def get_intersection_volume(roilist, xvoxel=1., yvoxel=1.):
# There is probably a clever way to compute this by constructing
# an "intersection contour" for each layer: for each contour, keep only
# points that are inside all other contours in the list. But is tough to then
# put those points in the right order.
# Instead we'll just make a grid of points and get the volume of the combined mask.
# With xvoxel and yvoxel the caller can tweak the voxel size of the mask in x and y.
# In z the voxel size is given by the incoming ROIs.
dz = min([r.dz for r in roilist])
assert (dz > 0)
assert (xvoxel > 0)
assert (yvoxel > 0)
bb = bounding_box(bb=roilist[0].bb)
for roi in roilist[1:]:
bb.intersect(roi.bb)
if bb.empty:
# too bad
return 0.
spacing = np.array([xvoxel, yvoxel, dz], dtype=float)
bb.add_margins(2 * spacing)
dimsize = np.array(np.round((bb.maxcorner - bb.mincorner) / spacing),
dtype=int)
#img = sitk.Image(dimsize,sitk.sitkUInt8)
img = itk.image_from_array(np.zeros(dimsize[::-2], dtype=np.uint8))
img.SetOrigin(bb.mincorner)
img.SetSpacing(spacing)
itkmask = itk.array_from_image(roilist[0].get_mask(img))
for roi in roilist[1:]:
itkmask *= itk.array_from_image(roi.get_mask(img))
return np.sum(itkmask) * np.prod(spacing)
def test_three_float_3D_images(self):
logger.info('Test_Product test_three_float_3D_images')
nx, ny, nz = 2, 3, 4
minlog, maxlog = -5., 5.
spacing = (421., 214., 142.)
origin = (421421., 214214., 142142.)
thirteen = 13.333
imglistF = [
itk.image_from_array(
np.logspace(minlog, maxlog,
nx * ny * nz).reshape(nx, ny,
nz).astype(np.float32)),
itk.image_from_array(
np.logspace(minlog, maxlog, nx * ny * nz)[::-1].reshape(
nx, ny, nz).astype(np.float32)),
itk.image_from_array(thirteen * np.ones(
(nx, ny, nz), dtype=np.float32))
]
for imgF in imglistF:
imgF.SetSpacing(spacing)
imgF.SetOrigin(origin)
imgprodF = image_product(input_list=imglistF)
self.assertTrue(np.allclose(itk.array_from_image(imgprodF), thirteen))
self.assertTrue(itk.array_from_image(imgprodF).shape == (nx, ny, nz))
self.assertTrue(np.allclose(imgprodF.GetSpacing(), spacing))
self.assertTrue(np.allclose(imgprodF.GetOrigin(), origin))
self.assertTrue(type(imgprodF) == itk.Image[itk.F, 3])
def faf_ACGM_image(gm, acf, factor=4.168696975):
if gm.GetImageDimension() != 2:
print("gm image dimension (" + str(gm.GetImageDimension()) +
") is not 2")
sys.exit(1)
if acf.GetImageDimension() != 2:
print("acf image dimension (" + str(acf.GetImageDimension()) +
") is not 2")
sys.exit(1)
resampleACF = gt.applyTransformation(input=acf,
like=gm,
force_resample=True,
pad=-1)
resampleACFArray = itk.array_from_image(resampleACF)
gmArray = itk.array_from_image(gm)
acgmArray = np.zeros(gmArray.shape)
negativeACFIndex = np.where(resampleACFArray == -1)
positiveACFIndex = np.where(resampleACFArray != -1)
acgmArray[negativeACFIndex] = factor * gmArray[negativeACFIndex]
acgmArray[positiveACFIndex] = resampleACFArray[positiveACFIndex] * gmArray[
positiveACFIndex]
acgmImage = itk.image_from_array(acgmArray)
acgmImage.CopyInformation(gm)
return acgmImage
def test_itk_registration(self):
import os
os.environ["FOOTSTEPS_NAME"] = "test"
import footsteps
icon_registration.test_utils.download_test_data()
model = icon_registration.pretrained_models.OAI_knees_registration_model(
pretrained=True)
image_A = itk.imread(
str(icon_registration.test_utils.TEST_DATA_DIR /
"knees_diverse_sizes" /
#"9126260_20060921_SAG_3D_DESS_LEFT_11309302_image.nii.gz")
"9487462_20081003_SAG_3D_DESS_RIGHT_11495603_image.nii.gz"))
image_B = itk.imread(
str(icon_registration.test_utils.TEST_DATA_DIR /
"knees_diverse_sizes" /
"9225063_20090413_SAG_3D_DESS_RIGHT_12784112_image.nii.gz"))
print(image_A.GetLargestPossibleRegion().GetSize())
print(image_B.GetLargestPossibleRegion().GetSize())
print(image_A.GetSpacing())
print(image_B.GetSpacing())
phi_AB, phi_BA = icon_registration.itk_wrapper.register_pair(
model, image_A, image_B)
assert (isinstance(phi_AB, itk.CompositeTransform))
interpolator = itk.LinearInterpolateImageFunction.New(image_A)
warped_image_A = itk.resample_image_filter(
image_A,
transform=phi_AB,
interpolator=interpolator,
size=itk.size(image_B),
output_spacing=itk.spacing(image_B),
output_direction=image_B.GetDirection(),
output_origin=image_B.GetOrigin())
plt.imshow(
np.array(itk.checker_board_image_filter(warped_image_A,
image_B))[40])
plt.colorbar()
plt.savefig(footsteps.output_dir + "grid.png")
plt.clf()
plt.imshow(np.array(warped_image_A)[40])
plt.savefig(footsteps.output_dir + "warped.png")
plt.clf()
reference = np.load(icon_registration.test_utils.TEST_DATA_DIR /
"warped.npy")
np.save(footsteps.output_dir + "warped.npy",
itk.array_from_image(warped_image_A)[40])
self.assertLess(
np.mean(
np.abs(reference - itk.array_from_image(warped_image_A)[40])),
1e-6)
def test_eight_3D_images(self):
logger.info('Test_MinMax test_eight_3D_images')
nx,ny,nz = 30,40,50
dmin,dmax = np.float32(-20.5), np.float32(31230.5)
spacing = (321.,213.,132.)
