def test_volume_source_morph_round_trip(
tmpdir, subject_from, subject_to, lower, upper, dtype, morph_mat,
monkeypatch):
"""Test volume source estimate morph round-trips well."""
import nibabel as nib
from nibabel.processing import resample_from_to
src = dict()
if morph_mat:
# ~1.5 minutes with pos=7. (4157 morphs!) for sample, so only test
# morph_mat computation mode with a few labels
label_names = sorted(get_volume_labels_from_aseg(fname_aseg))[1:2]
if 'sample' in (subject_from, subject_to):
src['sample'] = setup_volume_source_space(
'sample', subjects_dir=subjects_dir,
volume_label=label_names, mri=fname_aseg)
assert sum(s['nuse'] for s in src['sample']) == 12
if 'fsaverage' in (subject_from, subject_to):
src['fsaverage'] = setup_volume_source_space(
'fsaverage', subjects_dir=subjects_dir,
volume_label=label_names[:3], mri=fname_aseg_fs)
assert sum(s['nuse'] for s in src['fsaverage']) == 16
else:
assert not morph_mat
if 'sample' in (subject_from, subject_to):
src['sample'] = mne.read_source_spaces(fname_vol)
src['sample'][0]['subject_his_id'] = 'sample'
assert src['sample'][0]['nuse'] == 4157
if 'fsaverage' in (subject_from, subject_to):
# Created to save space with:
#
# bem = op.join(op.dirname(mne.__file__), 'data', 'fsaverage',
# 'fsaverage-inner_skull-bem.fif')
# src_fsaverage = mne.setup_volume_source_space(
# 'fsaverage', pos=7., bem=bem, mindist=0,
# subjects_dir=subjects_dir, add_interpolator=False)
# mne.write_source_spaces(fname_fs_vol, src_fsaverage,
# overwrite=True)
#
# For speed we do it without the interpolator because it's huge.
src['fsaverage'] = mne.read_source_spaces(fname_fs_vol)
src['fsaverage'][0].update(
vol_dims=np.array([23, 29, 25]), seg_name='brain')
_add_interpolator(src['fsaverage'])
assert src['fsaverage'][0]['nuse'] == 6379
src_to, src_from = src[subject_to], src[subject_from]
del src
# No SDR just for speed once everything works
kwargs = dict(niter_sdr=(), niter_affine=(1,),
subjects_dir=subjects_dir, verbose=True)
morph_from_to = compute_source_morph(
src=src_from, src_to=src_to, subject_to=subject_to, **kwargs)
morph_to_from = compute_source_morph(
src=src_to, src_to=src_from, subject_to=subject_from, **kwargs)
nuse = sum(s['nuse'] for s in src_from)
assert nuse > 10
use = np.linspace(0, nuse - 1, 10).round().astype(int)
data = np.eye(nuse)[:, use]
if dtype is complex:
data = data * 1j
vertices = [s['vertno'] for s in src_from]
stc_from = VolSourceEstimate(data, vertices, 0, 1)
with catch_logging() as log:
stc_from_rt = morph_to_from.apply(
morph_from_to.apply(stc_from, verbose='debug'))
log = log.getvalue()
assert 'individual volume morph' in log
maxs = np.argmax(stc_from_rt.data, axis=0)
src_rr = np.concatenate([s['rr'][s['vertno']] for s in src_from])
dists = 1000 * np.linalg.norm(src_rr[use] - src_rr[maxs], axis=1)
mu = np.mean(dists)
# fsaverage=5.99; 7.97 without additional src_ras_t fix
# fsaverage=7.97; 25.4 without src_ras_t fix
assert lower <= mu < upper, f'round-trip distance {mu}'
# check that pre_affine is close to identity when subject_to==subject_from
if subject_to == subject_from:
for morph in (morph_to_from, morph_from_to):
assert_allclose(
morph.pre_affine.affine, np.eye(4), atol=1e-2)
# check that power is more or less preserved (labelizing messes with this)
if morph_mat:
if subject_to == 'fsaverage':
limits = (18, 18.5)
else:
limits = (7, 7.5)
else:
limits = (1, 1.2)
stc_from_unit = stc_from.copy().crop(0, 0)
stc_from_unit._data.fill(1.)
stc_from_unit_rt = morph_to_from.apply(morph_from_to.apply(stc_from_unit))
assert_power_preserved(stc_from_unit, stc_from_unit_rt, limits=limits)
if morph_mat:
fname = tmpdir.join('temp-morph.h5')
morph_to_from.save(fname)
morph_to_from = read_source_morph(fname)
assert morph_to_from.vol_morph_mat is None
morph_to_from.compute_vol_morph_mat(verbose=True)
morph_to_from.save(fname, overwrite=True)
morph_to_from = read_source_morph(fname)
assert isinstance(morph_to_from.vol_morph_mat, csr_matrix), 'csr'
# equivalence (plus automatic calling)
assert morph_from_to.vol_morph_mat is None
monkeypatch.setattr(mne.morph, '_VOL_MAT_CHECK_RATIO', 0.)
