def feature_map(self, feature, imPath=None, pw=0.95):
"""
Given a set of feature values, produce a feature map,
assuming that one feature corresponds to one region
Parameters
----------
feature, array of shape (self.k) : the information to map
imPath=None, string yielding the output image path
if not None
pw=0.95: volume of the Gaussian ellipsoid associated with the ROIs
Returns
-------
The image object
"""
if np.size(feature)!=self.k:
raise ValueError, 'Incompatible feature dimension'
from nipy.io.imageformats import save, Nifti1Image
label = self.map_label(self.generate_coordinates(), pval=pw)
label = np.reshape(label, self.shape)
values = np.zeros(self.shape)
values[label>-1] = feature[label[label>-1].astype(np.int)]
wim = Nifti1Image(values, self.affine)
wim.get_header()['descrip']='feature image'
if imPath!=None:
save(wim,imPath)
return wim
def make_image(self, path):
"""
write a int image where the nonzero values are the ROIs
Parameters
----------
path: string, the desired image path
Note
----
the background values are set to -1
the ROIs values are set as [0..self.k-1]
"""
if self.shape == None:
raise ValueError, "Need self.shape to be defined"
data = -np.ones(self.shape, np.int)
for k in range(self.k):
dk = self.xyz[k].T
data[dk[0], dk[1], dk[2]] = k
wim = Nifti1Image(data, self.affine)
header = wim.get_header()
header["descrip"] = "Multiple ROI image"
save(wim, path)
def make_image(self, path=None):
"""
write a int image where the nonzero values are the ROIs
Parameters
----------
path: string, optional
the desired image path
Returns
-------
brifti image instance
Note
----
the background values are set to -1
the ROIs values are set as [0..self.k-1]
"""
if self.shape==None:
raise ValueError, 'Need self.shape to be defined'
data = -np.ones(self.shape,np.int)
for k in range(self.k):
dk = self.xyz[k].T
data[dk[0], dk[1], dk[2]] = k
wim = Nifti1Image(data, self.affine)
header = wim.get_header()
header['descrip'] = "Multiple ROI image"
if path!=None:
save(wim, path)
return wim
def main():
# create the parser
parser = argparse.ArgumentParser()
# add the arguments
parser.add_argument("in_filenames", type=str, nargs="+", help="3D image filenames")
parser.add_argument("--out-4d", type=str, help="4D output image name")
parser.add_argument(
"--check-affines",
type=bool,
default=True,
help="False if you want to ignore differences "
"in affines between the 3D images, True if you "
"want to raise an error for significant "
"differences (default is True)",
)
# parse the command line
args = parser.parse_args()
# get input 3ds
filenames = args.in_filenames
# affine check
check_affines = args.check_affines
# get output name
out_fname = args.out_4d
if out_fname is None:
pth, fname = os.path.split(filenames[0])
froot, ext = os.path.splitext(fname)
if ext in (".gz", ".bz2"):
gz = ext
froot, ext = os.path.splitext(froot)
else:
gz = ""
out_fname = pjoin(pth, froot + "_4d" + ext + gz)
img4d = do_3d_to_4d(filenames, check_affines=check_affines)
nii.save(img4d, out_fname)
def ffx( maskImages, effectImages, varianceImages, resultImage=None):
"""
Computation of the fixed effecst statistics
Parameters
----------
maskImages, string or list of strings
the paths of one or several masks
when several masks, the half thresholding heuristic is used
effectImages, list of strings
the paths ofthe effect images
varianceImages, list of strings
the paths of the associated variance images
resultImage=None, string,
path of the result images
Returns
-------
the computed values
"""
# fixme : check that the images have same referntial
# fixme : check that mask_Images is a list
if len(effectImages)!=len(varianceImages):
raise ValueError, 'Not the correct number of images'
tiny = 1.e-15
nsubj = len(effectImages)
mask = intersect_masks(maskImages, None, threshold=0.5, cc=True)
effects = []
variance = []
for s in range(nsubj):
rbeta = load(effectImages[s])
beta = rbeta.get_data()[mask>0]
rbeta = load(varianceImages[s])
varbeta = rbeta.get_data()[mask>0]
effects.append(beta)
variance.append(varbeta)
effects = np.array(effects)
variance = np.array(variance)
effects[np.isnan(effects)] = 0
effects[np.isnan(variance)] = 0
variance[np.isnan(variance)] = tiny
variance[variance==0] = tiny
t = effects/np.sqrt(variance)
t = t.mean(0)*np.sqrt(nsubj)
#t = np.sum(effects/variance,0)/np.sum(1.0/np.sqrt(variance),0)
nim = load(effectImages[0])
affine = nim.get_affine()
tmap = np.zeros(nim.get_shape())
tmap[mask>0] = t
tImage = Nifti1Image(tmap, affine)
if resultImage!=None:
save(tImage, resultImage)
return tmap
def save_results(ppm, id_algo, noise):
savedir = os.path.join(baseres, subjects[s_idx])
tag = id_algo
if noise == 'laplace':
tag += '_laplace'
for i in range(ntissues):
im = Image(ppm[:,:,:,i], affine)
fname = id_posterior + '_' + regs[r_idx] + '_' + tag + '_' + str(i) + '.nii'
save(Nifti1Image(im), os.path.join(savedir, fname))
def intersect_masks(input_masks, output_filename=None, threshold=0.5, cc=True):
"""
Given a list of input mask images, generate the output image which
is the the threshold-level intersection of the inputs
Parameters
----------
input_masks: list of strings or ndarrays
paths of the input images nsubj set as len(input_mask_files), or
individual masks.
