List of Papers Browse by Subject Areas Author List
Abstract
This study addresses the critical challenges of accurate tumor localization in minimally invasive surgery (MIS) of the liver, where limited visibility and the absence of tactile feedback complicate surgery. The study focuses on integrating all three standard modalities: preoperative 3D models, laparoscopic ultrasound (LUS), and MIS images. Unlike previous approaches, our method exploits the interrelationships among all these modalities, without relying on markers or external sensors, to maximize applicability. It uses an advanced geometric model to integrate the existing registration constraints between pairs of modalities, such as the anatomical landmarks, with new spatial constraints, including the contact of the LUS transducer with the liver and the agreement of the LUS and the preoperative tumor profiles. Experimental validation on phantoms and patient data shows that the method boosts accuracy.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2910_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2025/supp/2910_supp.zip
Link to the Code Repository
https://github.com/MMKalantari/MultiModalLiverReg
Link to the Dataset(s)
https://encov.ip.uca.fr/ab/code_and_datasets/
BibTex
@InProceedings{KalMoh_Stronger_MICCAI2025,
author = { Kalantari, Mohammad Mahdi and Ozgur, Erol and Alkhatib, Mohammad and Rabbani, Navid and Espinel, Yamid and Modrzejewski, Richard and Le Roy, Bertrand and Buc, Emmanuel and Mezouar, Youcef and Bartoli, Adrien},
title = { { Stronger Together: Registering Preoperative Imagery, LUS, and MIS Liver Images } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15970},
month = {September},
page = {556 -- 566}
}
Reviews
Review #1
- Please describe the contribution of the paper
The main contribution of this paper is a deformable optimisation framework that unifies registration between a pre-operative CT, Laparoscopic Ultrasound (LUS) and Laparoscopic Video of the liver. Authors use previously published methods to initialise LUS to video and Video to CT, and refine the result using new LUS probe contact constraints and a tumour outline detected in the LUS image.
- 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.
- Very few frameworks address the joint LUS-video-CT registration problem in laparoscopic surgery.
- The rationale of the constraints is interesting, and have some novelty in this field.
- Validation is performed both on phantom and clinical data.
- 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.
- Authors claim their method is the first to do this integration, but this has been previously attempted (see comments below).
- The method is limited to images where the tumour is clearly visible in the ultrasound images.
- Results on real data are not clearly superior to previous approaches.
- 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
I believe this paper is of interest to the MICCAI community, and that the presentation of a framework integrating all 3 modalities in laparoscopic surgery holds promise. However, I have the following concerns with this submission:
- Authors claim this is the first framework to jointly solve the LUS-video-CT problem. However, this has been attempted in [A], by using a self-supervised framework where CT to US and video to CT are jointly predicted. Authors mention reference [20] in their paper as a closest work, but such work is only EM tracked LUS to CT registration.
- On Related work, reference [5] does not use optically tracked LUS as such is not easy to deploy (LUS probes are inside the abdomen and have a bendable tip that can not be tracked with an optical system). Reference [21] refers to 3D Intra-operative Ultrasound in open liver surgery, not LUS.
- On the public benchmark, improvement in TRE versus similar deformable registration methods is observed in 2 cases, where the tumour did not change size. How was TRE measured and what are the sizes of the tumours? If the LUS method used the tumour outline in C7, it seems that reporting tumour TRE is simply presenting the residual of the registration of the tumour - all other methods have to predict tumour positioning from optimising functions liver landmarks. A fairer comparison on accuracy would require other targets. Also, in Table 2, creating two means on a sample of 3 does not provide much additional information.
- How much additional deformation occurs on the tumour due to probe contact with the tissue? Can this have an effect on the
- The method rely on a match between the tumour outline in CT and Ultrasound. However, it is well known that liver tumours in ultrasound are not easily visible in some cases, hence the need for US or LUS to CT/MR algorithms to predict their position using other features. In the case where the tumour visibility is a problem, how would the method perform?
- Clinically, authors claim AR is not appropriate in the case where there are mismatches between intra-operative and pre-operative tumours. However, AR has value in displaying vascular content as well which is critical for safe and successful liver surgery. Even if tumours have different sizes, a matched representation can help surgeons.
- 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?
