| Summary: | 博士 === 元智大學 === 機械工程學系 === 95 === Despite Cisatracurium’s long-onset time, it is still widely used in regular surgical operations. A complete solution in terms of the hardware, software, and control methodology for controlling the degree of muscle relaxation is presented in this thesis. Our development procedure includes three tasks; the first task was to collect the clinical data using three different control methods – intermittent bolus control, intensive manual control, and automatic control. Intermittent bolus control was used on 13 patients. Intensive manual control was used on 15 patients. Automatic control was used on 15 patients. These control results showed that the mean (SD) of the mean error for each method was 8.76 (1.46), 1.65 (1.67), and 0.48 (1.43), respectively. Although these statistics showed that the automatic control method was not significantly different to the intensive manual control method, it was more accurate.
The second task was to build a patient model for evaluating the performance of the controllers. The method for building a compartmental numerical model, with two compartments based on pharmacokinetics and one compartment based on pharmacodynamics, is presented. The coefficients for the pharmacokinetic model were identified and based on Kisor’s research results. The pharmacokinetic model result showed that our model was 4.6% different to Kisor’s. The pharmacodynamic model result also showed that the two parameters of Hill Equation (50% of the maximum effect and an eliminating constant) were 70.93 ± 36.84 and 1.24 ± 0.29, respectively.
The final task was to extract the control rules from the clinical manual control data and to mimic the anesthesiologist’s behavior via the automatic control method. Completion of the patient model helped us with the final design of the controller. We have described how to extract the control rules using fuzzy modeling method. This thesis also presents two rule-bases: one from the fuzzy modeling method and the other from anesthesiologists’ clinical experience. They were compared with four tests: the different set points, the control interval strategy, the tolerance of noise effect, and the delay time effect.
The simulation showed that the fuzzy modeling algorithm could successfully extract the fuzzy rules from the clinical data, and its control error was smaller than the anesthesiologist’s rules for different set point tests. However, the control error increased and became worse when the set points were raised. It meant that these two rule-bases were not apt to control the higher set points (i.e. T1% of 40 or higher). The t-test also showed that these two rule-bases performance at different set points had significant differences (p < 0.05). Moreover, the results for the control interval tests showed that strategy had significant influence, especially in reducing standard deviation of control errors. However, these two rule-bases were not affected by noise disturbance, and the delay time only affected the overshoot for these two rule-bases in simulations.
Furthermore, for improving the performance of the extracted rule-base controller, we used the probability-type fuzzy rules extraction method, which each decision point having different weighting factor based on its control performance, and obtained a new rule-base. The simulation results show that it was not better than the previous controller. The reason is the mean of the control error for each anesthesiologist can not represent their overall performance. If the weighting factor could decided by the control result of each decision point, the performance may be better than it decided by the mean of the control error of each patient.
In this study, we not only established a muscle relaxation control system but also provided the control strategies for the drug Cisatracurium. These control strategies were obtained by the anesthesiologists, fuzzy modeling method, and probability-type fuzzy modeling method. Although the performance of the probability-type fuzzy modeling method is not significantly different with the self-organizing fuzzy modeling method, both these two fuzzy modeling methods can obtain the linguistic rules which is directly able to assist the user in realizing the features or related information of the analyzed data.
|
|---|