output_filename, string:
Path of the output image, if None no file is saved.
threshold: float within [0, 1], optional
gives the level of the intersection.
threshold=1 corresponds to keeping the intersection of all
masks, whereas threshold=0 is the union of all masks.
cc: bool, optional
If true, extract the main connected component
Returns
-------
grp_mask, boolean array of shape the image shape
"""
grp_mask = None
for this_mask in input_masks:
if isinstance(this_mask, basestring):
# We have a filename
this_mask = load(this_mask).get_data()
if grp_mask is None:
grp_mask = this_mask.copy().astype(np.int)
else:
grp_mask += this_mask
grp_mask = grp_mask>(threshold*len(input_masks))
if np.any(grp_mask>0) and cc:
grp_mask = largest_cc(grp_mask)
if output_filename is not None:
if isinstance(input_masks[0], basestring):
nim = load(input_masks[0])
header = nim.get_header()
affine = nim.get_affine()
else:
header = dict()
affine = np.eye(4)
header['descrip'] = 'mask image'
output_image = nifti1.Nifti1Image(grp_mask.astype(np.uint8),
affine=affine,
header=header,
)
save(output_image, output_filename)
return grp_mask>0
def to_image(self, path=None):
"""
Write itself as an image, and returns it
"""
data = np.zeros(self.shape).astype(np.int8)
data[self.ijk[:,0], self.ijk[:,1], self.ijk[:,2]] = 1
nim = Nifti1Image(data, self.affine)
nim.get_header()['descrip'] = 'mask image'
if path is not None:
save(nim, path)
return nim
def save(filename, obj):
""" Save an nipy image object to a file.
"""
obj = as_volume_img(obj, copy=False)
hdr = imageformats.Nifti1Header()
for key, value in obj.metadata.iteritems():
if key in hdr:
hdr[key] = value
img = imageformats.Nifti1Image(obj.get_data(),
obj.affine,
header=hdr)
imageformats.save(img, filename)
def threshold_z_image(iimage, oimage=None, correction=None, pval=None, smin=0,
nn=18, mask_image=None, method=None):
"""
this function takes a presumably gaussian image threshold and a
size threshold and gives as output an image where only the
suprathreshold component of size > smin have not been thresholded
out This corresponds to a one-sided classical test the null
hypothesis can be take to be the standard normal or the empiricall
null.
Parameters
----------
iimage, string, the path of a presumably z-variate input image
oimage=None, string, the path of the output image
correction=None, string the correction for multiple comparison method
correction can be either None or 'bon' (Bonferroni) or 'fdr'
pval=none, float, the desired classical p-value.
the default behaviour of pval depends on correction
if correction==None, pval = 0.001, else pval = 0.05
smin=0, int, the cluster size threshold
mask_image=None, string path of a mask image to determine
where thresholding is applies
if mask_image==None, the function is implied
on where(image)
method=None: model of the null distribution:
if method==None: standard null
if method=='emp': empirical null
Returns
-------
oimage: the output image
"""
#?# 1. read the image(s)
if mask_image==None:
m = None
else:
mask = load(mask_image)
m = mask.get_data()
nim = load(iimage)
x = nim.get_data()
thx = threshold_z_array(x, m, correction, pval, smin, nn, method)
ref_dim = nim.get_shape()
result = np.zeros(ref_dim)
result[m>0] = thx
onim = Nifti1Image(result.T, nim.get_affine())
onim.get_header()['descrip']= "thresholded image, threshold= %f,\
cluster size=%d"%(thx, smin)
if oimage !=None:
save(onim, oimage)
return onim
def parcellation_output_with_paths(Pa, mask_images, group_path, indiv_path):
"""
Function that produces images that describe the spatial structure
of the parcellation. It mainly produces label images at the group
and subject level
Parameters
----------
Pa : Parcellation instance that describes the parcellation
mask_images: list of images paths that define the mask
coord: array of shape (nvox,3) that contains(approximated)
MNI-coordinates of the brain mask voxels considered in the
parcellation process
group_path, string, path of the group-level parcellation image
indiv_path, list of strings, paths of the individual parcellation images
fixme
-----
the referential-defining information should be part of the Pa instance
"""
nsubj = Pa.nb_subj
mxyz = Pa.ijk
# write the template image
tlabs = Pa.group_labels
rmask = load(mask_images[0])
ref_dim = rmask.get_shape()
grid_size = np.prod(ref_dim)
affine = rmask.get_affine()
Label = np.zeros(ref_dim)
