MRIPredict is aimed to be easy to use and user friendly. It has a Matlab version, with a graphical user interface to be used as an SPM toolbox.
The R version may be also called from bash to yield a flavor similar to that of FSL and software alike..
Written in Matlab, it has a GUI easy to use and understand for every user. To be used as an SPM toolbox.
Written in R, can also be called from bash to yield a flavor similar to that of FSL and software alike.
mripredict model -mri mri_paths_file -data data_table_file -resp response_var -cov covariatesto set a new model
mripredict_cv modelto cross-validate the model
mripredict_fit modelto fit the model
mripredict_predict model -mri mri_paths_file -data data_table_file -pred prediction_fileto predict from new data
mripredict(mri_paths_file, data_table_file, response_var, covariates)to set a new model
mripredict_cv(model)to cross-validate the model
mripredict_fit(model)to fit the model
mripredict_load(xml_file)to load a model
mripredict_predict(model, mri_paths_file, data_table_file, prediction_file)to predict from new data
mripredict_save(model, xml_file)to save the model
Please note that the software downloadable from this preliminary website is a beta version created for the referees of the manuscript "Evaluation of machine learning algorithms and structural features for optimal MRI-based diagnostic prediction in psychosis".
The software is fully functional and may be already used within SPM / Matlab, FSL / bash or R, but the interface is being further developed to improve the user experience. Currently, it has the following limitations, which will be addressed promptly:
The software is currently optimized for predicting binary variables from VBM data, which must be registered to the MNI space (we also recommend smoothing). For each fold of the cross-validation as well as for the final fitting, the software creates a mask of the voxels with >0.01 gray matter density, voxelwise applies a linear model to remove the effects of covariates, and includes the residuals into a lasso regression using the lambda that gives the minimum error in a cross-validation. The intercept of the lasso regression, as well as the MNI coordinates, the covariates coefficients and the lasso coefficient of the voxels with non-null lasso coefficients are saved.