List of Papers Browse by Subject Areas Author List
Abstract
Automated video-based surgical skill assessment enables objective feedback for training, especially in resource-limited settings. However, most existing methods rely on generic spatiotemporal modeling (e.g., absolute positional encoding or simple temporal aggregation) and overlook large execution rhythm variability across surgeons. For instance, experts tend to operate smoothly and efficiently, whereas novices exhibit slower and irregular patterns, which ultimately limits skill discrimination. To address this, we propose R-MSA (Rhythm–Morphology Synchronous Assessment), a rhythm-aware framework that explicitly models execution rhythm differences. Specifically, a Rhythm Morphology Encoder performs duration-conditioned rhythm encoding to generate rhythm-adaptive representations for variable-length videos. A Morphology-Synchronous Decoder with learnable queries aligns latent morphology primitives (e.g., needle driving, knot tying) across rhythm-normalized sequences. To refine procedural correspondence, a Surgical Sequence Alignment Module integrates a Gaussian temporal prior into co-attention for localized temporal alignment. Finally, an Uncertainty-Calibrated Interpretable Tree produces relative skill scores with calibrated uncertainty through reparameterized prediction. Experiments on JIGSAWS, HeiChole, and our proposed Kangduo datasets show that R-MSA consistently outperforms state-of-the-art methods in ranking correlation and error metrics. Code will be released upon acceptance.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2933_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/DeDe5333/R-MSA
Link to the Dataset(s)
Kangduo dataset: https://drive.google.com/drive/folders/1YI_gTlgrRx93FQdwAmcVZg93xe2DEo8Z?usp=sharing
BibTex
@InProceedings{KeXia_RMSA_MICCAI2026,
author = { Ke, Xiao AND Zheng, Yining AND Xu, Huangbiao},
title = { { R-MSA: Rhythm-Conditioned Surgical Video Skill Assessment via Morphology Synchronization and Uncertainty Calibration } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16892},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes R-MSA, a video-based surgical skill assessment framework built around the idea that execution rhythm is an important but under-modeled source of variation in surgical performance. The method combines four parts: pairwise sequence alignment with a Gaussian temporal prior, duration-conditioned rhythm encoding to normalize videos with different execution tempos, a learnable-query decoder intended to extract rhythm-invariant morphology primitives, and a soft decision tree that outputs relative skill scores with uncertainty estimates. Besides, the paper introduces a new Kangduo dataset, consisting of 104 HD robotic suturing videos with expert scores.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.The central observation that skill is reflected not only in what actions occur but also in how rhythmically and regularly they unfold is well motivated. The expert/novice contrast shown on page 2 is intuitive and clinically plausible: experts tend to be smoother and more regular, whereas novices are slower and more erratic. This is a sensible perspective for surgical skill assessment and a useful framing contribution.
2.The overall pipeline on page 4 is coherent: first align videos, then inject rhythm conditioning, then extract synchronized morphology primitives, then predict with uncertainty. Even if some components are incremental individually, the paper presents a clear systems-level design rather than a loose collection of tricks.
3.The reported performance is good. On JIGSAWS, the method achieves the best average SROCC across LOSO, LOUO, and 4-fold protocols, and competitive-to-best MAE. On HeiChole and Kangduo, it also reports the strongest SRCC and lowest MAE among compared methods.
4.A new dataset with 104 HD robotic suturing videos and larger duration variability is potentially important, especially since the paper argues that older benchmarks such as JIGSAWS do not sufficiently stress rhythm-aware modeling. I recommend releasing this dataset publicly if possible, as it could be a useful resource for the community.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- Some claimed novelty is overstated, and several modules are adaptations of familiar ideas. While the overall combination is interesting, the paper does not clearly establish a strong leap over recent surgical skill assessment baselines such as ViSA, MT-ViT, Contra-Sformer, T2CR, and RICA² cited in the paper. In particular, the move from generic video temporal modeling to rhythm-aware conditioning is the main conceptual novelty, but the implementation feels somewhat heuristic.
