List of Papers Browse by Subject Areas Author List
Abstract
MRI provides excellent soft-tissue contrast without ionizing radiation, but long acquisition times increase patient discomfort while also raising exam costs and limiting scanner throughput. A common approach to reduce scan time is to acquire fewer measurements, which yields an ill-posed linear inverse problem; recovering diagnostic-quality images therefore requires incorporating prior knowledge beyond the measured data.
In follow-up exams, the most recent prior scan of a patient can provide a highly informative subject-specific context, but practical use is complicated by temporal changes (including pathology progression), misalignment between scans, and protocol drift across acquisitions.
In this work, we introduce L-TGVN, a Longitudinal Trust-Guided Variational Network that leverages prior scans as side information to reconstruct the current scan from heavily undersampled measurements. Crucially, L-TGVN constrains the influence of prior scans to be consistent with the acquired measurements. Unlike many existing longitudinal reconstruction methods, it does not require explicit pre-registration between prior and current scans. It further accommodates differences in acquisition protocols across visits (e.g., changes in sequence parameters).
We evaluate L-TGVN against matched-capacity baselines, including prior-guided methods and methods that do not use longitudinal priors, and observe consistent improvements in standard quantitative metrics together with better preservation of fine structures at challenging accelerations.
Source code is available at github.com/sodicksonlab/L-TGVN.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2497_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/sodicksonlab/L-TGVN
Link to the Dataset(s)
N/A
BibTex
@InProceedings{AtaArd_LTGVN_MICCAI2026,
author = { Atalık, Arda AND Chopra, Sumit AND Sodickson, Daniel K.},
title = { { L-TGVN: Leveraging Longitudinal Priors for Personalized Rapid MRI } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16887},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes a novel way of incorporating prior scans obtained during longitudinal studies as side information during image reconstruction, while maintaining data consistency with acquired measurements.
- 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 proposes an elegant solution to the problem of incorporating information from prior scans. This is achieved by introducing an additional penalty term to the optimization problem, which targets the ambiguous subspace of the forward operator. The benefits of the method compared to baseline methods are clear, which are reflected in both quantitative and qualitative results.
- 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.
Robustness to Forward Operator Discrepancies: The proposed method incorporates side information within the space spanned by the forward operator. It remains unclear how the model performs when the forward operator has reduced rank (e. g. , single-coil configurations common in portable MRI) or contains significant errors (e. g. , gross inaccuracies in coil sensitivity maps). A discussion regarding the trust guidance under these degraded conditions would significantly strengthen the paper.
Selection of Singular Value Threshold (delta): Could the authors clarify the selection process for the singular value threshold (delta)? Specifically, was this parameter manually selected, or was it learned during training?
Refinement of the H Function: Could the authors explain the rationale or the specific empirical benefits observed by going from H(s; gamma^t) definition of the original TGVN to H(s, x^t; gamma^t) definition.
- 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
I believe a high-impact future direction for this work lies in its application to portable, low-field MRI scanners, where SNR and image quality are inherently constrained.
Specifically, this method could be a key enabler for longitudinal radiological surveillance (e. g. , monitoring tumor progression). By effectively leveraging side information from high-field baseline scans, this approach could allow patients to transition to low-field portable scanners for follow-up sessions. This would not only improve patient comfort and accessibility but also optimize the utilization of high-field resources for diagnostic cases that require maximum resolution.
- 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.
(6) Strong Accept — must be accepted due to excellence
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The manuscript presents a compelling approach with strong experimental results. The methodology is well-constructed; however, a few technical details regarding the robustness and architectural choices require further clarification.
- 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 paper introduces a Longitudinal Trust-Guided Variational Network (L-TGVN) that leverages a prior scan as side information to reconstruct the current scan from heavily undersampled measurements (x20, x15, x10). It extends the previously proposed TGVN method, which uses side information to eliminate the undesirable solutions from the ambiguous space of the forward operator, and shows the effectiveness of the approach in leveraging longitudinal prior scans, without the need of explicit pre-registration.
- 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 work applies the TGVN framework to leverage longitudinal prior scans to help reconstruct highly undersampled MRI data and specifically handle the problem of misregistration (and acquisition protocol difference), which is an important question in leveraging longitudinal information, and demonstrated effectiveness in challenging undersampling rates (x10-20) with impressive results.
- 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.
- Method-wise, the innovation in comparison with the prior work TGVN is relatively incremental. The key method is still based on the TGVN framework. The new parts include handling mis-registration by incorporating multiple slices from the prior scan and adding the current reconstruction iterate as an additional input to the H network, which is the network used to encode the information from prior scan, so that an implicit registration may be done inside the network itself.
- In the evaluation, the two baseline methods that leverage longitudinal priors (MTrans and DMSI) show relatively bad performance, even worse than the E2E-VN without any side information, making them less convincing as baselines. Authors may consider designing additional baselines based on their E2E-VN network but with simpler conditioning mechanism (e.g., adding encoded side information directly into the reconstruction network itself) to highlight the advantages of the TGVN framework (i.e., having the projection to the ambiguous space). (This is a suggestion, not a request for additional experiments.)
- 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?
The work shows a nice application of the TGVN framework, targeting an interesting problem of leveraging longitudinal priors, with new designs to tackle the specific challenges (misregistration) in this application. It showed convincing results, with both impressive reconstructed images at very high undersampling rates and comparison to baselines. The paper is well-written and clear.
- 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 introduces L-TGVN (Longitudinal Trust-Guided Variational Network), a method for personalized rapid MRI reconstruction that leverages longitudinal priors. The main contributions of this work are: (1) utilizing a patient’s most recent prior scan as side information, (2) introducing a trust-guidance mechanism that restricts the influence of the prior to ensure it remains consistent with the current undersampled k-space measurements, and (3) designing a framework that successfully operates without a separate registration pipeline while tolerating protocol differences across acquisitions.
