List of Papers Browse by Subject Areas Author List
Abstract
Early diagnosis of neurodegenerative diseases is crucial for effective intervention and treatment planning. However, conventional screening tests such as Mini-Mental State Examination (MMSE) often produce false-negative issues. While electroencephalogram (EEG) signals contain valuable neurophysiological information, multi-class classification remains challenging due to subtle differences between conditions, with existing methods achieving around 50-60% accuracy. Therefore, we propose SSPNet, a novel deep learning framework for multi-class classification of neurodegenerative diseases using spatio-spectral portraits derived from EEG signals. Our approach extracts spatio-spectral images that maximize neurophysiological differences between Alzheimer’s disease, frontotemporal dementia (FTD), and cognitively normal subjects, utilizing minimal frequency bands encoded through specialized asymmetric convolutional blocks and attention mechanisms. To our knowledge, this represents the first attempt to use EEG spatio-spectral portraits for multi-class classification of neurodegenerative diseases. The proposed SSPNet significantly improves accuracy to 72.22% compared to existing EEG-based methods for multi-class classification. It also demonstrates notably lower false-negative rates for FTD patients compared to MMSE, thus accelerating practical clinical application.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/5358_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)
Alzheimer’s disease, Frontotemporal dementia and Healthy subjects dataset: https://openneuro.org/datasets/ds004504/versions/1.0.8
BibTex
@InProceedings{KimHo_SSPNet_MICCAI2025,
author = { Kim, Ho-Jung and Park, Dogeun and Jang, Jeong-Woo and Ju, Young-Gi and Won, Dong-Ok},
title = { { SSPNet: Towards Feasible Spatio-Spectral Portraits-Based Deep Learning Framework for Neurodegenerative Disease Multi-Classification } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15963},
month = {September},
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors have proposed a method based on CNNs to classify Alzheimer’s disease (AD) from frontotemporal lobe dementia (FTD) using alpha and delta EEG spectrograms. They also employ a cross-attention mechanism to enhance feature extraction. The dataset includes 36 individuals in the AD group, 23 in the FTD group, and 29 in the control group.
- 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 use of a dual-band CNN that processes the Alpha and Delta bands separately.
- The implementation of a cross-attention mechanism.
- 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.
-
Contrary to the authors’ claim, the approach is not novel. In fact, there are several references demonstrating similar or even better performance, e.g.: A Novel CNN-Based Framework for Alzheimer’s Disease Detection Using EEG Spectrogram Representations (https://www.mdpi.com/2075-4426/15/1/27).
-
The justification for including a third cohort, the control group is missing
-
- 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.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The authors’ approach lacks novelty, as similar methods have already been explored in the literature. For instance, the study titled “A Novel CNN-Based Framework for Alzheimer’s Disease Detection Using EEG Spectrogram Representations” demonstrates a comparable methodology with validated performance. Without clear advancements or differentiating factors, IMO does not offer a significant contribution beyond what is already established in the field.
- 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 study proposes a spatio-spectral image based deep learning framework (SSPNet) specifically targeting EEG multi-class classification of neurodegenerative diseases.
- 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.
- Incorporating several attention modules to capture relevant information in the spatio-spectral data of the two selected EEG bands.
- Subject-based dataset splitting in training/testing
- Considering multiclass EEG classification which is a scarcely addressed topic.
- 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.
-
Authors mention FTD/AD/CN results in literature were limited to 50-60%, but they are missing studies that gave results as high as 80%. Examples: a) https://link.springer.com/article/10.1007/s11571-024-10153-6 b) https://ieeexplore.ieee.org/abstract/document/10308628 c) https://iopscience.iop.org/article/10.1088/1741-2552/ac05d8 Authors are advised to enhance the comparison table or explain why they didn’t compare to such works from literature
-
Missing elaborate intuition and proper citation as to why specifically the alpha and delta bands were chosen
-
Was EfficientNet used with Transfer learning? Please clarify in text.