2.The paper claims rhythm is distinct from absolute duration or speed, but the actual rhythm encoding appears to be based on binning videos into 4 groups according to execution duration, then scaling sinusoidal frequencies by the bin index. That makes the practical definition of rhythm quite close to coarse duration conditioning, not a richer measurement of rhythm regularity, periodicity, pauses, or local tempo variation. This mismatch between concept and implementation weakens the core claim.
3.The method regresses relative skill scores from video pairs, using target-exemplar representations. That can work, but the paper does not explain enough about how exemplars are selected, how inference behaves under different exemplar choices, whether performance is sensitive to the reference pool, and how this compares to direct absolute score regression.
4.Kangduo dataset may be a strong contribution, but the paper gives only limited details on the scoring rubric, inter-rater reliability, participant expertise distribution, acquisition conditions, train/test subject separation, and whether the dataset will actually be released.
5.The paper says it uses an I3D backbone and compares with prior methods, but for some baselines it is not completely clear whether results were reproduced under the same preprocessing, splits, and supervision conditions (or instead copied from prior publications). Also, Table 1 mixes methods using video only, video + kinematics, and sometimes extra annotations, which makes the comparison not perfectly clean.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This is an interesting work overall. The paper addresses an important problem, proposes a coherent rhythm-aware framework, and reports very good results across three datasets. The ablations are meaningful, the qualitative visualizations support the technical claims, and the new Kangduo dataset could be a practical contribution for the community.
However, my concerns are mostly about overclaiming. The paper’s core concept of rhythm is interesting, but the actual implementation looks closer to duration-conditioned encoding than a deep model of rhythm. Likewise, the uncertainty calibration and interpretability claims are not fully validated. Several components are also adapted from familiar ideas.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
This paper proposes R-MSA, a rhythm-aware framework for pairwise surgical video skill assessment that aligns two videos, conditions features on execution rhythm/duration, extracts morphology primitives, and predicts a relative skill score with an uncertainty-aware tree. The authors report improved results over prior methods on JIGSAWS, HeiChole, and Kangduo (their newly proposed dataset).
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- Strong methodological novelty
- Strong experimental design and results
- Good presentation
- Contribution of a new dataset (will it be released?)
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
- The notation needs to be clarified. In Section 2.1, Vk and Vl appear to denote the two input videos, but it is not stated explicitly whether these correspond to the target and exemplar/reference videos. Likewise, N is introduced in the feature tensor shapes but is never defined.
- There seems to be a conceptual disagreement between Sections 2.1 and 2.2.Authors write, “Surgical videos of the same task exhibit large duration variance due to differing execution rhythms”. If that is the case, wouldn’t it contradict what you said about “comparable procedural phases tend to appear in temporally neighboring segments after uniform sampling”?
- The loss formulation is underspecified and difficult to follow. Please define L_ranking, L_uncertainty, L_DAE, and L_var explicitly, including all terms and supervision signals. It is also unclear whether mu denotes the relative skill score and whether sigma is directly optimized.
- The inference procedure is currently too vague. The paper says that mu is used as the skill estimate at inference, but the full test-time setup is missing: is prediction performed against a single exemplar/reference video or multiple exemplars, and how is a final absolute or per-video score obtained from pairwise predictions?
- The uncertainty interpretation is unclear. As defined, sigma appears to quantify uncertainty of the pairwise relative prediction, rather than uncertainty intrinsic to the target video itself. This seems closer to data uncertainty for the pair, so the manuscript should explain this distinction more carefully and justify the interpretability claims.
- Also, the practical use of uncertainty for interpretability is not sufficiently demonstrated. It is unclear how these uncertainty estimates would actually be used by clinicians or trainees in practice.
- For consistency, it would be preferable to report multiple-seed results on JIGSAWS as well, not only on HeiChole and Kangduo. The current experimental section states that multiple random seeds are used for the latter two datasets, but not for JIGSAWS.
