Multi Criteria Evaluation of Sleep and Anesthesia by Neural Networks, Fuzzy Rules, Evolutionary Algorithms and Support Vector Machines

Segments of EEG recorded during sleep and anesthesia are classified by evolutionary optimized populations of neural networks, adapted fuzzy rules and support vector machines. Most of the features extracted from the frontal EEG contribute to the separation of the different stages of both sleep and anesthesia. Neurophysiological research supports the idea to regard sleep and anaesthesia classification from a common point of view. According to the unbalanced number of stage specific EEG segments the concordance between expert scoring and automatically generated classification had to be evaluated by a set of criteria. The training of the networks and the adaptation of the fuzzy rules was controlled by multi criteria optimization. The results presented by confusion matrices let us conclude that the optimized populations of neural networks are more robust to the inter-individual differences than the adapted fuzzy rules. But on the other hand the fuzzy rules can be compared with the rules used by the expert in a more direct way and they can serve the improvement of visual scoring too. Approaches of support vector machines showed the highest efficiencies in the non-linear separation of two classes. The results are used to implement the classifiers in embedded systems which perform an efficient sleep diagnosis and support the anesthesiologist in controlling the drug administration.