List of Papers Browse by Subject Areas Author List
Abstract
Positive margins are common in head and neck squamous cell carcinoma, yet intraoperative re-resection is often imprecise because margin locations are typically communicated verbally from pathology. We present an all-in-one augmented reality (AR) system that relocalizes positive margins from a resected specimen to the resection bed and visualizes them in situ using HoloLens 2 depth sensing and fully automated markerless surface registration. In a silicone phantom study with six medical trainees, markerless registration achieved target registration errors comparable to a marker-based baseline (median 1.8 mm vs. 1.7 mm; maximum < 4 mm). In a margin relocalization task, AR guidance reduced error from verbal guidance (median 14.2 mm) to a few millimeters (median 3.2 mm), with all AR localizations within 5 mm error. These results support the feasibility of markerless AR margin guidance for more precise intraoperative re-resection.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/5292_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{YanYue_AllinOne_MICCAI2026,
author = { Yang, Yue AND Chabanas, Matthieu AND Reale, Carrie AND Benson, Annie AND Slagle, Jason AND Weinger, Matthew AND Topf, Michael AND Wu, Jie Ying},
title = { { All-in-One Augmented Reality Guided Head and Neck Tumor Resection } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16893},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The document describes a markerless augmented reality guidance system that assists in tumor resection by precise in-situ visualization. A phantom study involving six participants providing qualitative and quantitative data shows that the markerless registration is, above all, objectively on par with the baseline using markers.
- 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 entire paper is well written, and the novelty of markerless registration for in-situ AR visualization is clearly explained and demonstrated in a thorough evaluation. It is commendable that the evaluation does not only consider technical aspects but also subjective user feedback.
- 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.
After reading the document, I wonder how the results presented are to be contextualized in relation to similar work, including markerless approaches. In my opinion, the discussion does not go far enough in this regard. In particular, a connection should be established with the following two aspects, respectively references: Regarding deformable registration (see section on future work): Yang, Q. , Li, F. , Xu, J. , Liu, Z. , Sridhar, S. , Jin, W. , . . . & Wu, J. Y. (2025, September). Augmented Reality-Based Guidance with Deformable Registration in Head and Neck Tumor Resection. In International Conference on Medical Image Computing and Computer-Assisted Intervention (pp. 45-54). Cham: Springer Nature Switzerland. Regarding markerless situs registration: Liebmann, F. , von Atzigen, M. , Stütz, D. , Wolf, J. , Zingg, L. , Suter, D. , . . . & Fürnstahl, P. (2024). Automatic registration with continuous pose updates for marker-less surgical navigation in spine surgery. Medical Image Analysis, 91, 103027. In addition, I spotted redundant information which should be shortened (e. g. , quartile range in the text and in plots), and the statistical reports should be updated according to APA standards (i. e. , including all measures such as effect size). Non-significant aspects should be given less weight, especially given the relatively small sample size. In Figure 3, a significance bar could help to visualize differences, and in Figure 4, the similar color coding for different aspects creates potential for misunderstanding. I recommend that all metrics used should be listed before the Results section; otherwise, the questionnaires used simply appear surprisingly and confuse the reader. Also, the single qualitative quote in the Discussion and Conclusion sections is somewhat surprising; I assume it would be better to have a separate paragraph addressing this feedback within the Results section.
- Please rate the clarity and organization of this paper
Satisfactory
- 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?
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?
Given that it is clear, technically sound, and timely, I consider this submission highly suitable for the MICCAI conference. However, its somewhat unclear positioning relative to existing work and the need for minor optimization somewhat detract from the otherwise positive impression it makes. I hope my comments will help the authors address these points in a rebuttal, and I therefore recommend a “weak accept.”
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
Having the read the authors’ rebuttal and the other reviews, my main points of criticism have been addressed. But I also recognize that R2’s and R3’s concerns remain significant and substantial (particularly regarding novelty and methodology). Since, to my understanding, the authors have addressed the ambiguities, I recommend, by a narrow margin, to accept the paper. However, I would also be okay with rejecting the paper if R2 and R3, who are more experts in this field, disagree.
Review #2
- Please describe the contribution of the paper
This paper proposes a HoloLens 2 depth-sensor-based registration method to align specimen scans for in-situ visualization of positive margins, with the goal of guiding tumor resection in head and neck squamous cell carcinoma. The method was evaluated on a cranial phantom with six users. Compared with verbal guidance, the proposed method improved re-localization accuracy, but only showed limited benefit in system usability and workload reduction.
- 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 paper focuses on specimen re-localization, which is a clinically relevant application scenario that has rarely been addressed in previous AR navigation studies.
- 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.
