Search-MIND: Training-Free Multi-Modal Medical Image Registration

Boya Wang; Chao Chen; Ruizhe Li; Xin Chen

arxiv: 2604.09743 · v1 · submitted 2026-04-10 · 📡 eess.IV · cs.CV

Search-MIND: Training-Free Multi-Modal Medical Image Registration

Boya Wang , Ruizhe Li , Chao Chen , Xin Chen This is my paper

Pith reviewed 2026-05-10 17:46 UTC · model grok-4.3

classification 📡 eess.IV cs.CV

keywords multi-modal image registrationtraining-free optimizationmutual informationMIND descriptorsdeformable registrationmedical imagingcoarse-to-fine alignment

0 comments

The pith

Search-MIND registers multi-modal medical images without training by optimizing two new loss functions in a coarse-to-fine pipeline.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes that instance-specific multi-modal registration can be solved reliably through iterative optimization rather than learned models. This matters because non-linear intensity differences and local optima often trap classical methods while deep networks collapse on unseen modality pairs. The approach first performs hierarchical coarse alignment then refines with deformable registration. Two custom losses drive the process: one weights informative tissue regions to reduce background interference and the other expands the search range for structural matches. Tests on liver and abdominal challenge datasets show the method exceeds both classical baselines and foundation-model alternatives in accuracy and stability.

Core claim

Search-MIND is a training-free framework that combines a hierarchical coarse alignment stage with deformable refinement. It employs Variance-Weighted Mutual Information to prioritize tissue regions over uniform background areas and Search-MIND to enlarge the local search range of structural descriptors, thereby widening the basin of convergence for multi-modal cases.

What carries the argument

Variance-Weighted Mutual Information (VWMI) and Search-MIND (S-MIND) loss functions, which respectively emphasize informative tissues and expand structural descriptor search ranges to stabilize optimization across intensity relationships.

Load-bearing premise

That VWMI and S-MIND broaden the convergence basin and shield alignment from background noise without introducing new biases or requiring undisclosed modality-specific parameter tuning.

What would settle it

Registration errors that remain higher than ANTs or DINO-reg on a held-out set of multi-modal scans with varied noise levels or intensity non-linearities would show the claimed gains do not generalize.

Figures

Figures reproduced from arXiv: 2604.09743 by Boya Wang, Chao Chen, Ruizhe Li, Xin Chen.

read the original abstract

Multi-modal image registration plays a critical role in precision medicine but faces challenges from non-linear intensity relationships and local optima. While deep learning models enable rapid inference, they often suffer from generalization collapse on unseen modalities. To address this, we propose Search-MIND, a training-free, iterative optimization framework for instance-specific registration. Our pipeline utilizes a coarse-to-fine strategy: a hierarchical coarse alignment stage followed by deformable refinement. We introduce two novel loss functions: Variance-Weighted Mutual Information (VWMI), which prioritizes informative tissue regions to shield global alignment from background noise and uniform regions, and Search-MIND (S-MIND), which broadens the convergence basin of structural descriptors by considering larger local search range. Evaluations on CARE Liver 2025 and CHAOS Challenge datasets show that Search-MIND consistently outperforms classical baselines like ANTs and foundation model-based approaches like DINO-reg, offering superior stability across diverse modalities.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Search-MIND is a training-free coarse-to-fine registration tweak that adds variance-weighted MI and enlarged-search MIND losses, but the abstract supplies no numbers to back the outperformance claim.

