Abstract

Reliable prediction of seizure outcomes after surgical intervention before ablative surgery could play a critical role for tailoring epilepsy treatment. However, for diverse patient populations, accurate and personalized predic-tions remain challenging with traditional methods. Current methods rely heavily on clinical expertise and experience, and data driven tools may help in supporting clinicians to make more informed surgical decisions. This study presents a novel deep learning-based spatio-temporal graph neural network (ST-GNN) model to predict reduction in seizure frequency utiliz-ing high-quality stereo electroencephalography (sEEG) and structural mag-netic resonance imaging (MRI) data. sEEG and MRI data are curated from patients with pharma-coresistant refractory epilepsy and suspected wide/complex seizure networks or multifocal epilepsy. A total of 10 pedi-atric patients with sEEG contacts in the thalamus were considered, where data from multiple ictal events was used to train the model. Our ST-GNN model integrates local and global connectivity using graph convolutions with multi-scale attention mechanisms to capture patterns between diffi-cult-to-study regions such as the thalamus and cortical/subcortical regions, both from MRI and sEEG. The model achieved an accuracy of 90.4%, and 75.4% in predicting seizure outcomes for seizure-wise and patient-wise prediction respectively. Edge-level connectivity analysis highlighted the thalamus and mid insula regions as key regions. Our findings underscore the potential of new connectivity-based deep learning models leveraging multimodal data for enhancing the prediction of seizure outcomes and tailoring treatment planning for epilepsy. Our multi-modal approach can help inform AI-assisted personalized epilepsy treatment planning. Code is avail-able on our GitHub.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4834_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{AhaArt_MultiModal_MICCAI2025,
        author = { Aharonyan, Artur Arturi and Amir, Syeda Abeera and Wittayanakorn, Nunthasiri and Linguraru, Marius George and Oluigbo, Chima and Anwar, Syed Muhammad},
        title = { { Multi-Modal Graph-Based Machine Learning for Predicting Surgical Outcome in Epilepsy Patients } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15971},
        month = {September},

}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper presents a multimodal graph network-based method that integrates stereo-electroencephalography (sEEG) and structural MRI data to predict surgical outcome in epilepsy patients, and applies the model to analyze brain regions accordingly.

  • 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. Original Use of Multimodal Data.This paper demonstrates a 13.3% accuracy gain when combining sEEG with MRI over sEEG-only models, validating the added value of structural context for understanding seizure networks.
    2. Addressing Underexplored Brain Regions. This paper leverages rare thalamic sEEG recordings to study subcortical-cortical interactions, advancing understanding of poorly characterized seizure propagation pathways.
  • 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. Insufficient experimentation. First, this paper does not provide a baseline comparison experiment. Even though, as the authors state, this is a pioneering pipeline for predicting the outcome of epilepsy surgery, no comparative experiments have been able to prove that the GAT-GNN is the best framework. Second, the absence of ablation experiments prevent me from confirming the validity of the model structure. For example, the model architecture proposed by the authors includes two layers of GAT. However, is it designed in such a way that it could be analyzed using the attention mechanism in subsequent feature analyses, or does GAT have better performance than GCN in this task?

    2. Model performance. This paper evaluated the model performance using data from 1) ictal events and 2) patient-wise approach. In the ictal events, the accuracy of seizure-wise classification using multimodal data is 90.4%. The authors do not clarify whether such experiments are In-subject or Cross-subject tasks. In the case of the In-subject task, much existing work [1] has been able to achieve very high accuracy in seizure prediction relying on scalp EEG alone. This greatly affects the value of this paper。

    [1] Syed Muhammad Usman, Shehzad Khalid and Zafar Bashir, Epileptic seizure prediction using scalp electroencephalogram signals[J]. Biocybernetics and Biomedical Engineering, 41(1):211-220. [2] U. Rajendra Acharya, Shu Lih Oh, Yuki Hagiwara, Jen Hong Tan and Hojjat Adeli, Deep convolutional neural network for the automated detection and diagnosis of seizure using EEG signals[J]. Computers in Biology and Medicine, 100:270-278.

  • 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?

