def test_Fiber_unique_coords():
"""
Test class method Fiber.unique_coords
"""
arr = np.array([[1, 1, 1], [2, 2, 2], [3, 3, 3]])
npt.assert_equal(mtf.Fiber(arr).unique_coords, np.array([[1], [2], [3]]))
arr = np.array([[3, 2, 3], [2, 2, 2], [10, 10, 10]])
npt.assert_equal(
mtf.Fiber(arr).unique_coords, np.array([[3, 2], [2, 2], [10, 10]]))
for x in range(1000):
x1 = np.random.randn()
x2 = np.random.randn()
y1 = np.random.randn()
y2 = np.random.randn()
z1 = np.random.randn()
z2 = np.random.randn()
# So the next line isn't too long:
npta = npt.assert_almost_equal
npta(mtf.Fiber([[x1, x2, x1], [y1, y2, y1], [z1, z2,
z1]]).unique_coords,
np.array([[x1, x2], [y1, y2], [z1, z2]]),
decimal=4)
arr2d = np.array([[1, 2, 1], [3, 4, 3], [5, 6, 5]])
f1 = mtf.Fiber(arr2d)
npt.assert_equal(f1.unique_coords, np.array([[1, 2], [3, 4], [5, 6]]))
def test_pdb_from_fg():
"""
Test writing a fiber-group to file
"""
coords1 = np.arange(900).reshape(3, 300)
coords2 = np.arange(900).reshape(3, 300) + 100
fiber_stats = dict(foo=1, bar=2)
node_stats = dict(ecc=np.arange(300))
fg = mtf.FiberGroup([
mtf.Fiber(coords1, fiber_stats=fiber_stats, node_stats=node_stats),
mtf.Fiber(coords2, fiber_stats=fiber_stats, node_stats=node_stats)
])
temp_dir = tempfile.gettempdir()
mio.pdb_from_fg(fg, os.path.join(temp_dir, 'fg.pdb'))
# Test that the properties are preserved upon reloading:
fg2 = mio.fg_from_pdb(os.path.join(temp_dir, 'fg.pdb'))
npt.assert_equal(fg2[0].coords, fg[0].coords)
npt.assert_equal(fg2[1].coords, fg[1].coords)
npt.assert_equal(fg2[0].node_stats, fg[0].node_stats)
npt.assert_equal(fg2[1].node_stats, fg[1].node_stats)
npt.assert_equal(fg2.fiber_stats, fg.fiber_stats)
def test_Fiber_xform():
arr2d = np.array([[1, 2], [3, 4], [5, 6]])
affine1 = np.eye(4)
f1 = mtf.Fiber(arr2d, affine=affine1)
f1.xform()
npt.assert_equal(f1.coords, arr2d)
f2 = f1.xform(inplace=False)
npt.assert_equal(f2.coords, f1.coords)
f1_noaff = mtf.Fiber(arr2d, affine=None)
f1_noaff_notinplace = f1_noaff.xform(affine=None, inplace=False)
# Should give you back a different object:
npt.assert_equal(not f1_noaff_notinplace is f1_noaff, True)
# But with equal coords:
npt.assert_equal(f1_noaff_notinplace.coords, f1_noaff.coords)
# Keep everything the same, but translate the x coords by 1 downwards
# http://en.wikipedia.org/wiki/Transformation_matrix#Affine_transformations:
affine2 = np.matrix([[1, 0, 0, -1], [0, 1, 0, 0], [0, 0, 1, 0],
[0, 0, 0, 1]])
f3 = mtf.Fiber(arr2d, affine=affine2)
f3.xform()
npt.assert_equal(f3.coords[0], arr2d[0] - 1)
# This one rotates about the x axis by 90 degrees:
pi_2 = np.pi / 2
affine3 = np.matrix([[1, 0, 0, 0], [0, np.cos(pi_2), -np.sin(pi_2), 0],
[0, np.sin(pi_2), np.cos(pi_2), 0], [0, 0, 0, 1]])
f4 = mtf.Fiber([0, 1, 0], affine=affine3)
f4.xform()
# Rotating about the x axis should move all of the length of the vector to
# the z axis
npt.assert_almost_equal(f4.coords, [0, 0, 1])
# Next time you call xform should bring you back to where you were to begin
# with
f4.xform()
npt.assert_almost_equal(f4.coords, [0, 1, 0])
# If you assign into a new fiber:
f5 = f4.xform(inplace=False)
# This one should have an affine which is the inverse of your original
# affine:
npt.assert_equal(f5.affine, affine3.getI())
# xform-ing twice gives you back the same thing:
npt.assert_equal(
f5.xform(inplace=False).xform(inplace=False).coords, f5.coords)
npt.assert_equal(
f5.xform(inplace=False).xform(inplace=False).affine, f5.affine)
def test_FiberGroup():
"""
Testing intialization of FiberGroup class.
