Designing a Fuzzy Rule Based System to Estimate Depth of Anesthesia V. Esmaeili a , A. Assareh b , M. B. Shamsollahi a , M. H. Moradi b and N. M. Arefian c a Biomedical Signal and Image Processing Laboratory (BiSIPL), School of Electrical Engineering, Sharif University of Technology b Faculty of Biomedical Engineering, Amirkabir University of Technology c Department of Anesthesia, Shahid Beheshti University Abstract- Estimating the depth of anesthesia (DOA) is still a challenging area in anesthesia research. The objective of this study was to design a fuzzy rule based system which integrates electroencephalogram (EEG) features to quantitatively estimate the DOA. The proposed method is based on the analysis of single-channel EEG using frequency and time domain features as well as Shannon entropy measure. The fuzzy classifier is trained with features obtained from four subsets of data comprising well- defined anesthesia states: awake, moderate, general anesthesia, and isoelectric. The classifier extracts efficient fuzzy if-then rules and the DOA index is derived between 100 (full awake) to 0 (isoelectric) using fuzzy inference engine. To validate the proposed method, a clinical study has conducted on 22 patients to construct 4 subsets of reference states and also to compare the results with CSM monitor (Danmeter, Denmark), which has revealed satisfactory correlation with clinical assessments. I. Introduction Depth of anesthesia assessment has remained a challenging problem for several decades. It is because none of the parameters used to this aim has satisfactorily described the complexity of the system. Patient hemodynamics like blood pressure, heart rate, tearing and sweating can not avoid awareness and movements during surgery. Neither plasma nor the effect site concentration of the drug can measure clinical effects directly. Solving this problem the Central Nervous System, the main target for anesthetic agents, has received a great deal of attention and EEG-based methods have been widely used for estimating the anesthetic depth. Various types of features have been extracted from the electroencephalogram to predict depth of anesthesia. Early studies have used spectral edge frequency, median frequency and the relative and total power in the classical frequency bands [1]-[3]. Using parameters based on bispectrum made a progress in EEG-based anesthesia monitoring. The bispectrum power is said to indicate the presence of quadratic phase-coupling between different frequencies within the signal. Recently some researchers have used EEG entropy measures as an indicator of depth of anesthesia [4]-[7]. The concept behind this is that EEG becomes more regular as the anesthetic depth increases. Also Lempel-Ziv complexity of EEG has shown good correlation with increasing the anesthetic depth [8]. Although these parameters can distinguish well between awake and anesthetized states, they don’t behave monotonically during transition from wakefulness to deep isoelectric states [2]. So we can’t utilize them individually to continuously monitor anesthetic state changes during different phases of anesthesia. Concerning this, some efforts have been made to combine these features using computational intelligence techniques such as neural networks and neuro-fuzzy inference systems [5], [9-10]. In the present study we have proposed a rule-based fuzzy logic system merging different EEG derived measures to obtain an index for the depth of anesthesia. Several studies have introduced adaptive neuro-fuzzy inference system (ANFIS) as a powerful tool for classifying DOA. Although it led to good compatibility with clinical assessments, but the black box nature of neural learning makes these systems rigid to importing knowledge from human expert that may improve the performance of the system. Moreover, it is difficult to drive knowledge from artificially made rules of these systems. Considering the above mentioned problem we decided to use a fuzzy inference system (FIS) that was initially established by human expert and then optimized by machine learning procedures. Two trends can be observed in development of anesthesia monitors. Some algorithms put more emphasis on some advanced parameters like bispectrum or entropy, while the others (like CSM, Cerebral State Monitor) combine some well-known spectral ratios and time domain characteristic of EEG applying them to a classification algorithm. CSM (Danmeter, Denmark) is a recently developed depth of anesthesia monitor having good correlation with clinical assessments. It uses 3 later defined spectral ratios: alpha- ratio, beta-ratio and difference between them, which is called theta ratio in this paper, accompany with burst- suppression, a time domain feature relating to deep iso- electric states. Each of these components is affecting in a specific range of anesthetic level where they perform best. Adaptive Neural Fuzzy Inference System (ANFIS) is used to calculate the CSI which is a scalar index changing between 0 and 100. In this study we utilized features used in calculating CSI and also SEF and Shannon entropy in 681 Proceedings of the 2007 IEEE Symposium on Computational Intelligence and Data Mining (CIDM 2007) 1-4244-0705-2/07/$20.00 ©2007 IEEE