Label[Pa.ijk[:,0],Pa.ijk[:,1],Pa.ijk[:,2]]=tlabs+1
wim = Nifti1Image (Label, affine)
hdr = wim.get_header()
hdr['descrip'] = 'group_level Label image obtained from a \
parcellation procedure'
save(wim, group_path)
# write subject-related stuff
for s in range(nsubj):
# write the images
labs = Pa.label[:,s]
Label = np.zeros(ref_dim).astype(np.int)
Label[Pa.ijk[:,0],Pa.ijk[:,1],Pa.ijk[:,2]]=labs+1
wim = Nifti1Image (Label, affine)
hdr = wim.get_header()
hdr['descrip'] = 'individual Label image obtained \
from a parcellation procedure'
save(wim, indiv_path[s])
def threshold_scalar_image(iimage, oimage=None, th=0., smin=0, nn=18,
mask_image=None):
"""
this function takes a 'grey level' threshold and a size threshold
and gives as output an image where only the suprathreshold component
of size > smin have not been thresholded out
Parameters
----------
iimage, string, path of a scalar input image
oimage=None, string, path of the scalar output image
if None the output image is not written
th=0., float, the chosen trheshold
smin=0, int, cluster size threshold
nn=18, int spatial neighboring system: 6,18 or 26
mask_image=None: a mask image to determine where in image this applies
if mask_image==None, the function is implied on
where(image)
Returns
-------
output, image: the output image object
Note, the 0 values of iimage are not considered so far
"""
# FIXME: add a header check here
# 1. read the input image
if mask_image==None:
m = None
else:
mask = load(mask_image)
m = mask.get_data()
inim = load(iimage)
x = inim.get_data()
thx = threshold_array(x, m, th, smin, nn=nn)
ref_dim = inim.get_shape()
result = np.zeros(ref_dim)
result[m>0] = thx
onim = Nifti1Image(result.T,inim.get_affine())
onim.get_header()['descrip']= "thresholded image, threshold= %f,\
cluster size=%d"%(th,smin)
if oimage !=None:
save(onim, oimage)
return onim
def main():
# create the parser
parser = argparse.ArgumentParser()
# add the arguments
parser.add_argument('filename', type=str,
help='4D image filename')
# parse the command line
args = parser.parse_args()
img = nii.load(args.filename)
imgs = nii.four_to_three(img)
froot, ext = os.path.splitext(args.filename)
if ext in ('.gz', '.bz2'):
froot, ext = os.path.splitext(froot)
for i, img3d in enumerate(imgs):
fname3d = '%s_%04d.nii' % (froot, i)
nii.save(img3d, fname3d)
def test_mask_files():
with InTemporaryDirectory():
# Make a 4D file from the anatomical example
img = nii.load(anatfile)
arr = img.get_data()
a2 = np.zeros(arr.shape + (2,))
a2[:, :, :, 0] = arr
a2[:, :, :, 1] = arr
img = nii.Nifti1Image(a2, np.eye(4))
a_fname = "fourd_anat.nii"
nii.save(img, a_fname)
# check 4D mask
msk1 = nnm.compute_mask_files(a_fname)
# and mask from identical list of 3D files
msk2 = nnm.compute_mask_files([anatfile, anatfile])
yield assert_array_equal, msk1, msk2
def make_image(self, image_path):
"""
write a binary nifty image where the nonzero values are the ROI mask
Parameters
-----------
image_path: string
the desired image name
"""
if self.shape == None:
raise ValueError, "self.shape has to be defined"
data = np.zeros(self.shape)
data[self.discrete] = 1
wim = Nifti1Image(data, self.affine)
wim.get_header()["descrip"] = "ROI image"
save(wim, image_path)
def ffx_from_stat( maskImages, statImages, resultImage=None):
"""
Computation of the fixed effects statistics from statistic
Parameters
----------
maskImages, string or list of strings
the paths of one or several masks
when several masks, the half thresholding heuristic is used
statImages, list of strings
the paths ofthe statitsic images
resultImage=None, string,
path of the result images
Returns
-------
the computed values
"""
# fixme : check that the images have same referntial
# fixme : check that mask_Images is a list
nsubj = len(statImages)
mask = intersect_masks(maskImages, None, threshold=0.5, cc=True)
t = []
for s in range(nsubj):
rbeta = load(statImages[s])
beta = rbeta.get_data()[mask>0]
t.append(beta)
t = np.array(t)
t[np.isnan(t)] = 0
t = t.mean(0)*np.sqrt(nsubj)
nim = load(statImages[0])
affine = nim.get_affine()
tmap = np.zeros(nim.get_shape())
tmap[mask>0] = t
tImage = Nifti1Image(tmap, affine)
if resultImage!=None:
save(tImage,resultImage)
return tmap
def mask_parcellation(mask_images, nb_parcel, output_image=None):
"""
Performs the parcellation of a certain mask
Parameters
----------
mask_images: list of strings,
paths of the mask images that define the common space.
