def __init__(self, image, bbox=None, mask=False,
grid_spacing=None, interpolation_order=3):
"""
Creates a new SampledVolumeSlicer
Parameters
----------
image : a NIPY Image
The image to slice
bbox : iterable (optional)
The {x,y,z} limits of the enrounding volume box. If None, then
slices planes in the natural box of the image. This argument
is useful for overlaying an image onto another image's volume box
mask : bool or ndarray (optional)
A binary mask, with same shape as image, with unmasked points
marked as True (opposite of MaskedArray convention)
grid_spacing : iterable (optional)
New grid spacing for the sliced planes. If None, then the
natural voxel spacing is used.
"""
xyz_image = ni_api.Image(
np.asarray(image),
image.coordmap.reordered_range(xipy_ras)
)
## reorder_output(image.coordmap, 'xyz'))
self.coordmap = xyz_image.coordmap
self.raw_image = xyz_image
nvox = np.product(self.raw_image.shape)
# if the volume array of the map is more than 100mb in memory,
# better use a memmap
self._use_mmap = nvox*8 > 100e6
self.interpolator = ImageInterpolator(xyz_image,
order=interpolation_order,
use_mmap=self._use_mmap)
## if mask is True:
## mask = compute_mask(np.asarray(self.raw_image), cc=0, m=.1, M=.999)
if type(mask) is np.ndarray:
self.update_mask(mask, positive_mask=True)
else:
self._masking = False
self.raw_mask = None
if grid_spacing is None:
self.grid_spacing = list(vu.voxel_size(image.affine))
else:
self.grid_spacing = grid_spacing
if bbox is None:
self._nominal_bbox = vu.world_limits(xyz_image)
else:
self._nominal_bbox = bbox
# defines {x,y,z}shape and {x,y,z}grid
self._define_grids()
def update_mask(self, mask, positive_mask=True):
cmap = self.coordmap
if not positive_mask:
# IE, if this is a MaskedArray type mask
mask = np.logical_not(mask)
## fmask = np.array([ndimage.binary_fill_holes(m) for m in mask], 'd')
self.m_interpolator = ImageInterpolator(ni_api.Image(mask, cmap),
order=3,
use_mmap=self._use_mmap)
self.raw_mask = mask
self._masking = True
def resample(image, target, mapping, shape, order=3, **interp_kws):
"""
Resample an image to a target CoordinateMap with a "world-to-world" mapping
and spline interpolation of a given order.
Here, "world-to-world" refers to the fact that mapping should be
a callable that takes a physical coordinate in "target"
and gives a physical coordinate in "image".
Parameters
----------
image : Image instance that is to be resampled
target :target CoordinateMap for output image
mapping : transformation from target.function_range
to image.coordmap.function_range, i.e. 'world-to-world mapping'
Can be specified in three ways: a callable, a
tuple (A, b) representing the mapping y=dot(A,x)+b
or a representation of this in homogeneous coordinates.
shape : shape of output array, in target.function_domain
order : what order of interpolation to use in `scipy.ndimage`
interp_kws : keyword arguments for ndimage interpolator routine
Returns
-------
output : Image instance with interpolated data and output.coordmap == target
"""
if not callable(mapping):
if type(mapping) is type(()):
A, b = mapping
ndimout = b.shape[0]
ndimin = A.shape[1]
mapping = np.zeros((ndimout+1, ndimin+1))
mapping[:ndimout,:ndimin] = A
mapping[:ndimout,-1] = b
mapping[-1,-1] = 1.
# image world to target world mapping
TW2IW = AffineTransform(target.function_range, image.coordmap.function_range, mapping)
else:
TW2IW = CoordinateMap(mapping, target.function_range, image.coordmap.function_range)
function_domain = target.function_domain
function_range = image.coordmap.function_range
# target voxel to image world mapping
TV2IW = compose(TW2IW, target)
# CoordinateMap describing mapping from target voxel to
# image world coordinates
if not isinstance(TV2IW, AffineTransform):
# interpolator evaluates image at values image.coordmap.function_range,
# i.e. physical coordinates rather than voxel coordinates
grid = ArrayCoordMap.from_shape(TV2IW, shape)
interp = ImageInterpolator(image, order=order)
idata = interp.evaluate(grid.transposed_values, **interp_kws)
del(interp)
else:
TV2IV = compose(image.coordmap.inverse(), TV2IW)
if isinstance(TV2IV, AffineTransform):
A, b = affines.to_matrix_vector(TV2IV.affine)
data = np.asarray(image)
idata = affine_transform(data, A,
offset=b,
output_shape=shape,
output=data.dtype,
order=order,
**interp_kws)
else:
interp = ImageInterpolator(image, order=order)
grid = ArrayCoordMap.from_shape(TV2IV, shape)
idata = interp.evaluate(grid.values, **interp_kws)
del(interp)
return Image(idata, target)
class SampledVolumeSlicer(VolumeSlicerInterface):
"""
This object cuts up an image along the axes defined by its
CoordinateMap target space. The SampledVolumeSlicer provides slices
through an image such that the cut planes extend across the three
{x,y,z} planes of the target space. Each plane is sampled from the
original image voxel array by spline interpolation.
"""
def __init__(self, image, bbox=None, mask=False,
grid_spacing=None, interpolation_order=3):
"""
Creates a new SampledVolumeSlicer
Parameters
----------
image : a NIPY Image
The image to slice
bbox : iterable (optional)
The {x,y,z} limits of the enrounding volume box. If None, then
slices planes in the natural box of the image. This argument
is useful for overlaying an image onto another image's volume box
mask : bool or ndarray (optional)
A binary mask, with same shape as image, with unmasked points
marked as True (opposite of MaskedArray convention)
grid_spacing : iterable (optional)
New grid spacing for the sliced planes. If None, then the
natural voxel spacing is used.
