List of Papers Browse by Subject Areas Author List
Abstract
Consumer-grade RGB-D imaging for intraoperative orthopedic tissue tracking is a promising method with high translational potential. Unlike bone-mounted tracking devices, markerless tracking can reduce operating time and complexity. However, its use has been limited to cadaveric studies. This paper introduces the first real-world clinical RGB-D dataset for spine surgery and develops SpineAlign, a system for capturing deformation between preoperative and intraoperative spine states. We also present an intraoperative segmentation network trained on this data and introduce CorrespondNet, a multi-task framework for predicting key regions for registration in both intraoperative and preoperative scenes.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1396_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2025/supp/1396_supp.zip
Link to the Code Repository
https://github.com/condog101/SpineAlign
Link to the Dataset(s)
RGB-D spine surgical dataset: https://huggingface.co/datasets/zcbecda/SpineAlign/tree/main
BibTex
@InProceedings{DalCon_Towards_MICCAI2025,
author = { Daly, Connor and Marconi, Elettra and Riva, Marco and Ekanayake, Jinendra and Elson, Daniel S. and Rodriguez y Baena, Ferdinando},
title = { { Towards Markerless Intraoperative Tracking of Deformable Spine Tissue } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15968},
month = {September},
page = {630 -- 640}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper introduces a framework for markerless registration of spinal anatomy using intraoperative RGB-D imaging and preoperative CT/MRI segmentation. It presents a clinical RGB-D dataset collected during surgery, showing exposed lumbar spine regions with annotated anatomical landmarks. The authors develop a pipeline that registers preoperative segmented data with intraoperative RGB-D point clouds using both rigid and non-rigid alignment methods. The system includes a process to isolate key anatomical regions, such as vertebral segments, to improve registration performance. The dataset also includes labeled examples of different spinal poses, which can be used to study how pose changes affect registration. These components are intended to support the development and evaluation of markerless tracking systems for surgical applications.
- 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 paper introduces a clinical-grade RGB-D dataset acquired intraoperatively, which includes real surgical exposures of the lumbar spine. This is a valuable contribution, as datasets combining RGB-D imaging with anatomical ground truth in surgical settings are rare.
2) The paper includes annotations for different spinal poses in the dataset, allowing the authors to evaluate how registration performance is affected by changes in anatomical configuration.
3) Although the method has not yet been used in live patients, the data is collected in a clinical setting with real anatomical exposure, increasing the practical relevance of the work.
- 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 paper does not clearly explain how its registration strategy is novel in terms of algorithmic formulation or optimization.
-
The dataset is limited to the lumbar spine which may not generalize to other spinal levels or surgical views (e.g., thoracic or sacral regions, lateral access).
-
The paper does not provide a baseline comparison to traditional marker-based navigation systems, which are currently used in clinical spine surgery. Including such a baseline (e.g., optical tracking accuracy or fiducial-based registration error) would help quantify the tradeoffs between markerless and conventional systems and highlight areas where the proposed approach offers advantages or disadvantages.
-
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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/2025/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.
(3) Weak Reject — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I recommend a weak reject for this paper. The work addresses an important problem in markerless intraoperative registration and contributes a clinically relevant RGB-D dataset along with a pipeline for aligning preoperative segmented anatomy with intraoperative point clouds. However, the registration approach follows established methods without clear novelty, and the evaluation lacks sufficient quantitative analysis or comparison to baseline techniques. Additionally, while the dataset is collected in a clinical setting, the system has not been tested in live surgical workflows, limiting its demonstration of clinical feasibility. The paper shows promise, but would benefit from stronger validation or a clearer technical contribution.
- Reviewer confidence
Somewhat confident (2)
- [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
The authors proposed a deep learning based approach to segment key regions of interest in the spine anatomy from both preoperative and intraoperative data. The method can be used for registration to account for spine motion due to change between preoperative and intraoperative poses.
- 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 motivation is clear - to enable markerless tracking during open spine surgery. The proposed semi-automatic labeling approach is novel, as it utilizes a kinematic model to deform the spine before rigid alignment, and leverages existing CT/MR spine segmentation algorithms to create labels, and allows for fine adjustments in case of failed registration/labeling. The segmentation architecture is also novel, as it utilizes both preoperative and intraoperative data and leverages the pointNet++ network. The dataset used for training and testing are clearly stated in the paper and an ablation study was included.
- 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.
-
First, the baseline needs further clarification for both preoperative and intraoperative segmentation. The authors state for intraoperative segmentation, the baseline is a random neighborhood; for preoperative segmentation, the baseline is a sampling of the preoperative mesh. The current baseline model underperforms to the extent that it failed to provide a meaningful reference point for evaluating the proposed approaches (Table 2). The improvements cannot be properly interpreted.
-
The paper lacks discussion and interpretation of the results for clinical feasibility. For example, what is an acceptable threshold for the metric reported in Table 2. Does the segmentation provide reliable data for further registration?
-
To improve clarity and interpretability of the reported metrics, a figure that visually illustrates each metric in the results (IoU, Chamfer, Hausdorff) using a representative test frame/case should be included.
-
Please clearly define terms (e.g., scene, mesh, point cloud, target points, region of interest, etc) and use consistent terms throughout the paper. It can be difficult to appreciate the nuances between the similar terms especially for readers unfamiliar with the specific context.
-
- 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.