origin = (321321.,213213.,132132.)
alist = [ np.random.uniform(dmin,dmax,nx*ny*nz).astype(np.float32) for i in range(8)]
indices = np.arange(nx*ny*nz,dtype=np.uint32)
imglist=list()
for (j,a) in enumerate(alist):
a[indices%8 == j] = dmin
a[indices%8 == (j+4)%8] = dmax
img=itk.image_from_array(a.reshape(nx,ny,nz).copy())
img.SetSpacing(spacing)
img.SetOrigin(origin)
imglist.append(img)
imgmin = image_min(imglist)
imgmax = image_max(imglist)
self.assertTrue( type(imgmin) == itk.Image[itk.F,3])
self.assertTrue( type(imgmax) == itk.Image[itk.F,3])
self.assertTrue( np.allclose(itk.array_from_image(imgmin),dmin))
self.assertTrue( np.allclose(itk.array_from_image(imgmax),dmax))
self.assertTrue( np.allclose(imgmin.GetSpacing(),spacing))
self.assertTrue( np.allclose(imgmax.GetSpacing(),spacing))
self.assertTrue( np.allclose(imgmin.GetOrigin(),origin))
self.assertTrue( np.allclose(imgmax.GetOrigin(),origin))
def dilat(input, output):
'''
Doc todo
'''
print(input)
ctvi = itk.imread(input)
ImageType = type(ctvi)
Dimension = ctvi.GetImageDimension()
radiusValue = 1
StructuringElementType = itk.FlatStructuringElement[Dimension]
structuringElement = StructuringElementType.Ball(radiusValue)
grayscaleFilter = itk.GrayscaleDilateImageFilter[
ImageType, ImageType, StructuringElementType].New()
grayscaleFilter.SetInput(ctvi)
grayscaleFilter.SetKernel(structuringElement)
grayscaleFilter.Update()
o = grayscaleFilter.GetOutput()
o = itk.array_from_image(o)
c = itk.array_from_image(ctvi)
o[c > 0.0] = 0
o = c + o
o = itk.image_from_array(o)
o.CopyInformation(ctvi)
itk.imwrite(o, output)
def test_five_int_3D_images(self):
logger.info('Test_Product test_five_int_3D_images')
nx,ny,nz = 30,40,50
spacing = (321.,213.,132.)
origin = (321321.,213213.,132132.)
pval=np.ones(5)/5.0
p2=np.random.multinomial(1,pval,(nz,nx,ny)).swapaxes(0,3).copy()
p3=np.random.multinomial(2,pval,(nz,nx,ny)).swapaxes(0,3).copy()
p7=np.random.multinomial(2,pval,(nz,nx,ny)).swapaxes(0,3).copy()
p13=np.random.multinomial(1,pval,(nz,nx,ny)).swapaxes(0,3).copy()
p37=np.random.multinomial(1,pval,(nz,nx,ny)).swapaxes(0,3).copy()
a0,a1,a2,a3,a4 = 2**p2*3**p3*7**p7*13**p13*37**p37
imglist = list()
for a in (a0,a1,a2,a3,a4):
img = itk.image_from_array(a.astype(ctypes.c_ulong))
img.SetSpacing( spacing )
img.SetOrigin( origin )
imglist.append(img)
imgprodUS = image_product(input_list=imglist)
answer = 424242
self.assertTrue( type(imgprodUS) == itk.Image[itk.UL,3])
self.assertTrue( (itk.array_from_image(imgprodUS) == answer).all() )
self.assertTrue( itk.array_from_image(imgprodUS).shape == (nx,ny,nz))
self.assertTrue( np.allclose(imgprodUS.GetSpacing(),spacing) )
self.assertTrue( np.allclose(imgprodUS.GetOrigin(),origin) )
def _mwr_with_loops(dose, mass, newgrid):
"""
Reference implementation, only for testing.
This function computes a dose distribution using the geometry (origin,
size, spacing) of the `newgrid` image, using the energy deposition and mass
with some different geometry. A typical use case is that a Gate simulation first
computes the dose w.r.t. a patient CT (exporting also the mass image),
and then we want to resample this dose distribution to the geometry of the
new grid, e.g. from the dose distribution computed by a TPS.
