ANTsR summary

08 January 2014

R is a powerful statistical tool that helps one systematize reproducible scientific studies. By interfacing ANTs with R, we hope to provide a modern framework for medical analytics, with a focus on imaging-assisted prediction and statistical power. Thus, ANTsR.

I am currently experimenting with distributing ANTsR binaries in order to lower the barrier to entry for new collaborators.

The osx mavericks x86_64 binaries or for yosemite

The linux x86_64 binaries

After you download these, you can install to R from command line with:

$ R CMD INSTALL ANTsR_binaryname.tar.gz

Read ANTsR Quick Start

Complete usage and documentation available at: ANTsR site or through the standard R interface e.g. ?antsRegistration

“By hand” ANTsR installation

If you install ANTsR from source, instead of the binaries, you also get ANTs and ITK installed. This enables one to use command line tools from ANTs for pre-processing and ANTsR for the final statistical study.

This content is contained in more detail in the script here

First, install R, cmake and git. In osx, you need xcode as well.

In R, install dependencies:

install.packages("Rcpp",type="source")
mypkg<-c("signal","timeSeries","mFilter","MASS","robust","magic","knitr","pixmap","rgl","misc3d")
for ( x in mypkg ) install.packages(x)

I like to install Rcpp from source to get the most up-to-date version.

Then, outside of R , do

git clone https://github.com/stnava/ANTsR.git
R CMD INSTALL ANTsR

And you should then have ITK, ANTs and ANTsR.

Sample ANTsR Session

We believe that variable collections, acting synergistically, may better predict outcomes than individual variables. This idea captures the difference between univariate and multivariate analysis. ANTsR gives one a variety of tools for building multivariate models which we will detail over time via this site and academic papers.

library(ANTsR)
# assess whether values from one image relate to those of another
im1<-antsImageRead( getANTsRData('r16') ,2)
im2<-antsImageRead( getANTsRData('r64') ,2)
mat1<-as.matrix( im1 )
mat2<-as.matrix( im2 )
sd1<-apply(mat1,FUN=sd,MARGIN=2)
sd2<-apply(mat2,FUN=sd,MARGIN=2)
mat1<-mat1[, sd1 > 0  ]
mat2<-mat2[, sd2 > 0 ]
ff<-sparseDecom2( inmatrix=list(mat1,mat2), inmask=list(NA,NA),
   sparseness=c(0.02,0.02) ,nvecs=4 , its=21, perms=0,
   smooth=0, cthresh=c(0,0), ell1 = -10 )
print( summary( lm(  ff$projections[,1] ~ as.matrix( ff$projections2 ) ) ) )
print( summary( lm(  ff$projections[,2] ~ as.matrix( ff$projections2 ) ) ) )

This is a toy example that illustrates image I/O, dimensionality reduction via sparse canonical correlation analysis and statistical inference with R’s lm function.

To-Do

ANTsR does not pass R CMD check because of a variety of documentation issues.

software 2