Forecasting Expert System Based On Artificial Neural Network Technology Intubation, Cormack Grade

Objective: An airway physical examination should be conducted, prior to the initiation of anesthetic care and airway management in all patients. However, most of the specialists agree that prediction of a difficult airway may be improved by the assessment of multiple features versus single features of the airway physical examination. Cormack class is one of the most important and the ultimate indicator for difficult airway prediction. An improved detection of the Cormack class degree and preparatory efforts enhance success and minimize risk when a difficult airway is encountered.In this paper, the author presents a Cormack class predication system based on artificial neural network (ANN) using a augmented error back propagation (BP) algorithm for intelligent real-time decision support of the tracheal evaluation prior to anesthesia.All of the medical records have been analyzed by statistical methods to acquire the Chinese difficult airway predication parameter and value.Method: A total of 824 medical records of patients have been collected to train and test the system, including 461 Cormack classⅠ, 158 classⅡ, 189 classⅢ, 16 classⅣ.The author established a predication system based on the multilayer perceptron neural network architecture. In particular. 13 input variables related to the predication of the Cormack class of interest are categorized into three groups and then encoded using the proposed coding schemes, the output variables are 4(Cormack classⅠ～Ⅳ) and the hidden laver node are 9. The system is trained using a BP algorithm augmented with the momentum term, the adaptive learning rate, the forgetting mechanics, and an optimized algorithm based on conjugate gradients method. All of the records were divided into four subsets to train and test the system using an assessment cross validation method.Two new groups were built on original Cormack classⅠaddⅡand classⅢaddⅣapparently to perform statistical analysis include descriptive statistics for quantitative and qualitative variable, one-way analysis of variance and multivariate discriminant analysis.Result: The system constructed can achieve high predication accuracy: 91.9% according to average accuracy, 93.5% of classⅠaddⅡgroup. 86.3% of classⅢ, 85.7% of classⅣaccording to procedure accuracy, and 96.9% of classⅠaddⅡgroup, 76.7% of classⅢ. 75% of classⅣ, according to user accuracy. Another new group of 12 records of patients were collected to test the system extensive performance.The significative and unsignificative parameters for predicating Chinese difficult airway were detected. The clinical reference value were defined according to the 95% confidence limit of arithmetic rnean or quartile 25 of the group of classⅢaddⅣ. The discriminant function constructed by multivariate discriminant analysis, were assessed by cross validation and the result show that the predication accuracy is lower than ANN system.Conclusion: Artificial neural network technology has been applied in clinical decision support system tbr tracheal intubation Cormack class predication, showing great application prospect of tracheal evaluation prior to anesthesia.