"""
assert (equal_geometry(dose, mass))
if equal_geometry(dose, newgrid):
# If input and output geometry are equal, then we don't need to do anything, just copy the input dose.
newdose = itk.image_from_array(itk.array_from_image(dose))
newdose.CopyInformation(dose)
return newdose
if not enclosing_geometry(dose, newgrid):
# In a later release we may provide some smart code to deal with dose resampling outside of the input geometry.
raise RuntimeError("new grid must be inside the old one")
t0 = datetime.now()
xol, yol, zol = [
_overlaps(*xyz)
for xyz in zip(dose.GetOrigin(), dose.GetSpacing(),
dose.GetLargestPossibleRegion().GetSize(),
newgrid.GetOrigin(), newgrid.GetSpacing(),
newgrid.GetLargestPossibleRegion().GetSize())
]
nxyz = np.array(dose.GetLargestPossibleRegion().GetSize())
mxyz = np.array(newgrid.GetLargestPossibleRegion().GetSize())
mzyx = mxyz[::-1].tolist()
adose = itk.array_from_image(dose)
amass = itk.array_from_image(mass)
anew = np.zeros(mzyx, dtype=float)
wsum = np.zeros(mzyx, dtype=float)
N_ops = 0
# loop over nonzero overlaps of x-internvals
for (ixs, ixd) in zip(*np.nonzero(xol)):
dx = xol[ixs, ixd]
# loop over nonzero overlaps of y-internvals
for (iys, iyd) in zip(*np.nonzero(yol)):
dy = yol[iys, iyd]
# loop over nonzero overlaps of z-internvals
for (izs, izd) in zip(*np.nonzero(zol)):
dz = zol[izs, izd]
w = dx * dy * dz * amass[izs, iys, ixs]
anew[izd, iyd, ixd] += adose[izs, iys, ixs] * w
wsum[izd, iyd, ixd] += w
N_ops += 1
mask = (wsum > 0)
anew[mask] /= wsum[mask]
newdose = itk.image_from_array(anew)
newdose.CopyInformation(newgrid)
t1 = datetime.now()
dt = (t1 - t0).total_seconds()
logger.debug(
f"resampling using explicit loops over nonzero voxel overlaps took {dt:.3f} seconds"
)
return newdose
def test_big(self):
resampled_loops = _mwr_with_loops(self.dose, self.mass, self.newdose)
resampled = mass_weighted_resampling(self.dose, self.mass,
self.newdose)
self.assertTrue(equal_geometry(resampled_loops, self.newdose))
self.assertTrue(equal_geometry(resampled, self.newdose))
ar0 = itk.array_from_image(resampled_loops)
ar1 = itk.array_from_image(resampled)
self.assertTrue(np.allclose(ar0, ar1))
def faf_ACF_image(image, ctCoeff, spectCoeff, weight=None):
if image.GetImageDimension() != 3:
print("Image dimension (" + str(image.GetImageDimension()) +
") is not 3")
sys.exit(1)
if ctCoeff is None:
print("ctCoeff is mandatory")
sys.exit(1)
elif len(ctCoeff) != 2:
print("ctCoeff size (" + str(len(ctCoeff)) + ") is not 2")
sys.exit(1)
if weight is None:
weight = [1]
nbPeak = len(weight)
if spectCoeff is None:
print("spectCoeff is mandatory")
sys.exit(1)
elif len(spectCoeff) != 3 * nbPeak:
print("ctCoeff size (" + str(len(spectCoeff)) + ") is not 3*nbPeak (" +
str(3 * nbPeak) + ")")
sys.exit(1)
ctArray = itk.array_from_image(image)
ctPositiveValueIndex = np.where(ctArray > 0)
ctNegativeValueIndex = np.where(ctArray < 0)
attenuation = np.zeros(ctArray.shape)
for i in range(nbPeak):
attenuationTemp = np.zeros(ctArray.shape)
attenuationTemp[ctNegativeValueIndex] = spectCoeff[3 * i + 1] + (
spectCoeff[3 * i + 1] -
spectCoeff[3 * i]) / 1000.0 * ctArray[ctNegativeValueIndex]
attenuationTemp[ctPositiveValueIndex] = spectCoeff[
3 * i + 1] + ctCoeff[0] / (ctCoeff[1] - ctCoeff[0]) * (
spectCoeff[3 * i + 2] -
spectCoeff[3 * i + 1]) / 1000.0 * ctArray[ctPositiveValueIndex]
attenuation = attenuation + weight[i] * attenuationTemp
attenuation[attenuation < 0] = 0
attenuationImage = itk.image_from_array(attenuation)
attenuationImage.CopyInformation(image)
projectionAxis = 1
projection = image_projection.image_projection(attenuationImage,
projectionAxis)
acfArray = np.exp(image.GetSpacing()[projectionAxis] / (2.0 * 10.0) *
itk.array_from_image(projection))
acfImage = itk.image_from_array(acfArray)
acfImage.CopyInformation(projection)
flipFilter = itk.FlipImageFilter.New(Input=acfImage)
flipFilter.SetFlipAxes((False, True))
flipFilter.Update()
acfImage = flipFilter.GetOutput()
return acfImage
def read_images(folder, scale):
print(folder)
f = os.path.join(folder, 'dose-Edep.mhd')
img = itk.imread(f)
data = itk.array_from_image(img)
data = data * scale
f = os.path.join(folder, 'dose-Edep-Uncertainty.mhd')
img_s = itk.imread(f)
data_s = itk.array_from_image(img_s)
return img, data, data_s
def image_filter_wrapper(*args, **kwargs):
have_array_input = False
have_xarray_input = False
args_list = list(args)
