162 lines
4.5 KiB
R
162 lines
4.5 KiB
R
# Lab 11 for the University of Tulsa's CS-6643 Bioinformatics Course
|
|
# Introduction to fMRI Analysis and ICA
|
|
# Professor: Dr. McKinney, Fall 2022
|
|
# Noah L. Schrick - 1492657
|
|
|
|
## Set Working Directory to file directory - RStudio approach
|
|
setwd(dirname(rstudioapi::getActiveDocumentContext()$path))
|
|
|
|
#### Part A: Haemodynamic response functions (HRF) and block design
|
|
## Plot Basic HRF from 0-20s
|
|
hq <- function(t,q=4){
|
|
# q=4 or 5, where 5 has more of a delay
|
|
return (t^q * exp(-t)/(q^q * exp(-q)))
|
|
}
|
|
|
|
# use seq to create vector time and use hq to create hrf vectors
|
|
time <- seq(0,20)
|
|
hrf1 <- hq(time)
|
|
hrf2 <- hq(time, 5)
|
|
|
|
# plot
|
|
plot(time,hrf1,type="l")
|
|
lines(time,hrf2,col="red")
|
|
|
|
## Deconvolve with task onset times
|
|
# grabbed from afni c code
|
|
# basis_block_hrf4 from 3dDeconvolve.c
|
|
HRF <- function(t, d){
|
|
if (t<0){
|
|
y=0.0
|
|
}else{
|
|
y = 1/256*exp(4-t)*(-24-24*t-12*t^2-4*t^3-t^4 + exp(min(d,t))*(24+24*(t-min(d,t)) + 12*(t-min(d,t))^2+4*(t-min(d,t))^3+(t-min(d,t))^4))
|
|
}
|
|
return(y)
|
|
}
|
|
|
|
t=seq(0,360,len=360)
|
|
onsets=c(14,174,254)
|
|
blocks.model = double()
|
|
for (curr_t in t){
|
|
summed_hrf=0.0
|
|
for (start in onsets){
|
|
summed_hrf=summed_hrf+HRF(curr_t-start,20)
|
|
}
|
|
blocks.model = c(blocks.model,summed_hrf)
|
|
}
|
|
|
|
plot(blocks.model,type="l")
|
|
|
|
|
|
## Plot voxel time series data and the block design curve
|
|
voxel.data <- read.delim("059_069_025.1D")
|
|
plot(seq(1,2*dim(voxel.data)[1],by=2),t(voxel.data),
|
|
type="l", xlab="time",ylab="intensity")
|
|
# normalize the height of the blocks model
|
|
blocks.normal <- max(voxel.data)*blocks.model/max(blocks.model)
|
|
lines(blocks.normal,type="l",col="red")
|
|
|
|
# regression
|
|
length(blocks.normal) # too long
|
|
dim(voxel.data)[1]
|
|
# grab elements from blocks.normal to make a vector same as data
|
|
blocks.norm.subset <- blocks.normal[seq(1,length(blocks.normal),len=dim(voxel.data)[1])]
|
|
length(blocks.norm.subset)
|
|
|
|
voxel.data.vec <- matrix(unlist(voxel.data),ncol=1)
|
|
|
|
# use lm to create voxel.fit <- lm...
|
|
voxel.fit <- lm(voxel.data.vec ~ blocks.norm.subset)
|
|
|
|
|
|
plot(blocks.norm.subset,voxel.data.vec,
|
|
xlab="block model",ylab="voxel data",main="regression fit")
|
|
|
|
# use abline(voxel.fit) to overlay a line with fit coefficients
|
|
abline(voxel.fit)
|
|
|
|
|
|
#### Part B: Resting state fMRI visualization, multidimensional arrays and independent component analysis (ICA).
|
|
## example: multi-dimensional array
|
|
# 2 3x4 matrices
|
|
multArray <- array(1:24,dim=c(3,4,2))
|
|
dim(multArray)
|
|
multArray[,,1] # matrix 1
|
|
multArray[,,2] # matrix 2
|
|
image(multArray[,,1]) # plot slice 1
|
|
image(multArray[,,2]) # plot slice 2
|
|
|
|
## Neuroimaging Informatics Technology Initiative
|
|
if (!require("fastICA")) install.packages("fastICA")
|
|
if (!require("fmri")) install.packages("fmri")
|
|
library(fastICA)
|
|
library(fmri)
|
|
|
|
# read in 4d nifti
|
|
# uses fmri library, takes about 3min to load
|
|
img <- read.NIFTI("rest_res2standard.nii")
|
|
|
|
mask <- img$mask # Boolean mask for brain voxels
|
|
dim(mask)
|
|
ttt <- extractData(img) # extract 4d data cube
|
|
numScans <- dim(ttt)[4]
|
|
|
|
# plot a voxel's time series
|
|
plot(ttt[30,30,30,],type="l",xlab="time",ylab="activity")
|
|
|
|
## Plot a 2D Slice
|
|
yslice <- 35
|
|
scan2dslice <- ttt[,yslice,,50] # grab 2d slice at t=50
|
|
image(scan2dslice,main="no masking") # no mask
|
|
# mask it off
|
|
slice.mask <- mask[,yslice,]
|
|
scan2dslice[slice.mask] <- NA # NA's become white
|
|
image(scan2dslice,main="masked")
|
|
|
|
## ICA
|
|
t1 <- Sys.time() # for timing purposes
|
|
dataMat <- NULL
|
|
for(t in seq(1,numScans)){
|
|
scan <- ttt[,,,t]
|
|
# stretched out the 61x73x61 3d matrix into one row
|
|
# apply mask and stack
|
|
dataMat <- rbind(dataMat,scan[mask])
|
|
}
|
|
t2 <- Sys.time()
|
|
difftime(t2,t1) # 3 minutes
|
|
dim(dataMat)
|
|
|
|
## ICA analysis
|
|
# input X: rows observations (voxels) and cols variables (time)
|
|
X <- t(dataMat)
|
|
m <- 20 # specify number of ICA components
|
|
t1<-Sys.time()
|
|
f<-fastICA(X,n.comp=m,method="C")
|
|
t2<-Sys.time()
|
|
difftime(t2,t1) # 1.23min
|
|
# S=XKW,
|
|
# K is a pre-whitening PCA matrix (components by time)
|
|
# S is the matrix of m ICAs (columns of S are spatial signals)
|
|
# S has dimensions voxel x components
|
|
S<-f$S # you can find K and W with $
|
|
|
|
ica.comp <- 5 # look at 5th component
|
|
# plot the 5th time ICA
|
|
plot(f$K[,ica.comp],type="l",xlab="time",ylab="signal",main="ICA component")
|
|
|
|
# threshold S matrix
|
|
theta <- 2
|
|
S[S<=theta] <- NA
|
|
|
|
# turn S back into 4d multidimensional array
|
|
xdim<-dim(ttt)[1]
|
|
ydim<-dim(ttt)[2]
|
|
zdim<-dim(ttt)[3]
|
|
ica.4dArray <- array(matrix(S,ncol=1),dim=c(xdim,ydim,zdim,m))
|
|
dim(ica.4dArray) # 61x73x61x10
|
|
yslice <- 35
|
|
# grab 2d slice at y=yslice and ica 5
|
|
ica2dslice <- ica.4dArray[,yslice,,ica.comp]
|
|
image(ica2dslice,main="ica component (spatial locations)")
|
|
|