Related Experiment Videos
Adaptable preprocessing units and neural classification for the segmentation of EEG signals
1Institute of Medical Statistics, Computer Science and Documentation, Friedrich Schiller University Jena, Germany.
Methods of Information in Medicine
|October 16, 1999
Summary
This study presents a novel method for simultaneously adapting preprocessing units and neural classifiers for time series classification. This approach enhances generalization and reduces computational load for electroencephalogram (EEG) analysis.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Machine Learning
Background:
- Time series classification often requires careful selection of preprocessing units (PPUs) and neural classifiers.
- Existing methods may suffer from sensitivity to parameter choices or limited generalization capabilities.
Purpose of the Study:
- To develop a methodology for simultaneous adaptation of PPUs and neural classifiers for improved time series classification.
- To enhance generalization and reduce computational effort compared to purely neural systems.
Main Methods:
- An extension of the backpropagation algorithm was used to adapt preprocessing parameters.
- Quadratic filters with adaptable transmission bands served as PPUs for feature extraction.
- The method was applied to segment discontinuous neonatal EEG and EEG during deep sedation.
Main Results:
- Simultaneous adaptation of PPUs and classifiers led to improved generalization.
- Reduced input dimensionality decreased numerical effort.
- The method demonstrated robustness to parameter choices compared to fixed-parameter PPUs.
Conclusions:
- The developed methodology offers an effective approach for time series classification, particularly for complex biological signals like EEG.
- Adaptable preprocessing units combined with neural classifiers provide a powerful tool for analyzing discontinuous EEG signals.