    This article is not innovative in methodology, but rather applies graph neural networks combined with multimodal data to postoperative epilepsy outcome prediction. This article is confusing in its writing, e.g., the abstract states that STGCN was used, while later in the article it is referred to as GAT-GNN. The experimental design and results were also inadequate.

  • 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

    The proposed method is a graph convolutional neural network for predicting epilepsy progression with the intention of identifying the seizure onset zone(s) using the network’s own attention modules. This is a good application of existing machine learning methods for binary/categorical classification to a more clinically relevant problem (determining SOZs for resection, for example). The ultimate take-away from the paper is the confirmation of the role of the thalamus in seizure pathways, which has been known in the literature for quite some time, enough to motivate thalamic sEEG and DBS, the former being used in this paper.

  • 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.
    • Multiple types of validation are well-appreciated, especially the “patient-wise” cross-validation which best reflects use of the method on new, unseen patients without patient-specific retraining.
    • Figures 3 and 4 largely verify that the network is indeed looking at the relevant areas of the brain where the thalamic sEEG electrodes have been implanted.
    • A number of the paper’s limitations are clearly and succinctly described on page 8.
    • The pre-processing of the MRI and sEEG information is largely well-described although more information for sEEG temporal correlations would be warranted.
  • 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 is slightly vague as to what is meant by “seizure outcomes” and the type of epilepsy surgery used, especially given that one type (ANT DBS) can be relatively patient-constant and others (e.g. focal resection) are more fundamentally patient-variable. This would have a large effect on how the results of the classification algorithm (and therefore what it attends to) would be interpretted.
    • Figures 3 and 4 demonstrate the general plausibility of the network on the patient population (or large subgroups) but do not show how this could apply to an individual patient. Showing an individual patient’s results would show the clinical utility of this attention-evaluation approach to finding the SOZ.
    • The authors may have some data leakage in the multi-modal and sEEG-only experiments as if an individual patient’s data is split into multiple folds, it is hard to say that the information used in the training folds can be collected without leaking information from the testing fold (i.e. there is some patient-specific connectivity fingerprint to memorise (encoded in the MRI distance information) rather than learn something more epilepsy-specific) or be infeasible (e.g. if we have to wait six-months to get the post-op results, then the network can’t be meaningfully re-trained for each patient on-the-fly)
    • Given the low number of patients (10) and the relatively unbalanced data (7 vs. 3) it is unclear if the results in Table 1 are better than non-informative models (58% for uniformly random guesses with distribution 70% (Group 0) & 30% (Group 1) or 70% accuracy for always guessing Group 0).
    • Page 5: “Cov” (with quotations) in Pearson Correlation equation
    • Inconsistent train-test splits between Figure 1 (80-20) and the ten-fold CV described on page 6 (90-10)
  • 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.

    (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?

    I think that this paper has the kernel of good ideas that could be interesting to the community as well as shown the application of MIC-style paradigms (e.g. graph-based machine learning for medical image computing) for a distinctly CAI-focused application. I am thus leaning towards acceptance, despite the weaknesses outlined above. It is also a clearly and concisely written paper than has largely focused on reproducibility and validity above technical novelty, which would be a good counter-balance to MICCAI’s increasing focus on new machine learning architectures.

  • 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

    This paper introduced a spatio-temporal graph neural network (ST-GNN) approach to integrate anatomical MRI and sEEG data from epilepsy patients to predict surgical outcomes.

  • 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 proposed method is potentially useful for treatment planing for epilepsy patients to predict the outcome of surgery. 2) The method is developed based on high quality sEEG data with thalamic implantation. 3) The ST-GNN integrates both anatomical MRI and sEGG. 4) The results were avalidated using LOOCV.

  • 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) Relavtively few data sets that raised a concern about the geralizability of the method to out-of-the-domain data. 2) Intuitively, the outcome may also depends on the resected brain region. But the resected brain regions are not considered in the model, which may be limited by the sample size. 3) More details about the ST-GNN are needed. Some notations in Fig. 1, e.g. X, Y, are not explained.

  • 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.

  • 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

    Adding more details about the ST-GNN is helpful to understand the method. The notations in Fig. 1 need to be explained.

  • 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 fits in both MIC and CAI and introduces a novel and reasonable method to predict treatment outcome using anatomical MRI and sEEG.

  • 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

N/A




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



back to top