for index, arg in enumerate(args):
if _HAVE_XARRAY and isinstance(arg, xr.DataArray):
have_xarray_input = True
image = itk.image_from_xarray(arg)
args_list[index] = image
elif is_arraylike(arg):
have_array_input = True
array = np.asarray(arg)
image = itk.image_view_from_array(array)
args_list[index] = image
potential_image_input_kwargs = ('input', 'input1', 'input2', 'input3')
for key, value in kwargs.items():
if (key.lower() in potential_image_input_kwargs
or "image" in key.lower()):
if _HAVE_XARRAY and isinstance(value, xr.DataArray):
have_xarray_input = True
image = itk.image_from_xarray(value)
kwargs[key] = image
elif is_arraylike(value):
have_array_input = True
array = np.asarray(value)
image = itk.image_view_from_array(array)
kwargs[key] = image
if have_xarray_input or have_array_input:
# Convert output itk.Image's to numpy.ndarray's
output = image_filter(*tuple(args_list), **kwargs)
if isinstance(output, tuple):
output_list = list(output)
for index, value in output_list:
if isinstance(value, itk.Image):
if have_xarray_input:
data_array = itk.xarray_from_image(value)
output_list[index] = data_array
else:
array = itk.array_from_image(value)
output_list[index] = array
return tuple(output_list)
else:
if isinstance(output, itk.Image):
if have_xarray_input:
output = itk.xarray_from_image(output)
else:
output = itk.array_from_image(output)
return output
else:
return image_filter(*args, **kwargs)
def faf_lutetium_calibration(spect, ct, planar, injected_activity, delta_time):
if spect.GetImageDimension() != 3:
print("spect image dimension (" + str(spect.GetImageDimension()) + ") is not 3")
sys.exit(1)
if ct.GetImageDimension() != 3:
print("ct image dimension (" + str(ct.GetImageDimension()) + ") is not 3")
sys.exit(1)
if planar.GetImageDimension() != 3:
print("planar image dimension (" + str(planar.GetImageDimension()) + ") is not 3")
sys.exit(1)
if planar.GetLargestPossibleRegion().GetSize()[2] != 8:
print("planar image dimension (" + str(planar.GetLargestPossibleRegion().GetSize()[2]) + ") is not 8")
sys.exit(1)
planarArray = itk.array_from_image(planar)
tempArray1 = planarArray[2:4,:,:]
tempArray2 = planarArray[6:8,:,:]
planar208Array = np.concatenate([tempArray1, tempArray2], axis=0)
planar208Image = itk.image_from_array(planar208Array)
planar208Image.SetSpacing(planar.GetSpacing())
planar208Image.SetOrigin(planar.GetOrigin())
gmImage = faf_create_planar_geometrical_mean.faf_create_planar_geometrical_mean(planar208Image)
registeredGmImage = faf_register_planar_image.faf_register_planar_image(gmImage, spect)
acfImage = faf_ACF_image.faf_ACF_image(ct, [0.2068007, 0.57384408], [0.00014657, 0.13597229, 0.24070651])
acgmImage = faf_ACGM_image.faf_ACGM_image(registeredGmImage, acfImage)
calibratedSpectImage, fafFactor = faf_calibration.faf_calibration(spect, acgmImage, injected_activity, 6.647*24,delta_time, 900, True)
print("Calibration factor with FAF (Bq/count): " + str(fafFactor))
return calibratedSpectImage
def read_file(self, path):
image = itk.imread(path, itk.F)
data = itk.array_from_image(image) # type: np.ndarray
data = data.astype(np.int)
dim_x, dim_y, dim_z = data.shape
return dim_x, dim_y, dim_z, data
def test_faf_calibration(self):
gm = np.ones((32, 12)) * 11.2
spect = np.ones((16, 16, 6)) * 0.33
gmImage = itk.image_from_array(gm)
gmImage.SetOrigin(np.array([-6.0, -16.0]))
spectImage = itk.image_from_array(spect)
spectImage.SetOrigin(np.array([-3.0, -8.0, -8.0]))
calibratedSpectImage, calibrationFactor = faf_calibration(spectImage,
gmImage,
1.0,
half_life=4,
delta_time=4,
verbose=True)
calibratedSpectArray = itk.array_from_image(calibratedSpectImage)
self.assertTrue(
calibratedSpectImage.GetLargestPossibleRegion().GetSize()[0] == 6)
self.assertTrue(
calibratedSpectImage.GetLargestPossibleRegion().GetSize()[1] == 16)
self.assertTrue(
calibratedSpectImage.GetLargestPossibleRegion().GetSize()[2] == 16)
theoreticalcalibrationFactor = 1 / (6 * 16 * 16 * 0.33 /
(0.5 * 0.25)) * 1000000
self.assertTrue(
np.allclose(calibrationFactor, theoreticalcalibrationFactor))
self.assertTrue(
np.allclose(calibratedSpectArray[4, 12],
0.33 * theoreticalcalibrationFactor / 1000000))
def Import_data(frames_path, masks_path):
print("Import raw frames ...")
# Read input image
itk_image = itk.imread(frames_path)
raw_frames = itk.array_from_image(itk_image).astype(np_.uint8)
print("Import masks ...")