"""
xyz_image = ni_api.Image(
np.asarray(image),
image.coordmap.reordered_range(xipy_ras)
)
## reorder_output(image.coordmap, 'xyz'))
self.coordmap = xyz_image.coordmap
self.raw_image = xyz_image
nvox = np.product(self.raw_image.shape)
# if the volume array of the map is more than 100mb in memory,
# better use a memmap
self._use_mmap = nvox*8 > 100e6
self.interpolator = ImageInterpolator(xyz_image,
order=interpolation_order,
use_mmap=self._use_mmap)
## if mask is True:
## mask = compute_mask(np.asarray(self.raw_image), cc=0, m=.1, M=.999)
if type(mask) is np.ndarray:
self.update_mask(mask, positive_mask=True)
else:
self._masking = False
self.raw_mask = None
if grid_spacing is None:
self.grid_spacing = list(vu.voxel_size(image.affine))
else:
self.grid_spacing = grid_spacing
if bbox is None:
self._nominal_bbox = vu.world_limits(xyz_image)
else:
self._nominal_bbox = bbox
# defines {x,y,z}shape and {x,y,z}grid
self._define_grids()
def _define_grids(self):
dx, dy, dz = self.grid_spacing
xbox, ybox, zbox = map( lambda x: np.array(x).reshape(2,2),
vu.limits_to_extents(self._nominal_bbox) )
# xbox is [ <ylims> , <zlims> ], so xgrid is [ypts.T, zpts.T]
xr = np.diff(xbox)[:,0]
xspacing = np.array([dy, dz])
self.xshape = (xr / xspacing).astype('i')
aff = ni_api.AffineTransform.from_start_step('ij', 'xy',
xbox[:,0], xspacing)
self.xgrid = ni_api.ArrayCoordMap(aff, self.xshape).values
# ybox is [ <xlims>, <zlims> ], so ygrid is [xpts.T, zpts.T]
yr = np.diff(ybox)[:,0]
yspacing = np.array([dx, dz])
self.yshape = (yr / yspacing).astype('i')
aff = ni_api.AffineTransform.from_start_step('ij', 'xy',
ybox[:,0], yspacing)
self.ygrid = ni_api.ArrayCoordMap(aff, self.yshape).values
# zbox is [ <xlims>, <ylims> ], so zgrid is [xpts.T, ypts.T]
zr = np.diff(zbox)[:,0]
zspacing = np.array([dx, dy])
self.zshape = (zr / zspacing).astype('i')
aff = ni_api.AffineTransform.from_start_step('ij', 'xy',
zbox[:,0], zspacing)
self.zgrid = ni_api.ArrayCoordMap(aff, self.zshape).values
bb_min = [b[0] for b in self._nominal_bbox]
lx = self.yshape[0]
ly = self.xshape[0]
lz = self.yshape[1]
bb_max = [bb_min[0] + dx*lx, bb_min[1] + dy*ly, bb_min[2] + dz*lz]
self.bbox = zip(bb_min, bb_max)
def update_mask(self, mask, positive_mask=True):
cmap = self.coordmap
if not positive_mask:
# IE, if this is a MaskedArray type mask
mask = np.logical_not(mask)
## fmask = np.array([ndimage.binary_fill_holes(m) for m in mask], 'd')
self.m_interpolator = ImageInterpolator(ni_api.Image(mask, cmap),
order=3,
use_mmap=self._use_mmap)
self.raw_mask = mask
self._masking = True
def _update_grid_spacing(self, grid_spacing):
self.grid_spacing = grid_spacing
self._define_grids()
## def update_mask_crit(self, crit, thresh):
## self._masking = True
## self._mcrit = crit
## self._thresh = thresh
def _closest_grid_pt(self, coord, ax):
"""For a given coordinate value on an axis, find the closest
fixed grid point as defined on the complementary grid planes.
This is to ensure consistency at the intersection of
interpolated planes
"""
ax_min = np.asarray(self.bbox)[ax,0]
ax_step = self.grid_spacing[ax]
# ax_min + n*ax_step = coord
new_coord = round ( (coord-ax_min) / ax_step ) * ax_step + ax_min
return new_coord
def _cut_plane(self, ax, coord, **interp_kw):
"""
For a given axis in {SAG, COR, AXI}, make a plane cut in the
volume at the coordinate value.
Parameters
----------
ax : int
axis label in {SAG, COR, AXI} (defined in xipy.slicing)
coord : float
coordinate value along this axis
interp_kw : dict
Keyword args for the interpolating machinery
(ie, ndimage.map_coordinates keyword args)
Returns
_______
plane : ndarray
The transverse plane sampled at the grid points and fixed axis
coordinate for the given args
"""
# a little hokey
grid_lookup = {SAG: ('xshape', 'xgrid'),
COR: ('yshape', 'ygrid'),
AXI: ('zshape', 'zgrid')}
sname, gname= grid_lookup[ax]
ii, jj = transverse_plane_lookup(ax)
grid = getattr(self, gname)
shape = getattr(self, sname)
coords = np.empty((3, grid.shape[0]), 'd')
coords[ax] = self._closest_grid_pt(coord, ax)
coords[ii] = grid[:,0]; coords[jj] = grid[:,1]
pln = self.interpolator.evaluate(coords, **interp_kw).reshape(shape)
if self._masking:
m_pln = self.m_interpolator.evaluate(coords, mode='constant',
cval=-10).reshape(shape)
return np.ma.masked_where(m_pln < 0.5, pln)
return pln
def update_target_space(self, coreg_image):
raise NotImplementedError('not sure how to do this yet')
