List of Papers Browse by Subject Areas Author List
Abstract
Microvascular remodeling is a critical hallmark of early-stage tumorigenesis. While Confocal Laser Endomicroscopy (CLE) enables \textit{in vivo} “optical biopsy” of the microvasculature, its restricted field of view (FoV) limits the assessment of macroscopic pathological continuity. To bridge this gap, we propose Angio-Stitch, an unsupervised deep learning framework tailored for multi-image stitching of hepatic microvascular sequences. Unlike existing deep learning methods primarily optimized for dual-image natural scenes—which are prone to cumulative errors in long sequences—our model adopts a robust coarse-to-fine alignment strategy. Specifically, we utilize a ResNet-based encoder integrated with a Vascular-aware Perception Transformer (VPT) to extract robust semantic features amidst speckle noise. A mesh-based deformation network is then employed to correct non-rigid tissue distortions. To explicitly preserve microvascular topology, we introduce a novel Vascular Structure Loss ($\mathcal{L}_{vsl}$) based on multi-scale Hessian responses. We evaluate our approach on the CLE-LMV dataset, established via a custom-developed NIR-CLE system. Experimental results demonstrate that Angio-Stitch achieves state-of-the-art performance in geometric accuracy and structural fidelity. This approach effectively provides clinicians with a wide-field, seamless visualization for reliable diagnostic support.Code available at: https://github.com/xiaoshihu8/Angio-Stitch
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2300_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/xiaoshihu8/Angio-Stitch
Link to the Dataset(s)
N/A
BibTex
@InProceedings{HuXia_AngioStitch_MICCAI2026,
author = { Hu, Xiaoshi AND Zhang, Huahui AND Deng, Xiang AND Liu, Xiaoyue AND Wang, Peng AND Ye, Xuesong},
title = { { Angio-Stitch: An Unsupervised Coarse-to-Fine Framework for Sequential Stitching of Confocal Laser Endomicroscopy Images } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16895},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes Angio-Stitch, an unsupervised coarse-to-fine framework for confocal endomicroscopy image stitching. The method introduces a vessel-perception Transformer module, namely VPT, and a vessel-structure loss to strengthen the constraints on vascular morphological continuity and structural consistency. On the authors’ CLE-LMV dataset, the proposed method achieves the best or second-best performance compared with existing competing methods.
- 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 The method is specifically designed for CLE image stitching, and both the proposed VPT module and the Vascular Structure Loss are well motivated by the characteristics of the target scenario. 2 The overall technical pipeline is relatively complete, with a reasonable correspondence among the method motivation, module design, and experimental validation. The comparison studies and ablation experiments are generally well designed and provide support, to a certain extent, for the effectiveness and necessity of the proposed innovations. 3 The authors present the CLE-LMV dataset to evaluate the method in this specific application scenario, which is of certain value for validating the practical applicability of the proposed approach.
- 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 overall unsupervised coarse-to-fine framework is relatively similar to existing work, such as Parallax, and the main improvements are more reflected in the scene-specific module and loss design. Therefore, from the perspective of the overall methodological framework, the novelty is somewhat limited. 2 The construction of the masks are insufficiently described. It is unclear how the masks are generated and whether they are directly provided by the dataset. This to some extent affects the reproducibility of the method and the interpretability of the experimental results. 3 The stitching success rate used in the paper is to some extent subjective, and the authors are encouraged to make this point clearer in the manuscript.
- 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.
(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 addresses a practically meaningful problem and demonstrates a good understanding of the low image contrast and non-rigid deformation characteristics in the target stitching scenario. The method introduces scene-specific module and loss designs, and its effectiveness is supported by relatively sufficient experimental validation. The main contributions of the paper are clear, and the experimental results are reasonably convincing.
- 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
1.This paper proposes an unsupervised, coarse-to-fine image stitching framework specifically for the microvascular reconstruction task 2.brand-new microvascular microscopic dataset for mice livers, named CLE-LMV, is constructed. 3.On this dataset, the stitching success rate of the proposed method is significantly higher than that of other comparison methods.
- 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.The research motivation is clear and reasonable. The traditional stitching methods perform poorly on medical images. The liver tissue is prone to non-rigid deformation, and the image texture is monotonous with low contrast. Most existing stitching methods are designed for natural images, and there is a lack of research on medical microvascular images. 2.By adopting a coarse-to-fine registration process, it can effectively reduce the error accumulation problem in long sequence images. The loss function is designed to meet the requirement of preserving the vascular structure, and the overall logic is clear. 3.A dedicated medical dataset are constructed.
- 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 method lacks innovation: Transitioning from rough to refined is a very classic strategy in the field of image stitching. 2.The core modules VPT and VSL do not provide intuitive visual experiments, making it impossible to clearly demonstrate their actual effects in preserving microvascular topology and reducing the accumulation of sequence errors.
- 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.
- 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 work is the first dedicated research on the microvascular stitching task. It not only proposed a dedicated dataset, but also the compared baseline methods are relatively cutting-edge.
