Search

In this research article a dynamic estimation of neuronal activity and brain dynamics from electroencephalographic (EEG) signals is presented using a dual Kalman filter. The dynamic model for brain behavior is evaluated using physiological-based linear models. Filter performance is analyzed for simulated and clinical EEG data, over several noise conditions. As a result a better performance on the solution of the dynamic inverse problem is achieved, in case of time varying parameters compared with the system with fixed parameters and the static case. An evaluation of computational load is performed when predicted dynamic cases, estimated using the Kalman filter, are up to ten times faster than the static case.

In this research article a dynamic estimation of neuronal activity and brain dynamics from electroencephalographic (EEG) signals is presented using a dual Kalman filter. The dynamic model for brain behavior is evaluated using physiological-based linear models. Filter performance is analyzed for simulated and clinical EEG data, over several noise conditions. As a result a better performance on the solution of the dynamic inverse problem is achieved, in case of time varying parameters compared with the system with fixed parameters and the static case. An evaluation of computational load is performed when predicted dynamic cases, estimated using the Kalman filter, are up to ten times faster than the static case. En este artículo de investigación se presenta la estimación de la actividad neuronal y la dinámica del cerebro a partir de señales electroencefalográficas (EEG) usando un filtro de Kalman dual. El comportamiento dinámico del cerebro se representa a partir de modelos lineales con base en restricciones fisiológicas. Como resultado, se obtiene un mejor desempeño en la solución en el caso de modelos variantes con el tiempo, al compararla con modelos invariantes en el tiempo, y con la solución estática. Se presenta una evaluación de la carga computacional donde es claro que la estimación de la actividad neuronal con el filtro de Kalman basada en modelos dinámicos lineales, es hasta 10 veces más rápida, que la solución para el caso estático.