Developing Connectivity Model from Neuroelectrical Brain Activities based on Graph Theory (Çizgi Kuramı Ağlarıyla NöroElektriksel Beyin Aktivitelerinden Bağlantısallık Modeline Geçiş)

Aydin S. (Executive), Dağ O.

Project Supported by Higher Education Institutions, 2020 - 2021

  • Project Type: Project Supported by Higher Education Institutions
  • Begin Date: October 2020
  • End Date: August 2021

Project Abstract

In the present study, contrasting emotional states (happiness vs sadness, amusement vs disgust, excitement vs calmness, calmness vs anger, fear vs anger) are classified with Support Vector Machines with respect to emotional brain network measures in terms of segregation (clustering coefficients (CC), transitivity (T)), integration (global efficiency (GE), local efficiency (LE)), and modularity (Q). Hemispheric dependency levels are estimated by applying Pearson Correlation (PC) and Spearmann Correlation (SC) to both shorter (6 sec) and longer (12 sec) EEG segments across the whole cortex. Binary adjacency matrixes are obtained from dependency estimations with respect to two distinct thresholds as (1st) 60% of maximum value in dependency matrix and (2nd) the mean value of the dependency matrix. In addition to statistical one-way Anova tests, step-wise logistic regression modelling has also been applied to resulting brain network measures (CC, T, GE, LE, Q). Emotional EEG data was downloaded from publicly available database called DREAMER. The higher classification performances were obtained with 2nd threshold for longer segmentation. Five brain network measures all were found to be highly included in logistic regression modelling in comparing not only un-pleasant but distinct emotions as fear (emotional withdrawal) vs anger (emotionally empowering), but also negative and positive emotions as amusement vs disgust. In other contrasting emotional states, the impact of each brain network index varied in accordance with segmentation and method in use (PC vs SC, however the best results are provided by PC