Research in recent years has shown a correlation between stress level and several nonlinear measures of the Electroencephalogram. We use this to estimate the stress level of a person and help him lower it using biofeedback.
Abstract
Research in recent years has shown a correlation between stress level and several nonlinear measures of the Electroencephalogram. We use this to estimate the stress level of a person and help him lower it using biofeedback.
Introduction
EEG waves are though to be non-linear signals. Recent work has shown a connection between the fractal dimension of specific brain signals to the mental condition of people during several tasks (for example; during meditation). We tried to use this approach to estimate mental stress of people in general in order to help them lower it by learning to control a parameter they are otherwise unaware of, namely the fractal dimension of the EEG.
In order to achieve this we selected a suitable algorithm for estimating the fractal dimension of sampled EEG signals, and adjusted its’ parameters for our purposes. The algorithm was first run offline on MATLAB using data from previously sampled signals. We then implemented the algorithm in a real-time environment (VC++) with a graphic user interface. This enabled us to show users the fractal dimension of their EEG graphically. The application was tested on several people and was proved efficient.
Basic Approach
This is a diagram of the basic algorithm we used:
Signals are sampled from the subject’s head after amplification. These signals are then filtered in order to remove common noises, and then reconstructed in the signal space according to a time delay parameter. The optimal delay parameter was estimated using the Takens delay theory. This reconstruction results in a signal of a higher dimension, which is the first step in dynamic systems analysis of time series data.
The next step of the algorithm is another filtering stage called Singular Value Decomposition (SVD). This stage decomposes the higher-dimensional signal into eigenvalues and eigenvectors. Filtering is achieved by removing eigenvectors associated with small eigenvalues.
Using this filtered signal, the algorithm then estimates the fractal dimension using a method proposed by Grassberger and Procaccia. Finally, the fractal dimension is shown to the user by means of a colored line on the screen. The length of the line is proportional to the fractal dimension.
System validation
In order to test the system and see if a person can indeed learn to control the fractal dimension of his EEG, we tested the system on four people in different days and mental conditions.
One of our subjects is prone to migraines. While testing the system it was found that the fractal dimension of his EEG was indeed much lower on days he had a migraine, compared to days when he did not.
It was also discovered that people could indeed learn to control the fractal dimension of their EEG. The following graph shows the fractal dimension of a person over time during a typical session with the system. One can see the fractal dimension rising progressively over time when using biofeedback.
Conclusions
Our approach was successful when used as biofeedback instrument, enabling people to relax by making them aware of parameters they are otherwise oblivious to.
Acknowledgments
We would like to thank Elad Yom-Tov for his assistance, guidance and patience and the Ollendorff Research Fund which supported the project.