Name

db_corr - Correlate DNN activation with brain response

Synopsis

db_corr [-h] -act Activation [-layer Layer [Layer ...]]
        [-chn Channel [Channel ...]] [-dmask DnnMask] [-iteraxis Axis]
        -resp Response [-bmask BrainMask] [-roi RoiName [RoiName ...]]
        -out Output

Arguments

Required Arguments

Argument	Discription
act	DNN activation file
resp	a .roi.h5/.nii file. If it is .nii file, -roi will be ignored. All voxels’ activation will be a correlate.
out	output directory.

Optional Arguments

Argument	Discription
layer	layer names of interest
chn	channel numbers of interest Default is using all channels of each layer specified by -layer.
dmask	a .dmask.csv file in which layers of interest are listed with their own channels, rows and columns of interest.
iteraxis	choices = (channel, row_col) Iterate along the specified axis. channel: Summarize the maximal pearson r for each channel. row_col: Summarize the maximal pearson r for each position (row_idx, col_idx). default: Summarize the maximal pearson r for the whole layer.
bmask	Brain mask is used to extract activation locally. Voxels with non-zero value will be regarded as correlates. Only used when the response file is .nii file.
roi	Specify ROI names as the correlates. Default is using all ROIs in .roi.h5 file.

Outputs

Maximal pearson r and its location will be saved.

Different layers’ output is stored in different folders.

Examples

Calculate pearson r correlation between DNN unit and brain voxel.

For each voxel, find the unit with the maximal correlation within each layer.

The DNN units are from layer conv5 and fc1.

The brain voxels are those with non-zero value in bmask.nii.gz.

Maximal pearson r and its location will be saved at out_dir

db_corr -act test.act.h5 -layer conv5 fc1 -resp resp.nii.gz -bmask bmask.nii.gz -out out_dir