1.Methodological concerns a. Practical usability: specimen deformation and shrinkage Specimen deformation and shrinkage are major issues in the operating room. Although the authors mention these as future work, they are in fact central challenges for any real clinical deployment. Treating the specimen registration problem as rigid severely limits the practical usability of the proposed method. b. Practical usability in iterative resection workflows Re-localization is not a one-time procedure. In real surgical workflows, resection may be iterative. The proposed registration-based strategy appears applicable only to the initially resected specimen and does not seem able to support repeated or iterative re-localization during subsequent resections. c. Depth sensing accuracy of the AHAT camera Previous studies have shown that the depth error of the HoloLens 2 AHAT sensor is substantial and varies across different surface types [Ref 1]. This issue is not considered in the current study, although it may significantly affect both registration accuracy and visualization reliability. d. Dependence on HoloLens SLAM stability The proposed multi-frame stacking approach relies on HoloLens 2 pose estimation, which means that the overall system depends heavily on the stability of the device SLAM for both registration and in-situ visualization. In realistic scenarios, head movement may introduce spatial drift, which could substantially reduce usability and reliability. e. Limited methodological novelty Using a depth camera for point cloud registration has already been widely adopted in previous work [Ref 1]. The pipeline of FPFH-based coarse registration followed by ROI selection and ICP refinement is also a common approach in point-cloud-based registration pipelines [Ref 2]. Registration method difference between this paper and others are not clearly presented. Therefore, the technical novelty of the method itself appears limited, aside from the specific clinical use case. 2.Experimental concerns a. Lack of clarity in the experimental setup The paper does not clearly explain how marker-based registration was performed. As a result, it is difficult to understand the actual difference between the marker-based and markerless registration settings. b. Lack of objective ground truth for registration accuracy Since the authors used a cranial phantom with retro-reflective markers, it should be feasible to establish objective ground truth using CT scanning or an optical scanner. However, no direct objective evaluation of registration accuracy is provided. c. Inappropriate use of expert performance as ground truth The mean performance of experts was used as the ground truth in the evaluation. This is not appropriate for accuracy assessment, because it conflates user error with system error. Ground truth in an accuracy study should be defined independently and objectively. d. Insufficient comparison baseline In Experiment 2, the markerless AR method is compared only with verbal guidance. However, in actual practice, surgeons may also rely on specimen photographs or non-in-situ 3D holographic visualization. These would be more meaningful and clinically relevant baselines for comparison. 3.Writing a. Writing The paper is generally clearly written. However, the comparison with previous studies is insufficient. The manuscript does not clearly articulate its innovation beyond introducing a relatively uncommon use case. b. Figures The figures do not sufficiently illustrate the proposed method. For example, the paper does not clearly show what the AHAT point cloud looks like, how the ROI is extracted, or how the registration pipeline operates in practice. In contrast, there are too many figures dedicated to NASA-TLX and SUS results. Given that only six users were included and limited distribution information is shown, these results could be more effectively summarized in tables rather than multiple separate figures. ______________ Suggestions for improvement 1.The ground truth should be defined objectively. Expert performance should not be used as the ground truth in an accuracy evaluation. 2.The proposed method should be compared against stronger and more clinically relevant baselines or current standard workflows, rather than only against a very naive verbal-guidance condition. 3.Since the automatic registration procedure occupies a large portion of the method section, if the authors consider automation to be a core contribution, they should: a. clearly explain why automatic registration is particularly important in this application; b. clarify what limitations or ambiguities remain in prior automated approaches; c. provide a clearer demonstration of how the proposed automated method works in practice; d. objectively evaluate the automation result against a non-automated alternative, including both accuracy and efficiency; e. analyze how much time is saved and how much accuracy is preserved over time, particularly under spatial drift or repeated use. ______________ Reference 1.Li, H., Yan, W., Liu, D., Qian, L., Yang, Y., Liu, Y., … & Wang, G. (2024). EVD surgical guidance with retro-reflective tool tracking and spatial reconstruction using a head-mounted augmented reality device. IEEE Transactions on Visualization and Computer Graphics, 31(9), 6497-6511.2.Xiao, Z. (2024, November). ICP Registration Based on Overlapping Region. In 2024 International Conference on Image Processing, Computer Vision and Machine Learning (ICICML) (pp. 980-984). IEEE.
- Please rate the clarity and organization of this paper
Satisfactory
- 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 provide sufficient information for 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?
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.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This paper focused on specimen in-situ visualization for positive margins, which is rarely mentioned in previous works. However, they over simplify the problem on phantom and limited innovation is made on the algorithms for better usability in this new use case. The experiment performed failed to get an objective standard and use user result as GT, which is a severe mistake. As there would be limited time for them to optimize the method and experiment during rebuttal, I suggest to reject this paper.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Reject
- [Post rebuttal] Please justify your final decision from above.
