Rule Extraction By Fuzzy Modeling Algorithm For Fuzzy Logic Control Of Cisatracurium As A Neuromuscular Block

摘要

This paper provides two rule bases to control administration of cisatracurium, a non-depolarizing neuromuscular blocking agent. One rule base is extracted from the objective approach of fuzzy modeling algorithm (FMA), and the other is from the subjective approach of experts' clinical experience. First, we established the data-acquisition system to record the manual neuromuscular block control during surgery. After collecting 15 patients data control by cisatracurium, we extracted six rules from these data via FMA. Another rule base also had six rules from experts with clinical anesthesia experience. Each rule-base was combined with three rules regarding the safety of the fuzzy controller. To compare their performance through simulations, we used the patient model established by our previous study which is a combination model consisting of a three-compartment mathematical model based on pharmacokinetics, and the Hill equation based on pharmacodynamics. In order to test the differences between these two rule-bases, the simulation used four disturbances: the different set points, the control interval strategy, the tolerance of noise effect, and the tolerance of delay time effect. The simulation shows that the FMA could successfully extract the fuzzy rules from the clinical data, and its control error is smaller than expert rules for different set point tests. However, the control error is increased and becomes worse when the set points are raised, which means that these two rule-bases are not appropriate to control the higher set points (i.e. T1% of 40 or higher). The t-test also shows that these two rule-bases performance of different set points have significant differences (p<0.05). Moreover, the results for control interval tests show that strategy has a significant influence, especially in reducing the standard deviation of control error. However, in simulations, these two rule-bases are not affected by noise disturbance, and the delay time affects only the overshoot for these two rule-bases.