

    Axonem dataset: 3d axon instance segmentation of brain cortical regions, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer

    Wei, D., Lee, K., Li, H., Lu, R., Bae, J.A., Liu, Z., Zhang, L., dos Santos, M., Lin, Z., Uram, T., et al., 2021. Axonem dataset: 3d axon instance segmentation of brain cortical regions, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 175–185. doi:10.1007/978-3 -030-87193-2_17

    doi:10.1007/978-3 2021

  50. [50]

    Wu, Z., Zhao, Q., Hu, M., Li, Y., Xue, H., Jiang, Z., Stefanidis, A., Wang, Q., Razzak, I., Ge, Z., et al.,

  51. [51]

    Resect: a clinical database of pre-operative mri and intra-operative ultrasound in low-grade glioma surgeries

    Xiao, Y., Fortin, M., Unsg ˚ ard, G., Rivaz, H., Reinertsen, I., 2016. Resect: a clinical database of pre-operative mri and intra-operative ultrasound in low-grade glioma surgeries. (No Title) doi:10.11582/2017.00004

    arXiv 2016

  52. [52]

    Deeplesion: automated mining of large-scale lesion annotations and universal lesion detection with deep learning

    Yan, K., Wang, X., Lu, L., Summers, R.M., 2018. Deeplesion: automated mining of large-scale lesion annotations and universal lesion detection with deep learning. Journal of medical imaging 5, 036501–036501. doi:10.1117/1.jmi.5.3.036501

    work page doi:10.1117/1.jmi.5.3.036501 2018

  53. [53]

    Yang, J., Gu, S., Wei, D., Pfister, H., Ni, B., 2021. Ribseg dataset and strong point cloud baselines for rib segmentation from ct scans, in: International conference on medical image computing and computer-assisted intervention, Springer. pp. 611–621. doi:10.1007/978-3-030-87193-2_58

    work page doi:10.1007/978-3-030-87193-2_58 2021

  54. [54]

    Data from lung ct segmentation challenge (lctsc)(version 3)[data set]

    Yang, J., Sharp, G., Veeraraghavan, H., Van Elmpt, W., Dekker, A., Lustberg, T., Gooding, M., 2017. Data from lung ct segmentation challenge (lctsc)(version 3)[data set]. The Cancer Imaging Archive doi:10.7937/ K9/TCIA.2017.3R3FVZ08

    2017

  55. [55]

    Benchmarking the cow with the topcow challenge: Topology- aware anatomical segmentation of the circle of willis for cta and mra

    Yang, K., Musio, F., Ma, Y., et al., 2025a. Benchmarking the cow with the topcow challenge: Topology- aware anatomical segmentation of the circle of willis for cta and mra. doi:10.48550/arXiv.2312.17670, arXiv:2312.17670

    Pith/arXiv arXiv doi:10.48550/arxiv.2312.17670

  56. [56]

    Vivim: a video vision mamba for ultrasound video segmentation

    Yang, Y., Xing, Z., Yu, L., Fu, H., Huang, C., Zhu, L., 2025b. Vivim: a video vision mamba for ultrasound video segmentation. IEEE Transactions on Circuits and Systems for Video Technology doi:10.1109/tcsvt. 2025.3563411

    arXiv 2025

  57. [57]

    Coffee-break lung ct collection with scan images reconstructed at multiple imaging parameters

    Zhao, B., Schwartz, L.H., Kris, M.G., Riely, G.J., 2015. Coffee-break lung ct collection with scan images reconstructed at multiple imaging parameters. (No Title) doi:10.7937/tcia.2020.jit9grk8

    work page doi:10.7937/tcia.2020.jit9grk8 2015

  58. [58]

    Nis3d: A completely annotated benchmark for dense 3d nuclei image segmentation

    Zheng, W., Peng, J.C., Hou, Z., Lyu, B., Wang, M., Mi, X., Qiao, S., Wan, Y., Yu, G., 2023. Nis3d: A completely annotated benchmark for dense 3d nuclei image segmentation. doi:10.52202/075280-0211

    work page doi:10.52202/075280-0211 2023

  59. [59]

    Automatic segmentation of mitochondria and endolysosomes in volumetric electron microscopy data

    ˇZerovnik Mekuˇ c, M., Bohak, C., Hudoklin, S., Kim, B.H., Romih, R., Kim, M.Y., Marolt, M., 2020. Automatic segmentation of mitochondria and endolysosomes in volumetric electron microscopy data. Computers in Biology and Medicine 119, 103693. doi:10.1016/j.compbiomed.2020.103693. 9

    arXiv 2020

  60. [2025]

    Mswal: 3d multi-class segmentation of whole abdominal lesions dataset, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer. pp. 378–388. doi:10.1007/97 8-3-032-04937-7_36

    doi:10.1007/97