I insist that this manuscript should be rejected for following reasons: 1.Iterative operation and deformation are claimed to be supported but yet not clear how. These parts are the actually difficulties towards an all-in-one solution. The author did not prove why the engineering contribution of “all-in-one” construction worth study if the downstream methodologies are unclear or even be not addressable. 2.Though author claimed that the verbal based method is proposed by an experienced surgeon, it can not prove it is a correct baseline for fair comparison.
After all, I agree that there should be quite some hard work behind this paper. If the author spend more time in polishing the contribution and a) Address the case specific problem b) Apply to different cases to prove it’s broadly applicable, it will be suitable for some journals. Considering that author can not add experiment during review / rebuttal, and the limited space of the paper, I do not recommend accept under current status.
Review #3
- Please describe the contribution of the paper
This paper presents an all-in-one augmented reality system designed to assist in head and neck squamous cell carcinoma resections. The system uses a Microsoft HoloLens 2 and its depth sensor to perform fully automated, markerless surface registration. By leveraging curvature-based keypoint sampling, FPFH descriptors, and a TEASER++ based coarse-to-fine registration pipeline, the system eliminates the need for manual region-of-interest selection or fiducial markers.
- 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 proposed technical pipeline successfully removes the workflow burden of manual region-of-interest selection and physical markers. The combination of curvature-biased sampling, TEASER++ for robust global alignment, automatic bounding box construction for cropping, and point-to-plane ICP yields an impressive median TRE of 1.8 mm. 2.The study provides a well-rounded assessment by incorporating human factors. The inclusion of a margin relocation task with six participants, alongside validated workload and usability metrics, strengthens the paper’s claims regarding the system’s practical viability.
- 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.
1.The current evaluation is conducted entirely on a rigid silicone face phantom. In actual surgical scenarios, tissue deformation, bleeding, and shrinkage of the resected specimen will heavily impact the rigid surface registration assumptions. 2.The user study relies on a very small sample size of six participants. The absence of experienced attending surgeons limits the generalizability of the baseline “verbal guidance” error.
- Please rate the clarity and organization of this paper
Satisfactory
- 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?
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 all-in-one AR navigation system is clinically valuable, especially for its automated registration that removes manual steps. But, the study’s conclusions are less compelling due to the small sample size and the use of a single rigid phantom.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
I would recommend acceptance, because the rebuttal clarifies the objective landmark-based evaluation, markerless registration baseline, AHAT correction procedure, and iterative workflow applicability, while the proposed all-in-one HoloLens-only AR system remains clinically relevant and demonstrates promising preliminary accuracy despite limitations in phantom-based validation and sample size.
Author Feedback
We thank the reviewers.
Novelty We agree prior deformable and markerless navigation work needs clearer contrast (R1, R2-1e, R2-3a). Our contribution is an all-in-one AR margin relocalization system using only the HoloLens POV AHAT sensor, without external RGB-D cameras, fiducials, or manual ROI selection (R1, R2-1e). This lowers OR burden by avoiding sensor installation (R1). Yang et al. model deformation but use ArUco and manual landmarks for AR registration [3], and Liebmann et al. study spine navigation with external sensing [4]. R2 Ref.1 concerns AHAT error correction with markers, not markerless specimen-to-bed registration [5]. Our Sec.2 uses curvature-biased sampling, TEASER++-style global alignment, automatic coarse-pose ROI cropping, and ICP for small HN specimens (R2-1e, R3). These choices gave markerless TRE comparable to a marker baseline, median 1.8 mm vs 1.7 mm and max <4 mm (R1, R3). Although we demonstrated our method on HN, our method is broadly applicable to open surgery solid malignancies.
Clinical scope We agree deformation and shrinkage are crucial (R2-1a, R3-1). Clinical deployment needs deformation modeling plus markerless AR relocalization of the deformed model (R2-1a). This paper evaluates the second part, still required after deformation modeling (R2-1a). Iterative resections are supported because after any later positive-margin report, the same scan-register-visualize workflow can be rerun (R2-1b). The goal is to reduce repeated re-resections caused by coarse path-to-surgeon communication. [1,2] (R2-1b, R2-2d).
