A Diffusion Model with Spatial Filters and Multi-Scale Attention for EEG Abnormality Detection
Poster No:
1329
Submission Type:
Abstract Submission
Authors:
Jae-Pil Lee1, Jin-Young Chung1, Hyun-Chul Kim1
Institutions:
1Kyungpook National University, Daegu, Korea, Republic of
First Author:
Co-Author(s):
Introduction:
Over recent years, diffusion models have demonstrated remarkable success across various domains from image generation to speech synthesis and signal processing. Notably, Denoising Diffusion Probabilistic Model (DDPM) (Ho et al., 2020) has been proven particularly promising through its iterative process of noise addition and removal, effectively learning underlying data distributions during the denoising process. However, their application to Electroencephalography (EEG) for abnormality detection is not sufficiently explored. Therefore, the aim of this study is to extend the capabilities of diffusion-based approach to enhance EEG abnormality classification by integrating spatial filtering and attention mechanisms to better capture the complex characteristics of neurological signals.
Methods:
We utilized the Temple University Hospital Abnormal EEG Corpus, a dataset containing 2,993 recordings (Normal: N =1,521; Abnormal: N =1,472) collected from 2,130 patients (Normal: N =1,237; Abnormal: N =893) whose EEG patterns were clinically evaluated. For preprocessing, we extracted the initial 120 seconds from each recording and applied time-reverse augmentation (Albaqami et al., 2023) to enhance model robustness. Building upon Diff-E (Kim et al., 2023) as our foundation, we designed an improved DDPM architecture by integrating specialized modules. Specifically, we integrated a learnable spatial filter layer with singular value decomposition initialization to better capture spatial relationships in EEG signals, along with Convolutional Block Attention Modules (CBAM) that implement dual attention mechanisms at multiple scales. Each CBAM processes signals via channel attention with both max-pooling and average-pooling operations, and spatial attention. Additionally, we incorporated stacked denoising autoencoders to progressively refine the signal representations by learning noise-invariant features. To systematically validate experimental reproducibility across different training instances, we performed 10 independent training runs with varying random seeds, each consisting of 100 epochs. Furthermore, we conducted extensive comparisons with established EEG signal processing architectures, including EEGNet (Lawhern et al., 2018), Deep4Net (Schirrmeister et al., 2017), ChronoNet (Roy et al., 2019), and TCN (Bai et al., 2018), alongside the baseline Diff-E model. We compared the performance across models using bootstrapping analysis.
Results:
Our evaluation demonstrated consistent performance advantages across multiple metrics compared to the existing EEG classification architectures. The proposed model achieved the highest accuracy (83.88% ± 0.44), surpassing other models, including Diff-E (average ± standard error; 82.65% ± 0.09%), EEGNet (78.01% ± 0.36%), Deep4Net (82.46% ± 0.56%), ChronoNet (77.46% ± 1.62%), and TCN (79.89% ± 0.12%; Fig. 2a). Additionally, the proposed model recorded the best recall performance (83.45% ± 0.50%) compared to other models (Diff-E: 81.80% ± 0.12%, EEGNet: 77.72% ± 0.41%, Deep4Net: 81.95% ± 0.51%, ChronoNet: 76.47% ± 1.75%, TCN: 79.29% ± 0.15%; Fig. 2b). Furthermore, the evaluation of specificity revealed significant improvements (proposed model: 78.49% ± 1.70%) over other models (Diff-E: 72.06% ± 0.83%, EEGNet: 74.44% ± 1.39%, Deep4Net: 76.03% ± 1.14%, ChronoNet: 65.00% ± 3.39%, TCN: 72.38% ± 1.05%; Fig. 2c).
Conclusions:
Our experimental results confirm that integrating spatial filtering, attention mechanisms, and diffusion-based denoising significantly enhances EEG abnormality detection. This substantial improvement over existing methods demonstrates the framework's effectiveness in handling complex EEG signals. Moving forward, future work will explore advanced visualization techniques such as Grad-CAM (Selvaraju et al., 2017) and layer-wise relevance propagation (Binder et al., 2016) to provide model interpretability and temporal feature learning.
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
EEG/MEG Modeling and Analysis 1
Methods Development
Multivariate Approaches
Keywords:
Electroencephaolography (EEG)
Other - Abnormality Detection; Attention Mechanism; Diffusion Model; Spatial Filtering; Stacked Denoising Autoencoders; Time-reverse Augmentation
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Which processing packages did you use for your study?
Provide references using APA citation style.
2. Bai, S., Kolter, J. Z., & Koltun, V. (2018). An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271.
3. Binder, A., Montavon, G., Lapuschkin, S., Müller, K. R., & Samek, W. (2016). Layer-wise relevance propagation for neural networks with local renormalization layers. In Artificial Neural Networks and Machine Learning–ICANN 2016: 25th International Conference on Artificial Neural Networks, Barcelona, Spain, September 6-9, 2016, Proceedings, Part II 25 (pp. 63-71). Springer International Publishing.
4. Ho, J., Jain, A., & Abbeel, P. (2020). Denoising diffusion probabilistic models. Advances in neural information processing systems, 33, 6840-6851.
5. Kim, S., Lee, Y. E., Lee, S. H., & Lee, S. W. (2023). Diff-E: Diffusion-based learning for decoding imagined speech EEG. arXiv preprint arXiv:2307.14389.
6. Lawhern, V. J., Solon, A. J., Waytowich, N. R., Gordon, S. M., Hung, C. P., & Lance, B. J. (2018). EEGNet: a compact convolutional neural network for EEG-based brain–computer interfaces. Journal of neural engineering, 15(5), 056013.
7. Roy, S., Kiral-Kornek, I., & Harrer, S. (2019). ChronoNet: A deep recurrent neural network for abnormal EEG identification. In Artificial Intelligence in Medicine: 17th Conference on Artificial Intelligence in Medicine, AIME 2019, Poznan, Poland, June 26–29, 2019, Proceedings 17 (pp. 47-56). Springer International Publishing.
8. Schirrmeister, R. T., Springenberg, J. T., Fiederer, L. D. J., Glasstetter, M., Eggensperger, K., Tangermann, M., ... & Ball, T. (2017). Deep learning with convolutional neural networks for EEG decoding and visualization. Human brain mapping, 38(11), 5391-5420.
9. Selvaraju, R. R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., & Batra, D. (2017). Grad-cam: Visual explanations from deep networks via gradient-based localization. In Proceedings of the IEEE international conference on computer vision (pp. 618-626).
Acknowledgement: This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (No. RS-2022-00166735 & No. RS-2023-00218987).