-
Please mention how all the 19 EEG channels were used for creating the spatio-spectral images and give more details on multitaper PSD estimation
-
Experiment 2 details are not clear enough in the paper. Please clarify how the ability to identify false-negative FTD cases was tested in this experiment.
-
- 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 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
-
‘As shown in Table2, most models, including EfficientNet and other CNN-based architectures, struggled with FTD classification.’ » only EfficientNet and EEGNet were used, what is most models referring to?
-
Please report other papers that used PSD images for disease classification using EEG and indicate how this work is different.
-
- 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?
Some experimental details are not clear in the paper’s current form and require further clarification.
- 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 authors proposed a novel deep learning model called Spatio-Spectral Portraits Network (SSPNet) to classify and differentiate between two neurodegenerative disorders, Alzheimer’s disease (AD) and Frontotemporal Dementia (FTD), as well as cognitively normal (CN) individuals. The model takes as input 2D RGB topographic maps generated from the alpha and delta frequency bands of the preprocessed EEG signals. Each frequency band is processed independently: vertical asymmetric convolutions are applied to the alpha band maps, while horizontal asymmetric convolutions are used for the delta band maps. These two streams of features are then merged through a cross-attention block, which captures inter-band relationships by learning both shared and contrastive representations. The resulting high-level features are concatenated and further fused using convolutional layers to effectively integrate the unique and complementary EEG characteristics present in the alpha and delta bands. The model was evaluated in two key experiments: a multiclass classification task involving the three groups (AD, FTD, and CN), and the identification of FTD cases that received false-negative results on the Mini-Mental State Examination (MMSE). In both cases, SSPNet outperformed existing state-of-the-art models such as EEGNet and EfficientNet, as well as traditional machine learning classifiers like support vector machines (SVM). These results suggest that spatio-spectral EEG features, as modeled by SSPNet, capture subtle yet meaningful differences in brain activity associated with neurodegenerative disorders more effectively than existing approaches.
- 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 proposed model incorporates prior neuroscientific knowledge about EEG spatial patterns: The model processes alpha band data using vertical convolutional kernels to exploit the known upper-lower asymmetry commonly observed in alpha activity. In contrast, delta band data, which exhibits left-right asymmetry, is processed using horizontal kernels. This tailored convolutional approach aligns the model’s structure with established neurophysiological characteristics of the respective frequency bands.
- The inclusion of a cross-attention block: This block enables the model to extract high-level representations by learning both shared and distinctive features across the alpha and delta bands, effectively capturing complementary patterns essential for distinguishing between neurodegenerative disorders.
- 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.
- It is unclear why the authors encoded the spatio-spectral images in RGB: The authors state that “Spatio-spectral images are constructed using Power Spectral Density (PSD) information, encoded in RGB color channels to retain frequency-based characteristics of EEG signals.”. However, they do not provide any justification for how this approach retains such information.
- The authors mention the importance of early diagnosis for neurodegenerative disorders. However, this work does not contribute to providing an early detection mechanism since the used data is from subjects who were already diagnosed. Thus, the introduction is somewhat misleading for the reader.
- Lack of comparison to other neuroimaging approaches: There is a lack of comparison to other neuroimaging methods or at least a justification for why using EEG instead of other neuroimaging data. For example, Nguyen et al, (2023) successfully differentiated between AD-TFD-NC with an accuracy of 86%. Nguyen, H. D., Clément, M., Planche, V., Mansencal, B., & Coupé, P. (2023). Deep grading for MRI-based differential diagnosis of Alzheimer’s disease and Frontotemporal dementia. Artificial Intelligence in Medicine, 144. https://doi.org/10.1016/j.artmed.2023.102636
- 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?
The authors proposed a novel approach that is capable of extracting spatio-spectral features from the alpha and delta bands of EEG data and manages to successfully improve the classification ability of neurodegenerative diseases. However, there are a few things that the authors must address mainly referred to in the weaknesses section.
- 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.