"""
arr2d = np.array([[1, 2], [3, 4], [5, 6]])
arr1d = np.array([5, 6, 7])
f1 = mtf.Fiber(arr2d, fiber_stats=dict(a=1, b=2))
f2 = mtf.Fiber(arr1d, fiber_stats=dict(a=1))
fg1 = mtf.FiberGroup([f1, f2])
npt.assert_equal(fg1.n_fibers, 2)
# We have to sort, because it could also come out as ['b', 'a']:
npt.assert_equal(np.sort(fg1.fiber_stats.keys()), ['a', 'b'])
# The number of nodes is just the sum of nodes/fiber:
npt.assert_equal(fg1.n_nodes, f1.n_nodes + f2.n_nodes)
def fg_from_trk(trk_file, affine=None):
"""
Read data from a trackvis .trk file and create a FiberGroup object
according to the information in it.
"""
# Generate right away, since we're going to do it anyway:
read_trk = tv.read(trk_file, as_generator=False)
fibers_trk = read_trk[0]
# Per default read from the affine from the file header:
if affine is not None:
aff = affine
else:
hdr = read_trk[1]
aff = tv.aff_from_hdr(hdr)
# If the header contains a bogus affine, we revert to np.eye(4), so we
# don't get into trouble later:
try:
np.matrix(aff).getI()
except np.linalg.LinAlgError:
e_s = "trk file contains bogus header, reverting to np.eye(4)"
warnings.warn(e_s)
aff = np.eye(4)
fibers = []
for f in fibers_trk:
fibers.append(ozf.Fiber(np.array(f[0]).T, affine=aff))
return ozf.FiberGroup(fibers, affine=aff)
def test_FiberGroup_xform():
"""
Test affine transformation method of FiberGroup
"""
# This affine rotates vectors 90 degrees around the x axis:
pi_2 = np.pi / 2
affine1 = np.matrix([[1, 0, 0, 0], [0, np.cos(pi_2), -np.sin(pi_2), 0],
[0, np.sin(pi_2), np.cos(pi_2), 0], [0, 0, 0, 1]])
y = [0, 0, 1]
x = [0, 1, 0]
f1 = mtf.Fiber(x, affine=affine1)
f2 = mtf.Fiber(y, affine=affine1)
fg1 = mtf.FiberGroup([f1, f2])
fg1.xform()
# The first fiber's coordinates should be like the second one originally:
npt.assert_almost_equal(fg1.fibers[0].coords, y)
f3 = mtf.Fiber(x)
f4 = mtf.Fiber(y)
fg2 = mtf.FiberGroup([f3, f4], affine=affine1)
fg2.xform()
# Same should be true when the affine is associated with the FiberGroup:
npt.assert_almost_equal(fg2.fibers[0].coords, y)
# And the transformtation should have mutated the original object:
npt.assert_almost_equal(f3.coords, y)
f5 = mtf.Fiber(x)
f6 = mtf.Fiber(y)
fg3 = mtf.FiberGroup([f5, f6])
fg3.xform(affine1)
# Same should be true when the affine is provided as input:
npt.assert_almost_equal(fg3.fibers[0].coords, y)
# And the transformtation should have mutated the original object:
npt.assert_almost_equal(f5.coords, y)
# This also attaches the inverse of this affine to the original object, so
# that you can always find your way back:
npt.assert_almost_equal(f5.affine, affine1.getI())
f7 = mtf.Fiber(x)
f8 = mtf.Fiber(y)
fg4 = mtf.FiberGroup([f7, f8])
fg4.xform()
npt.assert_equal(fg4.affine, None)
# The affine should 'stick':
fg4.xform(np.eye(4))
npt.assert_equal(fg4.affine, np.eye(4))
# Even to the fibers:
npt.assert_equal(f8.affine, np.eye(4))
def test_FiberGroup_unique_coords():
