Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Tunnel magnetoresistance (TMR) sensors have been recognized as a cost-effective alternative for measuring magnetocardiography (MCG) signals. However, their relatively high noise levels and susceptibility to contamination limit their practical clinical applications. To address these challenges, we propose a novel Multi-Level Gated U-Net (MGU-Net) model specifically designed for denoising long sequential MCG signals obtained from TMR sensors. The MGU-Net leverages the U-Net architecture to learn hierarchical representations, integrated with a novel Gated Linear Unit (GLU) module to capture the periodic pattern of Q, R, and S wave complex (QRS complex) from MCG. This design enhances periodic cardiac signatures and suppresses irregular noise components through adaptive gating mechanisms. We have developed a TMR-based MCG system and collected both simulated and real MCG data in a magnetically shielded environment. The results show that our method improve signal-to-noise ratio (SNR) from -2.142 dB to 10.505 dB on the simulated MCG dataset and from 3.958 dB to 14.514 dB on the real dataset, surpassing other state-of-the-art methods. Our model successfully recovers subtle P-wave and T-wave features from the noisy signals, illustrating a promising direction of using TMR-based systems for potential practical clinical applications.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3635_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/YorkXingZeyu/MCG-denoising-project.git

Link to the Dataset(s)

N/A

BibTex

@InProceedings{XinZey_MultiLevel_MICCAI2025,
        author = { Xing, Zeyu and Li, Hao and Dou, Hao and Zheng, Zhong and Dai, Jingguo and Wang, Chen and Cui, Jian and Zhang, Xin and Jiang, Tianzi},
        title = { { Multi-Level Gated U-Net for Denoising TMR Sensor-Based MCG Signals } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15962},
        month = {September},
        page = {433 -- 442}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    The paper focuses on denoising MCG signals. Unlike ECG signals, MCG signals have a 1/f (or pink noise) profile. As such, band filtering is less effective in MCG than in ECG. The deep learning approach is a standard U-Net architecture combined with Gated Linear Units, which reinforce/emphasize “long” periodic patterns. The investigators used both simulated data and human-subject data to analyze the performance of the proposed architecture. The results demonstrated an increase in performance overall compared to other models. Although MCG is a long way from ECG (in terms of clinical adoption), the results are interesting and warrant additional study.

  • 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 main strengths of the paper are 1) the introduct of the GLU (although it is difficult to clearly articulate how the GLU improves performance, and 2) the nicely organized study design including simulation and real data.

  • 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 primary weakness of the study is that it doesn’t go far enough into translation. Basic engineering validation is used as an endpoint. It is unclear if the method truly improves diagnostic performance.

  • 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 does not provide sufficient information for reproducibility.

  • 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/2025/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 — could be rejected, dependent on rebuttal

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

    The performance is better than other methods in the experiment, but it is not clear if that is sufficient to move forward with future studies. The investigators would benefit from clearly showing the diagnostic improvement. The investigators could also demonstrate the type of cases where this technology is practically more superior than existing technology (specifically in the context of diagnostics or maybe a demonstration of improved performacne in particularly bad data).

  • 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 authors propose a denoising method for MCG signals acquired by TMR, and experimentally verify the effectiveness of the method on simulated and real MCG signals.

  • 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.