read the letter

The core idea is straightforward: run a hierarchical optimization that first does coarse alignment then refines with two new losses, VWMI to down-weight uniform background and S-MIND to widen the capture range of structural descriptors. That directly targets the generalization problem in multi-modal work where retraining a network for every new scanner or contrast is impractical. The approach stays instance-specific and avoids learned models, which is a clean fit for clinical settings that need something that just runs on new data pairs without GPU time or labeled training sets. The choice to build on established MI and MIND rather than invent an entirely new descriptor keeps the method interpretable and easy to implement on top of existing toolkits like ANTs. Credit for spelling out the coarse-to-fine schedule and the two explicit loss modifications; those are the concrete increments over the cited baselines. The evaluation is framed on CARE Liver 2025 and CHAOS, which are reasonable public benchmarks for liver and abdominal multi-modal tasks. The stress-test note is right that nothing in the loss definitions or protocol looks internally contradictory or circular. The soft spot is the complete absence of any quantitative results, error bars, or statistical tests in the abstract. Without those, the statement that it “consistently outperforms” ANTs and DINO-reg cannot be checked, and it is impossible to judge whether the gains are large enough to matter in practice or whether they come at the cost of longer run times. If the full manuscript has solid tables and ablation runs, that gap disappears; if it does not, the central claim stays unverified. This paper is for readers who already work on classical or hybrid registration and want a lightweight, no-training option for new modality pairs. It is not yet a must-read for the broader field until the numbers appear. I would send it to peer review because the problem is real and the method is simple enough that referees can quickly assess whether the reported improvements hold.

Referee Report

0 major / 3 minor

Summary. The manuscript proposes Search-MIND, a training-free, instance-specific iterative optimization framework for multi-modal medical image registration. It employs a hierarchical coarse-to-fine pipeline (coarse alignment followed by deformable refinement) and introduces two novel loss functions: Variance-Weighted Mutual Information (VWMI) to prioritize informative tissue regions and mitigate background noise, and Search-MIND (S-MIND) to expand the convergence basin of structural descriptors via larger local search ranges. Evaluations on the CARE Liver 2025 and CHAOS Challenge datasets are presented as demonstrating consistent outperformance over classical baselines such as ANTs and foundation-model approaches such as DINO-reg, with improved stability across modalities.

Significance. If the performance claims hold under rigorous scrutiny, the work offers a practical training-free alternative to both traditional optimization-based and learning-based registration techniques, particularly valuable in clinical settings with unseen modalities or limited training data. The instance-specific optimization and the design of VWMI and S-MIND to address noise and local-optima issues represent a meaningful extension of established mutual-information and MIND concepts.

minor comments (3)

The abstract asserts consistent outperformance but does not include any quantitative metrics, statistical significance tests, error bars, or implementation details of the hierarchical strategy and loss functions; adding these would strengthen the presentation of the central claim.
Clarify the precise mathematical definitions of VWMI and S-MIND (including any weighting parameters or search-range hyperparameters) in the methods section to support reproducibility and to allow readers to verify the claimed broadening of the convergence basin.
Ensure that all experimental results (tables or figures) report standard deviation or confidence intervals across multiple runs or folds, and include direct comparisons with the same initialization and stopping criteria used for ANTs and DINO-reg.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary, significance assessment, and recommendation of minor revision. The report contains no specific major comments to address.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper describes a training-free instance-specific optimization pipeline using coarse-to-fine alignment with two introduced loss functions (VWMI and S-MIND). No equations, predictions, or first-principles results are presented that reduce by construction to fitted parameters or self-referential inputs on the evaluation data. The method is explicitly framed as iterative optimization grounded in standard registration concepts, with performance claims based on direct comparisons to external baselines (ANTs, DINO-reg) on CARE Liver 2025 and CHAOS datasets. No self-citation chains, ansatz smuggling, or renaming of known results appear in the provided text as load-bearing steps. The derivation remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review performed on abstract only; no explicit free parameters, axioms, or invented entities are stated. The framework implicitly rests on standard assumptions of mutual-information optimization and local structural similarity that are not audited here.

pith-pipeline@v0.9.0 · 5457 in / 1225 out tokens · 62542 ms · 2026-05-10T17:46:54.647177+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We introduce two novel loss functions: Variance-Weighted Mutual Information (VWMI)... and Search-MIND (S-MIND), which broadens the convergence basin of structural descriptors by considering larger local search range.
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Evaluations on CARE Liver 2025 and CHAOS Challenge datasets show that Search-MIND consistently outperforms classical baselines like ANTs and foundation model-based approaches like DINO-reg

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages · 1 internal anchor

[1]

Neuroimage54(3), 2033–2044 (2011)

Avants, B.B., Tustison, N.J., Song, G., Cook, P.A., Klein, A., Gee, J.C.: A repro- ducibleevaluationofantssimilaritymetricperformanceinbrainimageregistration. Neuroimage54(3), 2033–2044 (2011)

work page 2033
[2]