nb_parcel: int,
number of desired parcels
output_image: string, optional
path of the output image
Returns
-------
wim: Nifti1Imagine instance, the resulting parcellation
"""
from ..mask import intersect_masks
# compute the group mask
affine = load(mask_images[0]).get_affine()
shape = load(mask_images[0]).get_shape()
mask = intersect_masks(mask_images, threshold=0)>0
ijk = np.where(mask)
ijk = np.array(ijk).T
nvox = ijk.shape[0]
# Get and cluster coordinates
ijk = np.hstack((ijk,np.ones((nvox,1))))
coord = np.dot(ijk, affine.T)[:,:3]
cent, tlabs, J = kmeans(coord, nb_parcel)
# Write the results
label = -np.ones(shape)
label[mask]= tlabs
wim = Nifti1Image(label, affine)
wim.get_header()['descrip'] = 'Label image in %d parcels'%nb_parcel
if output_image is not None:
save(wim, output_image)
return wim
def save_volume(shape, path, affine, mask=None, data=None, descrip=None):
"""
volume saving utility for masked volumes
Parameters
----------
shape, tupe of dimensions of the data
path, string, output image path
affine, transformation of the grid to a coordinate system
mask=None, binary mask used to reduce the volume size
data=None data to be put in the volume
descrip=None, a string descibing what the image is
Fixme
-----
Handle the case where data is multi-dimensional
"""
volume = np.zeros(shape)
if mask== None:
print "Could not write the image: no mask"
return
if data == None:
print "Could not write the image: no data"
return
if np.size(data.shape) == 1:
volume[mask > 0] = data
else:
for i in range(data.shape[0]):
volume[i][mask[0] > 0] = data[i]
wim = Nifti1Image(volume, affine)
if descrip !=None:
wim.get_header()['descrip']=descrip
save(wim, path)
def Parcellation_output(Pa, mask_images, learning_images, coord, nbru,
verbose=1,swd = "/tmp"):
"""
Function that produces images that describe the spatial structure
of the parcellation. It mainly produces label images at the group
and subject level
Parameters
----------
Pa : Parcellation instance that describes the parcellation
mask_images: list of images paths that define the mask
learning_images: list of float images containing the input data
coord: array of shape (nvox,3) that contains(approximated)
MNI-coordinates of the brain mask voxels considered in the
parcellation process
nbru: list of subject ids
verbose=1 : verbosity level
swd = '/tmp': write directory
Results
-------
Pa: the updated Parcellation instance
"""
nsubj = Pa.nb_subj
Pa.set_subjects(nbru)
# write the template image
tlabs = Pa.group_labels
LabelImage = os.path.join(swd,"template_parcel.nii")
rmask = load(mask_images[0])
ref_dim = rmask.get_shape()
affine = rmask.get_affine()
Label = np.zeros(ref_dim)
Label[Pa.ijk[:,0],Pa.ijk[:,1],Pa.ijk[:,2]]=tlabs+1
wim = Nifti1Image (Label, affine)
hdr = wim.get_header()
hdr['descrip'] = 'group_level Label image obtained from a \
parcellation procedure'
save(wim, LabelImage)
# write subject-related stuff
Jac = []
if Pa.isfield('jacobian'):
Jac = Pa.get_feature('jacobian')
Jac = np.reshape(Jac,(Pa.k,nsubj))
for s in range(nsubj):
# write the images
labs = Pa.label[:,s]
LabelImage = os.path.join(swd,"parcel%s.nii" % nbru[s])
JacobImage = os.path.join(swd,"jacob%s.nii" % nbru[s])
Label = np.zeros(ref_dim).astype(np.int)
Label[Pa.ijk[:,0],Pa.ijk[:,1],Pa.ijk[:,2]]=labs+1
wim = Nifti1Image (Label, affine)
hdr = wim.get_header()
hdr['descrip'] = 'individual Label image obtained \
from a parcellation procedure'
save(wim, LabelImage)
if ((verbose)&(np.size(Jac)>0)):
Label = np.zeros(ref_dim)
Label[Pa.ijk[:,0],Pa.ijk[:,1],Pa.ijk[:,2]]=Jac[labs,s]
wim = Nifti1Image (Label, affine)
hdr = wim.get_header()
hdr['descrip'] = 'image of the jacobian of the deformation \
associated with the parcellation'
save(wim, JacobImage)
return Pa
grp_mask = Nifti1Image(mask, load(mask_images[0]).get_affine())
ijk = np.array(np.where(mask)).T
nvox = np.sum(mask)
# output dir
b_smooth = True
if b_smooth:
print "smoothed data"
threshold_path = 'volume_threshold_smooth.con'
swd = '/data/home/virgile/virgile_internship/group_analysis/smoothed_FWHM5'
else:
print "unsmoothed data"
threshold_path = 'volume_threshold.con'
swd = '/data/home/virgile/virgile_internship/group_analysis/smoothed_FWHM0'
save(grp_mask, op.join(swd,'grp_mask.nii'))
################################################################
# Load the effects and variance
################################################################