This paper attempts to solve a relevant problem in laparoscopic surgery with an interesting framework. However, its use is quite limited and reliant on both clinical conditions and success of previous methods.
- 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 apologise the authors for not providing the reference [A] in my review, I put it here now for clarity. I am not sure the authors still consider work 20 to the closest to their work, but I believe they will correct this and add [A] instead.
I understand that the method can still optimise even if the tumour is not visible, using mostly probe contact constraints, making it more feasible to use clinically.
I still have no clarity on how TRE is exactly measured, if there are correspondences on specific points, or the tumour contours of LUS are sampled and compared with a closest distance criterion to the resliced contour of the tumour predicted by the ground truth. Since C7 and C8 approximate the tumour contours, I do not think they are totally independent.
I am not sure what the authors mention by vessel augmentation. Regardless, I recommend the paper to be accepted for presentation at MICCAI, given the authors intention to change their novelty claims and final statement on how AR should be used.
[A] - Montaña-Brown, Nina, et al. “Towards multi-modal self-supervised video and ultrasound pose estimation for laparoscopic liver surgery.” International Workshop on Advances in Simplifying Medical Ultrasound. Cham: Springer International Publishing, 2022.
Review #2
- Please describe the contribution of the paper
This paper presents a method for intra-operative registration during MIS. The main contribution is to combine there modalities at once: pre-op CT, laparoscopic images and ultrasound. This is novel w.r.t to most existing methods that claim multimodality but register only two modalities at a time. Results are provided on phantoms and patient data.
- 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.
-Well-written paper and clear method -The method is described properly and technically sound. -Experiments on phantoms and benchmark dataset are appreciated, in particular for CAI papers.
- 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.
-Does assuming LUS and Lap being synchronized means the rigid transformation between is known. Please explain. -It is unclear if the US is 3D or reconstructed from slices, with or without EM. Please explain. -Does splitting the 8-terms minimization into two phase comes down to a part-wise two-modality registration? The initialization focuses on Prp-LUS only. The promise of a full three-modality registration is somehow diluted. Please explain. -The ablation study could have included all terms, instead of three models. -Figure 2 is not sufficient to appreciate the qualitative results on the benchmark. Note that the supplementary material was suppressed by meta-reviewer. -Release the phantom dataset publicly would benefit the community.
- 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?
There are some unclear assumptions that need to be explained. Beside that, a figure of the results on all cases is needed to appreciate the effectiveness of the method.
The phantom dataset should be made public.
- 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.
The rebuttal answered most of my questions.
Review #3
- Please describe the contribution of the paper
The authors propose a multi-modal registration framework for tumor localization guidance during laparoscopic surgery. The proposed framework integrates three image modalities typically available for laparoscopic guidance: preoperative CT/MRI, laparoscopic ultrasound (US) and laparoscopic image. Using the geometry of the liver (CT/MRI) and tumor (CT/MRI and US), the authors formulate the multi-modal registration problem as the minimization of an eight-term functional modeling the interrelations between the liver, tumor, US probe and laparoscopic view during surgery. The proposed framework is evaluated quantitatively using a publicly available phantom and real dataset, respectively; as well as qualitatively using an additional in-house real dataset. The authors report generally improved precision with respect to state-of-the-art methods, while maintaining acceptable computational times.
- 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 multi-modal registration problem addressed in this work is extremely challenging, as it involves 2D and 3D modalities simultaneously. The authors report tumor localization errors generally below 10mm, which is remarkable for this type of problem. Although the authors build upon previous work from the literature, they introduce simple, yet effective models of liver-probe interactions that generally improve tumor localization errors with respect to existent frameworks. Further, the proposed evaluation is particularly strong, using both real surgical and phantom data.
- 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.
I believe this paper has no major weaknesses.
- 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
- In the introduction, the fact of using only two from the three modalities do not change the ill-posedness of the registration problem
- In Table. 1, I find it easier to read the TRE directly, and not the reported percentages
- In Table 2., it would be informative to report all the cases and just clarify in the table which ones did not work
- 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?
The authors provide methodological contributions to the multi-modal registration problem, as well as a particularly strong evaluation both with real and phantom data.
- Reviewer confidence
Somewhat confident (2)
- [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 did not have any major issues with this paper in my initial review, and my understanding is that the authors provide satisfactory answers to the remarks raised by the other reviewers.
