List of Papers Browse by Subject Areas Author List
Abstract
Accurate, real-time monitoring of tissue ischemia is crucial
to understand tissue health and guide surgery. Spectral imaging shows great potential for contactless and intraoperative monitoring of tissue oxygenation. Due to the difficulty of obtaining direct reference oxygenation values, conventional methods are based on linear unmixing techniques. These are prone to assumptions and these linear relations may not always hold in practice. In this work, we present deep learning approaches for real-time tissue oxygenation estimation using Monte-Carlo simulated spectra. We train a fully connected neural network (FCN) and a convolutional neural network (CNN) for this task and propose a domain-adversarial training approach to bridge the gap between simulated and real clinical spectral data. Results demonstrate that these deep learning models achieve a higher correlation with capillary lactate measurements, a well-known marker of hypoxia, obtained during spectral imaging in surgery, compared to traditional linear unmixing. Notably, domain-adversarial training effectively reduces the domain gap, optimizing performance in real clinical settings.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2417_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{DeJen_Clinical_MICCAI2025,
author = { De Winne, Jens and Willems, Siri and Luthman, Siri and Babin, Danilo and Luong, Hiep and Ceelen, Wim},
title = { { Clinical Validation of Deep Learning for Real-Time Tissue Oxygenation Estimation Using Spectral Imaging } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15969},
month = {September},
page = {118 -- 125}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes a deep learning approach for real-time quantification of tissue oxygenation using spectral imaging data. By leveraging both simulated and real clinical data, the authors train fully connected and convolutional neural networks, incorporating domain-adversarial training to mitigate the domain gap. The proposed method shows improved correlation with capillary lactate measurements, a known marker of hypoxia, outperforming traditional linear unmixing techniques.
- 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.
- A key strength of this work lies in its generation of simulated and leveraging real clinical data. The real patient data was collected through multispectral imaging and corresponding capillary lactate measurements using a handheld analyzer. This creates a valuable dataset for model training and evaluation.
- 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 realism of the simulated data could be enhanced by incorporating temporal dynamics and organ motion, which are critical aspects of intraoperative environments. Including such factors would likely improve the fidelity and generalizability of the training data.
- The paper lacks clarity regarding the separation between training and test sets. Specifically, it is unclear whether the spectra used for testing were fully independent from the training data, as both are derived from the same group of 11 patients. This raises concerns about potential data leakage and overestimation of performance.
- Table 1 includes training and validation loss values, but these are difficult to interpret in isolation. Such metrics are better presented as learning curves to illustrate training dynamics over epochs, which would provide more insight into model convergence and potential overfitting.
- 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.
(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?
I recommend a weak rejection due to concerns about potential data leakage(data used both in training and testing), limited realism in simulated data.
- 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 #2
- Please describe the contribution of the paper
The authors propose a deep learning-based framework for real-time intraoperative tissue oxygenation estimation using spectral imaging. To address the lack of ground-truth oxygenation in clinical data, the authors generate 640,000 Monte Carlo-simulated spectra with known oxygenation labels. Two neural network architectures (FCN, CNN) are trained using this synthetic dataset, and a domain-adversarial training strategy is employed. In addition, real clinical validation is performed on data of 11 patients using capillary lactate measurements as a hypoxia reference. The proposed method achieves improved correlation with capillary lactate, as well as shorted inference time compared to conventional 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.
- Clinical Relevance: The task (intraoperative oxygenation monitoring) is of high clinical value, especially in surgical settings where tissue viability affects surgical outcomes. The proposed method aims to support real-time decision-making in this context.
- Clear Writing: The paper is well-written with a strong clinical motivation. The background and method is provided in details and are easy to follow.
- 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.
- Limited dataset: The clinical validation is based on only 11 patients with corresponding capillary lactate measurements and evaluated retrospectively. While the correlation results are promising, the small sample size limits the generalizability of the conclusions.
- Methodological Novelty: The proposed framework primarily combines established components—standard neural network architectures and domain-adversarial learning. The novelty lies more in the integration and clinical application than in algorithmic innovation.
- 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?
This paper addresses a clinically important challenge using a well-structured and technically sound pipeline. While the methodological novelty is moderate, the application of those techniques to real time spectral oxygenation estimation is meaningful and well-motivated. The use of capillary lactate as a reference for clinical validation further strengthens the contribution. The study is carefully conducted and results are clearly presented.
- 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
The authors propose a deep learning method instead of linear unmixing to learn tissue oxygenation. This is partially validated using lactate measurements, which have a relationship with tissue oxygenation measurements. A snapshot camera is used to acquire new clinical data and the neural network is compared against linear unmixing and DA networks.
- 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.
Overall this is an interesting paper, has a modest new computational contribution and some new data. I would be interested in hearing more about the work if accepted.
- 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.
There are certainly limitations when using unmixing and the authors note that ‘the pixel spectra are influenced by environmental conditions and sensor limitations, and non linear relationships are expected’. In fact, the linear unmixing model is limited by a large number of unmodelled phenomena, including optical scattering, and the fact that different wavelengths effectively sample different tissue volumes.
Another challenge mentioned is the noise and calibration when using different camera systems. Again this understates the problems, since the different measurement methods fundamentally affect what data is acquired and cannot be undone by something as simple as dealing with the measurement noise.
The correlation with lactate is questionable, but I accept that there are limited alternatives. However, the relationship in the plots in figure 2 is not very clear, and some of the well perfused data points have higher lactate, or low calculated oxygenation. The relationship with the DA method does appear to work better and the negative exponential fit appears reasonable.
At what point in the surgery does the surgeon identify and mark the regions?
Minor comments Top of page 3 – 20x20xD voxels are mentioned with a size of 0.01 mm. Should this be 1 mm?
- 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
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?
Achieving a relationship between the lactate and oxygenation in a clinical study using DA method is a useful finding.
- 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 the reviewers for their thorough and constructive feedback on our manuscript. We would like to address two specific points to provide clarification:
1) Region marking prior to spectral imaging The three regions (anastomosis site, well-perfused control, and ischemic control) are identified and marked immediately before cutting the stomach graft, which is part of the procedure to prepare it for anastomosis with the esophagus. The selection of these regions is based on the surgeon’s subjective assessment. Spectral recordings are then acquired immediately after marking. We will revise the manuscript to more clearly describe this step.
2) Independence of Training and Test Data: The training and test datasets are fully independent, as they were acquired from different patient cohorts. In total, spectral data was collected from 17 patients. Of these, 6 patients were used for training and validation of the domain-adversarial models, while the remaining 11 patients were used exclusively for testing. This information is provided in Sections 2.2 and 3.2, but we agree that this distinction should be emphasized more clearly. We will revise the text in the final version of the manuscript to make this more explicit..
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