def load_images(con_images, var_images):
"""
"""
nsubj = len(con_images)
beta = []
varbeta = []
tiny = 1.e-15
for s in range(nsubj):
rbeta = load(con_images[s])
temp = (rbeta.get_data())[mask]
drift_model='Cosine',
hfcut=128,
hrf_model=hrf_model,
paradigm=paradigm,
add_regs=motion,
add_reg_names=add_reg_names)
#######################################
# Get the FMRI data
#######################################
fmri_data = surrogate_4d_dataset(shape=shape, n_scans=n_scans)[0]
# if you want to save it as an image
data_file = op.join(swd, 'fmri_data.nii')
save(fmri_data, data_file)
########################################
# Perform a GLM analysis
########################################
# GLM fit
Y = fmri_data.get_data()
model = "ar1"
method = "kalman"
glm = GLM.glm()
mp.pcolor(X)
mp.show()
glm.fit(Y.T, X, method=method, model=model)
#explained = np.dot(X,glm.beta.reshape(X.shape[1],-1)).reshape(Y.T.shape).T
#residuals = Y - explained
########################################
paradigm = dm.EventRelatedParadigm(conditions, onsets)
X, names = dm.dmtx_light(frametimes, drift_model='Cosine', hfcut=128,
hrf_model=hrf_model, paradigm=paradigm, add_regs=motion, add_reg_names=add_reg_names)
#######################################
# Get the FMRI data
#######################################
fmri_data = surrogate_4d_dataset(shape=shape, n_scans=n_scans)[0]
# if you want to save it as an image
data_file = op.join(swd,'fmri_data.nii')
save(fmri_data, data_file)
########################################
# Perform a GLM analysis
########################################
# GLM fit
Y = fmri_data.get_data()
model = "ar1"
method = "kalman"
glm = GLM.glm()
mp.pcolor(X)
mp.show()
glm.fit(Y.T, X, method=method, model=model)
#explained = np.dot(X,glm.beta.reshape(X.shape[1],-1)).reshape(Y.T.shape).T
#residuals = Y - explained
def make_surrogate_array(nbsubj=10, dimx=30, dimy=30, sk=1.0,
noise_level=1.0, pos=pos, ampli=ampli,
spatial_jitter=1.0, signal_jitter=1.0,
width=5.0, out_text_file=None, out_image_file=None,
verbose=False, seed=False):
"""
Create surrogate (simulated) 2D activation data with spatial noise.
Parameters
-----------
nbsubj: integer, optionnal
The number of subjects, ie the number of different maps
generated.
dimx: integer, optionnal
The x size of the array returned.
dimy: integer
The y size of the array returned.
sk: float, optionnal
Amount of spatial noise smoothness.
noise_level: float, optionnal
Amplitude of the spatial noise.
amplitude=noise_level)
pos: 2D ndarray of integers, optionnal
x, y positions of the various simulated activations.
ampli: 1D ndarray of floats, optionnal
Respective amplitude of each activation
spatial_jitter: float, optionnal
Random spatial jitter added to the position of each activation,
in pixel.
signal_jitter: float, optionnal
Random amplitude fluctuation for each activation, added to the
amplitude specified by ampli
width: float or ndarray, optionnal
Width of the activations
out_text_file: string or None, optionnal
If not None, the resulting array is saved as a text file with the
given file name
out_image_file: string or None, optionnal
If not None, the resulting is saved as a nifti file with the
given file name.
verbose: boolean, optionnal
If verbose is true, the data for the last subject is plotted as
a 2D image.
seed=False: int, optionnal
If seed is not False, the random number generator is initialized
at a certain value
Returns
-------
dataset: 3D ndarray
The surrogate activation map, with dimensions (nbsubj, dimx, dimy)
"""
if seed:
nr = np.random.RandomState([seed])
else:
import numpy.random as nr
shape = (dimx, dimy)
ij = np.transpose(np.where(np.ones(shape)))
dataset = []
for s in range(nbsubj):
# make the signal
data = np.zeros(shape)
lpos = pos + spatial_jitter*nr.randn(1, 2)
lampli = ampli + signal_jitter*nr.randn(np.size(ampli))
for k in range(np.size(lampli)):
data = np.maximum(data,
_cone(shape, ij, lpos[k], lampli[k], width))
# make some noise
noise = nr.randn(dimx,dimy)
# smooth the noise
noise = nd.gaussian_filter(noise, sk)
noise = np.reshape(noise, (-1, 1))
noise *= noise_level/np.std(noise)
#make the mixture
data += np.reshape(noise, shape)
dataset.append(data)
if verbose:
import matplotlib.pylab as mp
mp.figure()
mp.imshow(data, interpolation='nearest')
mp.colorbar()
dataset = np.array(dataset)
if out_text_file is not None:
dataset.tofile(out_text_file)
