Biomedical Engineering
Phd Thesis at the Medical University of Vienna and the Austrian Institute of Technology.
An Electroencephalography-based Model of Daytime Vigilance Trends
The aim of this thesis is to objectively describe long-term daytime vigilance and sleepiness based on Electroencephalogram (EEG)-derived variables.
The homeostatic and circadian influences on daytime vigilance and sleepiness are well known. The thesis provides empirical evidence for these processes based on objective EEG-based variables. Such variables can be used in a model describing and predicting diurnal vigilance and sleepiness and to separate EEG data from subjects after sleep-deprivation from data after normal sleep. It is further shown that the vigilance trends of several variables can be used to classify daytime EEG data under sleep deprivation and after normal sleep.
An exploratory study was conducted under real-world conditions, including 26 healthy female and male subjects. For two 24 hour periods we recorded EEG channels using a mobile system, along with hourly subjective sleepiness ratings and reaction times. The sessions consisted of a night and a day under a sleep-deprived and a normal-sleep condition. The two sessions were embedded in 14 days of recording actigraphy and sleep quality ratings.
52 EEG-based variables were derived from the acquired data, such as frequency band po- wers, band ratios, complexity and entropy measures, and EEG events such as alpha events or diurnal sleep spindles. EEG artifacts are also used as variables.
The actigraphy data was used to estimate the individual circadian rhythm. For the analysis of circadian trends we scaled the data relative to the individual circadian phase, for the homeostatic analysis relative to the time of wake up.
Using statistical means such as correlation coefficients and paired t-tests we analyzed the variables’ homeostatic and circadian trends, compared them to the two-process model, and tested the variables’ ability to distinguish between the data under sleep deprivation and data after normal sleep. Based on the homeostatic trends of a combination of variables a model was build that is able classify EEG data according to the two conditions. The correctness was assessed by a k-fold cross-validation and the significance was tested based on a X2 test.
In the analysis several variables were discovered that show behavior significantly correlated to the two-process model. Especially the daytime-trends of EEG-artifacts provide an interesting insight, for instance the artifacts caused by eye movements correlate negatively with the homeostatic timescale. The standard deviation of the theta band is an examples for a variable with a strong circadian behavior.
Several variables are able to separate sleep-deprived data from data after normal sleep. Goodexamples for separating variables are the relative delta-band power or the (theta+alpha)/beta ratio. These variables can serve as descriptive and predictive biomarkers for diurnal vigilance and sleepiness in the EEG during daytime.
The homeostatic trends of a combination of variables were used as a classification model and showed significantly correct classification rates of 80%. The best performing variables were based on EEG activity in the theta-, very high beta-, and delta-frequency-bands as well as variables derived from eye and muscle artifacts.
We were able to provide significant empirical evidence for the circadian and homeostatic behavior of EEG-based variables. Our exploratory work is an important step towards objectively describing and predicting daytime vigilance and sleepiness. The positive results contribute to a better understanding of the representation of attentive processes in the EEG and encourage us to investigate the topic on a larger scale.
Distributed and Embedded Real-Time Networking
Master Thesis at the University of Salzburg
###