Abstract

Intraoperative adverse events (IAEs), such as bleeding or thermal injury, can lead to severe postoperative complications if undetected. However, their rarity results in highly imbalanced datasets, posing challenges for AI-based detection and severity quantification. We propose BetaMixer, a novel deep learning model that addresses these challenges through a Beta distribution-based mixing approach, converting discrete IAE severity scores into continuous values for precise severity regression (0-5 scale). BetaMixer employs Beta distribution-based sampling to enhance underrepresented classes and regularizes intermediate embeddings to maintain a structured feature space. A generative approach aligns the feature space with sampled IAE severity, enabling robust classification and severity regression via a transformer. Evaluated on the MultiBypass140 dataset, which we extended with IAE labels, BetaMixer achieves a weighted F1 score of 0.76, recall of 0.81, PPV of 0.73, and NPV of 0.84, demonstrating strong performance on imbalanced data. By integrating Beta distribution-based sampling, feature mixing, and generative modeling, BetaMixer offers a robust solution for IAE detection and quantification in clinical settings.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4934_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)

https://github.com/CAMMA-public/MultiBypass140

BibTex

@InProceedings{BosRup_Feature_MICCAI2025,
        author = { Bose, Rupak and Nwoye, Chinedu Innocent and Lazo, Jorge F. and Lavanchy, Joël L. and Padoy, Nicolas},
        title = { { Feature Mixing Approach for Detecting Intraoperative Adverse Events in Laparoscopic Roux-en-Y Gastric Bypass Surgery } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15970},
        month = {September},

}


Reviews

Review #1

  • Please describe the contribution of the paper

    The main contribution of this paper is the approach to detect and assess the severity of surgical complications in laparoscopic Roux-en-Y gastric bypass. The authors use a beta distribution-based method to transform the discrete labels into a continuous regression framework, which improves the granularity and precision of the complication severity analysis. In addition, this paper introduces a generative model for feature normalization to improve the robustness and reliability of the data processing pipeline.

  • 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 demonstrates superior performance compared to existing methods in the task of detecting and predicting the severity of IAE. By utilizing the beta distribution, the study effectively addresses the issue of data imbalance, ensuring better handling of minority class examples. It transforms the task from classification to regression, enabling a more precise and continuous assessment of severity levels.

  • 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 lacks a clear explanation of the rationale behind performing feature normalization. There is no justification provided for selecting laparoscopic Roux-en-Y gastric bypass surgery as the target procedure, which limits the contextual relevance of the study. The description of the proposed method is insufficiently detailed, making it challenging for others to reproduce the results. In Fig. 1, the Discriminator is incorrectly labeled as “H,” which could confuse readers and detract from the clarity of the illustration.

  • 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?

    Difficult to reproduce due to lack of explanation and rationale for choosing specific methods presented in the paper.

  • 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

    The paper presents a classification method for predicting intraoperative adverse events and their severity scores, leveraging a beta distribution-based approach alongside a feature normalization technique.

  • 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. Compared to most existing methods, the proposed approach also predicts severity scores, which may be useful for intraoperative adverse event (IAE) detection.

    2. The paper is well-structured and 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.
    1. The primary challenge in this task lies in data imbalance, a well-known issue in classification problems that has been extensively studied. However, the related work section lacks discussion of existing techniques specifically designed to address data imbalance. This omission makes it difficult for readers to understand the novelty and significance of the proposed beta distribution-based method. Currently, most cited works focus on architectural designs rather than imbalance handling. Incorporating recent advances in this area would help contextualize the contribution more effectively.

    2. The paper does not provide sufficient theoretical justification for the use of the beta distribution. It is unclear how mapping discrete severity levels to a continuous space using a beta distribution addresses the label imbalance problem. Additional clarification on the rationale and benefits of this approach would be valuable to the reader.

    3. The current explanation of how continuous severity levels are generated during training is unclear. A more detailed description of the generation process and how it is integrated into the training pipeline would help improve clarity and reproducibility.

  • 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?

    The major contribution of this paper lies in handling the imbalanced dataset via the proposed beta-based sampling method. However, related techniques to handle the imbalanced dataset are not introduced and compared, which raise the concern about the effectiveness and novelty of the major technical contribution.

  • Reviewer confidence

    Not confident (1)

  • [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 address the conerns of the reviewer well.



Review #3

  • Please describe the contribution of the paper

    The paper proposes a novel regression model ‘BetaMixer’ to predict the severity of interoperative adverse events, such as bleeding, thermal and mechanical injury, by transforming the discrete labels into a continuous space. The novel approach combines a feature normalization, beta distribution-based sampling and continuous feature space regularization, to address the data imbalance of rare events. The model is trained and evaluated on an extended version of the public dataset MultiBypass140.