"""
Test class method Fiber.unique_coords
"""
for x in range(1000):
x1 = np.random.randn()
x2 = np.random.randn()
y1 = np.random.randn()
y2 = np.random.randn()
z1 = np.random.randn()
z2 = np.random.randn()
# So the next line isn't too long:
npta = npt.assert_almost_equal
# Should work if both fibers have non-unique coords
npta(mtf.FiberGroup([
mtf.Fiber([[x1, x1, x2], [y1, y1, y2], [z1, z1, z2]]),
mtf.Fiber([[x1, x1, x2], [y1, y1, y2], [z1, z1, z2]])
]).unique_coords,
np.array([[x1, x2], [y1, y2], [z1, z2]]),
decimal=4)
# And also for extracting across fibers with unique coords
npta(mtf.FiberGroup(
[mtf.Fiber([[x1], [y1], [z1]]),
mtf.Fiber([[x2], [y2], [z2]])]).unique_coords,
np.array([[x1, x2], [y1, y2], [z1, z2]]),
decimal=4)
# And also for extracting across shared coords
npta(mtf.FiberGroup([
mtf.Fiber([[x1], [y1], [z1]]),
mtf.Fiber([[x2, x1], [y2, y1], [z2, z1]])
]).unique_coords,
np.array([[x1, x2], [y1, y2], [z1, z2]]),
decimal=4)
def test_Fiber():
"""
Testing initalization of the Fiber class
"""
# Providing a list as an input works:
arr1d = [1, 2, 3]
# This is the most basic example possible:
f1 = mtf.Fiber(arr1d)
# 2D arrays should be 3 by n:
arr2d = np.array([[1, 2], [3, 4], [5, 6]])
# So this is OK:
f2 = mtf.Fiber(arr2d)
# But this raises a ValueError:
npt.assert_raises(ValueError, mtf.Fiber, arr2d.T)
# This should also raise (first dim is 4, rather than 3):
npt.assert_raises(ValueError, mtf.Fiber, np.empty((4, 10)))
# This should also raise (funky affine):
npt.assert_raises(ValueError, mtf.Fiber, np.empty((3, 10)), np.eye(5))
# This should be OK:
f3 = mtf.Fiber(np.array(arr2d), affine=np.eye(4), fiber_stats=dict(a=1))
npt.assert_equal(f3.fiber_stats, {'a': 1})
def test_Fiber_tensors():
"""
Test generation of tensors from fiber coordinates
"""
bvecs = [[1, 0, 0], [0, 1, 0], [0, 0, 1]]
bvals = [1, 1, 1]
f1 = mtf.Fiber([[2, 2, 3], [3, 3, 4], [4, 4, 5]])
# Values for axial and radial diffusivity randomly chosen:
ad = np.random.rand()
rd = np.random.rand()
tensors = f1.tensors(ad, rd)
npt.assert_equal(tensors[0], np.diag([ad, rd, rd]).ravel())
npt.assert_equal(len(tensors), len(f1.coords))
def test_Fiber_predicted_signal():
"""
Test fiber prediction of the signal along its coordinates
"""
f1 = mtf.Fiber([[2, 2, 3, 5], [3, 3, 4, 6], [4, 4, 5, 7]])
bvecs = [[1, 0, 0], [0, 1, 0], [0, 0, 1]]
bvals = [1, 1, 1]
ad = np.random.rand()
rd = np.random.rand()
sig = f1.predicted_signal(bvecs, bvals, ad, rd)
# Or just one number:
S0 = np.random.rand()
sig = f1.predicted_signal(bvecs, bvals, ad, rd)
def track(model,
data,
sphere=None,
step_size=1,
angle_limit=20,
seeds=None,
density=[2, 2, 2],
voxel_size=[1, 1, 1]):
"""
Interface for tracking based on fiber ODF models
`model` needs to have a `fit` method, such that model.fit(data).odf(sphere)
is a legitimate ODF (that is has dimensions (x,y,z, n_vertices), where
n_vertices refers to the vertices of the provided sphere.