with h5py.File(masks_path, 'r') as f:
# List all groups
print("Keys: %s" % f.keys())
a_group_key = list(f.keys())[0]
# Get the data
data_ = list(f[a_group_key])
masks=[]
for mask in data_:
masks.append(np_.reshape(mask-1,(mask.shape)))
return raw_frames, masks
def label_ROI(roiDir, ROI_exp, allRoisView, roiBaseImage, label):
#for ROI in fnmatch.filter(os.listdir(roiDir), ROI_exp):
#print(f'{ROI_exp} {label}')
found = False
for ROI in os.listdir(roiDir):
#print('here ' + ROI + ' ' + ROI_exp)
if fnmatch.fnmatch(ROI.lower(), ROI_exp):
#print('yes ' + ROI)
print(os.path.join(roiDir, ROI))
roisImage = itk.imread(os.path.join(roiDir, ROI))
# resize
roisImage = gt.applyTransformation(input=roisImage,
like=roiBaseImage,
force_resample=True,
interpolation_mode='NN')
#itk.imwrite(roisImage, f'resized_{ROI}')
roisView = itk.array_from_image(roisImage)
allRoisView += roisView * label
if np.any(allRoisView > label):
f = open(f'{taskfolder_3D}/errorlog.txt', 'a')
f.write("Overlapping structures: " +
str(int(allRoisView[allRoisView > label][0] - label)) +
" and " + str(label) + "\n")
f.close()
print("Overlapping structures: " +
str(int(allRoisView[allRoisView > label][0] - label)) +
" and " + str(label))
allRoisView = np.where(allRoisView > label, label, allRoisView)
#return allRoisView, False
found = True #return allRoisView, True
return allRoisView, found
def faf_create_planar_geometrical_mean(image):
if image.GetImageDimension() != 3:
print("Image dimension (" + str(image.image.GetImageDimension()) + ") is not 3")
sys.exit(1)
if image.GetLargestPossibleRegion().GetSize()[2] != 4:
print("Image size (" + str(image.GetLargestPossibleRegion().GetSize()[2]) + ") is not 4")
sys.exit(1)
array = (itk.array_from_image(image)).astype(float)
arrayAnt = array[0, :, :] - 1.1*array[2, :, :]
arrayPost = array[1, :, :] - 1.1*array[3, :, :]
arrayPost = np.flip(arrayPost, 1)
arrayAnt[arrayAnt < 0] = 0
arrayPost[arrayPost < 0] = 0
outputArray = np.sqrt(arrayAnt*arrayPost)
outputImage = itk.image_from_array(outputArray)
spacing = np.array(image.GetSpacing())
spacing = np.delete(spacing, 2)
outputImage.SetSpacing(spacing)
origin = np.array(image.GetOrigin())
origin = np.delete(origin, 2)
outputImage.SetOrigin(origin)
return outputImage
def median_filter_with_mask(img, mask, radius_dilatation, radius_median):
# debug
debug = False
if debug:
input = img
itk.imwrite(img, 'ctvi_before.mhd')
ctvim = itk.median_image_filter(img, radius=radius_median)
itk.imwrite(ctvim, 'ctvi_median.mhd')
dimg = dilate_at_boundaries(img, radius_dilatation)
if debug:
itk.imwrite(dimg, 'ctvi_before_dilated.mhd')
imgm = itk.median_image_filter(dimg, radius=radius_median)
if debug:
itk.imwrite(imgm, 'ctvi_median_after_dilated.mhd')
# reapply mask after median
imgm = itk.array_from_image(imgm)
imgm = imgm.astype('float')
imgm[mask == 0] = 0
if debug:
imgm = imgm.astype(np.float32)
a = itk.image_from_array(imgm)
a.CopyInformation(input)
itk.imwrite(a, 'ctvi_median_final.mhd')
return imgm
def get_stats_struct(uncertpath, dicom_struct, struct_name):
"""Generate stats from a dose image and uncertainty image;
but only for pixels inside specified structure
"""
uncertimg = itk.imread(uncertpath)
uncert_flat = itk.array_from_image(uncertimg).flatten()
ds = pydicom.dcmread(dicom_struct)
aroi = roiutils.region_of_interest(ds, struct_name)
mask = aroi.get_mask(uncertimg, corrected=False)
mask_voxels_flat = itk.array_view_from_image(mask).flatten()
relevant_uncerts = []
for i, val in enumerate(mask_voxels_flat):
if val == 1:
relevant_uncerts.append(uncert_flat[i])
relevant_uncerts = np.array(relevant_uncerts)
print()
print("Number of voxels in {} = {}".format(struct_name,
len(relevant_uncerts)))
print("Mean uncertainty = {}".format(relevant_uncerts.mean()))
print("Median = {}".format(np.median(relevant_uncerts)))
print("Max = {}".format(relevant_uncerts.max()))
print("Min = {}".format(relevant_uncerts.min()))
print("Std dev = {}".format(relevant_uncerts.std()))
# Make histogram
ymax = int(relevant_uncerts.max() * 100 + 2)
binsize = 0.01
bins = [i * binsize for i in range(int(ymax / binsize))]
plt.hist(100 * relevant_uncerts, bins=bins)
plt.xlabel("Dose uncertainty (%)")
plt.show()
def add(self, dose_file):
lockfile = dose_file + ".lock"
dose = None
n_primaries = 0
t0 = datetime.now()
logger.debug("lockfile exists" if os.path.