- 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/2025/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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Markerless tracking removes the need for physical trackers and setup in the surgical field and allow flexibility of the procedure. The authors proposed a novel approach as a step towards the registration needed to accomplish markerless tracking, as the correspondNet segments regions of interest from both preoperative and intraoperative data for further registration. However, it is unclear if the resulted predications are reliable for registration due to the absence of a gold standard.
- 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
In this work, the authors present 4 major contributions: (1) a clinical RGB-D dataset for spine surgery including 27 subjects; (2) SpineAlign, a method to capture deformation between preoperative imaging and intraoperative surgical presentation of patients; (3) an intraoperative segmentation network adapted from PointNet++; (4) CorrespondNet, a framework to predict key regions for registration for both preop and intraop scenes.
- 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) Markerless registration (additionally with the use of RGB-D device, contact-less registration) for spine surgery is an important and interesting direction that can offer both cost and workflow benefits. (2) The authors noted that this work provides the first real-world clinical RGB-D dataset for orthopedic markerless tissue registration: such dataset represents a notable contribution to the field. (3) The proposed system combining deformable registration, segmentation via a flavor of PointNet++ and CorrespondNet to isolate key regions, is interesting especially comparing to commercial solutions—a notable comparator not mentioned in the paper but just to point out to authors is 7D by SeaSpine.
- 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) One major concern is that the results especially RMSE comparing alignment via rigid vs deformation are unconvincing in Table 1, which casts some questions over the value of SpineAlign. Moreover, there lacks an understanding if the reported RMSE e.g. is clinically acceptable: i.e. comparing ~7.14-7.16 mm RMSE for clinical data in Table 1 to the dimensions of lumbar vertebraes. (2) On data: (2a) an explanation is needed on the exclusion of thoracic and cervical cases; (2b) what’s the breakdown of CT and MR data for the 24 lumbar cases, especially imaging spacing, which can impact segmentation quality. (3) Why correspondence distance changed from 8 mm for phantom to 10 mm for clinical data?
- 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.
- 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/2025/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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This work offers 2 unique contributions: (1) a combination of methods leading to a markerless registration using workflow-friendly RGB-D device; (2) a unique dataset beneficial for the community.
The major concern is if this work needs further refinement for MICCAI publication: the RMSE result (a) is unconvincing for both potential clinical application; (b) casts questions over the contribution of deformation model proposed.
- 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
We thank the reviewers for their thoughtful feedback. We appreciate their recognition of our technical innovations (R1, R3) and the contribution we make to a relatively under-explored field (R2, R3). We address some of their specific comments below.
Regarding the lack of discussion on clinical feasibility (R1, R2): We will expand upon this in the additional space granted to us in the camera ready paper. We concur that the absence of a gold-standard is somewhat limiting for comparison against existing technologies, but capturing a true intraoperative gold standard requires further imaging and fiducial placement, adding radiation, time, and workflow disruption for the surgical team. This results in a disproportionate “ethical compliance” burden on the team and we therefore initially opted for a more convenient solution. On the basis of these promising results we have now secured an ethically-approved follow-up study beginning in July, which will integrate intraoperative X-ray with optical fiducial markers for definitive validation.
On the efficacy of the baseline model (R1): The Random Neighborhood baseline is intentionally simple and yields relatively low absolute scores. It provides a crucial sanity check: by significantly outperforming a random correspondence model in Dice and RMSE, CorrespondNet demonstrates that it learns true anatomical correspondences rather than producing results by chance. Other baseline methods to consider in future work may include pretrained networks such as ImageNet.
On generalizability to different viewpoints and regions of the spine (R2, R3): Lumbar surgeries dominate our dataset due to the caseload makeup of the hospital (as is the trend in most developed countries, lumbar spine surgery in elderly patients is becoming increasingly prevalent [1]). Accordingly it would be infeasible to train a reliable segmentation model on a very small number of samples. Further, as we mention in the paper, the lumbar spine is the region least affected by movement of the lungs. Thus we also limit to lumbar cases to ensure we can model the final registration approach as rigid, once the initial pre- to intraoperative deformation has been performed.
On lack of novelty in the registration process (R2): The reviewer does not specify which aspect they find lacking in novelty, and we assume they refer to our integration of techniques such as ICP. This was only one essential part of the clinical study pipeline, which encompasses technology that is easily accessible and a surgeon-friendly workflow. Beyond this, our multi-task segmentation pipeline is novel both in terms of architecture as well as the multi-modality segmentation goal. We will aim to make these technical novelties clearer in the camera ready version.
On the SpineAlign RMSE results (R3): We recognize that the difference on initial inspection appears small, however these values must be within the overall registration context. In our clinical videos, only the spinous processes and facets (<15% of the vertebral surface) are visible. After the initial rigid step, most points are already within a few mm. The RMSE is the mean over approximately 4k - 6k surface samples, so the local refinements introduced by SpineAlign are numerically diluted. Thus, the RMSE must be read in conjunction with the fitness score in Table 1 - showing the proportion of scene points falling inside the 10 mm correspondence window rises from 0.56 to 0.58 (+3.6%), while its s.d. drops 0.13 to 0.11, confirming that SpineAlign consistently converts out-of-tolerance points into valid correspondences and stabilises the alignment. On the fully exposed phantom, the benefit is clearer (fitness +0.04, RMSE -0.23 mm). We will make this clearer in the revised manuscript.
On the breakdown of CT and MRI for the lumbar cases (R3): Two cases have MRI preoperative imaging, the remainder have CT scans.
[1] Grotle, Margreth, et al. BMJ open 9.8 (2019): e028743.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A