Radial basis function neural networks versus fuzzy models to predict return of consciousness after general anesthesia

Abstract

Abstract-This paper presents three modeling techniques to predict return of consciousness (ROC) after general anesthesia, considering the effect concentration of the anesthetic drug at awakening. First, several clinical variables were statistically analysed to determine their correlation with the awakening concentration. The anesthetic and the analgesic mean dose during surgery, and the age of the patient, proved to have significantly high correlation coefficients. Variables like the mean bispectral index value during surgery, duration of surgery did not present a statistical relation with ROC. Stochastic regression models were built using the variables with higher correlation. Secondly, fuzzy models were built using an Adaptive Network-Based Fuzzy Inference System (ANFIS) also relating different sets of variables. Thirdly, radial basis function (RBF) neural networks were trained relating different sets of clinical values with the anesthetic drug effect concentration at awakening. Clinical data was used to train and test the models. The stochastic models and the fuzzy models proved to have good prediction properties. The RBF network models were more biased towards the training set. The best balanced performance was achieved with the fuzzy models.

How and to what end may consciousness contribute to action? Attributing properties of consciousness to an embodied, minimally cognitive artificial neural network

“…Dehaene and Changeux (2011) review the relevant models of networks that are supposed to simulate consciousness, including their own approach “global neural workspace” (GNW) (see also Seth, 2007 for a systematic summary). Of all models discussed by Dehaene and Changeux, GNW shows the largest overlap with reaCog.…”

Identificación de etapas de anestesia en registros electroencefalográficos mediante máquinas de soporte vectorial

“…Cuando se aplica anestésico-propofol , relajante muscular y analgésicoremifentanil, se pierde la claridad del grado de dolor y la certeza de un adecuado nivel de profundidad anestésica. Un experimento reportado con 20 registros de pacientes, con diferentes edades, el índice BIS, la dosis media del propofol y del remifentanil, y la duración de la cirugía; luego de entrenar la red, se tuvo error de 31 %, que fue una predicción de recuperación no confiable [32). Otro reporte de identificación de las etapas de anestesia con el uso de las RBF-NN y los anestésicos desfluorano y propofol; utilizando índices de entropía espectral SE y de los valores singulares al fijar el eigen-espectro espacial; ajustando y optimizando la NN con el método de validación cruzada, la prueba con 1 O pacientes registrados con desfluorano y 5 con propofol, reportó una tasa de clasificación de 98 ± 0.2% de las clases (baja, media y profunda anestesia), sin diferencias entre las NN en respuesta a los agentes anestésicos [39].…”

“…Los sistemas de Redes Adaptivas basadas en Inferencia Difusa, ANFIS, son otro recurso de clasificación que relaciona diferentes conjuntos de variables con el objetivo de identificar ya sea las etapas de anestesia o bien, la predicción de la concentración del anestésico intravenoso que dé lugar a la recuperación de conciencia al finalizar la cirugía [32].…”