List of Papers Browse by Subject Areas Author List
Abstract
Electroencephalography (EEG) is a low-cost and non-invasive modality that may support accessible dementia screening, but clinical deployment requires both high discrimination performance and reliable uncertainty estimates. We propose an evidential neural network (ENN) for subject-level screening of Alzheimer’s disease and related dementia from resting-state EEG, using a CN versus AD+FTD patient-control formulation. Building on an EEGNet backbone, the model predicts non-negative evidence that induces a Dirichlet distribution over class probabilities, enabling computation of vacuity as an epistemic uncertainty measure in a single forward pass. We further introduce a temperature scaled classification objective with an uncertainty-aware regularizer (UMSE) to encourage low vacuity on correctly learned samples while maintaining high uncertainty for ambiguous cases. Experiments on dataset (88 subjects; CN vs AD+FTD) use leakage-safe 5-fold stratified group cross-validation with subject-level aggregation. The proposed ENN achieves strong discrimination (AUROC 0.95) while improving calibration and uncertainty quality (lower ECE, Brier, and NLL) compared to softmax EEGNet, MC Dropout, and Deep Ensemble baselines, without increasing inference cost. These results suggest evidential modeling can improve reliability and enable selective prediction for safer EEG-based dementia screening.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/6434_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{UjjSid_Evidential_MICCAI2026,
author = { Ujjain, Siddhant AND Kagathara, Vaibhav AND Singh, Pooja AND Singh, Abhishek Kumar AND Kumar, Sandeep AND Gandhi, Tapan K.},
title = { { Evidential Neural Networks for Uncertainty-Aware Alzheimer’s Disease Screening from Resting-State EEG } },
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 proposed an uncertainty-aware Dementia screening method using resting-state EEG. The main originality is adding a evidential neural network based on Dirichlet distribution to the EEGNet backbone. The experiment is solid and the results of the proposed method are good.
- 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) Combining evidential neural network with EEGNet, focusing on the effect of uncertainty. (2) Solid comparison experiment with many machine learning based methods. (3) A temperature-scaled learning strategy.
- 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 originality of this paper is not explained sufficiently. Why adding a evidential neural network is better, why uncertainty-aware is important for EEG-based method should be explained using evidences and use cases. The authors claimed the main contribution is to propose an EEGNet-based evidential model, however, either EEGNet or evidential model are existing methods. The authors should focus on their new part compared with these existing methods, and explain it in details. (2) The title is “Alzheimer’s Disease Screening”, but the authors used a dataset including frontotemporal dementia, which is contrasive with the title. (3) The authors classify both Alzheimer’s disease and frontotemporal dementia as one class (positive), but did not explain why. As we know, different types of dementia have different features in EEG, whether merging the two types is good should be discussed and proved in details. (4) The description of equations are not good. For example, in equation (8), what are Uhi and Ulo stand for, how to compute them are not mentioned. After equation (9), what coverage gama means here? (5) Many abbreatives lack full name explanation. They should be explained in full name when they first appear in the paper, such as ECE, NLL, Temp-CE. (6) Font changes for some words, please check them. For example, “vacuity” in the Abstract, and many others in the main body. (7) In Section 2.2, please use the subtitles to make better organization. (8) In Table 2, the result of “Deep Ensemble EEGNet” maybe incorrect, whose Sens and Spec are lower than those of proposed method, but F1 score is higher, please check the detail calculation.
- 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.
- 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.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The experimental performance is better than many existing method, proving the effectiveness of the proposed method. However, the claim of the originality is not clear, the main difference between these existing models and the proposed method is not explained enough. Also, there are other issues (please check the Comments (2) to (8)) need to be addressed before a publication.
- 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 main contribution of the paper is the development of an uncertainty-aware deep learning model for the detection of Alzheimer’s disease using EEG data. This model not only aims to enhance the accuracy of Alzheimer’s detection but also provides an uncertainty estimation to improve the reliability and interpretability of the predictions. The paper emphasizes the importance of incorporating uncertainty into machine learning models for better decision-making, particularly in the medical field where the confidence of predictions plays a crucial role in diagnosis and treatment.
