List of Papers Browse by Subject Areas Author List
Abstract
Microwave ablation (MWA) is a well-established curative treatment for unresectable tumors. However, current clinical planning strategies rely predominantly on geometric heuristics that neglect patient-specific physiology, often resulting in incomplete tumor ablation or unintended damage to healthy tissues. While physics-based bioheat finite element simulations can model the underlying electromagnetic and thermal interactions, their computational cost renders them impractical for interactive clinical planning. We propose a novel, human-in-the-loop computational framework for interactive, patient-specific preoperative MWA planning. At its core there is a deep learning inverse model that directly infers clinically actionable ablation parameters (generator power, ablation duration, and axial depth) given a target tumor mask.
The model assesses with sub-second latency the feasibility of insertion paths proposed by the clinician. When viable, it predicts optimal settings to maximize tumor coverage while minimizing collateral damage. Our framework also supports multi-antenna strategies via a sequential optimization scheme. It has been validated against in vivo porcine experiments and tested on patient data. Our solution achieved an overall Technical Success Rate of 97% across 116 clinical cases and generated viable multi-antenna treatment plans with sub-second latency (~300 ms per antenna). This interactive framework bridges the gap between high-fidelity bioheat modeling and interactive clinical decision-making, enabling safe, efficient, and personalized MWA strategies.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/4853_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026/supp/4853_supp.zip
Link to the Code Repository
N/A
Link to the Dataset(s)
ircadb-01 dataset: https://www.ircad.fr/research/data-sets/liver-segmentation-3d-ircadb-01/
BibTex
@InProceedings{DetFra_Interactive_MICCAI2026,
author = { Dettori, Francesco AND Nahmed, Ilias AND Duprez, Michel AND Alvarez, Pablo AND Cotin, Stéphane},
title = { { Interactive MWA Planning via Probabilistic Inverse Inference } },
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 main contribution of this paper is a real-time, patient-specific planning framework for microwave ablation that directly infers optimal treatment parameters (power, duration, and insertion depth) from a target tumor using a probabilistic inverse model. By combining a fast surrogate model with forward-in-the-loop training, the method enables interactive, clinician-in-the-loop planning with sub-second latency while maintaining high tumor coverage and minimizing damage to healthy tissue.
- 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 direct inverse inference approach that predicts clinically actionable parameters (power, duration, depth) from the target tumor, avoiding iterative optimization. This is particularly novel for MWA, where the parameter space is complex and highly coupled, making traditional search-based methods impractical for real-time use.
2.The combination of a physics-informed forward surrogate model with a forward-in-the-loop training strategy is a strong contribution. Optimizing the inverse model based on predicted ablation outcomes aligns the learning objective with clinical goals, which is both elegant and practically meaningful.
3.The method is validated across synthetic data, in vivo porcine experiments, and a clinical dataset. This multi-tier evaluation strengthens the credibility of the approach, particularly the inclusion of in vivo validation and clinically meaningful metrics such as technical success rate and ablation efficiency.
- 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.While the use of synthetic FEM data is well-motivated, it remains unclear whether the dataset size (2400 samples) and its variability are sufficient to capture the full complexity of patient-specific anatomy and perfusion patterns. Additional analysis or ablation studies on data efficiency and diversity would help better support the robustness of the learned models.
2.The sequential multi-antenna strategy does not appear to account for tissue property changes induced by prior ablations (e. g. , reduced perfusion or altered thermal properties). In the current formulation, only the residual tumor mask is updated, while the underlying perfusion map remains unchanged. This simplification may limit the physical realism and accuracy of multi-antenna planning, and this limitation is not explicitly discussed.
3.the study lacks comparison against meaningful baselines. In particular, it would be valuable to compare the proposed method with standard clinical practice, such as manufacturer-provided power–time charts or heuristic-based planning. Such a comparison would help better quantify the practical benefits and clinical relevance of the proposed framework.
4.The reproducibility of the proposed framework is limited. Key components, such as the FEM-based data generation pipeline and training details, are not described in sufficient detail to enable full reproduction. In the absence of code or additional implementation details, it is challenging to assess whether the method can be reliably replicated.
