FastSurferCNN.reduce_to_aseg¶

FastSurferCNN.reduce_to_aseg.clean_segmentation(seg, vox_vol)[source]¶

Remove the labels outside the brain mask and flip disconnected white matter islands to their hemisphere.

Parameters:

segnp.ndarray: The aparc+aseg (or aseg) segmentation.
vox_volfloat: The volume of a voxel in mm^3.

Returns:

segnp.ndarray: The cleaned segmentation.
masknp.ndarray: The brain mask, see create_mask.

FastSurferCNN.reduce_to_aseg.create_mask(aseg_data, dnum, enum)[source]¶

Create dilated mask (works for all data orientations, LIA/etc).

Parameters:

aseg_dataint np.ndarray: The input segmentation data.
dnumint: The number of iterations for the dilation operation.
enumint: The number of iterations for the erosion operation.

Returns:

bool np.ndarray same shape as aseg_data: Returns mask.

FastSurferCNN.reduce_to_aseg.create_mask_and_save(seg, seg_affine, seg_header, filename=None)[source]¶: Convenience function for brainmask generation plus saving.

FastSurferCNN.reduce_to_aseg.flip_wm_islands(aseg_data)[source]¶

Flip labels of disconnected white matter islands to the other hemisphere (works for all data orientations, LIA/etc).

Parameters:

aseg_datanumpy.ndarray: The input segmentation data, an aseg or an aparc+aseg.

Returns:

flip_datanumpy.ndarray: The segmentation data with flipped WM labels.

FastSurferCNN.reduce_to_aseg.mask_segmentation(seg, mask, vox_vol, keep=())[source]¶

Remove the labels outside the mask, except those in keep, and warn if this removes much.

Labels outside the brain mask are errors of the segmentation, e.g. tissue in the neck. They would end up outside brainmask.mgz, where the surface pipeline finds intensity 0 for them.

Parameters:

segnp.ndarray: The segmentation.
masknp.ndarray: The mask, voxels > 0 are inside.
vox_volfloat: The volume of a voxel in mm^3.
keeplist of int, tuple of int, default=(): Labels to keep outside the mask.

Returns:

np.ndarray: The segmentation without labels outside the mask.

FastSurferCNN.reduce_to_aseg.options_parse()[source]¶

Command line option parser.

Returns:

options: Object holding options.

FastSurferCNN.reduce_to_aseg.reduce_to_aseg(data_inseg)[source]¶

Reduce the input segmentation to a simpler segmentation (for all data orientations, LIA/etc).

Parameters:

data_insegnp.ndarray, torch.Tensor: The input segmentation. This should be a 3D array where the value at each position represents the segmentation label for that position.

Returns:

data_insegnp.ndarray, torch.Tensor: The reduced segmentation.

FastSurferCNN.reduce_to_aseg.reduce_to_aseg_and_save(seg, seg_affine, seg_header, filename=None)[source]¶: Convenience function for reduce_to_aseg plus saving.