def dist_to_spine(self):
if self.spine_wvs is None:
self.get_spine()
ld = scipy.ndimage.morphology.distance_transform_edt(1 -
self.spine_wvs)
ld = ld * float(self.working_vs[0]) # convert distances to mm
return resize_to_shape(ld, self.orig_shape)
def get_lungs_orig(
self
): # TODO - this doesnt work correctly, is segmenting a lot of unneeded stuff
lungs = scipy.ndimage.filters.gaussian_filter(self.data3dr,
sigma=[4, 2, 2]) > -150
lungs[0, :, :] = 1
lungs = scipy.ndimage.morphology.binary_fill_holes(lungs)
labs, n = scipy.ndimage.measurements.label(lungs == 0)
cornerlab = [
labs[0, 0, 0], labs[0, 0, -1], labs[0, -1, 0], labs[0, -1, -1],
labs[-1, 0, 0], labs[-1, 0, -1], labs[-1, -1, 0], labs[-1, -1, -1]
]
lb = np.median(cornerlab)
labs[labs == lb] = 0
labs[labs == labs[0, 0, 0]] = 0
labs[labs == labs[0, 0, -1]] = 0
labs[labs == labs[0, -1, 0]] = 0
labs[labs == labs[0, -1, -1]] = 0
labs[labs == labs[-1, 0, 0]] = 0
labs[labs == labs[-1, 0, -1]] = 0
labs[labs == labs[-1, -1, 0]] = 0
labs[labs == labs[-1, -1, -1]] = 0
lungs = labs > 0
self.lungs = lungs
#self.body = (labs == 80)
return resize_to_shape(lungs, self.orig_shape)
def get_spine(self):
""" Should have been named get_bones() """
# prepare required data
if self.body is None:
self.get_body() # self.body
# filter out noise in data
data3dr = scipy.ndimage.filters.median_filter(self.data3dr, 3)
# tresholding
# > 240 - still includes kidneys and small part of heart
# > 280 - still a small bit of kidneys
# > 350 - only edges of bones
bones = data3dr > 320
del (data3dr)
bones[self.body == 0] = 0 # cut out anything not inside body
# close holes in data
bones = scipy.ndimage.morphology.binary_closing(
bones, structure=np.ones(
(3, 3, 3))) # TODO - fix removal of data on edge slices
# compute center
self.spine_center_wvs = np.mean(np.nonzero(bones), 1)
self.spine_center_mm = self.spine_center_wvs * self.working_vs / self.voxelsize_mm.astype(
np.double)
# self.center2 = np.mean(np.nonzero(bones), 2)
self.spine_wvs = bones
return resize_to_shape(bones, self.orig_shape)
def get_spine(self, skip_resize=False):
""" Should have been named get_bones() """
# prepare required data
if self.body is None:
self.get_body(skip_resize=True) # self.body
# filter out noise in data
data3dr = scipy.ndimage.filters.median_filter(self.data3dr, 3)
# tresholding
# > 240 - still includes kidneys and small part of heart
# > 280 - still a small bit of kidneys
# > 350 - only edges of bones
bones = data3dr > 320; del(data3dr)
bones[self.body==0] = 0 # cut out anything not inside body
# close holes in data
bones = scipy.ndimage.morphology.binary_closing(bones, structure=np.ones((3,3,3))) # TODO - fix removal of data on edge slices
# compute center
self.spine_center_wvs = np.mean(np.nonzero(bones), 1)
self.spine_center_mm = self.spine_center_wvs * self.working_vs / self.voxelsize_mm.astype(np.double)
# self.center2 = np.mean(np.nonzero(bones), 2)
self.spine_wvs = bones
if skip_resize:
return bones
else:
return resize_to_shape(bones, self.orig_shape)
def dist_diaphragm(self):
if self.diaphragm_mask is None:
self.get_diaphragm_mask()
dst = (scipy.ndimage.morphology.distance_transform_edt(
self.diaphragm_mask) -
scipy.ndimage.morphology.distance_transform_edt(
1 - self.diaphragm_mask))
return resize_to_shape(dst, self.orig_shape)
def weak_dist_bone(self, bone_hull, body):
""" Metoda pro zjisteni, jak blizko jsme stredu (nejvzdalenejsimu mistu od povrchu) """
blank_body = np.ones(body.shape)
bone_edge_dist = scipy.ndimage.morphology.distance_transform_edt(bone_hull)
bone_edge_dist_maximum = np.max(bone_edge_dist)
bone_edge_dist_focus = bone_edge_dist > 0*bone_edge_dist_maximum
blank_body[:] = bone_edge_dist_focus
ld = scipy.ndimage.morphology.distance_transform_edt(blank_body)
return resize_to_shape(ld, self.ss.orig_shape)
def weak_dist_bone(self, bone_hull, body):
""" Metoda pro zjisteni, jak blizko jsme stredu (nejvzdalenejsimu mistu od povrchu) """
blank_body = np.ones(body.shape)
bone_edge_dist = scipy.ndimage.morphology.distance_transform_edt(
bone_hull)
bone_edge_dist_maximum = np.max(bone_edge_dist)
bone_edge_dist_focus = bone_edge_dist > 0 * bone_edge_dist_maximum
blank_body[:] = bone_edge_dist_focus
ld = scipy.ndimage.morphology.distance_transform_edt(blank_body)
return resize_to_shape(ld, self.ss.orig_shape)
def dist_to_ribs(self):
"""
Distance to ribs.
