Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

SincVAE: A new semi-supervised approach to improve anomaly detection on EEG data using SincNet and variational autoencoder

Created by
  • Haebom

Author

Andrea Pollastro, Francesco Isgr o, Roberto Prevete

Outline

This paper proposes SincVAE, a novel semi-supervised deep learning method for epileptic seizure detection using electroencephalography (EEG) data. While existing supervised learning methods suffer from poor performance due to the difficulty of labeling epileptic seizure data and data imbalance, SincVAE overcomes these challenges by training using only seizure-free data. SincVAE integrates specialized bandpass filter array learning into the first layer of a variational autoencoder (VAE), simplifying the preprocessing step and enabling effective seizure detection and pre- and post-ictal monitoring.

Takeaways, Limitations

•
Takeaways:
◦
Improving the accuracy of epileptic seizure detection based on EEG data.
◦
Demonstrating the effectiveness of semi-supervised learning methods for solving data imbalance problems.
◦
Increased efficiency by streamlining preprocessing steps.
◦
Suggests the possibility of monitoring preictal and postictal stages.
•
Limitations:
◦
Lack of diversity in the models against which SincVAE's performance can be compared.
◦
Lack of validation in real clinical settings.
◦
Absence of performance evaluation on large datasets.
◦
Further research is needed on the generalization performance of the SincVAE model.
👍
Made with Slashpage