Evaluation, ground truth, and baselines We apologize for the wording “expert measurements provided ground truth” (R2-2b, R2-2c). Targets were small physical ink landmarks on the phantom that were also identified on the structured-light specimen model (R2-2b). They were hidden among black marks for blinding participants, while the expert who created them could easily identify them with AR (R2-2c). Expert touching was only a measurement step for known physical landmarks, repeated twice and averaged (for tracking noise), so expert performance was not used as clinical ground truth (R2-2c). The marker baseline used the same STTAR retro-reflective tracking used for the stylus [6] (R2-2a). The study therefore reports registration accuracy versus a fiducial baseline and preliminary margin relocalization accuracy (R2-2d). We agree the phantom task does not capture all intraoperative information, but the verbal script was designed with experienced HN surgeon, and studies show surgeons often rely on verbal pathology communication [1,2] (R2-2d, R3-2).
AHAT, SLAM, reporting We agree AHAT error is surface-dependent [5] (R2-1c). In our setup, an AHAT-tracked tool sampled the true local surface contour, nearest depth-sensor points were paired to tool points, and least-squares SVD corrected the local point cloud (R2-1c). This took <30 s and was applied once before experiments (R2-1c). For SLAM, acquisition took ~3s while users viewed the resection bed via AR, followed by noise reduction processing (R2-1d). The ground truth points were also collected right after each experiment to minimize drift. We will streamline statistics and figures, add effect sizes and APA reporting, deemphasize non-significant NASA-TLX and SUS results, clarify experiment setup, and release acquisition, depth correction, and ROI code (R1, R2-3b, R2 reproducibility).
[1] Quality in Surgical Pathology Communication and Reporting 2011 [2] How Often Is Cancer Present in Oral Cavity Re-resections After Initial Positive Margins? 2024 [3] Augmented Reality-Based Guidance with Deformable Registration in Head and Neck Tumor Resection 2025 [4] Automatic Registration with Continuous Pose Updates for Marker-less Surgical Navigation in Spine Surgery 2024 [5] EVD Surgical Guidance with Retro-reflective Tool Tracking and Spatial Reconstruction Using a Head-Mounted AR Device 2024 [6] STTAR Surgical Tool Tracking Using Off-the-Shelf AR HMDs 2023
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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.
This submission received mixed and polarizing reviews with two weak-accepts (R1, R3) and one reject (R2). R2 raises various valid concerns, and I recommend that the authors are invited to provide a rebuttal, not only to R2 but also to the concerns raised by R1 and R3.I suggest the authors focus on convincing responses to the critical items:
-
R2 section 7 (1a, 1b, 1c, 1d, 1e, 2b, 2c, 2d, 3a (insufficient SOTA comparison/positioning). The suggestions from the reviewer to improve the paper are valid, however, they constitute major revisions that are not possible post-rebuttal. Consequently, the authors should rebut the critisims based on how the manuscript stands in its current form.
- R3 section 7, item 1
- R1 section 7, novelty issues/positioning regarding 2 refs, and lack of statistical significance reporting.
-
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
This submission proposes a markerless AR system for tumor resection guidance. It is based on HoloLens 2 depth sensing, combined with a coarse-to-fine registration pipeline (curvature sampling, FPFH/TEASER++ initialization, ICP refinement). Evaluated was performed on a rigid phantom with a six-participant user study.
All reviewers recognize the clinical relevance and the value of eliminating markers and manual ROI selection. R1 and R3 additionally highlight manuscript clarity, automated workflow, user validation, and quantitative improvements over baselines./ R2 also acknowledges good motivation and potential clinical value.
The rebuttal clarified key methodological details: landmark-based evaluation, AHAT depth handling, markerless registration baselines, and workflow applicability.
However, reviewer concerns about methodological novelty, evaluation rigor, and clinical realism remain post-rebuttal. R2 argues that the method largely follows established depth-based registration pipelines with limited innovation. R2 also raises major issues about rigid phantom evaluation and weak baseline comparisons. R1 and R3 similarly note the small sample size and phantom-only validation l;imitations.
In summary, the rebuttal resolved several key issues, and the overall reviewer sentiment shifted positively. I recommend acceptance based on the clinical importance of the problem, the successful integration of a fully markerless AR workflow that removes manual registration steps, supporting evidence from the user validation. Of course, the work remains an early-stage CAI prototype and will require substantially broader validation under realistic clinical conditions, nevertheless, it represents a meaningful contribution.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
Following the rebuttal, two reviewers noted that the authors clarified the objective landmark-based evaluation, the markerless registration baseline, AHAT correction procedure, and the intended iterative workflow. However, one reviewer is still concerned regarding deformation and shrinkage, iterative clinical applicability, the appropriateness of the verbal-guidance baseline, and the limited phantom-based evaluation. Overall, while the current study remains preliminary and should more clearly position its contribution relative to prior AR registration work, the rebuttal addressed several major ambiguities and the proposed HoloLens-only workflow demonstrates promising accuracy and translational relevance. I therefore recommend acceptance.