exists(lockfile) else "lockfile does not exist")
lock = SoftFileLock(lockfile)
try:
# TODO: the length of the timeout should maybe be configured in the system configuration
with lock.acquire(timeout=3):
t1 = datetime.now()
logger.debug("acquiring lock file took {} seconds".format(
(t1 - t0).total_seconds()))
##########################
n_primaries = self.get_nprimaries(dose_file)
if n_primaries < 1:
logger.warn(
f"dose file seems to be based on too few primaries ({n_primaries})"
)
elif bool(self.mass) and bool(self.mask):
simdose = itk.imread(dose_file)
logger.debug(
"resampling dose with size {} using mass file of size {} to target size {}"
.format(itk.size(simdose), itk.size(self.mass),
itk.size(self.mask)))
dose = mass_weighted_resampling(simdose, self.mass,
self.mask)
del simdose
else:
dose = itk.imread(dose_file)
logger.debug("read dose with size {}".format(
itk.size(dose)))
t2 = datetime.now()
logger.debug(
"acquiring dose data {} file took {} seconds".format(
os.path.basename(dose_file),
(t2 - t1).total_seconds()))
if self.wmin > n_primaries:
self.wmin = n_primaries
if self.wmax < n_primaries:
self.wmax = n_primaries
except Timeout:
logger.warn(
"failed to acquire lock for {} for 3 seconds, giving up for now"
.format(dose_file))
return
if not bool(dose):
logger.warn("skipping {}".format(dose_file))
return
adose = itk.array_from_image(dose)
if adose.shape != self.dosesum.shape:
raise RuntimeError(
"PROGRAMMING ERROR: dose shape {} differs from expected shape {}"
.format(adose.shape, self.dosesum.shape))
self.dosesum += adose # n_primaries * (adose / n_primaries)
self.dose2sum += adose**2 / n_primaries # n_primaries * (adose / n_primaries)**2
self.weightsum += n_primaries
self.n += 1
def write_scaled_dose(mhdfile, output, scalefactor):
"""Scale provided dose image and save to output"""
img = itk.imread(mhdfile)
dose = itk.array_from_image(img)
dosescaled = dose * scalefactor
newimg = itk.image_view_from_array(dosescaled)
newimg.CopyInformation(img)
itk.imwrite(newimg, output)
def force_positive_directionality( image ):
"""Return altered mhd image with positive axes directionality
Accepts path to mhd image or itk image object
Returns mhd image
"""
#img = itk.imread(imgpath)
img = None
if type( image )==str:
#Assume we have file path
img = itk.imread( image )
else:
#Assume we have itk image object
img = image
spacing = img.GetSpacing()
origin = img.GetOrigin()
size = img.GetLargestPossibleRegion().GetSize()
direction = np.array( img.GetDirection()*[1,1,1] )
new_origin=[]
for d,o,sz,sp in zip( direction, origin, size, spacing ):
if d==-1:
orig = o - (sz-1)*sp
new_origin.append(orig)
else:
new_origin.append( o )
arr = itk.array_from_image( img )
rot_arr = None
if np.array_equal( direction, np.array([-1,-1,1]) ):
#(1,2 for z-axes) (2,1) opposite sense; take care for 90 degrees.
rot_arr = np.rot90( arr, k=2, axes=(1,2) )
elif np.array_equal( direction, np.array([1,-1,-1]) ):
rot_arr_1 = np.rot90( arr, k=2, axes=(1,2) )
rot_arr = np.rot90( rot_arr_1, k=2, axes=(0,2) ) #(0,2 for y-axis)
elif np.array_equal( direction, np.array([1,1,1]) ):
pass
else:
print("TransformMatrix (Direction) of image not handled correctly")
print("Exiting")
exit(0)
if rot_arr is not None:
#Image shape will change for 90 degree rotations (decubitis positions)
new_img = itk.image_from_array( rot_arr )
new_img.CopyInformation(img)
new_img.SetOrigin( new_origin )
new_img.SetDirection( np.array([[1,0,0],[0,1,0],[0,0,1]]) )
return new_img
else:
return img
def ctvi_slice_nbr_click(input, ct, mask, axis, slice_start, slice_step,
slice_stop, output):
'''
Doc todo
'''
# read images
ctvi_itk = itk.imread(input)
ct = itk.imread(ct)
mask = itk.imread(mask)
ctvi = itk.array_from_image(ctvi_itk)
ct = itk.array_from_image(ct)
mask = itk.array_from_image(mask)
print(ctvi.shape, ct.shape, mask.shape)
# images MUST be the same size
# spacing MUST be isotropic
# slice
if slice_start == -1:
slice_start = int(ctvi.shape[1] / 2)
if slice_stop == -1:
slice_stop = slice_start + 1
elif slice_stop == -2:
slice_stop = ctvi.shape[axis] - 1
print(slice_start, slice_step, slice_stop)
# normalisation ? 90% percentile like in Kipritidis2019 ?
print('min max', np.min(ctvi), np.max(ctvi))
t = np.quantile(ctvi[ctvi > 0], 0.9)
print('Q90%', t)
ctvi = ctvi / t
# get colormap
cmap, cmap_ct, norm = get_colormap1()
# loop for slice
s = slice_start
while s < slice_stop:
f = f'{output}_{s:03d}.png'
ctvi_s = get_slice(ctvi, axis, s)
ct_s = get_slice(ct, axis, s)
mask_s = get_slice(mask, axis, s)
save_img(f, ctvi_s, ct_s, mask_s, cmap, cmap_ct, norm)
s = s + slice_step
def test_five_2D_images(self):
logger.info('Test_Sum test_five_2D_images')
nx,ny = 4,5
hundred = 100
thousand = 1000
spacing = (42.,24.)
origin = (4242.,2424.)
# float images
imglistF = [ itk.image_from_array(np.arange(nx*ny,dtype=np.float32).reshape(nx,ny).copy()),
itk.image_from_array(np.arange(nx*ny,dtype=np.float32)[::-1].reshape(nx,ny).copy()),
itk.image_from_array(np.arange(0,nx*ny*hundred,hundred,dtype=np.float32).reshape(nx,ny).copy()),
itk.image_from_array(np.arange(0,nx*ny*hundred,hundred,dtype=np.float32)[::-1].reshape(4,5).copy()),
itk.image_from_array(thousand*np.ones((nx,ny),dtype=np.float32)) ]
for imgF in imglistF:
imgF.SetSpacing( spacing )
imgF.SetOrigin( origin )
imgsumF = image_sum(input_list=imglistF)
logger.debug("got image with spacing {}".format(imgsumF.GetSpacing()))
index = imgsumF.GetLargestPossibleRegion().GetSize() -1
logger.debug("get sum value {} while expecting {}".format(imgsumF.GetPixel(index),4.*5. -1.))