The distance is grater than zero inside of body and outside of body
"""
if self.ribs is None:
self.get_ribs()
ld = scipy.ndimage.morphology.distance_transform_edt(1 - self.ribs)
ld = ld * float(self.working_vs[0]) # convert distances to mm
return resize_to_shape(ld, self.orig_shape)
def dist_to_chest(self):
if self.chest is None:
self.get_ribs()
ld_positive = scipy.ndimage.morphology.distance_transform_edt(
self.chest)
ld_negative = scipy.ndimage.morphology.distance_transform_edt(
1 - self.chest)
ld = ld_positive - ld_negative
ld = ld * float(self.working_vs[0]) # convert distances to mm
return resize_to_shape(ld, self.orig_shape)
def dist_to_ribs(self):
"""
Distance to ribs.
The distance is grater than zero inside of body and outside of body
"""
if self.ribs is None:
self.get_ribs()
ld = scipy.ndimage.morphology.distance_transform_edt(1 - self.ribs)
ld = ld * float(self.working_vs[0]) # convert distances to mm
return resize_to_shape(ld, self.orig_shape)
def dist_axial(self):
if self.diaphragm_mask is None:
self.get_diaphragm_mask(skip_resize=True)
axdst = np.ones(self.data3dr.shape, dtype=np.int16)
axdst[0 ,: ,:] = 0
iz, ix, iy = np.nonzero(self.diaphragm_mask)
# print 'dia level ', self.diaphragm_mask_level
axdst = scipy.ndimage.morphology.distance_transform_edt(axdst) - int(self.diaphragm_mask_level)
axdst = axdst * self.working_vs[0]
return resize_to_shape(axdst, self.orig_shape)
def dist_axial(self):
if self.diaphragm_mask is None:
self.get_diaphragm_mask()
axdst = np.ones(self.data3dr.shape, dtype=np.int16)
axdst[0, :, :] = 0
iz, ix, iy = np.nonzero(self.diaphragm_mask)
# print 'dia level ', self.diaphragm_mask_level
axdst = scipy.ndimage.morphology.distance_transform_edt(axdst) - int(
self.diaphragm_mask_level)
return resize_to_shape(axdst, self.orig_shape)
def dist_to_surface(self):
"""
Positive values in mm inside of body.
:return:
"""
if self.body is None:
self.get_body(skip_resize=True)
# return self._resize_and_dist(self.body)
ld = scipy.ndimage.morphology.distance_transform_edt(self.body)
ld = ld*float(self.working_vs[0]) # convert distances to mm
return resize_to_shape(ld, self.orig_shape)
def get_body(self):
# create segmented 3d data
body = OrganDetectionAlgo.getBody(\
scipy.ndimage.filters.gaussian_filter(self.data3dr, sigma=2), \
self.working_vs)
self.body = body
# get body width and height
widths = []
heights = []
for z in range(body.shape[0]):
x_sum = np.sum(body[z, :, :], axis=0) # suma kazdeho sloupcu
x_start = next((i for i, x in enumerate(list(x_sum)) if x != 0),
None)
x_end = next(
(i for i, x in enumerate(reversed(list(x_sum))) if x != 0),
None)
if x_start is None or x_end is None:
width = 0
else:
width = (body.shape[2] - x_end) - x_start
widths.append(width)
y_sum = np.sum(body[z, :, :], axis=1) # suma kazdeho radku
y_start = next((i for i, y in enumerate(list(y_sum)) if y != 0),
None)
y_end = next(
(i for i, y in enumerate(reversed(list(y_sum))) if y != 0),
None)
if y_start is None or y_end is None:
height = 0
else:
height = (body.shape[1] - y_end) - y_start
heights.append(height)
# get value which is bigger then 90% of calculated values (to ignore bad data)
body_width = np.percentile(np.asarray(widths), 90.0)
body_height = np.percentile(np.asarray(heights), 90.0)
# convert to original resolution
body_width = body_width * (self.orig_shape[2] /
float(self.body.shape[2]))
body_height = body_height * (self.orig_shape[1] /
float(self.body.shape[1]))
# conver to mm
self.body_width = body_width * float(self.voxelsize_mm[2])
self.body_height = body_height * float(self.voxelsize_mm[1])
return resize_to_shape(self.body, self.orig_shape)
def dist_diaphragm(self):
"""
Get positive values in superior to (above) diaphragm and negative values inferior (under) diaphragm.
