EEG-DIM Visualizer

Yangfan Deng1,2, Hamad Albidah1, Helen X. Mao3, Ahmed Dallal1, Jijun Yin1, and Zhi-Hong Mao1
1University of Pittsburgh    2University of Maryland, College Park    3Yale University
arXiv Paper

Interactive visualization of EEG signals for sleep stages using 2D/3D PCA, t-SNE, and UMAP embeddings.

If you find our work useful, please . Thanks! :) 

@article{deng2024twostage,
  title={Two-Stage Hierarchical and Explainable Feature Selection Framework for Dimensionality Reduction in Sleep Staging},
  author={Deng, Yangfan and Albidah, Hamad and Mao, Helen X. and Dallal, Ahmed and Yin, Jijun and Mao, Zhi-Hong},
  journal={arXiv preprint arXiv:2409.00565},
  year={2024}
}

Pipeline Overview

1

NPZ upload

2

Spread-Spectral Features Calculation

3

PCA / t-SNE / UMAP

4

Interactive 2D/3D visualization

5

Top-5 feature contributions

Upload NPZ and Run EEG-DIM

Please upload an EEG sleep-stage NPZ file.

The file should contain an EEG signal matrix or feature matrix, with optional sleep-stage labels. Example public sleep datasets: ISRUC and PhysioNet Sleep-EDF.

A default visualization demo from the PhysioNet Sleep-EDF dataset is shown below; you can freely interact with the plot and switch between embedding views.

No file selected.

The API is called once per run. Plot tabs below use the cached response and do not re-run the backend.

Run Settings

Auto-search is optional and slower on CPU Basic. It is off by default for responsiveness.

Visualization Dashboard

Upload an NPZ file and run EEG-DIM to view embeddings.

Top-5 feature contributions

Rank Feature code Feature name Contribution
Run EEG-DIM to view top feature contributions.

Auto-search selected hyperparameters

method dimensions accuracy perplexity iterations learning_rate n_neighbors min_dist metric
Run EEG-DIM to view selected hyperparameters.