List of Papers Browse by Subject Areas Author List
Abstract
Unsafe surgical care is a critical health concern, often linked to limitations in surgeon experience, skills, and situational awareness. Integrating patient-specific 3D models into the surgical field can enhance visualization, provide real-time anatomical guidance, and reduce intraoperative complications. However, reliably registering these models in general surgery remains challenging due to mismatches between preoperative and intraoperative organ surfaces—such as deformations and noise.
To overcome these challenges, we introduce the first deep learning-based non-rigid point cloud registration method that is genuinely patient-specific, being both trained and tested on the same individual’s anatomy. Our approach combines a Transformer encoder-decoder architecture with overlap estimation and a dedicated matching module to predict dense correspondences, followed by a physics-based algorithm for registration. Experimental results on both synthetic and real data demonstrate that our patient-specific method significantly outperforms traditional agnostic approaches, achieving 45% Matching Score with 92% Inlier Ratio on synthetic data, highlighting its potential to improve surgical care.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3372_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)
IRCAD Liver 1 from: https://www.ircad.fr/research/data-sets/liver-segmentation-3d-ircadb-01/
DePoll dataset: https://www.ircad.fr/research/data-sets/respiratory-cycle-3d-ircadb-02-copy/
BibTex
@InProceedings{NerAlb_Towards_MICCAI2025,
author = { Neri, Alberto and Haouchine, Nazim and Penza, Veronica and Mattos, Leonardo S.},
title = { { Towards Patient-Specific Deformable Registration in Laparoscopic Surgery } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15968},
month = {September},
page = {641 -- 650}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors present a novel patient-specific deformable point cloud registration method for laparoscopic surgery based on a transformer architecture and a matching module. The authors also introduce a novel data generation strategy during training that generates patient-specific deformations on the fly. The results on synthetic and real data show superior performance compared to state-of-the-art approaches which are trained in a non patient-specific way.
- 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 authors address a relevant problem of high significance: inter-operative, cross-modality registration of deformable regions. The presented training augmentation based on real patient data that is deformed via a SOFA simulation is novel and innovative. The training pipeline consists of known building blocks but the design has been carefully crafted to address the main challenges of the deformable matching task. The results on simulated and real datasets show superior performance compared to state-of-the-art methods. A small ablation study highlights the influence of the chosen losses.
- 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 methods description lacks some important details to judge the validity of the approach: how are the 3D point clouds generated? Or are they already included in the datasets used? How were the ground truth values for the 3 losses obtained? Was the whole pipeline depicted in Fig. 1 trained end-to-end or where parts of it trained separately using the different losses. It would be helpful to explicitly state what the output of the model is (e.g. in Fig.1). Was the training exclusively performed with the 3D-IRCADb-01 dataset?
To better judge the validity of the claim that patient-specific training is necessary, a comparison to state-of-the-art approaches on the DePOLL dataset would be great. At least authors should more carfully analyse their results on this dataset. What were the deformation parameters present in the recorded porcine sequences (i.e. the actual deformations observed) and how does this compare to the simulation?
- 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.
- 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?
The authors present an innovative and novel approach to tackle an important problem in the field. However, the methods and results have to be more clearly described to be able to judge the validity of the approach.
- 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 #2
- Please describe the contribution of the paper
- The authors propose a point-cloud based method for registering patient-specific preoperative organ surface point clouds, extracted from imaging modalities such as CT, with intraoperative point clouds obtained from sources like stereo vision of the surgical field.
- The method introduces a novel patient-specific non-rigid point cloud registration approach that incorporates robust data augmentation strategies, including deformation and resampling to create incomplete shapes, ensuring improved model generalization.
- 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 study addresses an important clinical problem of aligning preoperative and intraoperative shapes to enhance clinical outcomes, which is well motivated.
- The novel focus on patient-specific augmentation for generalization represents a promising strategy to tailor alignment to individual samples.
- Utilizing point cloud-based methods is advantageous as they handle unstructured data and can process point clouds of varying sizes effectively.
- 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 manuscript is somewhat difficult to follow; the methods section would benefit from better organization. The figure requires a more descriptive caption, and the text should sequentially follow each component of the figure to improve clarity. Currently, the description of the architecture is fragmented and challenging to understand.
- The authors do not address how the proposed method manages the presence of surgical instruments within the field of view, which is a critical practical consideration.
- 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.
- 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
See weakness
- 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?
I recommend a weak accept as the work tackles a clinically relevant problem with a novel and promising approach. However, improvements in clarity and addressing practical challenges would strengthen the manuscript.
- 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 #3
- Please describe the contribution of the paper
This paper proposes a five-module pipeline for deformable registration to align a preoperative imaging model to a point cloud of a laparoscopic scene. The authors claim to achieve improved registration performance by taking a patient-specific approach, rather than taking an agnostic approach that is meant to generalize to new patient geometries.
- 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.