- 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 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.
(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 paper addresses an important and clinically relevant problem and proposes a novel inverse planning framework for real-time microwave ablation. The combination of a physics-informed surrogate model with direct inverse inference is well-motivated and enables interactive, clinician-in-the-loop planning with impressive computational efficiency. The work is further strengthened by comprehensive validation, including in vivo experiments and retrospective clinical evaluation, demonstrating promising performance and practical applicability.
While there are some limitations, such as insufficient methodological details, lack of strong baselines, and questions regarding generalization from synthetic data, these do not outweigh the overall novelty, technical soundness, and potential clinical impact of the work.
- 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 present a novel, deep-learning-based framework for real-time Microwave Ablation (MWA) planning. The core innovation lies in a dual-model architecture: 1) Forward Surrogate Model: Pre-trained on Finite Element Method (FEM) data to predict necrosis masks. It effectively accounts for heat-sink effects by integrating patient-specific perfusion and vascular maps. 2) Inverse Planning Model: Utilizes the surrogate model within its training loop to optimize antenna parameters and trajectory depth. By penalizing discrepancies between the predicted ablation zone and the target tumor mask, the model learns to derive optimal strategy parameters.
Based on the previous models, they developed an interactive planning where clinicians are able to plan the MWA tumour ablation and the platform suggests and provide the optimal planning parameters and the estimated ablation.
In general, the article is well written and can be easily followed.
- 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) Computational Efficiency: By replacing computationally expensive FEM simulations with deep learning surrogates, the tool achieves the high-speed execution necessary for an interactive clinical interface.
2) Biological Accuracy: The inclusion of perfusion maps ensures that the cooling effects of surrounding vasculature are factored into the plan, without relying only on geometric heuristics.
Validation: The study is well-structured and provides robust evidence through validation on open datasets and an in vivo assessment of the forward surrogate model.
- 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.
In the results, the authors presented the Ablation Efficiency score. For single antennas, they reported a smaller value (0.44) with respect to the overall (joint single antenna and multiple antenna plannings). The authors justified these results on sub-centimeters tumours, that are smaller than the minimum physical ablation volume achievable by a MWA device. From my point of view, the authors should provide: 1) the minimum achievable ablation volume; 2) the number of sub-centimeters tumours in the validation datasets. In my understanding, the cases where the tumour volume is smaller than the minimum ablation could be presented separately, providing a better understanding on the application.
As a minor weakness, especially in the introduction, the authors states frequently the need of real time during planning. While I understand the importance of a fast and responsive simulation in a planning applications, at first read I was misleaded thinking the clinical application had the need of a real time planning framework (e.g. intraoperative needs). I suggest to review the introduction and better explain that, from a usability point of view of a planning application, having a very fast simulation framework can be very beneficial. I suggest to change the term “real-time” with “high-performance” computing capabilities in this text.
- 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?
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.
(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?
I believe the paper is presenting a planning framework based on several innovations thay are also presenting at the MICCAI conference. This article explains the combination of different implementation in a real planning application, test on several cases. It translates the research into a software usable by clinicians.
- 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
Real-time (i.e., within clinical time constraints) computational framework for liver tumour microwave ablation planning based on deep learning and probabilistic inverse inference. The framework assesses the feasibility of clinician-proposed ablation antenna insertion paths and supports the selection of appropriate ablation parameters.
- 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 study aims to develop a solution that integrates well into the clinical workflow. Because clinician input is required to initiate the planning process, the proposed system is designed to assist clinicians rather than replace them.
2.The methods are described clearly and in sufficient detail, especially considering the space constraints of MICCAI papers.
3.The use of a synthetic dataset for training and validation is well motivated, with clearly defined ground truth generated using a physically plausible finite element (FE) model.
4.The validation is comprehensive, encompassing synthetic data, experimental animal data, and retrospective (publicly available) liver tumour patient data.
5.Physically meaningful accuracy metrics, including the Hausdorff distance, are employed in the validation.
