List of Papers Browse by Subject Areas Author List
Abstract
Efficiently and accurately removing noise from medical images is crucial for clinical diagnosis. Nevertheless, most deep learning-based medical images denoising methods are highly complex and inaccurate in preserving the edge and shape of different organs, resulting in suboptimal denoising performance. In our study, we propose a Human Visual System Inspired Lightweight Dual-Path Network for medical images denoising (VisNet), which can efficiently and accurately remove noise from different types of medical images. Specifically, to simulate the mechanism in the visual system where magnocellular and parvocellular pathways capture significant and subtle noise, respectively, we design a dual-path multi-scale perception module. Then, to simulate the function of the primary visual cortex, we propose an edge detection and shape adaptation module to preserve the structural information of the medical images. Finally, inspired by dorsal and ventral pathways, a spatial-semantic information extraction module is designed to enhance the main semantic information in the image through the interactive fusion between the spatial and semantic pathways. Experimental results demonstrate that VisNet achieves superior performance across three medical datasets compared to nine existing baselines, while maintaining minimal computational complexity. (Params=0.15, FLOPs=16.41). In addition, for brain tumor classification, using denoised images of VisNet as input significantly improves accuracy (87.5% vs 96.7%) and achieves performance comparable to noise-free images. Code of VisNet is available at https://anonymous.4open.science/r/VisNet-2270.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/0436_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/yuehailin/VisNet
Link to the Dataset(s)
N/A
BibTex
@InProceedings{YueHai_VisNet_MICCAI2025,
author = { Yue, Hailin and Kuang, Hulin and Ma, Lei and Liu, Jin and Li, Junjian and Cheng, Jianhong and Wang, Jianxin},
title = { { VisNet: A Human Visual System Inspired Lightweight Dual-Path Network for Medical Images Denoising } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15972},
month = {September},
page = {667 -- 677}
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors proposed a universal lightweight dual-path medical images denoising framework called VisNet, which simulates the mechanism of the human visual system with magnocellular and parvocellular pathways. Experimental results across three medical image datasets demonstrate the superiority of the proposed VisNet on both denoising performance and computational complexity.
- 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.
This work is inspired by the mechanism of the human visual system with magnocellular and parvocellular pathways, which provides a novel insight for lightweight medical image denoising.
- 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 authors have not provided any references about the human visual system with magnocellular and parvocellular pathways in Fig. 1 (A), and the interpretability of the proposed framework has not been proved effectively. In addition, there exist confusions of variable symbols in some equations, and conclusion is also too short.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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?
- The idea of the method design is innovative, but related references about the human visual system with magnocellular and parvocellular pathways in Fig. 1 (A) should be provided.
- The experiments lack interpretability analysis to prove the consistency of the mechanisms between the proposed method and the human visual system.
- Were noisy MRI images involved in the clinical application real rather than synthetic? The experiments on synthetic noisy MRI images does not necessarily demonstrate the effectiveness on real noisy MRI images in the clinical application.
- Confusions of variable symbols in some equations. For example, W in Eq. 2 has two meanings; C^k=n in Eq. 4 should be the convolution with kernel size n rather than the convolution kernel of size n.
- Conclusion is too short.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
This paper designs a multi-scale convolution method that can extract both global and local information from the image, under the inspiration of human visual pathway. The authors also propose an edge detection model and a spatial attention model that help the network extract information for high-accuracy image denoising.
- 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.
Results on n = 9648 images show that the proposed method outperforms all 9 comparison methods. The ablation studies also show that all changes to the network help increase denoising accuracy.
- 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 need to clarify why they believe two scales are sufficient for their network. As we know, most image processing networks use U-Net-based models, which typically have more than two resolution scales.
2.For the large-scale convolution, the authors need to explain why they only use a 3×3 convolution kernel. This is still considered local in other U-Net-based methods.
3.The authors only use three convolution layers in the edge detection model. They need to explain why they believe the information in the Edge Detection Block truly represents edge features, since there are no constraints on what information these layers extract.
4.What does the p in Equation 2 represent?
5.Why do the authors not compare their method with other methods in the clinical application section (Section 3.4)?
6.What are the future works? Although space is limited, it is better to include a discussion on future work.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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?
This paper has rich work in experiments, but I have some concerns about the method.
- 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
The authors propose a Human Visual System-Inspired Lightweight Dual-Path Network for medical image denoising (VisNet), which can efficiently and accurately remove noise from different types of medical images. Specifically, it takes the human visual system as the prototype and designs a dual-path multi-scale perception module, an edge detection and shape adaptation module, and a spatial semantic information extraction module. Experimental results demonstrate that VisNet achieves superior performance across three medical datasets compared to nine existing baselines, while maintaining minimal computational complexity. Moreover, when using the denoised images as input, the performance in brain tumor classification is significantly improved compared to using noisy images, demonstrating its potential in clinical diagnosis.
- 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.
Simplicity and Novelty: The architecture is biologically inspired and modular, offering a lightweight yet effective solution to medical image denoising. The dual-path design and multi-scale feature extraction strategies are well-motivated and practically relevant. Evaluation: The paper includes comprehensive quantitative and qualitative comparisons across three distinct datasets, as well as ablation studies that validate each module’s contribution. Clinical Relevance: The authors not only improve denoising performance, but also show that their method positively impacts a downstream clinical task (tumor classification), reinforcing its practical importance. Reproducibility: The authors provide clear architectural diagrams, release well-organized code, and include a detailed README, making it reproducible.
- 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.
Loss Function Description: The paper does not clearly specify the loss functions used to train the model. This information is essential for understanding the training dynamics and reproducibility. Figure 2 Clarity: Figure 2 lacks concise annotation or legend for key components such as GAP (Global Average Pooling) and GMP (Global Max Pooling). A brief explanation in the figure would improve interpretability. Noise Modeling for All Modalities: While the MRI dataset is described as having Gaussian and Poisson noise (with specified levels), it is unclear how noise was simulated or obtained for CT and X-ray images. The paper should clarify whether clean and noisy image pairs exist, or how the noise distributions were defined.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This paper presents a well-motivated, interpretable, and effective approach for medical image denoising that generalizes across modalities and contributes to downstream clinical applications. The proposed modules are novel and inspired by the human visual system, which adds to the theoretical depth of the work. Experimental results are comprehensive, and the provided code further supports reproducibility. Given the novelty, completeness of evaluation, and clear presentation, I recommend accept.
- 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
- 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