if out_image_file is not None:
from nipy.io.imageformats import save, Nifti1Image
save(Nifti1Image( dataset, np.eye(4)), out_image_file)
return dataset
R.set_source_fov(fixed_npoints=64**3) # Make Gaussian spline transform instance spacing = 16 slices = [slice(0,s.stop,s.step*spacing) for s in R._slices] cp = np.mgrid[slices] cp = np.rollaxis(cp, 0, 4) # Start with an affine registration A0 = Affine() A = R.optimize(A0) ###A = Affine() # Save affinely transformed target Jt = J.transform(A, reference=I) save(asNifti1Image(Jt), 'affine_anubis_to_ammon.nii') # Then add control points... T0 = SplineTransform(I, cp, sigma=20., grid_coords=True, affine=A) """ # Test 1 s = R.eval(T0) sa = R.eval(T0.affine) assert_almost_equal(s, sa) # Test 2 T = SplineTransform(I, cp, sigma=5., grid_coords=True, affine=A) T.param += 1. s0 = R.eval(T0) s = R.eval(T)
label = -np.ones(F.V)
nroi = hroi.NROI_from_field(F, affine, shape, xyz, 0, threshold, smin)
bmap = -np.zeros(F.V)
if nroi!=None:
idx = nroi.discrete_features['index']
for k in range(nroi.k):
label[idx[k]] = k
# saving the blob image,i. e. a label image
wlabel = -2*np.ones(shape)
wlabel[data!=0] = label
blobPath = os.path.join(swd,"blob.nii")
wim = Nifti1Image(wlabel, affine)
wim.get_header()['descrip'] = 'blob image extracted from %s'%InputImage
save(wim, blobPath)
# --- 2.b take blob labelled "1" as an ROI
roi = DiscreteROI( affine=affine, shape=shape)
roi.from_labelled_image(blobPath, 1)
roiPath2 = os.path.join(swd, "roi_blob_1.nii")
roi.make_image(roiPath2)
# --- 2.c take the blob closest to 'position as an ROI'
roiPath3 = os.path.join(swd, "blob_closest_to_%d_%d_%d.nii")%\
(position[0], position[1], position[2])
roi.from_position_and_image(blobPath, np.array(position))
roi.make_image(roiPath3)
# --- 2.d make a set of ROIs from all the blobs
mroi = MultipleROI( affine=affine, shape=shape)
def Parcellation_based_analysis(Pa, test_images, numbeta, swd="/tmp",
DMtx=None, verbose=1, method_id=0):
"""
This function computes parcel averages and RFX at the parcel-level
Parameters
----------
Pa Parcellation instance that is updated in this function
test_images: double list of paths of functional images used
as input to for inference.
Normally these are contrast images.
double list is
[number of subjects [number of contrasts]]
numbeta: list of int of the associated ids
swd='/tmp': write directory
DMtx=None: array od shape (nsubj,ncon)
a design matrix for second-level analyses
(not implemented yet)
verbose=1: verbosity level
method_id = 0: an id of the method used.
This is useful to compare the outcome of different
Parcellation+RFX procedures
Results
-------
Pa: the updated Parcellation instance
"""
nsubj = Pa.nb_subj
mxyz = Pa.ijk.T
mask = Pa.label>-1
nbeta = len(numbeta)
# 1. read the test data
# fixme: Check that everybody is in the same referential
Test = []
for s in range(nsubj):
beta = []
lxyz = mxyz[:,mask[:,s]]
lxyz = np.array(lxyz)
for b in range(nbeta):
# the raw contrast images
rbeta = load(test_images[s][b])
temp = rbeta.get_data()
temp = temp[lxyz[0,:],lxyz[1,:],lxyz[2,:]]
temp = np.reshape(temp, np.size(temp))
beta.append(temp)
temp[np.isnan(temp)]=0 ##
beta = np.array(beta)
Test.append(beta.T)
# 2. compute the parcel-based stuff
# and make inference inference (RFX,...)
prfx = np.zeros((Pa.k,nbeta))
vinter = np.zeros(nbeta)
for b in range(nbeta):
unitest = [np.reshape(Test[s][:,b],(np.size(Test[s][:,b]),1)) \
for s in range(nsubj)]
cname = 'contrast_%04d'%(numbeta[b])
Pa.make_feature(unitest, cname)
prfx[:,b] = np.reshape(Pa.PRFX(cname,1),Pa.k)
vinter[b] = Pa.variance_inter(cname)
vintra = Pa.variance_intra(Test)
if verbose:
print 'average intra-parcel variance', vintra
print 'average intersubject variance', vinter.mean()
# 3. Write the stuff
# write RFX images
ref_dim = rbeta.get_shape()
affine = rbeta.get_affine()
tlabs = Pa.group_labels
# write the prfx images
for b in range(len(numbeta)):
RfxImage = os.path.join(swd,"prfx_%s_%d.nii" % (numbeta[b],method_id))
if ((verbose)&(np.size(prfx)>0)):
rfx_map = np.zeros(ref_dim)
rfx_map[Pa.ijk[:,0],Pa.ijk[:,1],Pa.ijk[:,2]] = prfx[tlabs,b]
wim = Nifti1Image (rfx_map, affine)
hdr = wim.get_header()
hdr['descrip'] = 'parcel-based eandom effects image (in z-variate)'
save(wim, RfxImage)
return Pa
#########################################
# Estimate the contrasts
#########################################
print 'Computing contrasts...'
