In livioivil/eegusta: EEG Utilities and STatistical Analysis: eeguana-based tools for statistical analysis of EEG data

knitr::opts_chunk$set(echo = TRUE)

load data

library(eegUtils)
library(tidyverse)
library(data.table)
library(eeguana)

## see also Import_data_GroupLevel_eeguana_format.Rmd
load(file="../data-raw/all_erp_eeguana.Rdata")


## vedi anche https://bnicenboim.github.io/eeguana/
library(ggplot2)
all_erp %>%
  eeg_select(O1, O2, P7, P8) %>%
  ggplot(aes(x = .time, y = .value)) +
  geom_line(alpha = .1, aes(group = .id, color = Condition)) +
  stat_summary(
    fun = "mean", geom = "line", alpha = 1, size = 1.5,
    aes(color = Condition)
  ) +
  facet_wrap(~.key) +
  geom_vline(xintercept = 0, linetype = "dashed") +
  geom_vline(xintercept = .17, linetype = "dotted") +
  theme(legend.position = "bottom")


all_erp%>%
  eeg_filter(between(as_time(.sample, .unit = "milliseconds"), 140, 180)) %>%
  eeg_group_by(Condition) %>%
  eeg_summarize(across_ch(mean, na.rm = TRUE)) %>%
  plot_topo() +
  annotate_head() +
  geom_contour() +
  geom_text(colour = "black") +
  facet_grid(~Condition)

# pari a dx e dispari a sx
# anteriore sx F3, FC3 F7
# anteriore dx F4, FC4 F8
# posteriore sx P7, PO7 O1
# posteriore dx P8, PO8 O2



# https://bnicenboim.github.io/eeguana/articles/intro.html

ERP_faces <- all_erp %>%
  eeg_group_by(.sample, Condition) %>%
  eeg_summarize(across_ch(mean, na.rm = TRUE))

ERP_plot <- ERP_faces %>%
  ggplot(aes(x = .time, y = .value)) +
  geom_line(aes(color = Condition)) +
  facet_wrap(~.key) +
  theme(legend.position = "bottom") +
  ggtitle("Comparing Conditions") +
  theme_eeguana()

ERP_plot %>% plot_in_layout()

Crea Dataset Analysis

Crea Meta-canali (aree)

all_erp4 <-
all_erp %>%
  eeg_transmute(
    Ant_Left = chs_mean(across(c("F3","FC3","F7")), na.rm = TRUE), 
    Ant_Right = chs_mean(across(c("F4","FC4","F8")), na.rm = TRUE), 
    Post_Left = chs_mean(across(c("P7","PO7","O1")), na.rm = TRUE), 
    Post_Right = chs_mean(across(c("P8","PO8","O2")), na.rm = TRUE), 
  )

Valori medi per soggetto/condizione

D=all_erp4%>%
  eeg_filter(between(as_time(.sample, .unit = "milliseconds"), 140, 180)) %>%
  eeg_group_by(.id,Condition,Subj) %>%
  eeg_summarize(across_ch(mean, na.rm = TRUE))

head(D$.segments)
head(D$.signal)

Crea dataset

head(D$.segments)
head(D$.signal)
dati=tibble(merge(D$.signal,D$.segments,by=".id"))
dati$.recording <- dati$.sample <- dati$.id <- NULL

Wide to Long format

dati_long=
  dati %>%
   pivot_longer(1:4, names_to = 'Chan')

Anova per Misure ripetute

A VOI...

Linear Mixed Model

A VOI...

Test di permutazione per Misure Ripetute

Random coefficient analsys: Un modello per soggetto + Symmetry testing (i.e. 1 sample)

Potremmo usare i dati in formato Long con la stessa logica usata per i mixed models. Oppure possiamo analizzare i 4 (meta-)canali separatamente. In questa esercitazione scegliamo questa seconda possibilità

library(flip)


Y=as.matrix(dati[,-(5:6)])


dati$Condition=factor(dati$Condition)

dati$Scrambled=dati$Condition
levels(dati$Scrambled)[1:2]="NotScrambled"
levels(dati$Scrambled)[2:3]="Scrambled"
contrasts(dati$Scrambled)=contr.sum

dati$Cars_Faces=dati$Condition
levels(dati$Cars_Faces)=gsub("Scrambled","",levels(dati$Cars_Faces))
levels(dati$Cars_Faces)=gsub("ERP","",levels(dati$Cars_Faces))
contrasts(dati$Cars_Faces)=contr.sum

mod=obs2coeffWithin(Y~Cars_Faces*Scrambled, units=~Subj, data=dati)

Y=mod$coeffWithin
Intercette=Y[,grep("Intercept",colnames(Y))]
Y=Y[,-grep("Intercept",colnames(Y))]

Y=Y[,7:12]
main=(Y[,1:3]+Y[,4:6])
colnames(main)=gsub("^Post_Left:","",colnames(main))
interact=(Y[,1:3]-Y[,4:6])
colnames(interact)=gsub("^Post_Left:","Channel:",colnames(interact))


Y=cbind(main,Channel=Intercette[,3]-Intercette[,4],interact)
res=flip::flip(Y,perms = 10000)
summary(res)
#pvals=flip::t2p(res,obs.only = TRUE)[,]

#summary(pvals)

Permuco

library(permuco)

DD=dati[,-c(1:2,5)] %>%
  pivot_longer(c("Post_Left","Post_Right"), names_to = "Channel", values_to = "Y")

str(DD)
DD$Y=as.vector(DD$Y)

mod_permuco <- aovperm(Y~Scrambled*Cars_Faces*Channel + Error(Subj/(Scrambled*Cars_Faces*Channel)), data = DD, method = "Rd_kheradPajouh_renaud")

mod_permuco

livioivil/eegusta documentation built on May 6, 2024, 9:20 p.m.