Abstract

Accurate modeling of the slice selection profile is essential for super-resolution and resolution quantification in 2D-acquired MRI, where slice thickness and spacing are independent. Assuming that the slice spacing is equal to the slice thickness leads to systematic errors, yet estimating slice thickness from reconstructed images is ill-posed. We propose PRISM (Profile Recovery via Invariant Spectral Matching), a contrast-invariant method that estimates slice thickness by matching whole-volume spectral characteristics between a low-resolution acquisition and a degraded high-resolution reference. The reference image need not be the same contrast or subject and does not need registration nor voxel-wise correspondence to the acquired image. PRISM applies the rank transform to suppress contrast differences and compares the average 1D log power spectra along the slice axis. The thickness parameter is recovered via a 1D search without gradient-based optimization. Across controlled simulations, ablations, downstream super-resolution experiments, and real 2D-acquired data, PRISM accurately estimates slice thickness (within 0.09 mm for 3 mm data and within 0.64 mm for 10 mm data) and outperforms existing kernel estimation approaches, yielding reconstruction performance statistically indistinguishable from using the true slice profile. Code is available at https://github.com/sremedios/PRISM.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1956_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/sremedios/PRISM

Link to the Dataset(s)

OASIS-3 dataset: https://sites.wustl.edu/oasisbrains/home/oasis-3/ HCP-1200 dataset: https://www.humanconnectome.org/study/hcp-young-adult/document/1200-subjects-data-release

BibTex

@InProceedings{RemSam_ContrastInvariant_MICCAI2026,
        author = { Remedios, Samuel W. AND Wei, Shuwen AND Carass, Aaron AND Prince, Jerry L. AND Dewey, Blake E.},
        title = { { Contrast-Invariant Reference-Based Slice Thickness Estimation in MRI via Spectral Matching } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16896},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposes a simple yet efficient method named PRISM for accurately estimating the slice thickness of thick-slice MRI scans, independent of imaging contrast. The method minimizes the difference between power spectra via rank transformation and an exhaustive search strategy. Experimental results demonstrate the effectiveness of the proposed approach.

  • 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 proposed method is simple but highly effective, applicable across arbitrary imaging contrasts, and capable of utilizing isotropic imaging data from different subjects at varying resolutions for comparison. (2) The experimental setup is comprehensive, reporting necessary statistical significance tests and runtime information. (3) The effectiveness of the proposed method is validated across multiple datasets.

  • 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) In actual acquisitions, a certain degree of oblique scanning may occur. The paper needs to further discuss the applicability of the method under such conditions. (2) The paper should further explain and investigate whether the proposed method remains robust in the presence of noise interference or motion artifacts. (3) It remains unclear whether the proposed method is sensitive to the choice of reference data. For instance, the impact of using different reference HR volumes on the estimated thick-slice thickness should be explored.

  • 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

    (1) Beyond brain MRI, the theoretical applicability of this method to slice thickness estimation in MRI of other anatomical regions could be further discussed.

  • 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 method proposed in this paper is simple yet efficient, and the writing is clear. This method holds promise for improving existing super-resolution (SR) frameworks. However, a more in-depth discussion regarding the practical utility of the method is required to ensure its contribution to the field.

  • 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.

    Authors sufficiently address the points raised in the reviews.



Review #2

  • Please describe the contribution of the paper

    This paper proposes “PRISM,” a contrast-invariant method that estimates slice thickness by matching whole-volume spectral characteristics between a low-resolution acquisition and a degraded high-resolution reference. The main contributions of this work are: (1) proposing a frequency-domain formulation for slice thickness estimation based on invariant spectral attenuation, (2) enabling accurate slice thickness estimation without the need for anatomical alignment or paired training data, and (3) demonstrating improved performance in downstream super-resolution tasks.

  • 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.

    Contrast-Invariant Structural Representation: The use of the rank transform makes PRISM highly robust to global contrast changes. It addresses the practical variability often encountered in clinically acquired medical images.

    Flexibility and Efficiency: The method successfully estimates slice thickness without requiring anatomical alignment, paired data, or iterative optimization. This greatly enhances the practical applicability and efficiency of the proposed method compared to existing supervised or alignment-dependent approaches.

  • 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.

    Difficulty in Visual Assessment (Fig 3): While Figure 3 demonstrates the results of the downstream super-resolution experiment, it is challenging to visually recognize the improvement in the provided images. The authors could better highlight the superiority of their method by providing error maps or zoomed-in residual images.

