def twosample_test(data_images,
vardata_images,
mask_images,
labels,
stat_id,
permutations=0,
cluster_forming_th=0.01):
"""
Helper function for permutation-based mass univariate twosample group
analysis. Labels is a binary vector (1-2). Regions more active for group
1 than group 2 are inferred.
"""
# Prepare arrays
data, vardata, xyz, mask = prepare_arrays(data_images, vardata_images,
mask_images)
# Create two-sample permutation test instance
if vardata_images is None:
ptest = permutation_test_twosample(data[labels == 1],
data[labels == 2],
xyz,
stat_id=stat_id)
else:
ptest = permutation_test_twosample(data[labels == 1],
data[labels == 2],
xyz,
vardata1=vardata[labels == 1],
vardata2=vardata[labels == 2],
stat_id=stat_id)
# Compute z-map image
zmap = np.zeros(data_images[0].shape).squeeze()
zmap[list(xyz)] = ptest.zscore()
zimg = Image(zmap, get_affine(data_images[0]))
# Compute mask image
maskimg = Image(mask, get_affine(data_images[0]))
# Multiple comparisons
if permutations <= 0:
return zimg, maskimg
else:
# Cluster definition: (threshold, diameter)
cluster_def = (ptest.height_threshold(cluster_forming_th), None)
# Calibration
voxel_res, cluster_res, region_res = \
ptest.calibrate(nperms=permutations, clusters=[cluster_def])
nulls = {}
nulls['zmax'] = ptest.zscore(voxel_res['perm_maxT_values'])
nulls['s'] = cluster_res[0]['perm_size_values']
nulls['smax'] = cluster_res[0]['perm_maxsize_values']
# Return z-map image, mask image and dictionary of null
# distribution for cluster sizes (s), max cluster size (smax)
# and max z-score (zmax)
return zimg, maskimg, nulls
def onesample_test(data_images,
vardata_images,
mask_images,
stat_id,
permutations=0,
cluster_forming_th=0.01):
"""
Helper function for permutation-based mass univariate onesample
group analysis.
"""
# Prepare arrays
data, vardata, xyz, mask = prepare_arrays(data_images, vardata_images,
mask_images)
# Create one-sample permutation test instance
ptest = permutation_test_onesample(data,
xyz,
vardata=vardata,
stat_id=stat_id)
# Compute z-map image
zmap = np.zeros(data_images[0].shape).squeeze()
zmap[list(xyz)] = ptest.zscore()
zimg = Image(zmap, get_affine(data_images[0]))
# Compute mask image
maskimg = Image(mask.astype(np.int8), get_affine(data_images[0]))
# Multiple comparisons
if permutations <= 0:
return zimg, maskimg
else:
# Cluster definition: (threshold, diameter)
cluster_def = (ptest.height_threshold(cluster_forming_th), None)
# Calibration
voxel_res, cluster_res, region_res = \
ptest.calibrate(nperms=permutations, clusters=[cluster_def])
nulls = {}
nulls['zmax'] = ptest.zscore(voxel_res['perm_maxT_values'])
nulls['s'] = cluster_res[0]['perm_size_values']
nulls['smax'] = cluster_res[0]['perm_maxsize_values']
# Return z-map image, mask image and dictionary of null distribution
# for cluster sizes (s), max cluster size (smax) and max z-score (zmax)
return zimg, maskimg, nulls
def remake_images():
# Get group data
group_data = example_data.get_filename(
'neurospin', 'language_babies', 'offset_002.npz')
f = np.load(group_data)
data, vardata, xyz = f['mat'], f['var'], f['xyz']
dX = xyz[0].max() + 1
dY = xyz[1].max() + 1
dZ = xyz[2].max() + 1
aux = np.zeros([dX, dY, dZ])
data_images = []
vardata_images = []
mask_images = []
for i in range(data.shape[0]):
aux[list(xyz)] = data[i]
data_images.append(Image(aux.copy(), np.eye(4)))
aux[list(xyz)] = vardata[i]
vardata_images.append(Image(aux.copy(), np.eye(4)))
aux[list(xyz)] = 1
mask_images.append(aux)
return data_images, vardata_images, mask_images
def contrast(self, vector):
"""Compute images of contrast and contrast variance.
"""
# Compute the overall contrast across models
c = self.glm[0].contrast(vector)
for g in self.glm[1:]:
c += g.contrast(vector)
def affect_inmask(dest, src, xyz):
if xyz is None:
dest = src
else:
dest[xyz] = src
return dest
con = np.zeros(self.spatial_shape)
con_img = Image(affect_inmask(con, c.effect, self.xyz), self.affine)
vcon = np.zeros(self.spatial_shape)
vcon_img = Image(affect_inmask(vcon, c.variance, self.xyz),
self.affine)
z = np.zeros(self.spatial_shape)
z_img = Image(affect_inmask(z, c.zscore(), self.xyz), self.affine)
dof = c.dof
return con_img, vcon_img, z_img, dof
def linear_model_fit(data_images, mask_images, design_matrix, vector):
"""
Helper function for group data analysis using arbitrary design matrix
"""
# Prepare arrays
data, vardata, xyz, mask = prepare_arrays(data_images, None, mask_images)
# Create glm instance
G = glm(data, design_matrix)
# Compute requested contrast
c = G.contrast(vector)
# Compute z-map image
zmap = np.zeros(data_images[0].shape).squeeze()
zmap[list(xyz)] = c.zscore()
zimg = Image(zmap, get_affine(data_images[0]))
return zimg