    Model innovation: For the first time, U-Net architecture is combined with gated linear unit (GLU) to enhance the extraction of periodic heart features such as QRS wave groups and suppress aperitional noise through multi-scale feature extraction and adaptive gating mechanism. Technical breakthrough: Two gating modules (competitive gated CG and noise gated NG) were developed, which significantly improved the recovery ability of weak P /T waves in long sequence MCG signals and solved the problem of insufficient processing of non-stationary noise by traditional 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.
    1. The authors mention traditional methods such as EMD, SSS, etc., in the introduction, but do not discuss their effectiveness and performance in the experimental section. The manuscript only presents comparisons between IIR and FIR filters. Could the omission of traditional methods like EMD, VMD, EEMD, etc., in the comparative analysis be attributed to automation limitations?
    2. In the Methods section, the manuscript employs the classic U-Net architecture; however, the description lacks detail regarding critical aspects such as the size of the convolution kernels, the normalization methods used, and the activation functions. Please provide these details to fully understand the implemented architecture.
    3. Experimental details section: what is the signal-to-noise ratio of the model’s input noise-laden MCG signal? What is the ratio of division between training and test datasets? In addition, all experiments should be performed using the inter-patient paradigm.
    4. What are the noise characteristics of the input clean MCG signal? Is it a real noise signal or a simulated noise signal? The author needs to supplement the detailed construction process of the dataset.
    5. Experimental results section: the authors need to add a comparison of the clinical utility of MGU-Net, e.g., Params, FLOPs, or inference times.
  • 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.

  • 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/2025/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 — could be accepted, dependent on rebuttal

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
    1. The authors mention traditional methods such as EMD, SSS, etc., in the introduction, but do not discuss their effectiveness and performance in the experimental section. The manuscript only presents comparisons between IIR and FIR filters. Could the omission of traditional methods like EMD, VMD, EEMD, etc., in the comparative analysis be attributed to automation limitations?
    2. In the Methods section, the manuscript employs the classic U-Net architecture; however, the description lacks detail regarding critical aspects such as the size of the convolution kernels, the normalization methods used, and the activation functions. Please provide these details to fully understand the implemented architecture.
    3. Experimental details section: what is the signal-to-noise ratio of the model’s input noise-laden MCG signal? What is the ratio of division between training and test datasets? In addition, all experiments should be performed using the inter-patient paradigm.
    4. What are the noise characteristics of the input clean MCG signal? Is it a real noise signal or a simulated noise signal? The author needs to supplement the detailed construction process of the dataset.
    5. Experimental results section: the authors need to add a comparison of the clinical utility of MGU-Net, e.g., Params, FLOPs, or inference times.
  • 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.

    The authors claim to compare with other methods in the paper, and the details of the algorithm parameters are further disclosed.



Review #3

  • Please describe the contribution of the paper

    This paper proposes a Multi-Level Gated U-Net (MGU-Net) model specifically for denoising MCG signals acquired by TMR sensors. Quantitative experiments on both simulated and real-world datasets demonstrate the effectiveness of the proposed model. The proposed model has a promising direction of using TMR-based systems for potential practical clinical applications.

  • 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 major strengths are:

    1. A U-Net-based model for effectively denoising MCG signals acquired by TMR sensors.

    2. Quantitative experiments on both simulated and real-world datasets demonstrate the effectiveness of the proposed model.

    3. Open up a promising direction of using TMR-based systems for potential practical clinical applications.

  • 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 weaknesses are:

    1. The motivation for replacing the self-attention (SA) mechanism with the gated linear unit (GLU) is not clearly stated. The authors claimed that SA suffers from computational complexity, then making it computationally prohibitive and prone to outfitting. Then, the authors replace the SA with learnable linear projections; and this is implemented by GLU. From my understanding, these operators are inside the neural network, which is related to internal representations (hard to determine whether it should overfit or not), while the mentioned “prone to outfitting” is mainly about the denoised results for overfitting the noise. It is more appropriate to mention that learnable linear projections implemented by GLU follow the observation of periodic MCG signals. Then, an ablation study to compare SA to GLU can be utilized for justification.

    2. There is no alternative component to compare GLU in the ablation study. It is more appropriate to consider different alternatives to demonstrate the effectiveness of the GLU.

    3. The loss function is only MSE in this denoising task. It is more appropriate to incorporate other regularizations, such as Total Variation (TV), to achieve better performance and robustness. Otherwise, relying on early stopping to prevent overfitting is too empirical.

    4. The abbreviation “QRS” does not have a full name in the first place it appears (Abstract or Introduction). Its full name should be included appropriately for better clarity.