- Reviewer confidence
Somewhat confident (2)
- [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 Angio-Stitch, an unsupervised coarse-to-fine image stitching framework for sequential CLE hepatic microvasculature images. The main technical contributions are a Vascular-aware Perception Transformer (VPT) for robust matching under speckle noise and weak texture, a mesh/TPS-based fine deformation module for non-rigid alignment, and a vascular structure loss to better preserve vessel continuity. The paper also introduces evaluation on a custom CLE-LMV dataset and reports strong stitching performance after fine-tuning.
- 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 a meaningful medical imaging problem: CLE has a very limited field of view, and reliable stitching could improve tissue-level interpretation. -The method is well tailored to the task. The coarse-to-fine design, vessel-aware feature modeling, and topology-preserving loss are appropriate for noisy, low-contrast, deformable CLE images. -The approach is unsupervised, which is attractive because dense registration/stitching labels are difficult to obtain in this setting. -The reported quantitative results are promising, especially after fine-tuning, with improved mRMSE, mPSNR, mSSIM, and 100% success rate on the reported dataset. Qualitative results also look better than the compared baselines. -The paper is clinically motivated and focuses on a challenging but relevant endomicroscopy application rather than a generic benchmark.
- 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 dataset construction raises concerns about realism and independence. The paper expands 1,510 raw images into 10,000 five-image sets using cropping, distortion, and illumination simulation, but the split protocol is not sufficiently detailed. Possible leakage or high similarity between train and test sets should be clarified. -The evaluation protocol is not fully convincing. The paper reports mRMSE, mPSNR, and mSSIM, but the reference definition for stitched CLE sequences is not explained clearly enough. -Comparisons are mostly against natural-scene stitching methods. More discussion or comparison with prior medical/endomicroscopy mosaicking work would strengthen the novelty claim, for example prior endomicroscopy mosaicking methods such as Vercauteren et al. (MedIA 2006) and Gong et al. (Computers in Biology and Medicine 2022), both already cited by the authors. -The ablation study is incomplete relative to the final best model. Table 2 matches the non-fine-tuned performance, not the final fine-tuned result reported in Table 1.-The claim about robustness to cumulative error in long sequences is important, but experiments seem limited to five-frame sequences. Longer-sequence testing would be more convincing.
- 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
This is a promising and clinically relevant paper with a tailored technical design. The main points that need clarification are: dataset splitting and independence, how ground truth/reference stitching quality is defined, whether animal ethics approval was obtained for the ex vivo mouse data, and whether the method remains robust on longer and more realistic clinical sequences. A stronger discussion against prior medical/endomicroscopy mosaicking literature would also improve the paper.
- 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?
I scored this paper as a borderline Weak Accept because it tackles an important medical imaging problem and proposes a technically meaningful solution that is well matched to CLE data characteristics. The method is more specialized than generic stitching pipelines, and the reported results are strong on the presented dataset. However, the paper is held back by limited clinical realism, unclear dataset independence, insufficiently explained evaluation setup, and incomplete validation of long-sequence robustness. A solid rebuttal addressing these points would strengthen confidence in acceptance.
- 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 sincerely thank the reviewers for their time, constructive feedback, and valuable suggestions. Below, we address some reviewer’s concerns point by point. Q1: (@R1&R3) : Dataset construction and splitting issues. A1: We generated the 10,000 five-image sets from the 1,510 raw images using realistic spatial and photometric transformations (cropping, distortion, illumination changes) that mimic actual clinical scanning conditions. Regarding the split protocol, we guarantee strict source-level independence: the training and testing sets are derived from mutually exclusive raw image pools. There is absolutely no overlap or data leakage between the sets. Due to space constraints, we have briefly outlined the dataset construction and splitting in submission and we will comprehensively detail this process in the final version. Q2: (@R3) : Lack of comparison with prior medical or endomicroscopy mosaicking methods. A2: We thank the reviewer for pointing this out. We actually conducted experiments using prior medical and endoscopy mosaicking methods. However, their performance on our dataset was suboptimal. This is primarily because Confocal Laser Endomicroscopy (CLE) images exhibit significant differences from standard endoscopy images, presenting extreme feature sparsity, cellular-level noise, and specific microvascular topologies. Methods designed for standard endoscopy struggle to handle these CLE-specific characteristics. Q3: (@R3) : Limited evaluation on longer sequences. A3: We agree with the reviewer that evaluating longer sequences is an important and valuable direction. Most current state-of-the-art deep multi-image stitching methods [5, 22, 23, 24] primarily restrict their evaluations to short sequences (typically 2 to 5 frames). While our current framework focuses on effectively addressing the challenges of specific clinical scenarios, extending to significantly longer sequences involves broader global alignment tasks. We will explore advanced global optimization strategies to achieve robust long-sequence merging in our future work. All experiments were approved by the institutional ethics review committee and conducted in compliance with relevant ethical regulations.
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 presents a well-motivated and practically relevant contribution with promising task-specific innovations and solid empirical results. The main critics is that the work has limited methodological novelty. But the paper addresses a clinically meaningful CLE stitching problem with a well-designed, vessel-aware unsupervised framework and demonstrates strong empirical performance on a dedicated dataset. Overall, there is a consensus that the paper is marginally acceptable.