EEG spectral features provide basis for Artificial Neural Network comparison of anesthetics

Artificial Neural Networks (ANN) have proven useful in a wide variety of pattern recognition tasks in anesthesia monitoring research. In this study, ANN were used to analyze and compare dose-dependent EEG changes àuringsevoflurane and isoflumne anesthesia. Two categorization tasks were attempted: to differentiate between isoflurane and sevoflurane EEG; and to differentiate EEG at three anesthetic levels. The trained ANNs were unable to differentiate between sevoflurane and isoflurane at arty MAC levels. However, the ANNs trained with isoflurane data were able to correctly identify the anesthetic level of sevoflurane EEG with an accuracy of 75% comparable to the 77% accuracy achieved in categorizing isoflurane EEG. ANN may offer superior performance in categorization tasks when compared to statistical methods due to greater suitability for classifying nonlinear processes. In this study, ANN classification results offer evidence that sevoflurane and isoflurane have indistinguishable EEG spectral signatures.