This repository contains the code supplement for the paper
The Impact of Model Assumptions in Scalar-on-Image Regression
Clara Happ, Sonja Greven, Volker Schmid (Department of Statistics, LMU Munich, Munich, Germany)
for the Alzheimer's Disease Neuroimaging Initiative. Statistics in Medicine, 37(28): 4298-4317. The full article is available here.
- Usage examples for all models used in the paper
Rimplementations of methods, if not already available
Rfunctions for all measures developed in the paper
- ADNI roster IDs (RID) of the subjects used in the simulation settings (sample size 250 and 500) and in the application (sample size 754). We use slice
z = 75of each three-dimensional brain scan and select the coordinates
x = 30:93, y = 30:93to obtain the quadratic sub-images.
- Code for generating the beta-images for the simulation, together with csv-files containing the final images (bumpy, pca, smooth, sparse)
R functions are directly applicable. The
C implementation of the Bayesian GMRF models requires compilation. Change to the
C subdirectory and run the following code in the command line (tested under Linux only)
R CMD SHLIB utilities/*.c(compiles all utility functions)
R CMD SHLIB mainGibbs_GMRF.c utilities/*.o(compiles main for GMRF)
R CMD SHLIB mainGibbs_HyperparamsFixed.c utilities/*.o(compiles main for SparseGMRF)
Make sure that the
Makevars file is in the same directory as the main files.
Please use GitHub issues for reporting bugs or issues.