List of Papers Browse by Subject Areas Author List
Abstract
Autoregressive Initial Bits is a framework that integrates sub-image autoregression and latent variable modeling, demonstrating its advantages in lossless medical image compression. However, in existing methods, the image segmentation process leads to an even distribution of latent variable information across each sub-image, which in turn causes posterior collapse and inefficient utilization of latent variables. To deal with these issues, we propose a prediction-based end-to-end lossless medical image compression method named LVPNet, leveraging global latent variables to predict pixel values and encoding predicted probabilities for lossless compression. Specifically, we introduce the Global Multi-scale Sensing Module (GMSM), which extracts compact and informative latent representations from the entire image, effectively capturing spatial dependencies within the latent space. Furthermore, to mitigate the information loss introduced during quantization, we propose the Quantization Compensation Module (QCM), which learns the distribution of quantization errors and refines the quantized features to compensate for quantization loss. Extensive experiments on challenging benchmarks demonstrate that our method achieves superior compression efficiency compared to state-of-the-art lossless image compression approaches, while maintaining competitive inference speed. The code is at https://github.com/scy-Jackel/LVPNet.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2773_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/scy-Jackel/LVPNet
Link to the Dataset(s)
Chest X-Ray dataset: https://data.mendeley.com/datasets/rscbjbr9sj/2
CIFAR10 dataset: https://www.cs.toronto.edu/~kriz/cifar.html
ImageNet32 and ImageNet64 datasets: https://image-net.org/
BibTex
@InProceedings{SonChe_LVPNet_MICCAI2025,
author = { Song, Chenyue and Hui, Chen and Lin, Qing and Zhang, Wei and Li, Siqiao and Zhu, Haiqi and Zhang, Shengping and Li, Zhixuan and Liu, Shaohui and Jiang, Feng and Li, Xiang},
title = { { LVPNet: A Latent-variable-based Prediction-driven End-to-end Framework for Lossless Compression of Medical Images } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15967},
month = {September},
page = {288 -- 298}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes a latent-variable-based framework for lossless compression of medical images by integrating several modules such as sensing and compensation modules.
- 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 clinical motivation is clear. Lossless compression benefits how medical images are stored.
- 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 medical experiments are conducted on a single medical dataset (Chest X-ray), while there are plenty of other datasets available. Evaluation on additional datasets, especially different modalities such as MRI and CT, would be necessary to strengthen the claims. Also, there are no standard errors in any tables, so it is difficult to understand if the performance differences between LVPNet and the second-best method are statistically significant, or just within noise.
- 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.
(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 results look marginally better than the other methods (i.e., incremental improvements). Without providing standard errors or some statistical tests, it is difficult to conclude that the proposed method is better or clinically meaningful.
- Reviewer confidence
Somewhat confident (2)
- [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 clarifications on modality generalization, standard deviation reporting, and broader context have addressed my concerns.
Review #2
- Please describe the contribution of the paper
The paper proposes a lossless image compression method for medical images by leveraging learnable modules to capture; a) spatial dependencies, b) mitigate posterior collapse, and c) correct quantization error. The results shows an improved bits-per-pixel (bpp) performance.
- 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 proposed method is effective across both medical (Chest X-Ray) and natural image datasets (CIFAR10, ImageNet), and outperforms classical and deep learning-based compression methods. The paper is well written and relatively easy to follow. The architectural design choices are intuitive.
- 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.
a) Limited Medical Validation: While the evaluation on chest X-rays is solid, the model has not been tested on other medical modalities such as MRI or CT. Hence, the generalizability is currently an open question. b) How GMSM mitigates posterior collapse is not thoroughly analyzed or compared against alternative architectures.
- 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 introduces a technically sound and innovative framework with strong empirical results. While there are some limitations regarding validation on diverse medical modalities and a deeper theoretical analysis, the novel contributions support its acceptance.
- 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.
The authors answered adequately to my raised questions and promised to do necessary revisions in the paper.
Review #3
- Please describe the contribution of the paper
The authors present a latent-variable-based, prediction-driven end-to-end framework for lossless medical image compression, capable of handling both dataset-level and single-image scenarios. A key contribution lies in the design of the GMSM and QCM modules, which enhance multi-level feature extraction to alleviate posterior collapse and reduce quantization-induced information loss. The proposed approach achieves competitive compression ratios and demonstrates efficient inference across multiple medical imaging benchmarks.
- 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 introduces two novel modules: GMSM (Global Multi-Scale Module) and QCM (Quantization Compensation Module), which target key challenges in learned medical image compression. GMSM enhances multi-scale feature representation to mitigate posterior collapse, while QCM effectively addresses quantization-induced information loss. This modular innovation directly contributes to the improved performance of the proposed method. And this study also did broad experimental validation that the method LVPNet is evaluated across four diverse datasets (Chest X-ray, CIFAR10, ImageNet32, and ImageNet64), demonstrating superior or highly competitive bits-per-pixel (BPP) performance compared to both traditional compression algorithms (e.g., PNG, FLIF, JPEG-XL) and learned compression models (e.g., L3C, iFlow, BCM-Net, ArIB-BPS).
