Abstract

High-Intensity Focused Ultrasound (HIFU) is a non-invasive therapy, yet its safety is often degraded by severe acoustic interference during continuous ultrasound guidance. Conventional HIFU interference suppression methods heavily rely on proprietary raw Radio-Frequency (RF) data or complex hardware synchronization, limiting their clinical utility and preventing real-time implementation. To address this limitation, we propose Manifold-Constrained Hyper-Connections Diffusion (mHC-Diff), an image-domain diffusion framework for real-time interference suppression without specialized hardware synchronization, disentangling complex interference from anatomical structures while ensuring high reconstruction fidelity. To achieve clinical real-time application, our approach employs a two-stage strategy: (i) anatomy-aware prior acquisition, where a diffusion model is trained with multi-step UNet as a highfidelity Teacher; and (ii) efficiency distillation, where this prior is distilled into a one-step Student via knowledge distillation to achieve real-time throughput. Extensive validation on a clinically representative dataset across diverse therapeutic scenarios shows that mHC-Diff achieves superior restoration (26.65 dB PSNR), while enabling real-time inference (~20 FPS) on a single NVIDIA RTX 4090, providing a ~6.8x speedup over iterative diffusion baselines (e.g., HIFU-Diff). By eliminating the requirement for specialized hardware synchronization and proprietary RF access, this image-domain framework ensures compatibility and facilitates real-time interference suppression during ultrasound-guided HIFU interventions. The code will be released upon acceptance.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2888_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/caidejia/HIFU-mHC-Diff

Link to the Dataset(s)

N/A

BibTex

@InProceedings{CaiDej_RealTime_MICCAI2026,
        author = { Cai, Dejia AND Abdollahi, Ali AND Wang, Xi AND Yang, Kun AND Guo, Zhaohui AND Zhou, Xiaowei AND Chen, Hao},
        title = { { Real-Time Hardware-Free HIFU Interference Suppression via Teacher-Student Diffusion Framework } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16893},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper presents a deep learning-based method for suppressing interference during monitoring of high-intensity focused ultrasound (HIFU) treatment. While prior studies have addressed this problem, the key distinction of this work lies in the input data domain. Previous approaches typically relied on raw RF signals, whereas this study utilizes image-domain data, which is more practical for clinical applications given the limited access to raw data in commercial systems. Specifically, the authors employ a Distribution Matching Distillation (DMD)-based one-step diffusion model, combined with manifold-constrained hyper-connections (mHC), to enable real-time performance while enhancing model capacity.

  • 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.

    Methodological novelty and comprehensive validation: In contrast to many previous studies that address HIFU interference suppression in the raw RF domain, the proposed method operates in the image domain, making it more practical for clinical deployment. To achieve high-quality output, the authors employ a diffusion model. For real-time performance, the trained diffusion model serves as a teacher, and a one-step diffusion model is trained using a Distribution Matching Distillation (DMD) approach introduced at CVPR 2024.Additionally, a manifold-constrained hyper-connection (mHC) module is incorporated to enhance model capacity.

    The proposed method is comprehensively validated across various scenarios, including different beamforming strategies, HIFU power levels, and targets (ex vivo and in vivo), and is compared against both traditional non-learning-based methods and prior deep learning approaches. The results demonstrate significant performance improvements over the compared 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.

    Minor clarity issue: I wonder whether the proposed method has been validated on unseen domains. Specifically, it is unclear whether the model was trained on one domain (e. g. , ex vivo data) and tested on another (e. g. , in vivo data), which would better demonstrate its generalization capability.

    Additionally, I am concerned about the potential for data leakage between the training and test sets. Since multiple images (e. g. , 50 images) are acquired within a single session, there is a possibility that highly similar frames could appear in both sets. Clarifying how the data were split—particularly whether separation was performed at the session level—would help ensure a fair evaluation.

  • 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 mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?