:return:
"""
if self.diaphragm_mask is None:
self.get_diaphragm_mask(skip_resize=True)
dst = (scipy.ndimage.morphology.distance_transform_edt(
self.diaphragm_mask)#, sampling=self.voxelsize_mm)
-
scipy.ndimage.morphology.distance_transform_edt(
1 - self.diaphragm_mask)#, sampling=self.voxelsize_mm)
)
dst = dst * self.working_vs[0]
return resize_to_shape(dst, self.orig_shape)
def get_body(self, skip_resize=False):
# create segmented 3d data
body = OrganDetectionAlgo.getBody(\
scipy.ndimage.filters.gaussian_filter(self.data3dr, sigma=2), \
self.working_vs)
self.body = body
# get body width and height
widths = []; heights = []
for z in range(body.shape[0]):
x_sum = np.sum(body[z, :, :], axis=0) # suma kazdeho sloupcu
x_start = next((i for i, x in enumerate(list(x_sum)) if x!=0), None)
x_end = next((i for i, x in enumerate(reversed(list(x_sum))) if x!=0), None)
if x_start is None or x_end is None:
width = 0
else:
width = (body.shape[2]-x_end) - x_start
widths.append(width)
y_sum = np.sum(body[z, :, :], axis=1) # suma kazdeho radku
y_start = next((i for i, y in enumerate(list(y_sum)) if y!=0), None)
y_end = next((i for i, y in enumerate(reversed(list(y_sum))) if y!=0), None)
if y_start is None or y_end is None:
height = 0
else:
height = (body.shape[1]-y_end) - y_start
heights.append(height)
# get value which is bigger then 90% of calculated values (to ignore bad data)
body_width = np.percentile(np.asarray(widths), 90.0)
body_height = np.percentile(np.asarray(heights), 90.0)
# convert to original resolution
body_width = body_width * (self.orig_shape[2]/float(self.body.shape[2]))
body_height = body_height * (self.orig_shape[1]/float(self.body.shape[1]))
# conver to mm
self.body_width = body_width * float(self.voxelsize_mm[2])
self.body_height = body_height * float(self.voxelsize_mm[1])
if skip_resize:
return self.body
else:
return resize_to_shape(self.body, self.orig_shape)
def get_lungs_orig(self, skip_resize=False):
lungs_density_threshold_hu = -150
lungs = scipy.ndimage.filters.gaussian_filter(self.data3dr, sigma=[4, 2, 2]) > lungs_density_threshold_hu
if self.head_first:
bottom_slice_id = -1
else:
bottom_slice_id = 0
lungs[bottom_slice_id, :, :] = 1
lungs = scipy.ndimage.morphology.binary_fill_holes(lungs)
labs, n = scipy.ndimage.measurements.label(lungs==0)
cornerlab = [
labs[0,0,0],
labs[0,0,-1],
labs[0,-1,0],
labs[0,-1,-1],
labs[-1,0,0],
labs[-1,0,-1],
labs[-1,-1,0],
labs[-1,-1,-1]
]
lb = np.median(cornerlab)
labs[labs==lb] = 0
labs[labs==labs[0,0,0]] = 0
labs[labs==labs[0,0,-1]] = 0
labs[labs==labs[0,-1,0]] = 0
labs[labs==labs[0,-1,-1]] = 0
labs[labs==labs[-1,0,0]] = 0
labs[labs==labs[-1,0,-1]] = 0
labs[labs==labs[-1,-1,0]] = 0
labs[labs==labs[-1,-1,-1]] = 0
lungs = labs > 0
self.lungs = lungs
#self.body = (labs == 80)
if skip_resize:
return self.lungs
else:
return resize_to_shape(lungs, self.orig_shape)
def uncrop(data, crinfo, orig_shape, resize=False):
crinfo = fix_crinfo(crinfo)
data_out = np.zeros(orig_shape, dtype=data.dtype)
# print 'uncrop ', crinfo
# print orig_shape
# print data.shape
if resize:
data = resize_to_shape(data, crinfo[:, 1] - crinfo[:, 0])
startx = np.round(crinfo[0][0]).astype(int)
starty = np.round(crinfo[1][0]).astype(int)
startz = np.round(crinfo[2][0]).astype(int)
data_out[
# np.round(crinfo[0][0]).astype(int):np.round(crinfo[0][1]).astype(int)+1,
# np.round(crinfo[1][0]).astype(int):np.round(crinfo[1][1]).astype(int)+1,
# np.round(crinfo[2][0]).astype(int):np.round(crinfo[2][1]).astype(int)+1
startx:startx + data.shape[0], starty:starty + data.shape[1],
