def compare_volumes_boundingbox(vol1, vol2, voxelsize_mm):
import qmisc
crinfo = qmisc.crinfo_from_specific_data(vol1, [20, 20, 20])
vol1[
crinfo[0][0]:crinfo[0][1],
crinfo[1][0]:crinfo[1][1],
crinfo[2][0]:crinfo[2][1]
]=1
volume1 = np.sum(vol1 > 0)
volume2 = np.sum(vol2 > 0)
volume1_mm3 = volume1 * np.prod(voxelsize_mm)
volume2_mm3 = volume2 * np.prod(voxelsize_mm)
volume_avg_mm3 = (volume1_mm3 + volume2_mm3) * 0.5
df = vol1 - vol2
df1 = np.sum(df == 1) * np.prod(voxelsize_mm)
df2 = np.sum(df == -1) * np.prod(voxelsize_mm)
evaluation = {
'box_err1_mm3': df1,
'box_err2_mm3': df2,
'box_err1_percent': df1 / volume_avg_mm3 * 100,
'box_err2_percent': df2 / volume_avg_mm3 * 100,
}
return evaluation
def distance_matrics(vol1, vol2, voxelsize_mm):
# crop data to reduce comutation time
crinfo = qmisc.crinfo_from_specific_data(vol1, CROP_MARGIN)
logger.debug(str(crinfo) + ' m1 ' + str(np.max(vol1)) +
' m2 ' + str(np.min(vol2)))
logger.debug("crinfo " + str(crinfo))
vol1 = qmisc.crop(vol1, crinfo)
vol2 = qmisc.crop(vol2, crinfo)
border1 = get_border(vol1)
border2 = get_border(vol2)
#pyed = py3DSeedEditor.py3DSeedEditor(vol1, contour=vol1)
#pyed.show()
b1dst = scipy.ndimage.morphology.distance_transform_edt(
1 - border1,
sampling=voxelsize_mm
)
dst_b1_to_b2 = border2 * b1dst
#import ipdb; ipdb.set_trace() # BREAKPOINT
#pyed = py3DSeedEditor.py3DSeedEditor(dst_b1_to_b2, contour=vol1)
#pyed.show()
#print np.nonzero(border1)
# avgd = np.average(dst_b1_to_b2[np.nonzero(border2)])
avgd = np.average(dst_b1_to_b2[border2])
rmsd = np.average(dst_b1_to_b2[border2] ** 2)
maxd = np.max(dst_b1_to_b2)
return avgd, rmsd, maxd
def compare_volumes_boundingbox(vol1, vol2, voxelsize_mm):
import qmisc
crinfo = qmisc.crinfo_from_specific_data(vol1, [20, 20, 20])
vol1[
crinfo[0][0]:crinfo[0][1],
crinfo[1][0]:crinfo[1][1],
crinfo[2][0]:crinfo[2][1]
] = 1
volume1 = np.sum(vol1 > 0)
volume2 = np.sum(vol2 > 0)
volume1_mm3 = volume1 * np.prod(voxelsize_mm)
volume2_mm3 = volume2 * np.prod(voxelsize_mm)
volume_avg_mm3 = (volume1_mm3 + volume2_mm3) * 0.5
df = vol1 - vol2
df1 = np.sum(df == 1) * np.prod(voxelsize_mm)
df2 = np.sum(df == -1) * np.prod(voxelsize_mm)
evaluation = {
'box_err1_mm3': df1,
'box_err2_mm3': df2,
'box_err1_percent': df1 / volume_avg_mm3 * 100,
'box_err2_percent': df2 / volume_avg_mm3 * 100,
}
return evaluation
def distance_matrics(vol1, vol2, voxelsize_mm):
# crop data to reduce comutation time
crinfo = qmisc.crinfo_from_specific_data(vol1 + vol2, CROP_MARGIN)
logger.debug(str(crinfo) + ' m1 ' + str(np.max(vol1)) +
' m2 ' + str(np.min(vol2)))
logger.debug("crinfo " + str(crinfo))
vol1 = qmisc.crop(vol1, crinfo)
vol2 = qmisc.crop(vol2, crinfo)
border1 = get_border(vol1)
border2 = get_border(vol2)
# pyed = sed3.sed3(vol1, contour=vol1)
# pyed.show()
b1dst = scipy.ndimage.morphology.distance_transform_edt(
1 - border1,
sampling=voxelsize_mm
)
b2dst = scipy.ndimage.morphology.distance_transform_edt(
1 - border2,
sampling=voxelsize_mm
)
dst_b1_to_b2 = border2 * b1dst
dst_b2_to_b1 = border1 * b2dst
dst_12 = dst_b1_to_b2[border2]
dst_21 = dst_b2_to_b1[border1]
dst_both = np.append(dst_12, dst_21)
# sum_d12 = np.sum(dst_12)
# sum_d21 = np.sum(dst_21)
# len_d12 = len(dst_12)
# len_d21 = len(dst_21)
# import ipdb; ipdb.set_trace() # BREAKPOINT
# pyed = sed3.sed3(dst_b1_to_b2, contour=vol1)
# pyed.show()
# print np.nonzero(border1)
# avgd = np.average(dst_b1_to_b2[np.nonzero(border2)])
# import ipdb; ipdb.set_trace() # noqa BREAKPOINT
# avgd = (sum_d21 + sum_d12)/float(len_d21 + len_d12)
# rmsd = ((np.sum(dst_12**2) + dst_21**2)/float(len_d21 + len_d12))**0.5
avgd = np.average(dst_both)
