A software system based on Artificial Neuro-fuzzy hybrid technology will be developed for automatic detection of microsleep events from EEG data. The software system will be designed for used as a model-free and rule-free classification tool that achieves generalization power through learning from examples. The development of the software system will require a Graphical User Interface for data example selection, frequency-analytic preprocessing of EEG raw data, feature extraction for microsleep characterization, design and training of neural networks for single EEG channels, and a fuzzy system for contextual combination of network response for multiple EEG channels to a single system response. The training and testing of the neural networks will be based on a database of visually scored examples of microsleep and non-microsleep events from electrophysiological data, which will be randomly divided into training, validation and test sets. The performnance of the software system will be evaluated based on the false-positive and false-negative rate for the microsleep detection using data examples unknown to the system. The agreement rate between the combined network response and results from visual and conventional automatic scoring will be used as additional evaluation parameter.
The software system will be an attractive tool for researchers, medical and technical personal, industrial engineers. It enables the user to quantify alertness/sleepiness in studies on sleep disorders, shiftwork, drug effects and fatigue countermeasures. It will help reduce time-consuming visual scoring by human experts. In addition, it will widen our knowledge about the rapid transition events (microsleeps) between wake and sleep and can contribute to the development of alertness monitor systems.