  • 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.

  • 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/2025/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 — could be accepted, dependent on rebuttal

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

    Please refer to the major strengths and weaknesses.

  • 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 the reviewers for the constructive feedback. Here are our replies: Reply to Reviewer #1: The major concerns of R1 are lack of details on additional comparison methods (Q1), model architecture details (Q2), experimental specifications (Q3), noise characteristics (Q4) and clinical utility metrics (Q5). We will address these points and reflect them in the revised paper.

Re: We have compared with EMD, VMD, and EEMD and will update the paper. (Re to Q1); The convolutions with kernel sizes of 9 for the main branch, 7 for the gating branch, and 4 for down/upsampling are followed by RMSNorm and a SiLU activation. (Re to Q2); The original SNR values were shown in Table 2. Train/validation/test split is 7/1/2. (Re to Q3); The primary types of noise of clean signals from Kiel Cardio DB are thermal noise, optical system noise, and shot noise (<15 fT/√Hz), which are negligible after filtering, in comparison to the captured noise from TMR system with levels around 5 pT/√Hz. (Re to Q4). While clinical validation is ongoing, we will add in Section 3.2 Model parameters (16.13 M) and FLOPs (6.6G) vs. baseline comparisons and real-time inference capability (5.06 ms/sample on RTX 4090) (Re to Q5).

Reply to Reviewer #2 The reviewer raised concerns about the study’s translational depth, mentioning that the endpoints focused on basic engineering validation without clearly demonstrating diagnostic performance improvement.

Re: We appreciate the reviewer’s constructive feedback. Our study primarily aimed to address the critical challenge of noise reduction in TMR-based MCG signals, which is a prerequisite for enabling reliable clinical interpretation. While we did not explicitly quantify diagnostic outcomes, the denoising results on real-world MCG data highlight the model’s robustness in practical scenarios. For example, Fig. 3 illustrates the recovery of low-amplitude P-wave and T-waves in a high-noise real MCG recording, which is challenging for existing methods. These features are clinically essential for diagnosing arrhythmias, myocardial ischemia, and other cardiac abnormalities. By restoring these features from noisy signals, our method lays the groundwork for subsequent diagnostic applications. Future studies will explicitly link denoising efficacy to diagnostic accuracy, but this initial engineering validation is a necessary first step toward clinical translation.

Reply to Reviewer #3

  1. Motivation of using GLU over attention (Q1) Re: We agree with the reviewer that the replacement of SA with GLU is primarily driven by the periodic nature of MCG signals. As MCG exhibits strong self-correlation patterns and amplitude-dependent dependencies, GLU’s gating mechanism (via element-wise multiplication of two linear projections) inherently captures these global periodic features. In contrast, SA requires separate computation of Query (Q) and Key (K) matrices for similarity scoring, introducing redundant parameters (e.g., Q/K projection layers) that may over-parameterize the model. This redundancy could lead to suboptimal convergence, especially given MCG’s stable periodic patterns that do not require dynamic attention re-weighting. We will explicitly highlight these analyses and further compare their performance in the revised manuscript.

  2. Additional investigation on GLU (Q2) and loss function (Q3)，full name of QRS（Q4） Re: The implemented GLU formulation operates as a unified feature modeling unit without inherent submodules for partial ablation. Nevertheless, we will include a comparative analysis between GLU and self-attention (SA) mechanisms (Re to Q2); Integrating multiple loss terms requires careful balancing of weights and regularization strategies, which would exceed the page limit and necessitate extensive validation. We will discuss the limitation (Section 4.3) and will rigorously explore multi-loss frameworks in future work. (Re to Q3); “QRS complex” will appear in full as “Q, R, and S wave complex” in the paper (Re to Q4).



Meta-Review

Meta-review #1

  • Your recommendation

    Invite for Rebuttal

  • If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.

    N/A

  • 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    N/A



Meta-review #2

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    N/A

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    N/A



back to top