for index, contrast_id in enumerate(contrasts):
print ' Contrast % 2i out of %i: %s' % (index+1,
len(contrasts), contrast_id)
lcontrast = my_glm.contrast(contrasts[contrast_id])
#
contrast_path = op.join(swd, '%s_z_map.nii'% contrast_id)
write_array = mask_array.astype(np.float)
write_array[mask_array] = lcontrast.zscore()
contrast_image = Nifti1Image(write_array, fmri_image.get_affine() )
save(contrast_image, contrast_path)
affine = fmri_image.get_affine()
vmax = max(-write_array.min(), write_array.max())
plot_map(write_array, affine,
cmap=cm.cold_hot,
vmin=-vmax,
vmax=vmax,
anat=None,
figure=10,
threshold=2.5)
pylab.savefig(op.join(swd, '%s_z_map.png' % contrast_id))
pylab.clf()
def one_subj_parcellation(MaskImage, betas, nbparcel, nn=6, method='ward',
write_dir=None, mu=10., verbose=0, fullpath=None):
"""
Parcellation of a one-subject dataset
Return: a tuple (Parcellation instance, parcellation labels)
Parameters
----------
MaskImage: path to the mask-defining_image of the subject
betas: list of paths to activation images from the subject
nbparcel, int : number fo desired parcels
nn=6: number of nearest neighbors to define the image topology
(6, 18 or 26)
method='ward': clustering method used, to be chosen among
'ward', 'gkm', 'ward_and-gkm'
'ward': Ward's clustering algorithm
'gkm': Geodesic k-means algorithm, random initialization
'gkm_and_ward': idem, initialized by Ward's clustering
write_dir=None: write directory. If fullpath is None too, then no file output.
mu = 10., float: the relative weight of anatomical information
verbose=0: verbosity mode
fullpath=None, string,
path of the output image
If write_dir and fullpath are None then no file output.
If only fullpath is None then it is the write dir + a name
depending on the method.
Note
----
Ward's method takes time (about 6 minutes for a 60K voxels dataset)
Geodesic k-means is 'quick and dirty'
Ward's + GKM is expensive but quite good
To reduce CPU time, rather use nn=6 (especially with Ward)
"""
import nipy.neurospin.graph as fg
import nipy.neurospin.graph.field as ff
if method not in ['ward','gkm','ward_and_gkm','kmeans']:
raise ValueError, 'unknown method'
if nn not in [6,18,26]:
raise ValueError, 'nn should be 6,18 or 26'
nbeta = len(betas)
# step 1: load the data ----------------------------
#1.1 the mask image
nim = load(MaskImage)
ref_dim = nim.get_shape()
affine = nim.get_affine()
mask = nim.get_data()
xyz = np.array(np.where(mask>0)).T
nvox = xyz.shape[0]
if method is not 'kmeans':
# 1.2 get the main cc of the graph
# to remove the small connected components
g = fg.WeightedGraph(nvox)
g.from_3d_grid(xyz.astype(np.int),nn)
aux = np.zeros(g.V).astype('bool')
imc = g.main_cc()
aux[imc]= True
if np.sum(aux)==0:
raise ValueError, "empty mask. Cannot proceed"
g = g.subgraph(aux)
lmask = np.zeros(ref_dim)
lmask[xyz[:,0],xyz[:,1],xyz[:,2]]=aux
xyz = xyz[aux,:]
nvox = xyz.shape[0]
# 1.3 from vox to mm
xyz2 = np.hstack((xyz,np.ones((nvox,1))))
coord = np.dot(xyz2, affine.T)[:,:3]
# 1.4 read the functional data
beta = []
for b in range(nbeta):
rbeta = load(betas[b])
lbeta = rbeta.get_data()
lbeta = lbeta[lmask>0]
beta.append(lbeta)
beta = np.array(beta).T
#step 2: parcel the data ---------------------------
feature = np.hstack((beta, mu*coord/np.std(coord)))
if method is not 'kmeans':
g = ff.Field(nvox, g.edges, g.weights, feature)
if method=='kmeans':
cent, u, J = kmeans(feature, nbparcel)
if method=='ward':
u, J0 = g.ward(nbparcel)
if method=='gkm':
seeds = np.argsort(np.random.rand(g.V))[:nbparcel]
seeds, u, J1 = g.geodesic_kmeans(seeds)
if method=='ward_and_gkm':
w,J0 = g.ward(nbparcel)
seeds, u, J1 = g.geodesic_kmeans(label=w)
lpa = Parcellation(nbparcel, xyz, np.reshape(u,(nvox,1)))
if verbose:
pi = np.reshape(lpa.population(), nbparcel)
vi = np.sum(lpa.var_feature_intra([beta])[0], 1)
vf = np.dot(pi,vi)/nvox
va = np.dot(pi,np.sum(lpa.var_feature_intra([coord])[0],1))/nvox
print nbparcel, "functional variance", vf, "anatomical variance",va
# step3: write the resulting label image
Label = -np.ones(ref_dim,'int16')
Label[lmask>0] = u
if fullpath is not None:
LabelImage = fullpath
elif write_dir is not None:
if method=='kmeans':
LabelImage = os.path.join(write_dir,"parcel_kmeans.nii")
if method=='ward':
LabelImage = os.path.join(write_dir,"parcel_wards.nii")
elif method=='gkm':
LabelImage = os.path.join(write_dir,"parcel_gkmeans.nii")
elif method=='ward_and_gkm':
LabelImage = os.path.join(write_dir,"parcel_wgkmeans.nii")
else:
LabelImage = None
if LabelImage is not None:
wim = Nifti1Image(Label, affine)
hdr = wim.get_header()
hdr['descrip'] = 'Intra-subject parcellation image'
save(wim, LabelImage)
print "Wrote the parcellation images as %s" %LabelImage
return lpa, Label