Author Feedback
We thank the reviewers for their thorough feedback. We will carefully revise the paper accordingly. We organise the rebuttal along the 4 points in the meta-review and the remaining comments from the reviews.
1) Reproducibility. Only R2 rated as insufficient information for reproducibility. We recall that Alg 1 gives pseudo-code with implementation details in text (eg, calibration details, Sec 3.1; SciPy library, Sec 3.3; hyperparameter values, Sec 4). Datawise, the real patient benchmark and the phantom data are publicly available (not cited yet for anonymity). We will release our code.
2) Claimed novelty. R1: We will include ref [A] as closest work to ours (as [20], it indeed attempts LUS-video-CT registration) and update the claim about the first method in the conclusion. Importantly, the method in [A] is widely different from ours: [A] is learning-based, patient-specific (hence, it must be specifically re-trained for each patient), exploiting rendering; ours is patient-generic and grounded in a mathematical modelling of the interrelations between modalities. This offers the advantage of interpretability, allowing us to monitor and control the error at each step. In addition, ours exploits LUS even when the tumor is invisible (see point 4) below).
3) Incremental efficacy. R1: In addition to the substantial average improvement of 43% on the phantom dataset (Tab 1), our method achieves a 36% average TRE reduction for the three patient cases compared to the second-best method (Tab 2), with up to 50% for the challenging P3 case.
4) Clinical limitations. MR: Tumor visibility. Our method exploits the tumor but does not require its complete visibility or precise size, as it dynamically updates the correspondences (cost C7, eq 4, Alg 1). Fig 2c is an example with a partially visible tumor on the image edge; P4 is an example where the tumor changes shape between CT and LUS. Further, our method exploits LUS via the LUS probe contact constraint (costs C4, C5, eqs 1, 2), which improves registration TRE even when the tumor is invisible (ablation study in Tab 1). Exploiting LUS pose from the Lap image, even in the absence of the LUS image, is one of our key contributions. R1: TRE calculation and fairness. TRE calculation follows [15], which involves using printed markers on the LUS probe for calibration, and whose final step depends on the tumor contour from the LUS image to complete the evaluation. In our optimization pipeline, none of the preceding calibration data is used; only the tumor contour from the LUS image is used. So, cost C7 fundamentally differs from the residual of the evaluation method. The comparison with other methods is thus fair: the only additional (and optional) input is the LUS image. In other words, tumor localization is still performed by optimizing against liver landmarks, and LUS pose, including its plane orientation, is estimated solely based on this data, without relying on markers or any other external references.
5) Other comments. R1: Vessel augmentation. We will modify the sentence “if a mismatch exists between the preoperative and intraoperative ‘model’, AR should not be used”. R3: Synchronization. We meant temporal synchronization, not pose data, which merely allows us to make a temporally coherent Lap to LUS image association. R3: EM tracker. It is indicated that no additional sensor is used in our method (Abstract, Sec 1, 2, 5), and LUS pose estimation is performed from the Lap image (cost C3, Alg 1). R3: Multimodal registration dilution. Splitting cost terms into 2 groups is based on their nature (Sec 3.3), and full multimodal registration is performed in the end. R3: Ablation study. Cost terms are grouped as Contact and Tumor because they are dependent and must meaningfully be used in conjunction.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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”.
This submission received 3 somewhat polarizing reviews. The issues identified include: 1) reproducibility, 2) claimed novelty, and 3) incremental efficacy compared to other methods. Moreover, some of the required clinical conditions, such as the visibility of tumors in ultrasound needed to achieve requisitions, are often not met.
Nonetheless, this paper is of particular interest to the MICCAI community. In this regard, rebuttal is requested.
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
Novelty of application – three-way imaging and methods. The three reviewers agree that the paper is sufficiently meritorious despite drawbacks.
Meta-review #3
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
This paper presents a timely and clinically relevant contribution to CAI, proposing a unified framework for joint LUS-video-CT registration in laparoscopic liver surgery—an area with limited existing solutions. The integration of probe contact constraints and tumour outlines is a novel and promising approach, supported by validation on both phantom and clinical datasets. Despite some concerns around novelty claims and result interpretation, the reviewers found the rebuttal satisfactory and ultimately recommended acceptance, recognizing the framework’s potential interest on the MICCAI community.