- The paper would benefit from stating some limitations.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
Nice work and I appreciate that some omissions are because of space limitations. Please try to address my comments, I believe they will make the paper stronger - in particular, the comments on the loss functions, inference, and uncertainty.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
See major weaknesses
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
The manuscript addresses an important and timely problem in computer-assisted surgical education: automated assessment of surgical skill in robotic surgery. The proposed framework is technically well motivated. The paper is clearly written, although the methodological pipeline is a bit complex to follow easily. The empirical results are strong. Overall, I found this to be a solid and well-executed contribution.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The manuscript has several major strengths. First, the proposed framework appears technically well designed and well motivated. Second, the empirical results are strong, with high performance metrics that suggest the proposed approach is effective for the task.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
They mention a database they created for the task but it is not sufficiently discussed so its reproducibility is reduced.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
If the database they created poses an ethical breach is not commented whatsoever on the manuscript.
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
1.Put the meaning of LOSO and LOUO. 2.The explanation on how the dataset was created is minimal. I suggest expanding this section to better understand the differences between datasets.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
If the authors include more information on the dataset, the manuscript is indeed useful.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
Dear Area Chairs and Reviewers, We sincerely thank you for the positive feedback, constructive comments, and Early Accept recommendation. Your insights will significantly improve our camera-ready version. We address the main concerns below: 1.Dataset Ethics and Availability (R1, R2, R3) We apologize for the missing dataset details. The Kangduo suturing tasks were performed on surgical phantoms (simulated tissue), not human patients, thus involving no human subjects or privacy/ethical concerns. Regarding availability, while raw videos are restricted by our collaborating institution’s IP policies, we are committed to reproducibility. Upon publication, we will publicly release the extracted I3D visual features, expert annotations, and the complete R-MSA source code to allow full reproduction of our results. 2.Baselines and Input Modalities (R1, R2) We appreciate the reviewers noting the mixed input modalities (video-only vs. video+kinematics) in Table 1.Including baselines with extra kinematic data actually highlights R-MSA’s competitiveness. Achieving superior performance using only video data demonstrates that effectively modeling execution rhythm bridges (and even surpasses) the performance gap with models relying on privileged multi-modal sensors. We will explicitly clarify these modalities in the revised table. 3.Concept of “Rhythm” & Inference Procedure (R1, R2) We agree with R1 that our duration-binning is a heuristic proxy for surgical rhythm. We will refine our novelty claims and explicitly discuss this in the Limitations, noting future work will model continuous, fine-grained spatiotemporal dependencies to capture local rhythm variations beyond discrete duration groups. Regarding inference (R1, R2): Following established pairwise assessment protocols (e.g., CoRe, T2CR), a target video is compared against a fixed pool of expert exemplars from the training set. The absolute score is obtained by adding the predicted relative skill differences (mu) to the exemplars’ ground-truth scores. This pipeline will be detailed in the revision. 4.Notation, Loss Formulation, & Uncertainty (R2)We thank R2 for the meticulous check. We will explicitly define variables (Vk, Vl, N) and expand the loss formulations. We completely agree with R2 regarding uncertainty: our learned sigma primarily quantifies the pairwise predictive (data) uncertainty between target and exemplar, rather than the intrinsic absolute cognitive uncertainty of the target video alone. We will refine our interpretability claims to accurately reflect this critical distinction. 5.Additional Clarifications and Corrections All other minor issues will be carefully addressed in the camera-ready version. Specifically, we will clarify the temporal sampling strategy to resolve the apparent discrepancy between Sec. 2.1 and 2.2 (R2), define acronyms like LOSO/LOUO (R3), report multiple-seed variance results on JIGSAWS (R2), and expand our comprehensive discussion of limitations. We are fully committed to polishing the manuscript to meet MICCAI’s highest standards. Thank you again for your time and the recognition of our work.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
The paper has received 3 positive reviews with all reviewers suggesting acceptance. I broadly agree with their assessment but I find the weaknesses listed by Rev #1 and #2 very accurate and these undoubtledly diminish the contribution of the work in the its relevant area (surgical skill asssessment).
My recommendation is for the paper to be accepted but the authors are strongly advised to address the comments by Rev #1 and #2.