- 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.Although both the Dice coefficient and Hausdorff distance are used as validation metrics, it is unclear how the Hausdorff distance is computed in practice. In particular, the manuscript does not explain whether it is calculated on the boundaries (surfaces) of the ablation-induced necrotic regions.
2.Insertion of ablation antennas is likely to cause tissue deformation and potential target (tumour) displacement. While addressing this issue may be beyond the scope of the current study, acknowledging it explicitly as a limitation would strengthen the paper.
3.The Hausdorff distance is reported in millimetres; however, without information about the image resolution, these values are difficult to interpret. The manuscript should include information about image resolution to contextualise the reported distances.
- 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?
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.
(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?
Only limited additional information and revising section on the limitations are needed. This can be classified as a minor revision that can be addressed by adding few sentences to 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
Author Feedback
We thank the reviewers for their constructive assessment and for recognizing the novelty and potential clinical relevance of the work. We focus on the main points that require clarification. 1.Baseline comparison with current clinical practice (R1). R1 asked for a comparison with clinical practice. We believe this comparison is already included: the forward model was compared against manufacturer-provided power-time tables, which represent the standard planning reference and do not account for patient-specific anatomy or vascular heat-sink effects. In the in vivo validation, our method achieved a DSC of 0.82, compared with 0.59 for this baseline. 2.Role and size of the synthetic dataset (R1). R1 questioned whether 2400 synthetic samples are sufficient to capture patient-specific variability. These samples are not intended to form a complete atlas of human anatomical variability. Their role is to train the surrogate to reproduce the coupled electromagnetic-thermal response generated by the FEM solver under varying local conditions. The goal is not to memorize specific anatomies, but to approximate the physics of MWA across plausible tissue and vascular scenarios. We increased variability by randomizing vessel-generation parameters within literature-based ranges, perfusion values, and tumor morphology. We acknowledge that no finite synthetic dataset can exhaustively sample all tumor-vessel configurations, but our aim is to expose the model to diverse plausible cases so that it learns the simulated physical process rather than fixed geometries. 3.Ablation Efficiency in single-antenna cases (R2). R2 questioned the lower AE reported for single-antenna treatments. This reflects a physical constraint of MWA: even low settings (30 W, 180 s) create a minimum lesion size, typically on the order of 2 cm in diameter. For sub-centimeter tumors, covering the tumor plus the required oncological margin necessarily produces an ablation volume that is large relative to the target. AE therefore decreases because the minimum achievable lesion overshoots the small target. We will also include the number of sub-centimeters tumors in the validation dataset. 4.Use of “Real-Time” (R2). We agree that “real-time” may be misleading, as it could suggest an intraoperative requirement rather than the intended preoperative planning context. We have therefore removed “Real-Time” from the title. “Interactive” better reflects the intended clinical use: fast feedback allowing the clinician to explore and refine treatment plans during preoperative planning. 5.Sequential ablations, tissue changes, and deformation (R1 and R3). R1 noted that previous ablations modify tissue properties. Updating the tissue map after each virtual ablation would be straightforward within our framework. However, we did not include this step because there is no clear consensus on how tissue properties should be modified after necrosis, especially dielectric and thermal parameters. We therefore avoided introducing an assumption that could not be robustly justified. Currently, sequential planning updates the residual tumor mask while keeping pre-treatment tissue properties unchanged. R3 also noted that antenna insertion may cause deformation and target displacement. This falls outside the paper scope, but we agree that it can introduce prediction errors and will acknowledge it as a limitation. 6.HD95 computation and image resolution (R3). HD95 is computed on the surfaces of the predicted and reference necrotic regions. After thresholding the predicted probability map, we extract the boundary voxels of both masks and compute the 95th percentile of the bidirectional surface-to-surface distances. Distances are converted from voxel units to millimeters using image spacing. All volumes are resampled to an isotropic resolution of 0.6 x 0.6 x 0.6 mm^3 before training and evaluation, so HD95 values correspond to physical discrepancies in the resampled space.
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.
All reviewers found this a strong manuscripts and recommended acceptance.