INVESTIGADOR(ES) PRINCIPAL(ES):

Cristian Alejandro Torres Valencia

0 horas

CODIGO CIE

E6-18-1

Alvaro Angel Orozco Gutiérrez

Tutor

0 Horas

TIPO DE CONVOCATORIA

2017. Estudiantes De Posgrado

TIPO DE PROYECTO

Investigación Aplicada

OBJETIVO(S)

Objetivo General: Desarrollar un esquema de representación de señales EEG utilizando EHKR para el reconocimiento de patrones discriminantes de actividad eléctrica cerebral, con el fin de codificar información relevante en términos de criterios de clasificación e interpretabilidad de los datos con respecto a las dependencias espacio-temporales de los mismos. Objetivos específicos: 1) Desarrollar una metodología de representación en tiempo corto que permita codificar la no estacionariedad de señales EEG en tareas de clasificación de actividad eléctrica cerebral utilizando modelos discriminantes en EHKR. 2) Desarrollar una metodología de representación de señales EEG basada en EHKR que permita identificar dependencias espaciales relevantes y que considere la variabilidad entre sujetos en tareas de clasificación de actividad eléctrica cerebral. 3) Proponer una metodología de representación de señales EEG basada en EHKR que permita identificar dependencias espacio-temporales relevantes con el fin de codificar la no-estacionariedad de las series de tiempo y relaciones entre canales en tareas de clasificación de actividad eléctrica cerebral.

RESUMEN

Human brain controls, reflects, and regulates all the processes that occur in the bio-logical systems that compose the human body. So, the analysis of the brain electrical activity (BEA) can improve our understanding of several physiological processes as reflects of our body functioning [16]. The BEA is analyzed following different neuroimaging techniques, such as: Electroencephalography (EEG), Magnetoencephalography, Magnetic Resonance Imaging, functional MRI, among others. The EEG presents some advantages in comparison with other technologies; such as: low cost, less invasive scheme of acquisition, and higher temporal resolution. Following this advantages, EEG is preferred for BEA; nevertheless, there are some challenges associated with the processing of EEG signals, indeed, the non-stationarity nature of the recordings, the low spatial resolution, and the variability that exists between studies of different subjects with similar conditions, make the EEG processing task a very challenging one [26]. Some strategies for BEA processing has been proposed to deal with the aforementioned challenges. Most of them are based on data transformations, namely, temporal, spectral, and, time-frequency are employed to deal with the EEG non-stationarity behaviors [16]. However, classical spectral and time-frequency approaches provide redundant information that derives in suboptimal classification results [12]. On the other hand, the analysis of brain connectivity has emerged as a way to characterize and understand brain functions [6]. In this case, segregated regions of the brain are integrated to perform a specialized information processing and the resulting interactions between different areas are quantized. These measures of interdependences among channels can be understood as a spatio-temporal representation, which is used in further stages of BEA discrimination, i.e., emotion classification or motor tasks recognition [23]. In this work, we propose the development of an adaptive learning-based relevance analysis (ALRA) for BEA data in an emotion recognition context. We use brain connectivity measures to characterize the spatio-temporal dependencies of EEG channels. In particular, two functions of connectivity quantification are used: the correlation and the time-series generalized measure of association (TGMA) [7]. Besides, we carried out a segmentation stage of the BEA towards a sliding window. In addition, we use a physiological response (galvanic skin response (GSR) and the blood volume pressure (BVP)) to select relevant EEG time windows holding relevant connectivity patterns. So, an adaptive filter algorithm, termed the quantized Kernel mean least squares (QKLMS), is used for the physiological response estimation. The data of each window is used for the adjustment of the filter parameters using an online learning strategy, and a classifier is trained over the filter codebooks. Achieved results show an improvement using relevant spatio-temporal EEG dependencies for emotion classification in comparison to state-of-the-art approaches. The remainder of this paper includes a theoretical background in section 2.1, followed by a description of the experimental setup in section 4.1. Finally, a description of the obtained results is presented in section 4.2 and some remarks are included also in section 4.3.

ESTADO

Concluye Satisfactoriamente

FECHA DE INICIO

22/01/2018

FECHA DE FINALIZACION

22/01/2019

PRODUCTOS

A discriminative multi-output Gaussian processes scheme for brain electrical activity analysis

Artículos publicados en Revistas B, C ó D

Discriminant brain electrical activity analysis using Reproducing Hilbert Space representations

Doctorado

Emotion assesment using adaptative learning-based relevance analysis

Ponencia en evento especializado

Emotion assessment using adaptive learning-based relevance analysis

Artículos en revista A1 ó A2

URL

SOFTWARE: Python module for EEG data processing within RKHS scheme

Software