Abstract

Longitudinal volumetric tumour segmentation is critical for radiotherapy planning and response assessment, yet this problem is underexplored and most methods produce single-timepoint semantic masks, lack lesion correspondence, and offer limited radiologist control. We introduce LinGuinE (Longitudinal Guidance Estimation), a PyTorch framework that combines image registration and guided segmentation to deliver lesion-level tracking and volumetric masks across all scans in a longitudinal study from a single radiologist prompt. LinGuinE is temporally direction agnostic, requires no training on longitudinal data, and allows any registration and semi-automatic segmentation algorithm to be repurposed for the task. We evaluate various combinations of registration and segmentation algorithms within the framework. LinGuinE achieves state-of-the-art segmentation and tracking performance across four datasets with a total of 456 longitudinal studies. Tumour segmentation performance shows minimal degradation with increasing temporal separation.We conduct ablation studies to determine the impact of autoregression, pathology specific finetuning, and the use of real radiologist prompts. We release our code (https://github.com/n-e-garibli/LinGuinE) and substantial public benchmarking for longitudinal segmentation, facilitating future research.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3584_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/n-e-garibli/LinGuinE

Link to the Dataset(s)

4DCBCT: https://www.cancerimagingarchive.net/collection/4d-lung/ UniToChest: https://zenodo.org/records/5797912 APLCT: https://www.autopet.org/longitudinalct.html

BibTex

@InProceedings{GarNad_LinGuinE_MICCAI2026,
        author = { Garibli, Nadine AND Csiba, Bence AND Sidiropoulos, Konstantinos AND Wei, Yi AND Patwari, Mayank},
        title = { { LinGuinE: Longitudinal Guidance Estimation for Volumetric Tumour Segmentation } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16883},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper introduces LinGuinE, a framework for longitudinal guided segmentation, which can be used with any combination of registration and promptable segmentation algorithms.

  • 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 generalize the idea of earlier works like Hering et. al. and LesionLocator, building a plug and play framework, which works with any combination of registration and guided segmentation algorithm to track a tumor through the whole image sequence. By separating the temporal component in the form of simple registration, it enables training on single-timepoint data alone.

    The authors introduce an initial boosting method, which can make the model more resilient to slight registration errors.

    The authors use 4 datasets with multiple pathologies, showing the generalizability of the proposed framework.

    The authors demonstrate the capabilities of LinGuinE using the state-of-the-art in registration and promptable registration methods.

    The authors show additional improvements in performance by in-distribution finetuning of the promptable segmentation method used in the framework.

    The authors show the robustness to elapsed time between scans for their LinGuinE framework, compared to the baseline methods.

  • 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.

    As also noted by the authors, ED is suboptimal as a metric, when many lesions don’t have associated predictions. Therefore, it would have been interesting to see something like a detection rate as an additional metric to Dice.

    Minor: The authors claim that their framework is “the first temporally direction agnostic framework, allowing radiologists to provide prompts at any timepoint, facilitating retrospective analysis”. While this is true by design, e. g. LesionLocator can also be applied per timepoint in a non-autoregressive fashion.

  • 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.

  • 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 proposed framework shows strong performance, outperforming all evaluated baselines. While the paper doesn’t introduce major methodological novelty, the provided framework and insights are highly relevant in the context of tracking and longitudinal segmentation.

  • 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

    A Longitudinal segmentation method to track tumour growth and treatment response based on interactive segmentation approaches.

  • 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.

    Proposes an innovative solution to a relevant clinical problem

    Validates on multiple datasets

    Compares the performance of multiple backbones and compares the proposed approach to relevant baselines.

    Achieves state-of-the-art results.

    Performs relevant ablation 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.

    A significant limitation, as noted by the authors, is the potential for disappearing lesions. In a fully automated context, this scenario could lead the model to segment an unrelated structure, resulting in an inaccurate response assessment. However, this issue is a candidate for mitigation in future research.

  • 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 has provided an anonymized link to the source code, dataset, or any other dependencies.

  • 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?

    Interesting paper to read, creative idea, and innovative approach to a very relevant clinical problem, making use of novel interactive segmentation methods.

