Thesis Estudio comparativo sobre simulaciones de potenciales relacionados con eventos para P300
Loading...
Date
2026-07
Authors
Journal Title
Journal ISSN
Volume Title
Program
Ingeniería Civil Informática
Departament
Campus
Campus Casa Central Valparaíso
Abstract
El desarrollo de interfaces cerebro-computadora basadas en el potencial evocado P300 se ve limitado por la escasez de registros electroencefalográficos empíricos, cuya adquisición es costosa y lenta, lo que resulta insuficiente para entrenar modelos de aprendizaje profundo sin incurrir en sobreajuste. La generación de señales sintéticas surge como una alternativa para mitigar esta limitación, siempre que reproduzca con fidelidad las propiedades de los registros reales. Este trabajo presenta un estudio comparativo de tres modelos de propagación volumétrica de complejidad creciente para la simulación del componente P300: un dipolo en esfera conductora homogénea, el modelo de elementos de frontera (BEM) del NY Head implementado en SEREEGA sobre MATLAB, y el BEM del atlas poblacional fsaverage implementado en MNE-Python. Los tres enfoques se alimentaron con un conjunto unificado de parámetros de entrada, de modo que las diferencias observadas son atribuibles exclusivamente al mecanismo de propagación. Los datos sintéticos se validaron cuantitativamente contra registros empíricos del Dataset II del BCI Competition III, mediante métricas de morfología temporal, distribución topográfica y correlación de forma por canal. Los resultados muestran que el modelo de propagación condiciona de forma determinante la distribución espacial del P300 simulado: los modelos BEM reproducen correctamente la localización del máximo en el canal Cz observada en los datos reales, mientras que el modelo de esfera la sitúa erróneamente en Pz. El análisis identifica cinco brechas de realismo entre los datos sintéticos y los empíricos, que orientan las líneas de mejora del paradigma de simulación.
The development of brain-computer interfaces based on the P300 evoked potential is limited by the scarcity of empirical electroencephalographic recordings, whose acquisition is costly and time-consuming, resulting in insufficient data to train deep learning models without incurring overfitting. The generation of synthetic signals emerges as an alternative to mitigate this limitation, provided it faithfully reproduces the properties of real recordings. This work presents a comparative study of three volume conduction models of increasing complexity for simulating the P300 component: a dipole in a homogeneous conductive sphere, the Boundary Element Method (BEM) of the NY Head implemented in SEREEGA on MATLAB, and the BEM of the fsaverage population atlas implemented in MNE-Python. The three approaches were fed with a unified set of input parameters, so that the observed differences are exclusively attributable to the propagation mechanism. The synthetic data were quantitatively validated against empirical recordings from Dataset II of the BCI Competition III, using metrics of temporal morphology, topographic distribution, and shape correlation per channel. The results show that the propagation model deterministically conditions the spatial distribution of the simulated P300: the BEM models correctly reproduce the localization of the maximum in the Cz channel observed in the real data, while the sphere model erroneously places it in Pz. The analysis identifies five realism gaps between the synthetic an
The development of brain-computer interfaces based on the P300 evoked potential is limited by the scarcity of empirical electroencephalographic recordings, whose acquisition is costly and time-consuming, resulting in insufficient data to train deep learning models without incurring overfitting. The generation of synthetic signals emerges as an alternative to mitigate this limitation, provided it faithfully reproduces the properties of real recordings. This work presents a comparative study of three volume conduction models of increasing complexity for simulating the P300 component: a dipole in a homogeneous conductive sphere, the Boundary Element Method (BEM) of the NY Head implemented in SEREEGA on MATLAB, and the BEM of the fsaverage population atlas implemented in MNE-Python. The three approaches were fed with a unified set of input parameters, so that the observed differences are exclusively attributable to the propagation mechanism. The synthetic data were quantitatively validated against empirical recordings from Dataset II of the BCI Competition III, using metrics of temporal morphology, topographic distribution, and shape correlation per channel. The results show that the propagation model deterministically conditions the spatial distribution of the simulated P300: the BEM models correctly reproduce the localization of the maximum in the Cz channel observed in the real data, while the sphere model erroneously places it in Pz. The analysis identifies five realism gaps between the synthetic an
Description
Keywords
Interfaces cerebro-computadora, Potencial evocado P300, Electroencefalografía, Simulación de señales EEG, Modelos de propagación volumétrica