  • 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.
    • novel approach of combining feature normalization, beta mixing and converting targets into continuous space
    • extensive evaluation using F1, PPV, NPV, Recall, MSE and delay time
    • ablation study evaluating impact of feature normalization
    • Comparison against SotA
    • clearly structured and easy to read paper
  • 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.
    • why is the dataset split in a way such that not all classes are available in the testset?
    • How were the hyperparameters determined? Why did the authors choose e.g. a resolution of only 128x128?
    • the extensions of the public dataset are not released
    • no code available
  • 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

    I would like to encourage the authors to release the data and code. This would encourage the adaptation of this idea in the field.

  • 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 novel method for detecting IAEs that is of practical importance. Minor ambiguities should be answered in the rebuttal.

  • 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.

    All the concerns raised were properly addressed. My recommendation remains unchanged from pre-rebuttal.




Author Feedback

We sincerely thank all reviewers for their constructive feedback and are encouraged by their recognition of our contributions, particularly: the novelty and practicality of detecting & assessing the severity of IAEs using a beta distribution-based regression model (Review #1, #2, #3); the proposed combination of feature normalization, continuous space modeling, & data imbalance mitigation (Review #2); extensive evaluation with multiple metrics and ablation studies (Review #2); paper readability & clear structure (Review #2, #3); and superior performance over existing methods (Review #1, #3).

We address the primary concerns below:

  1. Rationale for Feature Normalization (Review #1): We clarify that our feature normalization step is not generic preprocessing but a core component of our model. As noted in Section 3.3 and further demonstrated in the ablation (Table 3), normalization is vital to guide the feature generator towards a smooth latent space distribution, enabling the regression task to capture subtle severity variations and better address class imbalance. This point was acknowledged positively by Review #2, confirming its clarity and impact.

  2. Theoretical Justification for Beta Distribution and Continuous Labeling (Review #3): The Beta distribution was chosen due to its mathematical properties and flexibility in modeling scores in the normalized [0,1] range. This allows us to model uncertainty and continuity in expert-provided severity annotations. Section 3.3 discusses how this supports a more granular representation than categorical labels. Review #1 noted that this approach improves “precision of the complication severity analysis” & handles minority examples more effectively.

  3. Clarity on Label Generation and Training Process (Review #3): Section 3.4 mentioned that continuous severity labels are learned via a Generator-Discriminator module. The generator converts backbone features to the severity score space via FCNN transformation, while the discriminator enforces adherence to the beta distribution.

  4. Dataset Choice and Split (Review #1, #2): Laparoscopic Roux-en-Y gastric bypass was chosen due to high clinical relevance, data availability, existing procedural standardization (SEVERE index) as stated in Section 3.2. The split follows the official partitioning of MultiBypass140, ensuring consistency with prior benchmarks and avoiding leakage in multi-task evaluations in future (e.g., phase recognition).

  5. Lack of Related Work on Imbalanced Data Techniques (Review #3) We acknowledge this oversight & will revise Section 2 to discuss key works such as focal loss, SMOTE, & label distribution learning. Nonetheless, our approach transforms discrete classification into a regression framework with label smoothing, which is orthogonal to typical resampling or reweighting techniques. This transformation provides a novel direction for mitigating imbalance.

  6. Image Resolution & Hyperparameter (Review #2): The 128×128 resolution offers a practical balance between performance & computational cost, with only marginal degradation compared to higher resolutions. Hyperparameters were tuned via grid or random search on validation data. We will include full training details in the revised manuscript.

  7. Reproducibility Concerns & Code/Dataset Release (Review #1, #2): Review #2 acknowledges that our submission provides a clear & detailed description of the algorithm to ensure reproducibility. The dataset extension was only withheld during submission to preserve anonymity. Given the strong community request, we consider releasing the code upon acceptance. Typo in Fig. 1 will be fixed (Review #1).

In summary, the reviews recognize the novelty, relevance, & thorough evaluation of our approach. Most concerns pertain to clarity and completeness rather than conceptual flaws. We are confident that with the proposed clarifications, the camera-ready manuscript will be significantly strengthened & beneficial to the MICCAI community.




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.

    Reject

  • Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’

    It is not clear how the claimed key contribution of the paper (BetaMixer based on Beta distribution) is actually operating. The authors do not discuss how the mapping of each discrete class distribution to the Beta distribution (Eq. 1) is operationalised as well as how the sampling works during training. In the absence of a) a theoretical motivation b) published source code and c) ablation study with respect to the mapping to the Beta distribution the recommendation is to reject this paper.



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’

    N/A



back to top