self.assertTrue( np.allclose(itk.array_view_from_image(imgsumF),nx*ny-1.+(nx*ny-1.)*hundred +thousand) ) # floats: approximate equality
self.assertTrue( itk.array_from_image(imgsumF).shape == (nx,ny))
self.assertTrue( np.allclose(imgsumF.GetSpacing(),spacing))
self.assertTrue( np.allclose(imgsumF.GetOrigin(),origin))
# unsigned short int images ("US" in itk lingo)
nx,ny = 40,50
ten = 10
thirteen = 13
spacing = (32.,23.)
origin = (3232.,2323.)
imglistUS = [ itk.image_from_array(np.arange(nx*ny,dtype=np.uint16).reshape(nx,ny).copy()),
itk.image_from_array(np.arange(nx*ny,dtype=np.uint16)[::-1].reshape(nx,ny).copy()),
itk.image_from_array(np.arange(0,ten*nx*ny,ten,dtype=np.uint16).reshape(nx,ny).copy()),
itk.image_from_array(np.arange(0,ten*nx*ny,ten,dtype=np.uint16)[::-1].reshape(nx,ny).copy()),
itk.image_from_array(thirteen*np.ones((nx,ny),dtype=np.uint16)) ]
for imgUS in imglistUS:
imgUS.SetSpacing( spacing )
imgUS.SetOrigin( origin )
imgsumUS = image_sum(input_list=imglistUS)
logger.debug("got image with spacing {}".format(imgsumUS.GetSpacing()))
logger.debug("get sum value {} while expecting {}".format(imgsumUS.GetPixel(index),40*50-1+10*40*50-10+13))
self.assertTrue( (itk.array_view_from_image(imgsumUS)==nx*ny-1+ten*nx*ny-ten+thirteen).all() ) # ints: exact equality
self.assertTrue( itk.array_from_image(imgsumUS).shape == (nx,ny))
self.assertTrue( np.allclose(imgsumUS.GetSpacing(),spacing))
self.assertTrue( np.allclose(imgsumUS.GetOrigin(),origin))
def update_plan_dose(pdd, label, beam_dose_image):
# 'pdd' is plan dose dictionary
# label will be "unresampled", "Physical" or "RBE"
if label in pdd:
# add dose image to dose image in the dict
a_plandose = itk.array_from_image(pdd[label])
a_beamdose = itk.array_view_from_image(beam_dose_image)
assert (a_plandose.shape == a_beamdose.shape)
a_plandose += a_beamdose
img_plandose = itk.image_from_array(a_plandose)
img_plandose.CopyInformation(pdd[label])
pdd[label] = img_plandose
else:
# copy dose image into dict
a_plandose = itk.array_from_image(beam_dose_image)
img_plandose = itk.image_from_array(a_plandose)
img_plandose.CopyInformation(beam_dose_image)
pdd[label] = img_plandose
def update_roi_characteristics(db, r, background=0):
'''
Update roi volume, density and mass
'''
im_roi = syd.find_one(db['Image'], roi_id=r['id'])
dicom_struct = syd.find_one(db['DicomStruct'], id=r['dicom_struct_id'])
if dicom_struct is not None:
im_ct = syd.find_one(db['Image'],
dicom_series_id=dicom_struct['dicom_series_id'])
else:
im_ct = syd.find_one(
db['Image'],
frame_of_reference_uid=r['frame_of_reference_uid'],
modality='CT')
file_img = syd.find_one(db['File'], id=im_roi['file_mhd_id'])
if im_ct is not None:
file_img_ct = syd.find_one(db['File'], id=im_ct['file_mhd_id'])
else:
print(f'Could not find the CT file image for the roi {r.id}')
return 0
filename_im = os.path.join(
db.absolute_data_folder,
os.path.join(file_img['folder'], file_img['filename']))
filename_ct = os.path.join(
db.absolute_data_folder,
os.path.join(file_img_ct['folder'], file_img_ct['filename']))
roi = itk.imread(filename_im)
ct = itk.imread(filename_ct)
array_im = itk.array_from_image(roi)
spacing = roi.GetSpacing()
### Volume ###
e = np.count_nonzero(array_im != background)
volume_elem = spacing[0] * spacing[1] * spacing[2] # spacing is in mm
volume_elem = volume_elem * 0.001 # convert mm3 to cm3 (/1000)
volume = e * volume_elem
r['volume'] = volume
syd.update_one(db['Roi'], r)
### Density ###
mask = gt.applyTransformation(input=roi,
like=ct,
force_resample=True,
interpolation_mode='NN')
stats = gt.imageStatistics(input=ct, mask=mask)
HU_mean = stats['mean']
density = 1 + (HU_mean / 1000)
r['density'] = density
syd.update_one(db['Roi'], r)
### Mass ###
mass = np.multiply(volume, density)
r['mass'] = mass
syd.update_one(db['Roi'], r)
return 1
def combine_let(dosefiles, letfiles, outname):
"""
Combine LET distributions from multiple simulations in a
dose-weighted fashion
NOTE: The order of dosefiles and letfiles must match,
i.e the index corresponds to a specific simulation!