- 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) While the proposed framework is promising, the motivation presented in the introduction could benefit from additional depth and specificity. Currently, the problem setting and context are described somewhat broadly. Strengthening this section with more concrete clinical scenarios or limitations of existing compression approaches would help readers better appreciate the need for the proposed method. 2) The GMSM and QCM modules are core components of the proposed architecture, and while they appear effective, their designs are relatively straightforward. It would be helpful to elaborate on the design choices or provide deeper insights into why these modules are particularly suited for addressing posterior collapse and quantization loss. This could enhance the perceived novelty and impact of the contributions. 3) Although LVPNet demonstrates strong overall performance, especially in BPP, the gains in inference time are relatively modest in some cases. For instance, as shown in Table 3, LVPNet does not always outperform the fastest baselines like ArIB-BPS. Similarly, Table 4 shows that adding QCM and GMSM increases inference time slightly. A more detailed discussion of these trade-offs would make the evaluation more compelling.
- 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?
The work addresses an important problem in medical image compression and presents a practical, end-to-end solution that performs well across multiple datasets. The proposed GMSM and QCM modules are straightforward but effective, and the authors conducted extensive experiments, including ablation studies, to validate their approach. However, the motivation could be more clearly articulated, regarding clinical relevance and how the proposed approach compares to existing methods in practical settings. While the architectural modules are novel, they are relatively simple, and inference time improvements, though present, are not always substantial. Overall, the method is solid and relevant, and I believe with stronger framing and deeper analysis, it can make a more meaningful contribution to the community.
- 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.
The authors provided a clear explanation of how GMSM and QCM are designed to address posterior collapse and quantization loss, which makes the overall methodology more coherent and convincing. And they also mentioned plans to expand the introduction by incorporating specific challenges faced in clinical workflows such as the storage of high-resolution medical images and real-time transmission in telemedicine. This additional context will help clarify the practical need for an efficient, lossless compression framework like LVPNet. Therefore, I recommend acceptance.
Author Feedback
We thank all three reviewers for their constructive comments and appreciations of our strengths such as ‘innovative framework with strong empirical results’(R1), ‘The clinical motivation is clear’(R3) and ‘demonstrating superior or highly competitive performance’(R5). To R1 Q1:Model has not been tested on other modalities like MRI or CT A1:Thanks for the suggestions. While our experiments are conducted on Chest X-rays, this modality is not only one of the most widely used in clinical practice but also exhibits key characteristics common to other medical modalities—such as high resolution, strong structural regularity, and clinically critical fine-grained details. Our model is modality-agnostic by design and does not rely on image-specific priors. GMSM and QCM are general-purpose, aiming to enhance multi-scale features and preserve details, which also apply to MRI and CT. Due to space and resource limits, we could not include results from other modalities. As requested, we tested model on CT and MRI, showing results consistent with Chest X-rays. Q2:How GMSM mitigates posterior collapse is not thoroughly analyzed or compared to other architectures A2:Thanks for the suggestions. Heavy reliance on a strong prior can cause latent variables to carry less information, leading to posterior collapse. GMSM is specifically designed to address this issue by reinforcing the encoder’s capacity to extract rich, hierarchical features and integrate global context. This architectural enhancement increases the utility of the latent variables, reducing the model’s over-reliance on the prior and thus mitigating posterior collapse. Our ablation study(Table4) confirms GMSM’s effectiveness, showing a performance drop when replaced by a CNN. We will clarify GMSM’s role in revision. To R3 Q3:Evaluation on other datasets like MRI and CT is needed to strengthen the claims A3:Thanks for the suggestions. Regarding the concern you mentioned, we’ve explained it in detail in our response to R1(A1). In that section, we offer a thorough explanation addressing this concern. Q4:There are no standard errors A4:Thanks for the suggestions. We trained LVPNet five times with different random seeds for both dataset-level and single-image compression settings, and reported the average results. The performance variance across runs is negligible compared to the performance gaps between LVPNet and the second-best methods, indicating the improvements are consistent rather than random. We will clarify this and include standard deviations in the revision. To R5 Q5:Problem setting and context are described somewhat broadly A5:Thanks for the suggestions. In the revision, we will expand the introduction by incorporating specific challenges faced in clinical workflows, such as storage of high-resolution medical images, real-time transmission in telemedicine. This context will help clarify the practical need for an efficient lossless compression framework like LVPNet. Q6:Why GMSM and QCM are suited to address posterior collapse and quantization loss A6:Thanks for the suggestions. The role of GMSM in mitigating posterior collapse has been explained in detail in our response to R1(A2). As for QCM, it’s designed to address the information loss from the floor operation during quantization. By using a lightweight CNN with residual blocks and a ReLU-activated output, it learns non-negative residuals to restore fine details and improve the predicted probabilities accuracy. Q7:Gains in inference time are modest A7:Thanks for the suggestions. Our design balances performance and efficiency. LVPNet offers faster inference on high-resolution datasets(Table3), which aligns well with clinical requirements where timely processing of large medical images is critical. While GMSM and QCM slightly increase inference time(Table4), they improve performance by mitigating posterior collapse and quantization loss. We believe this trade-off is reasonable and beneficial in practical medical scenarios.
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.
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 #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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
All reviewers agree that the article has merit and would be of interest to the MICCAI community. The rebuttal addressed all raised concerns precisely. I am inclined to recommend acceptance.