List of Papers Browse by Subject Areas Author List
Abstract
Optical coherence tomography angiography (OCTA) is an indispensable modality in ophthalmic imaging, providing high-resolution visualization of retinal microvasculature. Recently, deep learning approaches have been explored to reconstruct OCTA images; however, significant challenges persist, particularly the reliance on high-quality target data for model training, which is often impractical due to limitations in hardware and acquisition protocols. In this work, we present a novel pipeline for deep learning-based OCTA imaging from repeated OCT B-scans, circumventing the need for high-quality training labels. We introduce an Intra-View Enhancement (IVE) module together with a novel loss function Cross-View Matching (CVM) to improve the imaging. The proposed pipeline is evaluated on a local dataset, demonstrating a relative improvement of 4.97% and 27.42% in PSNR and CNR over state-of-the-art learning-based OCTA method respectively. Our results underscore the effectiveness and clinical viability of the proposed approach for OCTA images, highlighting its potential to advance imaging capabilities in challenging clinical environments.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/0696_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ZenJin_Intra_MICCAI2025,
author = { Zeng, Jingbo and Tan, Bingyao and Gu, Zaiwang and Gao, Shenghua and Schmetterer, Leopold and Cheng, Jun},
title = { { Intra- and Cross-View Enhancement for OCTA Imaging } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15972},
month = {September},
page = {306 -- 316}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper introduces a novel and effective pipeline for OCTA image reconstruction from repeated OCT B-scans without relying on high-quality training labels. By incorporating an Intra-View Enhancement module and a tailored Cross-View Matching loss function, the method robustly mitigates the impact of noisy labels, resulting in superior image clarity and structural fidelity.
- 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 presents several notable strengths. It introduces a novel framework for OCTA reconstruction without relying on high-quality ground truth, addressing a key challenge in clinical imaging. The proposed Intra-View Enhancement module leverages local structural similarity to improve feature representation, which is a relatively unexplored direction. Furthermore, the Cross View Matching loss, based on a generalized MSE formulation, is tailored for learning from noisy labels and demonstrates clear robustness. The data strategy using repeated B-scans to generate pseudo-labels offers a practical solution for settings where high-quality annotations are scarce.
- 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 only uses a locally collected dataset for evaluation, lacking validation on public datasets. It is recommended that the authors introduce standard public datasets for comparison. 2.Section 2.3 mentions that the labels are generated from OMAG processing of 1000 B-scan frames. Whether the quality of such results can replace ground truth remains unclear. The authors are advised to explain this in more detail. 3.Many of the baseline methods used in the paper are known to rely on high-quality labeled data for training, while the proposed method uses pseudo-labels generated from OMAG, which belongs to weak supervision. Are the datasets used in the baseline methods consistent with those used by the proposed method? If not, it would be difficult to demonstrate the superiority of the proposed approach. Direct comparison may lack fairness, and the authors should explain this issue. 4.The proposed CVM appears to be essentially a weighted reconstruction of similar blocks, a concept also found in other existing studies. It is unclear what the innovation of this idea is compared to previous methods. Could the authors elaborate on this specifically? 5.Although the proposed method is claimed to improve structural clarity under noisy conditions, there is a lack of interpretability analysis to support this claim. No attention maps, similarity visualizations, or block-wise enhancement comparisons are provided to explain how the model captures or preserves important vascular structures. It is strongly recommended that the authors include such visual analyses to enhance the understanding and credibility of the proposed method. 6.All the evaluation metrics used (PSNR, SSIM, CNR) are global image quality indicators, while structure-related metrics such as vessel continuity, detail fidelity, and anatomical accuracy, which are relevant for clinical diagnosis, are not evaluated. It is recommended to supplement with additional structural evaluation metrics. 7.Real-time performance and clinical deployability are important in the medical field. Could the authors further elaborate on the model size and inference speed, and whether it meets the real-time requirements of practical OCTA systems? 8.When reading Figure 1, it is necessary to repeatedly refer to the text and equations to understand the diagram. The correspondence between the structural diagram and the formulas is relatively unclear. It is suggested to add more structural annotations to facilitate readers’ understanding. 9.Figure 2 only shows a single example and lacks representative cases under different conditions. Could the authors provide multiple sample comparison results under varied scenarios? 10.It is unclear whether the paper constructs a novel theoretical framework. During the reading, no new visual modeling mechanisms such as new attention modules, regularization strategies, or learning paradigms were found. The method appears to be a combination of classical ideas applied to the OCTA image enhancement task, which lacks originality.
- Please rate the clarity and organization of this paper
Poor
- 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.
(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 paper is evaluated based solely on a locally collected dataset without validation on any standard public datasets, making it difficult to assess the generalizability of the proposed method. Additionally, the pseudo-labels used for training are generated via OMAG processing of 1000 B-scan frames. Whether such labels can be regarded as equivalent to ground truth remains unclear and requires further justification. Several of the baseline methods rely on high-quality annotations for training, whereas the proposed method uses weakly supervised pseudo-labels. This inconsistency in training data raises concerns about the fairness of the comparisons. Methodologically, the proposed CVM module essentially performs a weighted reconstruction based on similar blocks, a concept already explored in prior work, and its novelty relative to existing approaches is not clearly established. Moreover, the paper lacks interpretability analyses such as attention maps or structural visualizations to support the claim that the method preserves fine vascular structures under noisy conditions. All evaluation metrics are global image quality indicators, with no inclusion of clinically relevant structural assessments such as vessel continuity or anatomical accuracy. The paper also does not report model complexity or inference efficiency, leaving the practical deployability of the method in real-time clinical settings uncertain. Overall, while the approach has certain engineering merit, further improvements in methodological originality, fair benchmarking, and clinical relevance are necessary.