- 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.
Practical Clinical Utility: A major strength of L-TGVN is its ability to operate without a separate registration pipeline and its robustness to protocol differences across acquisitions. This makes the proposed method highly suitable for real-world clinical workflows, where acquiring perfectly aligned or protocol-matched longitudinal scans is rare.
Balanced Use of Priors: The trust-guidance mechanism is a strong conceptual approach. By intelligently restricting the influence of the longitudinal prior to remain consistent with current undersampled k-space measurements, the method effectively leverages side information while mitigating the risk of hallucinating outdated anatomical features.
- 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.
Need for Deeper Elaboration and Empirical Evidence on Trust Guidance: While the concept of Longitudinal Trust Guidance to guard against over-reliance is appealing, the current claims feel slightly overstated without sufficient elaboration and empirical backing. To convincingly demonstrate that this mechanism prevents over-reliance on the prior, the authors should provide both a deeper mechanistic explanation of how the guidance successfully operates and concrete empirical evidence.
Limited Methodological Novelty: Although the incorporation of longitudinal trust guidance is a practical contribution, the core architecture appears to be a straightforward extension of the existing TGVN (Reference 2). The authors should more clearly delineate the algorithmic innovations and unique technical contributions of L-TGVN compared to its predecessor.
Difficulty in Visual Assessment (Fig 2): Figure 2 illustrates the effect of using side information, but the visual improvements are difficult to recognize in certain slices. To better highlight the superiority of the proposed method, the authors should provide zoomed-in regions of interest or display residual images.
- 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.
(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?
Overall, this paper proposes a highly practical and clinically relevant approach (L-TGVN) for rapid MRI reconstruction. The ability to utilize longitudinal priors without requiring strict anatomical registration or identical imaging protocols is a significant strength that aligns perfectly with real-world clinical constraints.
While the clinical utility is strong and the proposed solution addresses a practical bottleneck in longitudinal MRI follow-ups, the major weaknesses of the study are the somewhat limited methodological novelty compared to the base TGVN framework and the need for stronger empirical evidence to fully support their claims regarding the trust-guidance mechanism.
- 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 R1 for the very positive assessment and constructive suggestions. Robustness to discrepancies. Thank you for raising this. In the single-coil regime the ambiguous subspace coincides exactly with the unacquired k-space lines, so trust guidance can only inject prior information into the missing lines, i.e., data consistency is preserved. For multi-coil reconstruction we follow the original TGVN and E2E-VarNet design and refine the sensitivity maps with a U-Net during reconstruction, which provides protection against gross sensitivity-map inaccuracies. Because trust guidance acts only inside the ambiguous space induced by the (refined) operator, L-TGVN inherits these safeguards and should remain non-inferior to a no-side-information baseline under degraded operators. We will add this discussion to the camera-ready. Selection of the singular-value threshold δ. The original TGVN work proposed an efficient conjugate-gradient inner loop for the trust-guidance projection, which makes δ fully differentiable; in our implementation δ is therefore a learned parameter of the network rather than a manually-tuned hyperparameter. We will state this explicitly in the camera-ready. Refinement of H. In our longitudinal setting, slice prescriptions of the current and prior scans can differ substantially and no explicit registration is performed. Conditioning H on the current iterate x^t provides an anatomical query that allows H to attend to the prior-volume slices most relevant to the evolving reconstruction — performing implicit, iterative slice selection / soft registration inside the network. The original H(s; γ^t) lacks this anchor, forcing a fixed prior-slice mapping ill-suited to protocol variability. Low-field MRI. We fully agree — this is an active follow-up direction; TGVN’s demonstrated robustness to contrast and low-SNR regimes supports the high-field-baseline and low-field-followup scenario.
We thank R2 for the supportive assessment. Baselines. MTrans and DMSI were chosen as the established open-source reconstruction architectures leveraging side information that could be adapted to our setting without non-trivial modifications — this remains an actively growing area. We agree that a simpler side-information conditioning mechanism would cleanly isolate the contribution of the trust-guidance projection; we will include such a baseline in the extended journal version of this work. Methodological novelty beyond TGVN. Our contributions specifically target the longitudinal setting: (i) tolerating slice-prescription and protocol differences without any explicit registration step, (ii) conditioning H on the current iterate x^t to enable implicit, anatomy-aware slice selection from the prior volume, and (iii) a multi-slice prior-conditioning strategy. Together these are what make the TGVN framework deployable in realistic longitudinal workflows.
We thank R3 for recognizing the practical clinical utility of L-TGVN. Empirical evidence on trust guidance. Ablations isolating the trust-guidance mechanism were reported in detail in the original TGVN work. Physically, this subspace contains image-domain components the forward operator constrains poorly — both patterns in unacquired k-space regions and aliased modes the coil geometry cannot disentangle (high g-factor directions of SENSE-style reconstruction). Trust guidance restricts the longitudinal prior’s influence to this subspace, regularizing only the ill-posed components while leaving well-conditioned ones to be determined by the data. Novelty. As detailed for R2, L-TGVN introduces critical longitudinal-specific innovations: (i) registration-free operation tolerating protocol differences, (ii) iterate-conditioned dynamic slice selection, and (iii) multi-slice priors. Together, these elevate the framework from a general concept to a clinically deployable longitudinal tool. Fig. 2.We will add zoomed-in ROIs in the camera-ready, per the half-page amendment policy.
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 acknowledge the significant practical contributions and strong empirical performance of this work.