We use a simulated data set provided by the cmR package and add random white noise.
data(cmrsim)
for (i in 1:dim(cmrdata_sim)[1])
for (j in 1:dim(cmrdata_sim)[2])
for (k in 1:3)
if (!is.na(cmrdata_sim[i,j,k,1]))
cmrdata_sim[i,j,k,]=cmrdata_sim[i,j,k,]+rnorm(30,0,sqrt(30))
Let’s have a look on the data:
The function cmr() does the analysis and provides two methods: “local”, which is a voxel wise fitting of the time series and “spatial”, which uses the spatial prior used in Schmid (2011): Voxel-Based Adaptive Spatio-Temporal Modelling of Perfusion Cardiovascular MRI. IEEE TMI 30(7) p. 1305 - 1313.
Here we start with the local analysis. The local analysis can be easily parallelised, each computer core can compute one voxel. Note: parallel computing cannot be used on Windows due to restrictions in the parallel package.
The model uses a Bayesian framework and credible intervals are also produced. We can have a look at the length of the credible intervals voxelwise:
Now let’s have a look at the spatial analysis. Here, parallelisation is not as easy as above, but is also possible to some extent.
Finally, we compare the true maximum response used for the simulation and the results from the local analysis and the spatial analysis. Here, we use the “pseudo bullseye” representation.