- 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
This paper proposes ICENet, a novel deep learning framework for OCTA image enhancement, featuring two key innovations: (1) an Intra-View Enhancement (IVE) module that leverages local patch similarity to suppress speckle noise while preserving microvascular structures, and (2) a Cross-View Matching (CVM) loss function designed for training with noisy labels, eliminating the need for high-quality ground truth. The architecture combines Swin-UNet’s global modeling with IVE’s local refinement, offering a practical solution for OCTA imaging in resource-constrained settings.
- 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 paper makes several significant contributions to OCTA reconstruction: (1) The Intra-View Enhancement (IVE) module that employs adaptive patch similarity matching to suppress speckle noise while preserving delicate microvascular structures - a improvement over generic denoising approaches; (2) It develops an innovative Cross-View Matching (CVM) loss function that enables effective training using only noisy labels.
- 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 proposed Cross-View Matching (CVM) loss appears to derive primarily from Jiang et al.’s framework [1], yet fails to adequately justify its incremental contribution beyond this existing work. While the patch-wise matching mechanism demonstrates potential, the technical description lacks critical implementation details that would enable proper evaluation. More fundamentally, the additive noise model (Eq. 5) represents a significant theoretical shortcoming, directly contradicting the well-established multiplicative Rayleigh statistics of OCT speckle noise [2]. This physical inconsistency raises serious concerns about the method’s validity, particularly in clinical scenarios where accurate noise modeling is crucial. [1] Jiang Z, Huang Z, Qiu B, et al. Weakly supervised deep learning-based optical coherence tomography angiography[J]. IEEE Transactions on Medical Imaging, 2020, 40(2): 688-698. [2] Liba O, Lew M D, SoRelle E D, et al. Speckle-modulating optical coherence tomography in living mice and humans[J]. Nature communications, 2017, 8(1): 15845.
- 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 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.
(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 phrase “Recent advances in learning” (P2) lacks specificity—it fails to distinguish whether this refers to supervised learning, self-supervised paradigms, or domain adaptation tailored for OCTA. (2)The manuscript oversimplifies ICENet’s target as generic “speckle noise” without addressing OCTA-specific challenges. (3)One of the paper’s core contributions is the proposed Cross-View Matching (CVM) loss for achieving higher image quality, whose formulation primarily builds upon the work “Weakly Supervised Deep Learning-Based Optical Coherence Tomography Angiography.” Specifically, the method enhances image representation by introducing a block-matching strategy. However, several critical aspects require clarification: (1) What is the exact methodology for selecting similar blocks? (2) What are the specific advantages of this strategy? (3) What is the underlying mechanism that enables this approach to effectively reduce noise? These questions need to be explicitly addressed to fully understand the novelty and effectiveness of the proposed technique. (4)Eq. 5’s additive noise model contradicts OCT physics.
- 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 #3
- Please describe the contribution of the paper
The paper presents ICENet, a novel pipeline for optical coherence tomography angiography (OCTA) imaging. This approach circumvents the need for high-quality training labels by leveraging noisy labels to enhance imaging. It introduces an Intra-View Enhancement (IVE) module that dynamically selects similar blocks to improve feature representation and a novel Cross-View Matching (CVM) loss function specifically designed to handle noisy labels. ICENet demonstrates significant improvements in image quality, achieving superior reconstruction accuracy compared to existing methods. The pipeline is evaluated on a local dataset, showing a relative improvement of 4.97% in PSNR and 27.42% in CNR over state-of-the-art learning-based OCTA 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.
Innovative approach: The introduction of the IVE module and CVM loss function provides a robust framework for handling noisy labels, enhancing image clarity in OCTA imaging. Strong evaluation: The method demonstrates significant improvements in PSNR, SSIM, and CNR metrics compared to existing methods. Practical applicability: The pipeline reduces the reliance on high-quality training data, which is often challenging to obtain in clinical settings.
- 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 paper could benefit from providing more detailed explanations of the computational complexity and efficiency improvements offered by the ICENet model. Additional comparisons with more diverse datasets could further establish the model’s generalizability across different imaging conditions.
- 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 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
ICENet represents a significant advancement in OCTA imaging, offering a robust solution for improving image quality without the need for high-quality labels. Future work could focus on expanding the dataset and exploring real-world clinical applications to further validate and enhance the model’s applicability.
- 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?
The paper presents a strong methodological contribution with the introduction of ICENet, which effectively addresses challenges in OCTA imaging using noisy labels. The model’s innovative integration of IVE and CVM loss function enhances image clarity and generalizability. Despite minor areas for improvement in computational complexity explanations and dataset diversity, the paper’s contributions and findings warrant a high acceptance score.
- 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
We thank all the reviewers for their valuable comments. Regarding the questions you raised, we will make modifications or supplements in camera-ready version. In subsequent research, we will also refer to these opinions to improve our future work.
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