The authors rebuttal is sufficient enough to justify acceptance. I would just like the authors incorporate the changes mentioned. Additionally, the authors should improve their justification in the manuscript for not comparing with other neuroimage techniques and their RGB encoding approach.
Author Feedback
We would like to express our sincere gratitude to the Reviewers for their constructive comments. We have prepared detailed responses to each comment and will incorporate these improvements into our revised manuscript. Our responses are organized as follows:
- Classification Target Groups and Performance Differences(R1): Our study used a public dataset (Miltiadous et al., 2023) for AD-FTD-CN classification. Direct comparisons with the referenced studies are limited due to differences in group composition or their focus on binary metrics. Our contribution extends beyond performance improvements; we present a framework differentiating between dementia types (AD and FTD) and normal cognition using EEG data—a clinically significant challenge. We will clarify these distinctions and emphasize clinical implications in our revision.
- Methodological Differentiation and Novelty(R2): Contrary to your concern, our EEG spatio-spectral portrait features utilize scalp topomaps that mapping frequency power to EEG channel locations, not spectrograms. Unlike spectrograms that represent signals in time-frequency domain, topomaps illustrate spatial power distribution across the entire scalp. Furthermore, Our novel deep learning framework incorporates asymmetric convolution blocks and attention mechanisms, demonstrating superior performance in AD-FTD-CN multiclass classification.
- Definition of Early Diagnosis(R4): We defined early diagnosis as cases where EEG-based models detect patient groups (e.g., FTD) missed by existing screening tests (MMSE), which has high false-negative rates for FTD. Our study evaluated EEG-based classification as a complementary tool using a dataset with MMSE scores and DSM-5 clinical diagnostic labels. We will clarify that our research aims to ‘complement the screening process’ and support the diagnostic process for neurodegenerative diseases.
- Justification for Using EEG Modality(R4): EEG is appropriate for early screening due to high temporal resolution, non-invasiveness, and portability. While neuroimaging offers diagnostic accuracy, it is limited by cost and accessibility. In addition, MMSE has limited sensitivity for FTD and EEG complements both methods for clinical workflow integration. We will incorporate neuroimaging comparisons in future research.
- Experimental Details for Revision 5-1. Alpha and Delta Bands Selection(R1): Alpha band shows decreased activity in occipito-parietal regions in AD and reflects cholinergic neurotransmission damage. Delta band displays increased frontal lobe activity in FTD (Jeong, 2004). Caso et al. (2011) showed EEG frequency characteristics effectively differentiate AD and FTD, and alpha/delta ratios improve disease monitoring and classification (Cassani et al., 2018). We will add citations to clarify this rationale. 5-2. EfficientNet Implementation(R1): EfficientNet was trained without transfer learning. 5-3. EEG Channel Usage and Multitaper PSD Estimation(R1): Our approach created 128×128 pixel images by spatially interpolating power values from 19 electrodes, visualizing brain activity patterns that capture disease-specific changes. We used MNE-Python’s multitaper PSD method with 7 DPSS tapers to balance frequency resolution and estimation variance. These details will be added to our revision. 5-4. Identifying False Negative FTD Cases(R1): In Experiment 2, we examined FTD patients with MMSE scores ≥24 who might be misclassified as normal. To evaluate whether our proposed framework could accurately identify false-negative cases of MMSE, we tested it on 9 FTD patients along with 9 randomly selected subjects from each of the AD and CN groups. We will clarify this methodology and its clinical relevance. 5-5. RGB Encoding for Spatio-spectral Images(R4): RGB encoding makes brain activity patterns more intuitive and enhances deep learning model training. Future work will compare RGB with various encoding methods including gray-scale performance.
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.
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 #3
- 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’
While some novelty in both conceptual approach and technical implementation is explained in the rebuttal, all reviewers come up with similar works that have not been cited or compared to. Therefore the contribution to the field therefore remains largely unclear. Also the clinical motivation and context (e.g. why is screening marker MMSE used) is not entirely clear.