IEEE transactions on medical imaging38(8), 1788–1800 (2019)

Balakrishnan, G., Zhao, A., Sabuncu, M.R., Guttag, J., Dalca, A.V.: Voxelmorph: a learning framework for deformable medical image registration. IEEE transactions on medical imaging38(8), 1788–1800 (2019)

work page 2019
[3]

Medical image analysis82, 102615 (2022)

Chen, J., Frey, E.C., He, Y., Segars, W.P., Li, Y., Du, Y.: Transmorph: Transformer for unsupervised medical image registration. Medical image analysis82, 102615 (2022)

work page 2022
[4]

Medical image analysis16(7), 1423–1435 (2012)

Heinrich, M.P., Jenkinson, M., Bhushan, M., Matin, T., Gleeson, F.V., Brady, M., Schnabel, J.A.: Mind: Modality independent neighbourhood descriptor for multi- modal deformable registration. Medical image analysis16(7), 1423–1435 (2012)

work page 2012
[5]

Physics in medicine & biology46(3), R1–R45 (2001)

Hill, D.L., Batchelor, P.G., Holden, M., Hawkes, D.J.: Medical image registration. Physics in medicine & biology46(3), R1–R45 (2001)

work page 2001
[6]

arXiv preprint arXiv:2410.14083 (2024)

Huang, S., Xu, T., Shen, Z., Saeed, S.U., Yan, W., Barratt, D., Hu, Y.: Sam- reg: Sam-enabled image registration with roi-based correspondence. arXiv preprint arXiv:2410.14083 (2024)

work page arXiv 2024
[7]

Advances in Neural Information Processing Systems36, 21285–21297 (2023)

Jiang, H., Salzmann, M., Dang, Z., Xie, J., Yang, J.: Se (3) diffusion model-based point cloud registration for robust 6d object pose estimation. Advances in Neural Information Processing Systems36, 21285–21297 (2023)

work page 2023
[8]

Medical image analysis69, 101950 (2021)

Kavur, A.E., Gezer, N.S., Barış, M., Aslan, S., Conze, P.H., Groza, V., Pham, D.D., Chatterjee, S., Ernst, P., Özkan, S., et al.: Chaos challenge-combined (ct- mr) healthy abdominal organ segmentation. Medical image analysis69, 101950 (2021)

work page 2021
[9]

In: Proceedings of the IEEE/CVF international conference on computer vision

Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4015–4026 (2023)

work page 2023
[10]

Neuroimage 46(3), 786–802 (2009)

Klein, A., Andersson, J., Ardekani, B.A., Ashburner, J., Avants, B., Chiang, M.C., Christensen, G.E., Collins, D.L., Gee, J., Hellier, P., et al.: Evaluation of 14 nonlin- ear deformation algorithms applied to human brain mri registration. Neuroimage 46(3), 786–802 (2009)

work page 2009
[11]

In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI)

Li, R., Figueredo, G., Auer, D., Wagner, C., Chen, X.: Mrregnet: Multi-resolution mask guided convolutional neural network for medical image registration with large deformations. In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI). pp. 1–5. IEEE (2024) 10 Boya Wang et al

work page 2024
[12]

Medical Image Analysis102, 103507 (2025)

Liu, Y., Gao, Z., Shi, N., Wu, F., Shi, Y., Chen, Q., Zhuang, X.: Merit: Multi- view evidential learning for reliable and interpretable liver fibrosis staging. Medical Image Analysis102, 103507 (2025)

work page 2025
[13]

Proceedings of the IEEE91(10), 1699–1722 (2003)

Maes, F., Vandermeulen, D., Suetens, P.: Medical image registration using mutual information. Proceedings of the IEEE91(10), 1699–1722 (2003)

work page 2003
[14]

Medical image analysis2(1), 1–36 (1998)

Maintz, J.A., Viergever, M.A.: A survey of medical image registration. Medical image analysis2(1), 1–36 (1998)

work page 1998
[15]

Computer methods and programs in biomedicine98(3), 278–284 (2010)

Modat, M., Ridgway, G.R., Taylor, Z.A., Lehmann, M., Barnes, J., Hawkes, D.J., Fox, N.C., Ourselin, S.: Fast free-form deformation using graphics processing units. Computer methods and programs in biomedicine98(3), 278–284 (2010)

work page 2010
[16]