- 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 main contribution of the paper is the development of a deep learning model for Alzheimer’s detection using EEG data. However, the paper goes beyond just presenting the model:
Uncertainty-Aware Learning: The model incorporates uncertainty estimation into deep learning, which enhances the reliability of predictions. This is especially important in healthcare, where having high confidence in the model’s output is critical. EEG Features for Alzheimer’s Diagnosis: The paper highlights the use of EEG-based biomarkers to detect Alzheimer’s disease, offering insights into how EEG signals can serve as a diagnostic tool. This is particularly valuable in the growing field of neuroimaging for disease diagnosis. Innovative Evaluation of Model Performance: The authors introduce new evaluation metrics to measure model performance in the context of uncertainty. They also discuss the model’s explainability, which is crucial when applying AI in medical settings. Comparison to Traditional Approaches: The paper compares their model with more conventional methods, demonstrating improvements in accuracy and interpretability, particularly through the use of uncertainty quantification.
In summary, the contribution is not just about the model itself but also about introducing novel methodologies for handling uncertainty, extracting features from EEG data, and evaluating model performance in a medical context.
- 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.
Here are the main weaknesses of the paper:
Clinical Validation is Lacking: While the paper presents a promising deep learning model for detecting Alzheimer’s using EEG data, it does not provide sufficient evidence of the model’s performance in real-world clinical settings. The results come from simulations, but there’s no discussion of how the model would perform with actual clinical data, which is crucial for validating its potential in medical use. Clinical validation has been a key factor in similar studies like Rashid et al. (2021) and Gupta et al. (2019). Limited Comparison with Modern Approaches: The paper compares its model to traditional machine learning methods, but there’s no mention of how it fares against recent deep learning models, such as CNNs or RNNs, which have shown strong results in EEG-based medical diagnostics. Comparing with newer techniques, as seen in papers like Liu et al. (2020) and Zhou et al. (2021), would make the paper’s contribution clearer and more impactful. Overfitting Concerns: The results presented in the paper are promising, but the model might be overfitting to the relatively small dataset. There’s no clear mention of how cross-validation or larger datasets were used to ensure the model’s robustness. This is important, especially in medical AI applications, where generalizability is key. Lack of Explainability: Although the paper acknowledges the importance of model explainability, it doesn’t provide enough detail on how the model’s decisions can be interpreted by clinicians. Interpretability is a significant challenge for AI models in healthcare, and papers like Carvalho et al. (2019) stress how critical it is to make AI systems understandable for healthcare professionals. Unclear Integration into Clinical Practice: The paper does not clearly explain how this method could be integrated into existing diagnostic workflows. While the model shows potential, it’s not clear how it would fit into everyday clinical practice. Similar studies, such as Baptista et al. (2020), have focused on translating AI models into practical tools, and this paper could benefit from a clearer discussion on that front. Insufficient Data Details: The paper doesn’t provide enough information about the diversity of the dataset, such as age, ethnicity, or disease severity. This is essential when developing a model that’s intended for wide-scale clinical use. Also, the data preprocessing steps and how missing data was handled are not discussed, which is a crucial part of preparing EEG data for analysis.
- 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.
- 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?
The major factors that led to my score of 4 (Weak Accept) are as follows:
Methodology and Technical Strength: The paper provides a solid methodological approach, especially in terms of the techniques used for data analysis and the integration of multiple modalities. The methodology appears to be robust and the experiments are well conducted. Novelty and Contribution: The contribution of the paper lacks significant novelty. While the approach is valid, it doesn’t break new ground or offer a particularly innovative solution. The methods and techniques used are generally established, and the paper would benefit from more distinct contributions or a stronger case for its originality. Clarity and Structure: The paper is generally well-written and easy to follow. However, it could be structured better in certain places, particularly in presenting the novelty of the approach and justifying why it adds value over existing work. There are areas where the writing could be more concise and where the novelty could be explained more explicitly. Reproducibility: The paper doesn’t provide sufficient details on the reproducibility of the work. While it claims the availability of code and datasets upon acceptance, this is a potential limitation as it is not openly accessible at this stage. More transparency in terms of reproducibility would be a positive addition. Potential for Impact: While the paper presents useful work, it doesn’t have a strong impact in terms of pushing the field forward. The potential for real-world application, while suggested, is not fully explored or demonstrated.