startz:startz + data.shape[2]] = data
return data_out
def get_diaphragm_mask(self, axis=0):
if self.lungs is None:
self.get_lungs()
ia, ib, ic = self._get_ia_ib_ic(axis)
data = self.lungs
ou = self.get_diaphragm_profile_image(axis=axis)
# reconstruction mask array
mask = np.zeros(data.shape)
for i in range(mask.shape[ia]):
if ia == 0:
mask[i, :, :] = ou > i
elif ia == 1:
mask[:, i, :] = ou > i
elif ia == 2:
mask[:, :, i] = ou > i
self.diaphragm_mask = mask
# maximal point is used for axial ze
# ro plane
self.diaphragm_mask_level = np.median(ou)
self.center0 = self.diaphragm_mask_level * self.working_vs[0]
return resize_to_shape(self.diaphragm_mask, self.orig_shape)
def get_diaphragm_mask(self, axis=0, skip_resize=False):
"""
Get False in inferior to diaphragm and True in superior to diaphragm.
:param axis:
:param skip_resize:
:return: Boolean 3D ndarray
"""
if self.lungs is None:
self.get_lungs()
ia, ib, ic = self._get_ia_ib_ic(axis)
data = self.lungs
ou = self.get_diaphragm_profile_image(axis=axis)
# reconstruction mask array
mask = np.zeros(data.shape)
for i in range(mask.shape[ia]):
if ia == 0:
mask[i,:,:] = ou > i
elif ia == 1:
mask[:,i,:] = ou > i
elif ia == 2:
mask[:,:,i] = ou > i
if not self.head_first:
# invert mask
mask = mask == False
self.diaphragm_mask = mask
# maximal point is used for axial ze
# ro plane
self.diaphragm_mask_level = np.median(ou)
self.center0 = self.diaphragm_mask_level * self.working_vs[0]
if skip_resize:
return self.diaphragm_mask
return resize_to_shape(self.diaphragm_mask, self.orig_shape)
def get_lungs_martin(self):
'''
Improved version of get.lungs function. Call with ss.get_lungs if get_lungs_martin is default.
first part checks if the CT-bed enables a clear segmentation of the background.
Then we threshold and label all cavities. After that, we compare the mean value of the real grayscale-image in
these labeled areas and only take the brighter ones. (Because in lung areas there are "white" alveoli)
'''
lungs = scipy.ndimage.filters.gaussian_filter(self.data3dr, sigma=[4, 2, 2]) > -450
lungs[0, :, :] = 1
lungs = scipy.ndimage.morphology.binary_fill_holes(lungs)
labs, n = scipy.ndimage.measurements.label(lungs == 0)
cornerlab = [
labs[0, 0, 0],
labs[0, 0, -1],
labs[0, -1, 0],
labs[0, -1, -1],
labs[-1, 0, 0],
labs[-1, 0, -1],
labs[-1, -1, 0],
labs[-1, -1, -1]
]
lb = np.median(cornerlab)
labs[labs == lb] = 0
labs[labs == labs[0, 0, 0]] = 0
labs[labs == labs[0, 0, -1]] = 0
labs[labs == labs[0, -1, 0]] = 0
labs[labs == labs[0, -1, -1]] = 0
labs[labs == labs[-1, 0, 0]] = 0
labs[labs == labs[-1, 0, -1]] = 0
labs[labs == labs[-1, -1, 0]] = 0
labs[labs == labs[-1, -1, -1]] = 0
lungs = labs > 0
self.lungs = lungs
# self.body = (labs == 80)
# new code transformed into function
segmented = self.data3dr < -400
preselection_cavities = lungs * segmented
#setting the first slice to one to close all of the cavities, so the fill holes works better
first_slice = copy.copy(preselection_cavities[-1, :, :]) # -1 means last slice, which is here the uppest in the image
preselection_cavities[-1, :, :] = 1
precav_filled = scipy.ndimage.morphology.binary_fill_holes(preselection_cavities)
precav_filled[-1, :, :] = first_slice
precav_erosion = scipy.ndimage.morphology.binary_erosion(precav_filled)
labeled = skimage.morphology.label(precav_erosion)
for f in range(1, np.max(labeled) + 1):
cavity = self.data3dr[labeled == f]
cavity_mean_intensity = np.std(cavity)
if cavity_mean_intensity > 50: #not too sure about the value of 50
#idea would be to take the mean value of the highest ones and set the little lower one as the limit
