Open data repository, Knab et al., Prediction of stroke outcome in mice based on non-invasive MRI and behavioral testing

README.txt

This information

dat

Contains MRI data in NIFTI format and secondary data from atlas registration. For documentation of atlas registration files see https://pubmed.ncbi.nlm.nih.gov/28829217/
Files used for the manuscript:
t2.nii: t2 weighted image acquired 24 h post stroke
masklesion.nii: manually delineated lesion
x_masklesion.nii: lesion in atlas space
ix_ANO.nii: Allen brain atlas in native space (i.e. matching t2.nii)
Lesion volume was calculated by volume of voxels unequal 0 in x_masklesion.nii
Overlap of regions defined by ix_ANO.nii with masklesion.nii were used for calculating percent damage in each atlas region

prediction_models

Contains separated training and test data as xlsx and csv files with lesion volumes in cubic mm of the Allen brain atlas space, percent damage per atlas region and behavioral data. The training data was used as input for training prediction models in MATLAB, the results were created using the test data.
The files have following sturcture:
Column 1: animal ID
Columns 2-537: MRI regions (column title corresponds to the region number as used in the Allen common coordinate framework)
Column 538: lesion volume
Column 539: initial performance (subacute deficit) = mean performance/deficit on days 2-6
Column 540: mean performance/deficit on days 2-6 = initial performance (subacute deficit) - this column equals column 539 but has different header which was used to train the residual from initial deficit
Column 541: residual performance/deficit
Column 542: test or training group
Consecutive rows contain data for each animal specified by the animal id

The repository also contains all trained models, prediction results for the test data and tables with resulting median absolute error (MedAE) and 5th, 25th, 75th and 95 absolute error quantiles for each model.
The model files end with '_models.mat' and contain 50 independently trained models each. Each model version is specified by number 1-50.
The result files end with '_test_results.mat' or '_test_results.xlsx', files with MedAE and quantiles end with '_test_errors.xlsx' or '_test_errors.csv. The common part of filenames specifies the used paradigm
Folder 'subacute deficit prediction' contains:
- initial_performance_from_lesion_volume: prediction of subacute deficit using lesion volume
- initial_performance_from_segmented_mri: prediction of subacute deficit using segmented mri
Folder 'long-term outcome prediction' contains:
- lesion_volume: prediction of residual deficit using lesion volume
- segmented_mri: prediction of residual deficit using segmented_mri
- initial_performance: prediction of residual deficit using subacute deficit
Folder 'mri_inc_oob_imp' contains models trained using increasing number of mri segments sorted according to the out-of-bag importance. The number of used segments is given in the file name. The models, results and errors are separated in subfolders.

Files with equal file name and different extension always contain the same data