DINOv2: Learning Robust Visual Features without Supervision

Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., et al.: Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193 (2023)

work page internal anchor Pith review Pith/arXiv arXiv 2023
[17]

In: International Con- ference on Medical Image Computing and Computer-Assisted Intervention

Song, X., Xu, X., Yan, P.: Dino-reg: General purpose image encoder for training- free multi-modal deformable medical image registration. In: International Con- ference on Medical Image Computing and Computer-Assisted Intervention. pp. 608–617. Springer (2024)

work page 2024
[18]

IEEE transactions on medical imaging32(7), 1153–1190 (2013)

Sotiras, A., Davatzikos, C., Paragios, N.: Deformable medical image registration: A survey. IEEE transactions on medical imaging32(7), 1153–1190 (2013)

work page 2013
[19]

Physics in Medicine & Biology50(12), 2887–2905 (2005)

Wang, H., Dong, L., O’Daniel, J., Mohan, R., Garden, A.S., Ang, K.K., Kuban, D.A., Bonnen, M., Chang, J.Y., Cheung, R.: Validation of an accelerated ‘demons’ algorithm for deformable image registration in radiation therapy. Physics in Medicine & Biology50(12), 2887–2905 (2005)

work page 2005
[20]

Medical image analysis 1(1), 35–51 (1996)

Wells III, W.M., Viola, P., Atsumi, H., Nakajima, S., Kikinis, R.: Multi-modal vol- ume registration by maximization of mutual information. Medical image analysis 1(1), 35–51 (1996)

work page 1996

[1] [1]

Neuroimage54(3), 2033–2044 (2011)

Avants, B.B., Tustison, N.J., Song, G., Cook, P.A., Klein, A., Gee, J.C.: A repro- ducibleevaluationofantssimilaritymetricperformanceinbrainimageregistration. Neuroimage54(3), 2033–2044 (2011)

work page 2033

[2] [2]

IEEE transactions on medical imaging38(8), 1788–1800 (2019)

Balakrishnan, G., Zhao, A., Sabuncu, M.R., Guttag, J., Dalca, A.V.: Voxelmorph: a learning framework for deformable medical image registration. IEEE transactions on medical imaging38(8), 1788–1800 (2019)

work page 2019

[3] [3]

Medical image analysis82, 102615 (2022)

Chen, J., Frey, E.C., He, Y., Segars, W.P., Li, Y., Du, Y.: Transmorph: Transformer for unsupervised medical image registration. Medical image analysis82, 102615 (2022)

work page 2022

[4] [4]

Medical image analysis16(7), 1423–1435 (2012)

Heinrich, M.P., Jenkinson, M., Bhushan, M., Matin, T., Gleeson, F.V., Brady, M., Schnabel, J.A.: Mind: Modality independent neighbourhood descriptor for multi- modal deformable registration. Medical image analysis16(7), 1423–1435 (2012)

work page 2012

[5] [5]

Physics in medicine & biology46(3), R1–R45 (2001)

Hill, D.L., Batchelor, P.G., Holden, M., Hawkes, D.J.: Medical image registration. Physics in medicine & biology46(3), R1–R45 (2001)

work page 2001

[6] [6]

arXiv preprint arXiv:2410.14083 (2024)

Huang, S., Xu, T., Shen, Z., Saeed, S.U., Yan, W., Barratt, D., Hu, Y.: Sam- reg: Sam-enabled image registration with roi-based correspondence. arXiv preprint arXiv:2410.14083 (2024)

work page arXiv 2024

[7] [7]

Advances in Neural Information Processing Systems36, 21285–21297 (2023)

Jiang, H., Salzmann, M., Dang, Z., Xie, J., Yang, J.: Se (3) diffusion model-based point cloud registration for robust 6d object pose estimation. Advances in Neural Information Processing Systems36, 21285–21297 (2023)

work page 2023

[8] [8]

Medical image analysis69, 101950 (2021)

Kavur, A.E., Gezer, N.S., Barış, M., Aslan, S., Conze, P.H., Groza, V., Pham, D.D., Chatterjee, S., Ernst, P., Özkan, S., et al.: Chaos challenge-combined (ct- mr) healthy abdominal organ segmentation. Medical image analysis69, 101950 (2021)

work page 2021

[9] [9]