# this sets not lung-areas to zero. Theoretically :D
precav_erosion[labeled == f] = 0
#print(cavity_mean_intensity)
precav_erosion = scipy.ndimage.morphology.binary_dilation(precav_filled) #dilation becuase of erosion before
# return precav_erosion
self.lungs = precav_erosion
return resize_to_shape(precav_erosion, self.orig_shape)
def _resize_to_orig_shape(self, data):
return resize_to_shape(data, self.orig_shape)
def dist_to_lungs(self):
if self.lungs is None:
self.get_lungs()
ld = scipy.ndimage.morphology.distance_transform_edt(1 - self.lungs)
ld = ld*float(self.working_vs[0]) # convert distances to mm
return resize_to_shape(ld, self.orig_shape)
def dist_to_spine(self):
if self.spine_wvs is None:
self.get_spine(skip_resize=True)
ld = scipy.ndimage.morphology.distance_transform_edt(1 - self.spine_wvs)
ld = ld*float(self.working_vs[0]) # convert distances to mm
return resize_to_shape(ld, self.orig_shape)
def dist_to_surface(self):
if self.body is None:
self.get_body()
ld = scipy.ndimage.morphology.distance_transform_edt(self.body)
ld = ld * float(self.working_vs[0]) # convert distances to mm
return resize_to_shape(ld, self.orig_shape)
def get_lungs_martin(self):
'''
Improved version of get.lungs function. Call with ss.get_lungs if get_lungs_martin is default.
first part checks if the CT-bed enables a clear segmentation of the background.
Then we threshold and label all cavities. After that, we compare the mean value of the real grayscale-image in
these labeled areas and only take the brighter ones. (Because in lung areas there are "white" alveoli)
'''
lungs = scipy.ndimage.filters.gaussian_filter(self.data3dr,
sigma=[4, 2, 2]) > -450
lungs[0, :, :] = 1
lungs = scipy.ndimage.morphology.binary_fill_holes(lungs)
labs, n = scipy.ndimage.measurements.label(lungs == 0)
cornerlab = [
labs[0, 0, 0], labs[0, 0, -1], labs[0, -1, 0], labs[0, -1, -1],
labs[-1, 0, 0], labs[-1, 0, -1], labs[-1, -1, 0], labs[-1, -1, -1]
]
lb = np.median(cornerlab)
labs[labs == lb] = 0
labs[labs == labs[0, 0, 0]] = 0
labs[labs == labs[0, 0, -1]] = 0
labs[labs == labs[0, -1, 0]] = 0
labs[labs == labs[0, -1, -1]] = 0
labs[labs == labs[-1, 0, 0]] = 0
labs[labs == labs[-1, 0, -1]] = 0
labs[labs == labs[-1, -1, 0]] = 0
labs[labs == labs[-1, -1, -1]] = 0
lungs = labs > 0
self.lungs = lungs
# self.body = (labs == 80)
# new code transformed into function
segmented = self.data3dr < -400
preselection_cavities = lungs * segmented
#setting the first slice to one to close all of the cavities, so the fill holes works better
first_slice = copy.copy(
preselection_cavities[-1, :, :]
) # -1 means last slice, which is here the uppest in the image
preselection_cavities[-1, :, :] = 1
precav_filled = scipy.ndimage.morphology.binary_fill_holes(
preselection_cavities)
precav_filled[-1, :, :] = first_slice
precav_erosion = scipy.ndimage.morphology.binary_erosion(precav_filled)
labeled = skimage.morphology.label(precav_erosion)
for f in range(1, np.max(labeled) + 1):
cavity = self.data3dr[labeled == f]
cavity_mean_intensity = np.std(cavity)
if cavity_mean_intensity > 50: #not too sure about the value of 50
#idea would be to take the mean value of the highest ones and set the little lower one as the limit
# this sets not lung-areas to zero. Theoretically :D
precav_erosion[labeled == f] = 0
#print(cavity_mean_intensity)
precav_erosion = scipy.ndimage.morphology.binary_dilation(
precav_filled) #dilation becuase of erosion before
# return precav_erosion
self.lungs = precav_erosion
return resize_to_shape(precav_erosion, self.orig_shape)