    Limitation in Evaluation Metrics: The paper primarily relies on PSNR and SSIM to quantify the results. However, considering the downstream task is super-resolution, it would be highly beneficial to evaluate or at least discuss the method’s ability to restore high-frequency components. Exploring metrics such as HFEN or LPIPS would make the evaluation much stronger. It would be helpful if the authors could discuss their rationale for metric selection or provide insights on high-frequency restoration in the rebuttal.

    Sensitivity of Rank Transform to Local Contrast: As briefly mentioned by the authors, while the rank transform is robust to global contrast differences, it can be sensitive to local contrast variations. For instance, in complex brain regions (e.g., near the gray matter and white matter boundary), the rank-transformed signal might experience a reversal in order between T1- and T2-weighted images. To fully justify the claim that the rank transform is more robust than other transformations, it would be beneficial for the authors to either provide visual examples of rank-transformed T1 and T2 images or provide a deeper discussion on how this local sensitivity impacts the overall performance.

    Limited Scope of Prospective Evaluation: The authors present a prospective result demonstrating that PRISM estimates a value closer to the target compared to other baseline techniques. While this is a positive and promising indicator, validating the method on such a specific, limited configuration is insufficient to fully demonstrate its prospective clinical utility and robustness. To convincingly support its clinical relevance, the evaluation should encompass a broader range of practical scenarios, or the authors should at least provide a deeper discussion or downstream super-resolution image.

  • 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 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.

    (3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    This paper proposes PRISM, an interesting and contrast-invariant approach for slice thickness estimation. The fact that the method circumvents the need for paired data or anatomical alignment is a clear methodological strength, and the core formulation using the rank transform is notable.

    However, there are concerns regarding the depth of its empirical validation and the robustness of its core formulation. While the authors claim contrast invariance through the rank transform, this property requires deeper consideration and discussion, particularly regarding its sensitivity to local contrast inversions at tissue boundaries. Furthermore, while the quantitative metrics such as PSNR and SSIM are promising, additional metrics such as HFEN would be required to fully validate the downstream performance.

    While the paper includes a prospective experiment showing improved thickness estimation, for a method aiming to solve practical bottlenecks, a more convincing demonstration of its prospective clinical utility is necessary.

    I believe the core idea has great potential, but the current manuscript requires stronger empirical validation and a clearer demonstration of its practical benefits. I look forward to the authors’ insights and further discussion on these points during the rebuttal.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.

    Reject

  • [Post rebuttal] Please justify your final decision from above.

    The rebuttal addresses my concerns at the surface level but does not fully resolve two issues.

    First, on the rank transform: Table 1 convincingly shows that rank-transform plus spectral-power-error dominates, but the rebuttal does not explain why structurally. Recasting the rank transform as “only a feature representation” also somewhat weakens the paper’s central framing and the novelty of using the rank transform. The dominance is empirically supported but not mechanistically justified.

    Second, on the SR evaluation: the authors frame it as a controlled forward-model ablation, which is partially reasonable. However, PSNR/SSIM are weak indicators of high-frequency fidelity.

    The core idea remains interesting and practically valuable, and the results are consistent across multiple datasets and settings. My remaining concerns are about depth of justification rather than correctness, and I would not mind if the paper were accepted.



Review #3

  • Please describe the contribution of the paper

    The authors propose PRISM, a contrast-invariant method designed to estimate slice thickness in 2D multi-slice MRI acquisitions. The approach matches the 1D log power spectrum of a low-resolution target volume to a high-resolution reference volume along the through-plane axis. To address potential contrast mismatches between the target and reference, the method applies a rank transform before spectral matching. The slice thickness parameter is recovered via an exhaustive 1D search over candidate Gaussian kernel standard deviations.

  • 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.

    Estimating slice thickness blindly is a severely ill-posed problem. The proposed method avoids the need for exact anatomical alignment or paired data by operating on whole-volume spectral statistics. This increases the method’s practical applicability across different subjects and contrasts. The use of the rank transform prior to spectral matching is a logical design choice. It suppresses global intensity and contrast variations while preserving the resolution-sensitive local orderings necessary to capture the blur signature. Furthermore, by framing the thickness estimation as a 1D exhaustive search over predefined standard deviations, the method sidesteps the convergence instability, degenerate solutions, and local minima that frequently plague gradient-based blind deconvolution algorithms. It is also computationally efficient, processing a full 3D brain volume in approximately 10 seconds.