    N/A

  • Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html

    It is notable that SwinUNet demonstrates very strong performance compared to previously developed models for HIFU interference suppression (e. g. , FUSNet, HIFU-Diff). It would be valuable for the authors to discuss why SwinUNet, which is not specifically designed for this task, achieves such competitive results, in some cases comparable to the proposed method.

    Additionally, clarification on the training setup for SwinUNet would be helpful. Specifically, was it trained in a manner similar to FUSNet, or was it trained end-to-end in the image domain using a standard loss function such as MSE? Understanding this would provide better insight into the fairness of the comparison and the relative strengths of each approach.

  • 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 proposed method combines two techniques (DMD-based one-step diffusion and mHC) for HIFU interference suppression. It is impressive that the method is able to reconstruct clear images from distorted inputs in real time. Moreover, the approach is validated through comprehensive experiments across various scenarios and compared with state-of-the-art methods. However, there is a possibility that this work may be perceived as a combination of existing techniques originally developed in other domains, adapted for medical imaging.

  • 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 #2

  • Please describe the contribution of the paper

    The paper proposes a two-stage teacher–student diffusion framework (mHC-Diff) for real-time suppression of HIFU interference in B-mode ultrasound images, without requiring raw RF data or specialized hardware synchronization. The method achieves high-fidelity lesion reconstruction across multiple imaging modes and power levels, validated on a large-scale dataset from ex vivo and in vivo experiments.

  • 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.Real-time hardware-free HIFU interference suppression without requiring raw RF data. 2.Experimental validation across diverse imaging modes and power levels. 3.Clear ablation clearlyto show the contribution of each component.

  • 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.The paper does not provide sufficient information about the ultrasound acquisition setup (e.g., transducer parameters, imaging settings, or modality-specific details), which makes reproduction difficult. 2.The paper mentions constructing a large-scale dataset as its contribution, but gives almost no details about its release. If the dataset is not publicly available, the contribution here is quite limited. 3.There are some small typos (“HFIU” in Fig. 2).

  • 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.

    (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, the reviewer thinks this is a solid and practical contribution. The idea of applying a diffusion-based framework for near real-time HIFU interference suppression is interesting, and the experimental results are convincing. That said, the paper would benefit from clearer descriptions of the imaging setup and dataset to improve reproducibility and transparency.

  • 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 proposes mHC-Diff, a two-stage teacher-student diffusion framework for real-time, hardware-free HIFU interference suppression in ultrasound imaging. The method first learns an anatomy-aware prior with a diffusion-based teacher and then distills it into a one-step student for fast inference. It also introduces an mHC-UNet backbone to improve interference-anatomy disentanglement and validates the approach on a large paired dataset collected under multiple imaging modes and HIFU power levels.

  • 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 an important and clinically relevant problem, since real-time interference suppression without RF access or hardware synchronization would be valuable for practical HIFU guidance. The method is well motivated, and the two-stage design achieves a reasonable balance between restoration quality and inference speed. The paper also includes a relatively large dataset, comparisons with several baselines, and clear quantitative and qualitative improvements over prior 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.

    The main weakness is that the necessity of the diffusion-based teacher is not fully justified, since the final deployed model is the one-step student, and it remains unclear whether a simpler restoration network could achieve a similar trade-off. In addition, for a paper emphasizing real-time and hardware-free deployment, the efficiency analysis is incomplete because it mainly reports inference time without comparing parameter count, FLOPs, or memory cost. The baseline comparison did not contain SOTA methods after 2024.

  • 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 study demonstrates a meaningful clinical problem and presents promising results, but I am not yet fully convinced by the practical and methodological justification of the proposed framework. In particular, the role of the diffusion teacher in the final deployment pipeline is not sufficiently validated, and the efficiency analysis does not fully support the hardware-friendly claim. Overall, the work is interesting and technically solid, but the current evidence is slightly below the acceptance threshold for me.

  • 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

N/A




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 paper proposes a manifold-constrained hyper-connected diffusion model to preserve anatomical priors and suppress HIFU interference in B-mode images. The submission is generally well written. However, it would be stronger with more details on the experimental setup and datasets, as well as clearer justification for the use of each modality.



back to top