This paper tackles the highly-relevant and challenging problem of deformable registration for image-guided surgery with a novel end-to-end pipeline. The paper is well organized and clearly describes the five modules in the pipeline. The experiments for evaluation are rigorous and appropriately designed to evaluate the method’s performance.
- 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.
My interpretation of a “patient-specific” strategy taken for this work is that the model was retrained independently for each specific patient geometry. While the authors claim that this is a feature of the work (being patient-specific), others would argue that it is actually data leakage across train-test sets. The LiverMatch method omitted the IRCAD-Liver1 geometry in their training set. The authors note that they have improved matching performance compared to LiverMatch on IRCAD-Liver1. However, that may be at least partially attributed to the fact that the authors trained their model specifically for that test geometry.
I would suggest making the following clarifications: (1) Include the time allotted for model training. If training is “patient-specific”, computationally intensive training times with the need to re-train on each patient’s geometry is a limitation. (2) Explicitly state that the model was retrained for each patient’s geometry. Leaving this detail as ambiguous can be misleading for readers.
Additionally, there are several state-of-the-art patient-specific nonrigid point cloud registration methods for liver registration that are highly relevant to this work. Despite limited space, I would suggest including summaries of some/all of these works in the introduction section for completeness. [1-5].
Finally, please revise this sentence in the abstract - “We introduce the first patient-specific nonrigid point cloud registration method…” This statement is not reflective of the current state of the field. As noted above, there have been several other patient-specific nonrigid point cloud registration methods, both deep-learning and mechanics-based, proposed specifically for liver registration.
[1] Pfeiffer M., et al. , “Non-rigid volume to surface registration using a data-driven biomechanical model,” Lect. Notes Comput. Sci. 12264, 724–734 (2020). [2] Jia M., Kyan M., “Improving intraoperative liver registration in image-guided surgery with learning-based reconstruction,” in IEEE Int. Conf. Acoust. Speech Signal Process. - Proc. (ICASSP), June, pp. 1230–1234 (2021). [3] Heiselman J. S., Jarnagin W. R., Miga M. I., “Intraoperative correction of liver deformation using sparse surface and vascular features via linearized iterative boundary reconstruction,” IEEE Trans. Med. Imaging 39(6), 2223–2234 (2020). [4] Mestdagh G., Cotin S., “An optimal control problem for elastic registration and force estimation in augmented surgery,” Lect. Notes Comput. Sci. 13437, 74–83 (2022). [5] Heiselman JS, et al., “The Image-to-Physical Liver Registration Sparse Data Challenge: comparison of state-of-the-art using a common dataset.” J Med Imaging (2024).
- 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.
- 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?
My main concern is the fact that the manuscript does not clearly state that the model was trained and tested on the same patient’s geometry. With revisions, I believe that this paper is a valuable contribution, as it details a new method for nonrigid registration in laparoscopic surgery. It demonstrates the performance improvements from a novel pipeline and patient-specific training.
- 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 are glad that reviewers found our work “novel” (R1, R2, R3) and “innovative” (R1), with a design “carefully crafted to address the main challenges” (R1) of the deformable registration in laparoscopic surgery, which is recognised as a “highly-relevant and challenging problem” (R3). The reviewers found that our experiments are “rigorous and appropriately designed” (R3), with results showing “superior performance compared to state-of-the-art methods” (R1). They also agree that the paper is “well organized and clear” (R1, R3).
We thank the reviewers for the precise questions, and we will incorporate the necessary clarifications into the paper’s final version accordingly.
TRAINING SET GENERATION AND PATIENT-SPECIFIC: We describe the point cloud generation in the “Patient-specific Training Dataset Generation” Section. Since our approach is patient-specific, we select one pre-operative point cloud and we apply our generation pipeline on-the-fly during training. For instance, we train our model end-to-end for each “new patient”. We first trained on liver 1 point cloud in the 3D-IRCADb-01 dataset. Then we retrained the model on the DePoll porcine liver for the registration test. As output, the model predicts point correspondences between the two point clouds, represented as a N_tgt×2 matrix whose rows denote the source and target point indices for each match. Because our training set is synthetically generated, we have access to all required ground truth data, including correspondence labels and overlap scores. We strongly believe that the patient-specific training is key to achieve optimal performances as shown in similar settings such as EndoNeRF.
PRACTICAL CHALLENGES: The focus of this work was exclusively on patient-specific point cloud registration. R2 mentioned the problem related to the “presence of surgical instruments within the field of view”. Indeed, handling surgical instruments in the field of view is a crucial practical challenge; however, it requires further research and development, falling beyond the scope of this work.
GENERAL: Given space constraints, in the final version we will: 1) Clarify that the model is both trained and tested on the same patient‐specific geometry. 2) Expand the caption of Figure 1 with a more descriptive explanation. 3) Report the total time required for model training. 4) Update and strengthen the state‐of‐the‐art discussion as suggested by R3.
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