  • 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 forward model explicitly parameterizes the continuous slice selection profile as a unit-area Gaussian. While this is a standard approximation in MRI super-resolution literature, true slice profiles—which are often generated by truncated sinc or SLR RF pulses—exhibit side lobes or flatter passbands. The paper would benefit from a simulation experiment evaluating the method’s robustness when the underlying ground-truth degradation deviates from a perfect Gaussian shape. Regarding the downstream super-resolution validation in Section 3.4, the experiments are performed exclusively using a specific pre-trained denoising diffusion null-space model. While the authors report that PRISM achieves PSNR and SSIM statistically indistinguishable from the true kernel, it remains unclear if the estimated kernels provide consistent improvements across other, more standard SR architectures, such as basic CNNs or TV-regularized reconstruction. Demonstrating that the estimated kernel acts as a consistent improvement across different solvers would strengthen the evaluation. In the real-world evaluation on OASIS-3 data (Section 3.5), the method estimates a mean thickness of 4.23 mm for data with a nominal 4 mm thickness. The authors hypothesize that this is due to PRISM capturing “additional contributions to the through-plane point spread function” relative to the reference. However, without an objective ground truth, this remains speculative. Including a physical phantom experiment where the true slice profile can be precisely measured would validate whether the method is accurately capturing scanner physics or merely exhibiting a systematic bias.

  • 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.

    (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?

    This paper addresses the problem of slice thickness estimation in 2D MRI. The proposed PRISM method is mathematically sound, computationally efficient, and avoids the pitfalls of explicit registration by leveraging invariant spectral matching. The ablation study supports the algorithmic design choices (rank transform and SPE) , and the downstream super-resolution results show improvement over baselines. However, the validation relies heavily on the assumption of ideal Gaussian degradation and a specific diffusion prior. Addressing the limits of the Gaussian assumption and generalizing the SR baseline in a minor revision would improve 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.

    Accept

  • [Post rebuttal] Please justify your final decision from above.

    I recommend accepting this paper because the authors convincingly addressed the reviewers’ main concerns in their rebuttal. The proposed PRISM method provides an elegant, contrast-invariant, and registration-free framework for MRI slice thickness estimation by leveraging the rank transform and 1D spectral matching. It effectively avoids the instability of gradient-based blind deconvolution through a highly efficient 1D search. In the rebuttal, the authors successfully clarified that their super-resolution experiment was a controlled forward-model ablation, and they reasonably explained that the slight thickness overestimation in real-world data usefully captures “effective through-plane blur.” While the Gaussian profile assumption remains a limitation, it is an acceptable standard approximation for this scope.



Author Feedback

We thank the reviewers and AC for their constructive feedback. We address the main concerns below. Robustness and applicability (R1, AC). While we did not sweep multiple reference images, the experiments intentionally use a difficult reference setting: T2-w target volumes are compared with held-out T1-w references from different subjects, across multiple slice-thickness/gap settings. Despite mismatch in subject, contrast, dataset, and resolution, PRISM remains close to the true thickness and outperforms the baselines. This supports the claim that PRISM does not require matched or registered references. For oblique acquisitions, PRISM should be applied along the physical slice-selection direction from acquisition orientation metadata. Since the comparison uses volume-level spectra, this does not require registration. We will clarify that the HR reference should have comparable directional/anatomical spectral statistics, and that severe motion, slice dropout, or interpolation artifacts may bias spectral estimation. Rank-transform assumption and spectral matching (R2, AC). We appreciate the reviewer’s point that T1/T2 contrast differences can cause local tissue-order changes. Our claim is that the rank transform is invariant to monotonic intensity transformations and robust to global contrast differences, not invariant to every local contrast reversal. In PRISM, the rank transform is only the feature representation; the final comparison uses averaged 1D log spectra along the slice direction over the full in-plane field of view. Local ordering changes may alter the baseline feature spectrum, but are unlikely to mimic the through-plane high-frequency attenuation caused by thicker slices. This is directly tested by the cross-subject, cross-contrast experiments, where PRISM remains accurate across slice settings. Table 1 further supports this design choice: rank transform with spectral power error is consistently best or statistically indistinguishable from best, while alternative features/metrics are less stable. We will clarify this scope of contrast invariance. Downstream SR, metrics, and Fig. 3 (R2, R4, AC). We agree that Fig. 3 is visually dense. We will revise it using the existing reconstructions, adding zoomed insets around resolution-sensitive regions. The SR experiment was designed as a controlled forward-model ablation. The prior, weights, and random seeds are fixed; only the degradation model changes. Under this comparison, PRISM achieves PSNR/SSIM statistically indistinguishable from using the true kernel, whereas the naive and ESPRESO kernels do not. This supports the paper’s claim that PRISM can recover the benefit of the true forward model in downstream SR. Additional solvers or metrics are useful follow-up work, but they are additional experiments and are not permitted in the rebuttal. We will clarify that the SR result evaluates the estimated slice profile, rather than benchmarking SR architectures or perceptual metrics. Real-world evidence and Gaussian model (R2, R4, AC). We agree that the OASIS-3 real-data experiment is not phantom validation of the physical RF slice-selection waveform. Its purpose is to test behavior on real 2D acquisitions with known nominal thickness and gaps. PRISM estimates 4.23±0.05 mm for nominal 4 mm data, closer to metadata than ESPRESO or the spacing-equals-thickness assumption. The slight overestimate is consistent with PRISM estimating effective through-plane blur relative to the HR reference, including contributions beyond the nominal RF slice profile. Similarly, the Gaussian model is used as a compact one-parameter effective FWHM approximation, common in MRI SR modeling, not as a claim that all RF slice profiles are exactly Gaussian. The framework could be extended to non-Gaussian or measured profile dictionaries, which we will note as future work. All experiments use publicly available, de-identified datasets, and no new human-subject data were collected; we will make this explicit.




Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • 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.

    This paper proposes PRISM, a simple and efficient contrast-invariant method for estimating slice thickness in thick-slice MRI by matching spectral characteristics between a low-resolution scan and a degraded high-resolution reference after rank transformation. The reviewers generally agree that this is a meaningful methodological contribution with practical potential, highlighting in particular that the method avoids the need for paired data or anatomical alignment, is computationally efficient, and may be useful for improving downstream super-resolution pipelines. At the same time, the current manuscript still requires clearer validation and discussion before it can be accepted. The main concerns focus on robustness under more realistic acquisition conditions, the assumptions behind the rank-transform formulation, and the strength of the downstream and prospective evidence. These issues do not appear fundamental and seem potentially addressable through clarification and better interpretation of the existing results. I therefore recommend invite for rebuttal. In the rebuttal, the authors should focus on clarifying the robustness and applicability of PRISM, discussing the limitations of the rank-transform assumption more explicitly, and better justifying the downstream and prospective validation.

  • 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.

    This paper proposes PRISM, a contrast-invariant reference-based slice thickness estimation method for MRI, aiming to estimate the effective slice thickness of thick-slice MRI scans by matching spectral characteristics between a low-resolution target volume and a degraded high-resolution reference. The method uses rank transformation and 1D spectral matching to avoid the need for paired data, anatomical alignment, or supervised training.

    The reviewers recognized that the proposed method is simple, efficient, and practically useful. In particular, they appreciated its contrast-invariant formulation, registration-free design, computational efficiency, and potential utility for improving downstream super-resolution pipelines. The experiments across multiple datasets, slice-thickness settings, and downstream SR evaluation were also considered supportive of the main claims.

    The reviewers raised concerns about the robustness of the rank-transform assumption, sensitivity to reference data, applicability under realistic acquisition conditions such as oblique scans, noise, motion, and non-Gaussian slice profiles, as well as the limited scope of downstream and prospective validation. In the rebuttal, the authors clarified that PRISM is intended to be robust to global contrast differences rather than all local contrast reversals, and that the rank transform serves as a feature representation before whole-volume spectral matching. They also explained how PRISM can be applied along the physical slice-selection direction for oblique acquisitions, clarified the interpretation of the real-world OASIS-3 result as estimating effective through-plane blur, and justified the downstream SR experiment as a controlled forward-model ablation rather than a full SR benchmark.

    Although some limitations remain, including the Gaussian slice-profile approximation, limited prospective validation, and lack of broader SR solvers or perceptual high-frequency metrics, these issues mainly define the scope of the current work rather than undermining the core contribution. The method is mathematically sound, computationally efficient, and empirically consistent across the evaluated settings. Therefore, I recommend acceptance.



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.

    Based on the positive post-rebuttal consensus where reviewers found their concerns sufficiently addressed, this paper is recommended for acceptance.



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.

    The authors convincingly addressed concerns by discussing limitations in formulation and applicability. I also find the concerns of R2 about limited evaluation scope to be unfounded, since the authors evaluate they method against relevant baselines in direct comparisons as well as for downstream SR scenarios, which is sufficient and meets the standards of the MICCAI conference. Overall, I recommend acceptance for this paper, since this mathematically sound method is novel and well-motivated and shows promising results in a thorough experimental setup.



back to top