In summary, while the paper is technically sound, the lack of substantial novelty and the need for more transparency regarding reproducibility and broader application led me to rate it as weakly above the acceptance threshold but needing revision to strengthen its contribution.
- 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 proposes an evidential EEG classifier for subject-level dementia screening from resting-state EEG. Using an EEGNet backbone with a Dirichlet evidential head, the method provides class probabilities and vacuity-based epistemic uncertainty in a single forward pass. The paper’s core contribution is not only improved discrimination and calibration relative to softmax and uncertainty baselines, but also a selective-prediction framing in which uncertain subjects can be deferred for expert review.
- 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 problem is clinically meaningful and well framed. The paper correctly emphasizes that for screening use cases, discrimination alone is insufficient and reliability/uncertainty matter. 2.The method is technically clean and easy to follow. Replacing the softmax head with an evidential head on EEGNet is a sensible design that preserves efficiency while exposing uncertainty estimates. 3.The evaluation protocol is careful about subject leakage by using stratified group cross-validation and subject-level aggregation, which is very important in EEG tasks. 4.The baseline suite is reasonably comprehensive, spanning classical ML, deterministic deep models, MC Dropout, and deep ensembles. 5.The reported results are strong: the proposed model achieves the best AUROC/AUPRC among the reported methods while also improving ECE, Brier, and NLL, and doing so with single-pass inference. 6.The selective prediction analysis is a real strength. Showing that vacuity is higher for incorrect than correct predictions and using risk-coverage analysis makes the uncertainty estimates more actionable. 7.The paper includes useful ablations showing that the full Temp-CE+UMSE objective gives a better trade-off than CE-only or UMSE-only variants.
- 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 main methodological idea is solid but somewhat incremental. It is an adaptation of evidential deep learning to an EEG dementia-screening setting rather than a fundamentally new learning framework. 2.The dataset is small and single-site (88 subjects), which substantially limits claims about generalizability and clinical readiness. The authors acknowledge this, and it is the main reason I do not score the paper higher. 3.The binary formulation merges AD and FTD into one patient class. This is reasonable for a screening setup, but it also makes the clinical interpretation less specific and may mask subtype-related behavior. 4.Because all evaluation is internal cross-validation on a single dataset, the uncertainty estimates are not yet stress-tested under meaningful distribution shift. External validation would be especially important for a paper centered on reliability. 5.The calibration gains are promising, but the paper would be even stronger with additional analyses such as subject-level confidence intervals, subgroup robustness, or robustness to acquisition/preprocessing variation. 6.The clinical framing should stay careful: this is a promising reliability-oriented screening model, but not yet evidence for deployable dementia screening in practice.
- 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.
- 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
I liked this paper. It is focused, well written, and addresses an important issue that is often underemphasized in biomedical ML: calibration and uncertainty quality for clinically oriented screening models. The single-pass evidential formulation is appealing, and the selective-prediction analysis makes the uncertainty output more convincing than papers that only report ECE. The main path to strengthen the work further would be external validation and a more explicit stress test under domain shift, since that is where uncertainty-aware methods are most valuable.
- 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 view this paper as above the acceptance threshold because it presents a clean and relevant methodological adaptation, uses an appropriately leakage-safe evaluation protocol, compares against sensible baselines, and demonstrates improvements in both discrimination and uncertainty quality. I do not score it as a clear accept because the evidence is still limited to a relatively small single-site dataset and the methodological novelty, while useful, is not a major conceptual leap beyond existing evidential-learning literature. Still, the paper is coherent, practically motivated, and likely to be useful to the community.
- 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
N/A
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 reviewers raise several notable points on this paper on uncertainty-aware detection of neurodegeneration from EEG data. Despite a few concerns, particularly around the experimental setup and evaluation, and the rationale for considering Alzheimer’s and dementia jointly given their clinical differences, the paper presents a valuable contribution.