  • 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 addresses longitudinal tumor segmentation and tracking from minimal radiologist input. The main contribution is LinGuinE, a flexible framework that combines image registration, prompt propagation, and guided volumetric segmentation to produce lesion-level segmentations across all scans in a longitudinal study from a single source-timepoint prompt. A key aspect of the work is that the framework is temporally direction agnostic, does not require training on longitudinal data, and can reuse existing registration and promptable segmentation algorithms. In addition to the framework itself, the paper provides a relatively broad benchmark across four datasets and analyzes practical aspects such as temporal degradation, autoregression, boosting, and the use of real radiologist prompts.

  • 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 addresses a clinically meaningful and underexplored problem. Longitudinal lesion tracking with segmentation is more relevant to treatment monitoring and response assessment than standard single-timepoint semantic segmentation.

    The proposed workflow is practical. Allowing a radiologist to provide guidance at one timepoint and propagate it across a longitudinal study is a reasonable and potentially useful clinical interaction model.

    Even though the method is conceptually simple, the results are strong, and the benchmarking itself is a valuable contribution to this task

  • 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 algorithmic novelty is moderate. At its core, the method is a combination of existing ingredients, image registration, propagated prompts, guided segmentation, and optional boosting. The main contribution is therefore more of a practical framework and benchmarking effort than a fundamentally new algorithm.

    While temporal direction agnosticism is useful in practice, it is partly a natural property of a registration-based prompt propagation pipeline, so I would not view it as a major conceptual novelty.

    A key practical dependency of the framework is the quality of the underlying registration. If registration is inaccurate, the propagated prompt may fall outside the target lesion or even near a different lesion, which can lead the guided segmentation stage to fail or segment the wrong object. The paper partially acknowledges this issue and introduces boosting as a mitigation, but it does not provide a systematic analysis of failure cases or robustness to registration inaccuracies. Given that real longitudinal scans often involve anatomical changes and imperfect alignment, this deserves more explicit discussion.

  • 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.

    (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?

    The main factors behind my score are the strong practical value of the problem setting, the clinically meaningful workflow, and the breadth of the empirical evaluation. I think the paper makes a useful contribution by addressing longitudinal lesion tracking and segmentation from minimal radiologist input, which is more relevant to real follow-up analysis than standard single-timepoint segmentation

    My main reservations are that the algorithmic novelty is moderate, since the framework mainly combines existing components, and that the method depends on registration quality without a deeper robustness analysis.

  • 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



Author Feedback

We thank all the reviewers for their feedback and careful consideration of our manuscript. We are grateful for the positive remarks and appreciate the thoughtful criticisms raised.

To R1 and R6: We agree that the key value of our contribution lies in the practicality and clinical relevance of the framework, along with the benchmarking efforts and new insights into an under-explored problem, rather than major algorithmic novelty. At the same time, we believe there is novelty in repurposing algorithms trained on single time-point data for longitudinal use, the boosting mechanism, and the plug-and-play nature of LinGuinE, which distinguishes it from prior work.

To R6: Regarding reproducibility, we would like to kindly point out that most of our results on UniToChest, 4DCBCT, and APLCT should be reproducible by the community since the data and many algorithms are public. We also pledged to release our code package and our ground truth masks for UniToChest (https://github.com/n-e-garibli/LinGuinE). We had not included links to the assets at the time of review to preserve anonymity.

To R6: Regarding the concern on the dependence of our framework on underlying registration quality. We considered this issue during development and explored robustness to out-of-tumour point propagation, including experiments with rigid registration. However, we prioritised including other analyses in the manuscript, such as experiments on time-direction agnosticism, the use of real radiologist prompts, and temporal degradation, because we felt these were both more novel and more clinically relevant. We also hope that the effect of registration quality is partially reflected in Table 1, where performance on the lung cancer datasets improves when LGI is used instead of UGI. More broadly, many modern guided segmentation methods are trained such that exact guidance location is not critical, and some closed-set algorithms can perform well as long as the guidance point is sufficiently close to the target object. In our experience, this means that even relatively simple registration can be effective in the LinGuinE setting, making the choice of segmentation algorithm more important than the choice of registration. Whilst we would like to support that claim empirically, unfortunately our understanding is that editing the manuscript to include additional results is not permitted.




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.

    The reviewers have a clear view on the paper as it provides a low-input solution for tracking lesions over time an area that has a clear clinical benefit. Moreover the presentation of the work and the evaluation and later discussion seem solid and were celebrated by the reviewers. While I am giving the paper a provisional accept following the reviewer’s recommendations, I do think it is more of an integration work with a good evaluation. I strongly recommend the authors to comment on the criticisms of reviewer 6.



back to top