In: Proceedings of the IEEE/CVF international conference on computer vision

Kirillov, A., Mintun, E., Ravi, N., Mao, H., Rolland, C., Gustafson, L., Xiao, T., Whitehead, S., Berg, A.C., Lo, W.Y., et al.: Segment anything. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 4015–4026 (2023)

work page 2023

[10] [10]

Neuroimage 46(3), 786–802 (2009)

Klein, A., Andersson, J., Ardekani, B.A., Ashburner, J., Avants, B., Chiang, M.C., Christensen, G.E., Collins, D.L., Gee, J., Hellier, P., et al.: Evaluation of 14 nonlin- ear deformation algorithms applied to human brain mri registration. Neuroimage 46(3), 786–802 (2009)

work page 2009

[11] [11]

In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI)

Li, R., Figueredo, G., Auer, D., Wagner, C., Chen, X.: Mrregnet: Multi-resolution mask guided convolutional neural network for medical image registration with large deformations. In: 2024 IEEE International Symposium on Biomedical Imaging (ISBI). pp. 1–5. IEEE (2024) 10 Boya Wang et al

work page 2024

[12] [12]

Medical Image Analysis102, 103507 (2025)

Liu, Y., Gao, Z., Shi, N., Wu, F., Shi, Y., Chen, Q., Zhuang, X.: Merit: Multi- view evidential learning for reliable and interpretable liver fibrosis staging. Medical Image Analysis102, 103507 (2025)

work page 2025

[13] [13]

Proceedings of the IEEE91(10), 1699–1722 (2003)

Maes, F., Vandermeulen, D., Suetens, P.: Medical image registration using mutual information. Proceedings of the IEEE91(10), 1699–1722 (2003)

work page 2003

[14] [14]

Medical image analysis2(1), 1–36 (1998)

Maintz, J.A., Viergever, M.A.: A survey of medical image registration. Medical image analysis2(1), 1–36 (1998)

work page 1998

[15] [15]

Computer methods and programs in biomedicine98(3), 278–284 (2010)

Modat, M., Ridgway, G.R., Taylor, Z.A., Lehmann, M., Barnes, J., Hawkes, D.J., Fox, N.C., Ourselin, S.: Fast free-form deformation using graphics processing units. Computer methods and programs in biomedicine98(3), 278–284 (2010)

work page 2010

[16] [16]

DINOv2: Learning Robust Visual Features without Supervision

Oquab, M., Darcet, T., Moutakanni, T., Vo, H., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., et al.: Dinov2: Learning robust visual features without supervision. arXiv preprint arXiv:2304.07193 (2023)

work page internal anchor Pith review Pith/arXiv arXiv 2023

[17] [17]

In: International Con- ference on Medical Image Computing and Computer-Assisted Intervention

Song, X., Xu, X., Yan, P.: Dino-reg: General purpose image encoder for training- free multi-modal deformable medical image registration. In: International Con- ference on Medical Image Computing and Computer-Assisted Intervention. pp. 608–617. Springer (2024)

work page 2024

[18] [18]

IEEE transactions on medical imaging32(7), 1153–1190 (2013)

Sotiras, A., Davatzikos, C., Paragios, N.: Deformable medical image registration: A survey. IEEE transactions on medical imaging32(7), 1153–1190 (2013)

work page 2013

[19] [19]

Physics in Medicine & Biology50(12), 2887–2905 (2005)

Wang, H., Dong, L., O’Daniel, J., Mohan, R., Garden, A.S., Ang, K.K., Kuban, D.A., Bonnen, M., Chang, J.Y., Cheung, R.: Validation of an accelerated ‘demons’ algorithm for deformable image registration in radiation therapy. Physics in Medicine & Biology50(12), 2887–2905 (2005)

work page 2005

[20] [20]

Medical image analysis 1(1), 35–51 (1996)

Wells III, W.M., Viola, P., Atsumi, H., Nakajima, S., Kikinis, R.: Multi-modal vol- ume registration by maximization of mutual information. Medical image analysis 1(1), 35–51 (1996)

work page 1996