with catch_logging() as log:
with pytest.warns(RuntimeWarning, match=r'calling morph\.compute'):
stc_from_rt_lin = morph_to_from.apply(
morph_from_to.apply(stc_from, verbose='debug'))
assert isinstance(morph_from_to.vol_morph_mat, csr_matrix), 'csr'
log = log.getvalue()
assert 'sparse volume morph matrix' in log
assert_allclose(stc_from_rt.data, stc_from_rt_lin.data)
del stc_from_rt_lin
stc_from_unit_rt_lin = morph_to_from.apply(
morph_from_to.apply(stc_from_unit))
assert_allclose(stc_from_unit_rt.data, stc_from_unit_rt_lin.data)
del stc_from_unit_rt_lin
del stc_from, stc_from_rt
# before and after morph, check the proportion of vertices
# that are inside and outside the brainmask.mgz
brain = nib.load(op.join(subjects_dir, subject_from, 'mri', 'brain.mgz'))
mask = _get_img_fdata(brain) > 0
if subject_from == subject_to == 'sample':
for stc in [stc_from_unit, stc_from_unit_rt]:
img = stc.as_volume(src_from, mri_resolution=True)
img = nib.Nifti1Image( # abs to convert complex
np.abs(_get_img_fdata(img)[:, :, :, 0]), img.affine)
img = _get_img_fdata(resample_from_to(img, brain, order=1))
assert img.shape == mask.shape
in_ = img[mask].astype(bool).mean()
out = img[~mask].astype(bool).mean()
if morph_mat:
out_max = 0.001
in_min, in_max = 0.005, 0.007
else:
out_max = 0.02
in_min, in_max = 0.97, 0.98
assert out < out_max, f'proportion out of volume {out}'
assert in_min < in_ < in_max, f'proportion inside volume {in_}'
def test_volume_source_morph_round_trip(tmpdir, subject_from, subject_to,
lower, upper):
"""Test volume source estimate morph round-trips well."""
import nibabel as nib
from nibabel.processing import resample_from_to
src = dict()
if 'sample' in (subject_from, subject_to):
src['sample'] = mne.read_source_spaces(fname_vol)
src['sample'][0]['subject_his_id'] = 'sample'
assert src['sample'][0]['nuse'] == 4157
if 'fsaverage' in (subject_from, subject_to):
# Created to save space with:
#
# bem = op.join(op.dirname(mne.__file__), 'data', 'fsaverage',
# 'fsaverage-inner_skull-bem.fif')
# src_fsaverage = mne.setup_volume_source_space(
# 'fsaverage', pos=7., bem=bem, mindist=0,
# subjects_dir=subjects_dir, add_interpolator=False)
# mne.write_source_spaces(fname_fs_vol, src_fsaverage, overwrite=True)
#
# For speed we do it without the interpolator because it's huge.
src['fsaverage'] = mne.read_source_spaces(fname_fs_vol)
src['fsaverage'][0].update(vol_dims=np.array([23, 29, 25]),
seg_name='brain')
_add_interpolator(src['fsaverage'], True)
assert src['fsaverage'][0]['nuse'] == 6379
src_to, src_from = src[subject_to], src[subject_from]
del src
# No SDR just for speed once everything works
kwargs = dict(niter_sdr=(),
niter_affine=(1, ),
subjects_dir=subjects_dir,
verbose=True)
morph_from_to = compute_source_morph(src=src_from,
src_to=src_to,
subject_to=subject_to,
**kwargs)
morph_to_from = compute_source_morph(src=src_to,
src_to=src_from,
subject_to=subject_from,
**kwargs)
use = np.linspace(0, src_from[0]['nuse'] - 1, 10).round().astype(int)
stc_from = VolSourceEstimate(
np.eye(src_from[0]['nuse'])[:, use], [src_from[0]['vertno']], 0, 1)
stc_from_rt = morph_to_from.apply(morph_from_to.apply(stc_from))
maxs = np.argmax(stc_from_rt.data, axis=0)
src_rr = src_from[0]['rr'][src_from[0]['vertno']]
dists = 1000 * np.linalg.norm(src_rr[use] - src_rr[maxs], axis=1)
mu = np.mean(dists)
assert lower <= mu < upper # fsaverage=7.97; 25.4 without src_ras_t fix
# check that pre_affine is close to identity when subject_to==subject_from
if subject_to == subject_from:
for morph in (morph_to_from, morph_from_to):
assert_allclose(morph.pre_affine.affine, np.eye(4), atol=1e-2)
# check that power is more or less preserved
ratio = stc_from.data.size / stc_from_rt.data.size
limits = ratio * np.array([1, 1.2])
stc_from.crop(0, 0)._data.fill(1.)
stc_from_rt = morph_to_from.apply(morph_from_to.apply(stc_from))
assert_power_preserved(stc_from, stc_from_rt, limits=limits)
# before and after morph, check the proportion of vertices
# that are inside and outside the brainmask.mgz
brain = nib.load(op.join(subjects_dir, subject_from, 'mri', 'brain.mgz'))
mask = _get_img_fdata(brain) > 0
if subject_from == subject_to == 'sample':
for stc in [stc_from, stc_from_rt]:
img = stc.as_volume(src_from, mri_resolution=True)
img = nib.Nifti1Image(_get_img_fdata(img)[:, :, :, 0], img.affine)
img = _get_img_fdata(resample_from_to(img, brain, order=1))
assert img.shape == mask.shape
in_ = img[mask].astype(bool).mean()
out = img[~mask].astype(bool).mean()
assert 0.97 < in_ < 0.98
assert out < 0.02