"""
# If no sphere is provided, we will use the dipy symmetrical sphere with
# 724 vertcies. That should be enough
if sphere is None:
sphere = dpd.get_sphere('symmetric724')
stepper = dpt.FixedSizeStepper(step_size)
interpolator = dpt.NearestNeighborInterpolator(data, voxel_size)
if seeds is None:
seeds = dpu.seeds_from_mask(mask, density, voxel_size)
pwt = dpt.ProbabilisticOdfWeightedTracker(model, interpolator, mask,
stepper, angle_limit, seeds,
sphere)
pwt_streamlines = list(pwt)
fibers = []
for f in pwt_streamlines:
fibers.append(ozf.Fiber(f))
def fg_from_pdb(file_name, verbose=True):
"""
Read the definition of a fiber-group from a .pdb file
Parameters
----------
file_name: str
Full path to the .pdb file
Returns
-------
A FiberGroup object
Note
----
This only reads Version 3 PDB. For the full file-format spec, see the
osmosis.io module top-level docstring
"""
# Read the file as binary info:
f_obj = file(file_name, 'r')
f_read = f_obj.read()
f_obj.close()
# This is an updatable index into this read:
idx = 0
# First part is an int encoding the offset to the fiber part:
offset, idx = _unpacker(f_read, idx, 1)
# Next bit are doubles, encoding the xform (4 by 4 = 16 of them):
xform, idx = _unpacker(f_read, idx, 16, 'double')
xform = np.reshape(xform, (4, 4))
# Next is an int encoding the number of stats:
numstats, idx = _unpacker(f_read, idx, 1)
# The stats header is a dict with lists holding the stat per
stats_header = dict(
luminance_encoding=[], # int => bool
computed_per_point=[], # int => bool
viewable=[], # int => bool
agg_name=[], # char array => string
local_name=[], # char array => string
uid=[] # int
)
# Read the stats header:
counter = 0
while counter < numstats:
counter += 1
for k in ["luminance_encoding", "computed_per_point", "viewable"]:
this, idx = _unpacker(f_read, idx, 1)
stats_header[k].append(np.bool(this))
for k in ["agg_name", "local_name"]:
this, idx = _unpacker(f_read, idx, 255, 'char')
stats_header[k].append(_word_maker(this))
# Must have integer reads be word aligned (?):
idx += 2
this, idx = _unpacker(f_read, idx, 1)
stats_header["uid"].append(this)
# We skip the whole bit with the algorithms and go straight to the version
# number, which is one int length before the fibers:
idx = offset - 4
version, idx = _unpacker(f_read, idx, 1)
if int(version) < 2:
raise ValueError("Can only read PDB version 2 or version 3 files")
elif verbose:
print("Loading a PDB version %s file from: %s" %
(int(version), file_name))
if int(version) == 2:
idx = offset
# How many fibers?
numpaths, idx = _unpacker(f_read, idx, 1)
if int(version) == 2:
pts = []
if verbose:
prog_bar = ProgressBar(numpaths[0])
f_name = inspect.stack()[0][3]
f_stats = []
n_stats = []
for p_idx in range(numpaths):
f_stats_dict = {}
n_stats_dict = {}
# Keep track of where you are right now
ppos = idx
path_offset, idx = _unpacker(f_read, idx, 1)
n_nodes, idx = _unpacker(f_read, idx, 1)
# As far as I can tell the following two don't matter much:
algo_type, idx = _unpacker(f_read, idx, 1)
seed_pt_idx, idx = _unpacker(f_read, idx, 1)
# Read out the per-path stats:
for stat_idx in range(numstats):
per_fiber_stat, idx = _unpacker(f_read, idx, 1, 'double')
f_stats_dict[stats_header["local_name"][stat_idx]] = \
per_fiber_stat
f_stats.append(f_stats_dict)
# Skip forward to where the paths themselves are:
idx = ppos
# Read the nodes:
pathways, idx = _unpacker(f_read, idx, n_nodes * 3, 'double')
pts.append(np.reshape(pathways, (n_nodes, 3)).T)
for stat_idx in range(numstats):
if stats_header["computed_per_point"][stat_idx]:
name = stats_header["local_name"][stat_idx]
n_stats_dict[name], idx = _unpacker(
f_read, idx, n_nodes, 'double')
n_stats.append(n_stats_dict)
fibers = []
# Initialize all the fibers:
for p_idx in range(numpaths):
this_fstats_dict = f_stats[p_idx]
f_stat_k = this_fstats_dict.keys()
f_stat_v = [this_fstats_dict[k] for k in f_stat_k]
this_nstats_dict = n_stats[p_idx]
n_stats_k = this_nstats_dict.keys()
n_stats_v = [this_nstats_dict[k] for k in n_stats_k]
fibers.append(
ozf.Fiber(pts[p_idx],
xform,
fiber_stats=dict(zip(f_stat_k, f_stat_v)),
node_stats=dict(zip(n_stats_k, n_stats_v))))
elif int(version) == 3:
# The next few bytes encode the number of points in each fiber:
pts_per_fiber, idx = _unpacker(f_read, idx, numpaths)
total_pts = np.sum(pts_per_fiber)
# Next we have the xyz coords of the nodes in all fibers:
fiber_pts, idx = _unpacker(f_read, idx, total_pts * 3, 'double')
# We extract the information on a fiber-by-fiber basis
pts_read = 0
pts = []
if verbose:
prog_bar = ProgressBar(numpaths[0])
f_name = inspect.stack()[0][3]
for p_idx in range(numpaths):
n_nodes = pts_per_fiber[p_idx]
pts.append(
np.reshape(fiber_pts[pts_read * 3:(pts_read + n_nodes) * 3],
(n_nodes, 3)).T)
pts_read += n_nodes
if verbose:
prog_bar.animate(p_idx, f_name=f_name)
f_stats_dict = {}
for stat_idx in range(numstats):
per_fiber_stat, idx = _unpacker(f_read, idx, numpaths, 'double')
# This is a fiber-stat only if it's not computed per point:
if not stats_header["computed_per_point"][stat_idx]:
f_stats_dict[stats_header["local_name"][stat_idx]] =\
per_fiber_stat
per_point_stat = []
n_stats_dict = {}
for stat_idx in range(numstats):
pts_read = 0
# If it is computer per point, it's a node-stat:
if stats_header["computed_per_point"][stat_idx]:
name = stats_header["local_name"][stat_idx]
n_stats_dict[name] = []
per_point_stat, idx = _unpacker(f_read, idx, total_pts,
'double')
for p_idx in range(numpaths):
n_stats_dict[name].append(
per_point_stat[pts_read:pts_read +
pts_per_fiber[p_idx]])
pts_read += pts_per_fiber[p_idx]
else:
per_point_stat.append([])
fibers = []
# Initialize all the fibers:
for p_idx in range(numpaths):
f_stat_k = f_stats_dict.keys()
f_stat_v = [f_stats_dict[k][p_idx] for k in f_stat_k]
n_stats_k = n_stats_dict.keys()
n_stats_v = [n_stats_dict[k][p_idx] for k in n_stats_k]
fibers.append(
ozf.Fiber(pts[p_idx],
xform,
fiber_stats=dict(zip(f_stat_k, f_stat_v)),
node_stats=dict(zip(n_stats_k, n_stats_v))))
if verbose:
print("Done reading from file")
name = os.path.split(file_name)[-1].split('.')[0]
return ozf.FiberGroup(fibers, name=name, affine=xform)