Se presenta un estudio de Audiogramas para el diagnostico automatico, realizando una extracción de características y discriminando entre tres patologías, y el caso normal. Las mediciones de perdida de audición de un órgano auditivo se ordenan en un vector, las cuales se ingresan aI sistema de reco nocimiento que se disenó mediante técnicas de Analisls Multiva riado de Datos. Para disminuír la gran cantidad de datos obtenI dos conservando la información mas relevante, se extrajeron las características esenciales mediante un Analisis en Componentes Principales. Para implementar el sistema de reconocimiento se aplicaron dos tecnicas alternativas de clasificación: Metodo de Bayes para Patrones Normales y Vecino Mas Cercano. El sistema desarrollado fue probado con algunos audiogramas suplementarios que tenían diagnóstico previo. Los resultados obtenidos reflejan que el metodo de Bayes presenta una mejor clasificación y ademas entrega información adicional que permite una evaluación mas precisa.

A system for automatic diagnostic in Otoneurology is presented. For this purpose, audiometries of a set of patients with their appropriate diagnostics were obtained. In this set three pathological groups and one normal group can be identified. The measurements of audition loss in each car of a patient are represented as a m-dimensional vector. By means of these vectors the essential features are extracted, using the statistical technique of Principal Components. In this way, the patients are represented in a space with low dimensionality which conserves the most important characteristics of the original data. To discriminate between the different groups, two alternative techniques of clasification have been developed: the Nearest Neighbor and Bayes Classification for Normal Patterns. With this, in the feature space, the regions where the groups lie are identified in such a way that if a newpatient is mapped, one can easily determine the possible pathology that the patient suffers. in the present work, the techniques were performed to be used in a micro-computer. The output shows the feuture space and indicates the pathological groups. The results suggest that Bayes Classification is better than the Nearest Neighbor because it also provides the probabilities for each group.