"""
#Store FLATTENED image arrays
dosearrays = []
letarrays = []
# Get shape and image info
img1 = itk.imread(dosefiles[0])
shape = itk.array_from_image(img1).shape
if len(dosefiles) != len(letfiles):
print("Unequal number of dose and LET files")
if len(dosefiles) == 0:
print("No files given to combine_let method")
else:
for f in dosefiles:
dosearrays.append(itk.array_from_image(itk.imread(f)).flatten())
for f in letfiles:
letarrays.append(itk.array_from_image(itk.imread(f)).flatten())
sumdose = sum(dosearrays)
#itk.imread(dosefiles[0])
totlet = np.zeros(len(letarrays[0]))
for i in range(len(totlet)):
tot = 0
for j in range(len(dosearrays)):
if sumdose[i] != 0: #TODO this ok for floats?
tot = tot + letarrays[j][i] * (dosearrays[j][i] / sumdose[i])
totlet[i] = tot
combinedlet = totlet.reshape(shape)
letimg = itk.image_from_array(combinedlet.astype(
np.float32)) ## ITK CANNOT WRITE DOUBLES, MUST CAST TO FLOAT
letimg.CopyInformation(img1)
itk.imwrite(letimg, outname)
def combine_uncertainty(dosefiles, dosesquaredfiles, statfiles, output):
"""
How to combine dose uncertainties from simulations?
"""
dosearrays = []
dosesq = []
#Gate uncertainty calculation uses NUMBER OF PRIMARIES in simulation,
#not the numberOfHits per voxel
N = get_tot_primaries(statfiles)
# Get shape and image info
img1 = itk.imread(dosefiles[0])
shape = itk.array_from_image(img1).shape
if len(dosefiles) != len(dosesquaredfiles):
print("Unequal number of dose and dosesquared files")
if len(dosefiles) == 0:
print("No files given to sum_uncertainty method")
else:
for f in dosefiles:
dosearrays.append(itk.array_from_image(itk.imread(f)).flatten())
for f in dosesquaredfiles:
dosesq.append(itk.array_from_image(itk.imread(f)).flatten())
sumdose = sum(dosearrays)
sumdosesq = sum(dosesq)
uncertainty = np.ones(len(dosearrays[0]))
for i in range(len(uncertainty)):
#Using Eq(2) from Chetty2006, "Reporting and analyzing statistical uncertainties in MC-based
#treatment planning" and dividing by dose/N for relative uncertainty...
if sumdosesq[i] != 0 and sumdose[i] != 0 and N > 1:
uncertainty[i] = math.sqrt(
1.0 / (N - 1) * (sumdosesq[i] / N -
(sumdose[i] / N)**2)) / (sumdose[i] / N)
combinedUncert = uncertainty.reshape(shape)
uncertimg = itk.image_from_array(combinedUncert.astype(
np.float32)) ## ITK CANNOT WRITE DOUBLES, MUST CAST TO FLOAT
uncertimg.CopyInformation(img1)
itk.imwrite(uncertimg, output)
def imageStatistics(input=None, mask=None, resample=False, histogramBins=1000):
if input is None:
logger.error("Set an input")
sys.exit(1)
inputArray = itk.array_from_image(input)
outputStats = {}
outputStats["nbPixel"] = inputArray.size
if not mask is None:
if resample:
mask = gt.applyTransformation(input=mask, like=input, force_resample=True)
if not np.allclose(mask.GetSpacing(), input.GetSpacing()):
logger.error("Input and mask do not have the same spacing")
sys.exit(1)
if not np.allclose(mask.GetOrigin(), input.GetOrigin()):
logger.error("Input and mask do not have the same origin")
sys.exit(1)
if not np.allclose(itk.array_from_matrix(mask.GetDirection()), itk.array_from_matrix(input.GetDirection())):
logger.error("Input and mask do not have the same direction")
sys.exit(1)
if not np.allclose(mask.GetLargestPossibleRegion().GetSize(), input.GetLargestPossibleRegion().GetSize()):
logger.error("Input and mask do not have the same size")
sys.exit(1)
maskArray = itk.array_from_image(mask)
if len(np.where(maskArray > 1)[0]) >0:
logger.error("The mask seems to be a non-binary image")
index = np.where(maskArray == 1)
outputStats["nbPixel"] = len(index[0])
inputArray = inputArray[index]
outputStats["minimum"] = np.amin(inputArray)
outputStats["maximum"] = np.amax(inputArray)
outputStats["sum"] = np.sum(inputArray)
outputStats["median"] = np.median(inputArray)
outputStats["mean"] = np.mean(outputStats["sum"]/outputStats["nbPixel"])
outputStats["variance"] = np.var(inputArray)
outputStats["sigma"] = np.sqrt(outputStats["variance"])
outputStats["hist"] = np.histogram(inputArray, histogramBins)
return outputStats
# test reading image series with `itk.imread()` and check that dimension is
# not increased if last dimension is 1.
image_series = itk.imread([image_series3d_filename, image_series3d_filename])
assert image_series.GetImageDimension() == 3
# pipeline, auto_pipeline and templated class are tested in other files
# BridgeNumPy
try:
# Images
import numpy as np
image = itk.imread(fileName)
arr = itk.GetArrayFromImage(image)
arr.fill(1)
assert np.any(arr != itk.GetArrayFromImage(image))
arr = itk.array_from_image(image)
arr.fill(1)
assert np.any(arr != itk.GetArrayFromImage(image))
view = itk.GetArrayViewFromImage(image)
view.fill(1)
assert np.all(view == itk.GetArrayFromImage(image))
image = itk.GetImageFromArray(arr)
image.FillBuffer(2)
assert np.any(arr != itk.GetArrayFromImage(image))
image = itk.GetImageViewFromArray(arr)
image.FillBuffer(2)
assert np.all(arr == itk.GetArrayFromImage(image))
image = itk.GetImageFromArray(arr, is_vector=True)
assert image.GetImageDimension() == 2
image = itk.GetImageViewFromArray(arr, is_vector=True)
assert image.GetImageDimension() == 2