""" print __doc__ import numpy as np import os from nipy.io.imageformats import load, save, Nifti1Image from nipy.neurospin.graph.field import Field import get_data_light import tempfile data_dir = get_data_light.get_it() # paths swd = tempfile.mkdtemp() input_image = os.path.join(data_dir, 'spmT_0029.nii.gz') mask_image = os.path.join(data_dir, 'mask.nii.gz') mask = load(mask_image).get_data()>0 ijk = np.array(np.where(mask)).T nvox = ijk.shape[0] data = load(input_image).get_data()[mask] image_field = Field(nvox) image_field.from_3d_grid(ijk, k=6) image_field.set_field(data) u = image_field.ward(100) label_image = os.path.join(swd, 'label.nii') wdata = mask - 1 wdata[mask] = u save(Nifti1Image(wdata, load(mask_image).get_affine()), label_image) print "Label image written in %s" % label_image
### Extract blobs maps as data arrays
blobs_labels = -np.zeros(domain.size)
blobs_means = -np.zeros(domain.size)
if nroi != None:
nroi.make_feature('activation', glm_data.ravel())
bfm = nroi.representative_feature('activation')
for k in range(nroi.k):
blobs_labels[nroi.label == k] = k
blobs_means[nroi.label == k] = bfm[k]
# saving the blobs image, i.e. a label image
label_data = np.zeros(mask_data.shape)
label_data[mask_data != 0] = (blobs_labels + 1)
label_image = Nifti1Image(label_data, mask.get_affine())
label_image.get_header()['descrip'] = 'blob image extracted from %s' \
%glm_data_path
save(label_image, os.path.join(swd,"blob.nii"))
print "Wrote the blobs label image in %s" \
%os.path.join(swd, "blob.nii")
# saving the image of the average-signal-per-blob
avg_data = np.zeros(mask_data.shape)
avg_data[mask_data != 0] = blobs_means
avg_image = Nifti1Image(avg_data, mask.get_affine())
avg_image.get_header()['descrip'] = 'blob average signal extracted from %s' \
%glm_data_path
save(avg_image, os.path.join(swd,"bmap.nii"))
print "Wrote the blobs average signal image in %s" \
%os.path.join(swd, "bmap.nii")
#------------------------------------------------------------
### Extract end-blobs (or leaves) maps as data arrays
leaves_labels = -np.zeros(domain.size)
def compute_mask_files(input_filename, output_filename=None,
return_mean=False, m=0.2, M=0.9, cc=1):
"""
Compute a mask file from fMRI nifti file(s)
Compute and write the mask of an image based on the grey level
This is based on an heuristic proposed by T.Nichols:
find the least dense point of the histogram, between fractions
m and M of the total image histogram.
In case of failure, it is usually advisable to increase m.
Parameters
----------
input_filename : string
nifti filename (4D) or list of filenames (3D).
output_filename : string or None, optional
path to save the output nifti image (if not None).
return_mean : boolean, optional
if True, and output_filename is None, return the mean image also, as
a 3D array (2nd return argument).
m : float, optional
lower fraction of the histogram to be discarded.
M: float, optional
upper fraction of the histogram to be discarded.
cc: boolean, optional
if cc is True, only the largest connect component is kept.
Returns
-------
mask : 3D boolean array
The brain mask
mean_image : 3d ndarray, optional
The main of all the images used to estimate the mask. Only
provided if `return_mean` is True.
"""
if isinstance(input_filename, basestring):
# One single filename
nim, vol_arr = get_unscaled_img(input_filename)
header = nim.get_header()
affine = nim.get_affine()
if vol_arr.ndim == 4:
if isinstance(vol_arr, np.memmap):
# Get rid of memmapping: it is faster.
mean_volume = np.array(vol_arr, copy=True).mean(axis=-1)
else:
mean_volume = vol_arr.mean(axis=-1)
# Make a copy, to avoid holding a reference on the full array,
# and thus polluting the memory.
first_volume = vol_arr[:,:,:,0].copy()
elif vol_arr.ndim == 3:
mean_volume = first_volume = vol_arr
else:
raise ValueError('Need 4D file for mask')
del vol_arr
else:
# List of filenames
if len(input_filename) == 0:
raise ValueError('input_filename should be a non-empty '
'list of file names')
# We have several images, we do mean on the fly,
# to avoid loading all the data in the memory
# We do not use the unscaled data here?:
# if the scalefactor is being used to record real
# differences in intensity over the run this would break
for index, filename in enumerate(input_filename):
nim = load(filename)
if index == 0:
first_volume = nim.get_data().squeeze()
mean_volume = first_volume.copy().astype(np.float32)
header = nim.get_header()
affine = nim.get_affine()
else:
mean_volume += nim.get_data().squeeze()
mean_volume /= float(len(input_filename))
del nim
if np.isnan(mean_volume).any():
tmp = mean_volume.copy()
tmp[np.isnan(tmp)] = 0
mean_volume = tmp
mask = compute_mask(mean_volume, first_volume, m, M, cc)
if output_filename is not None:
header['descrip'] = 'mask'
output_image = nifti1.Nifti1Image(mask.astype(np.uint8),
affine=affine,
header=header)
save(output_image, output_filename)
if not return_mean:
return mask
else:
return mask, mean_volume