# there is not clear what is correct
# rmsd = np.average(dst_both ** 2)**0.5
# rmsd = (np.average(dst_12) + np.average(dst_21))**0.5
rmsd = np.average(dst_both ** 2)
# rmsd = np.average(dst_b1_to_b2[border2] ** 2)
maxd = max(np.max(dst_b1_to_b2), np.max(dst_b2_to_b1))
# old
# avgd = np.average(dst_b1_to_b2[border2])
# rmsd = np.average(dst_b1_to_b2[border2] ** 2)
# maxd = np.max(dst_b1_to_b2)
return avgd, rmsd, maxd
def distance_matrics(vol1, vol2, voxelsize_mm):
# crop data to reduce comutation time
crinfo = qmisc.crinfo_from_specific_data(vol1 + vol2, CROP_MARGIN)
logger.debug(str(crinfo) + ' m1 ' + str(np.max(vol1)) +
' m2 ' + str(np.min(vol2)))
logger.debug("crinfo " + str(crinfo))
vol1 = qmisc.crop(vol1, crinfo)
vol2 = qmisc.crop(vol2, crinfo)
border1 = _get_border(vol1)
border2 = _get_border(vol2)
# pyed = sed3.sed3(vol1, contour=vol1)
# pyed.show()
b1dst = scipy.ndimage.morphology.distance_transform_edt(
1 - border1,
sampling=voxelsize_mm
)
b2dst = scipy.ndimage.morphology.distance_transform_edt(
1 - border2,
sampling=voxelsize_mm
)
dst_b1_to_b2 = border2 * b1dst
dst_b2_to_b1 = border1 * b2dst
dst_12 = dst_b1_to_b2[border2]
dst_21 = dst_b2_to_b1[border1]
dst_both = np.append(dst_12, dst_21)
# sum_d12 = np.sum(dst_12)
# sum_d21 = np.sum(dst_21)
# len_d12 = len(dst_12)
# len_d21 = len(dst_21)
# import ipdb; ipdb.set_trace() # BREAKPOINT
# pyed = sed3.sed3(dst_b1_to_b2, contour=vol1)
# pyed.show()
# print np.nonzero(border1)
# avgd = np.average(dst_b1_to_b2[np.nonzero(border2)])
# import ipdb; ipdb.set_trace() # noqa BREAKPOINT
# avgd = (sum_d21 + sum_d12)/float(len_d21 + len_d12)
# rmsd = ((np.sum(dst_12**2) + dst_21**2)/float(len_d21 + len_d12))**0.5
avgd = np.average(dst_both)
# there is not clear what is correct
# rmsd = np.average(dst_both ** 2)**0.5
# rmsd = (np.average(dst_12) + np.average(dst_21))**0.5
rmsd = np.average(dst_both ** 2)
# rmsd = np.average(dst_b1_to_b2[border2] ** 2)
maxd = max(np.max(dst_b1_to_b2), np.max(dst_b2_to_b1))
# old
# avgd = np.average(dst_b1_to_b2[border2])
# rmsd = np.average(dst_b1_to_b2[border2] ** 2)
# maxd = np.max(dst_b1_to_b2)
return avgd, rmsd, maxd
def train_one(self, data,voxelSize_mm):
"""
Trenovani shape modelu
data se vezmou a oriznou (jen jatra)
na oriznuta data je aplikovo binarni otevreni - rychlejsi nez morphsnakes
co vznikne je uhlazena cast ktera se odecte od puvodniho obrazu
cimz vzniknou spicky
orezany obraz se nasledne rozparceluje podle velikosti (shape) modelu
pokud pocet voxelu v danem useku prekroci danou mez, je modelu
prirazena nejaka hodnota. Meze jsou nasledujici:
0%-50% => 1
50%-75% => 2
75%-100% => 3
"""
crinfo = qmisc.crinfo_from_specific_data(data, margin=self.model_margin)
datacr = qmisc.crop(data, crinfo=crinfo)
dataShape = self.model.shape
datacrres = self.trainThresholdMap(datacr, voxelSize_mm, dataShape)
self.model += datacrres
self.data_number += 1
def train_one(self, data,voxelSize_mm):
"""
Trenovani shape modelu
data se vezmou a oriznou (jen jatra)
na oriznuta data je aplikovo binarni otevreni - rychlejsi nez morphsnakes
co vznikne je uhlazena cast ktera se odecte od puvodniho obrazu
cimz vzniknou spicky
orezany obraz se nasledne rozparceluje podle velikosti (shape) modelu
pokud pocet voxelu v danem useku prekroci danou mez, je modelu
prirazena nejaka hodnota. Meze jsou nasledujici:
0%-50% => 1
50%-75% => 2
75%-100% => 3
"""
crinfo = qmisc.crinfo_from_specific_data(data, margin=self.model_margin)
datacr = qmisc.crop(data, crinfo=crinfo)
dataShape = self.model.shape
datacrres = self.trainThresholdMap(datacr, voxelSize_mm, dataShape)
self.model += datacrres
self.data_number += 1
def _crinfo_from_specific_data(self, data, margin):
# hledáme automatický ořez, nonzero dá indexy
return qmisc.crinfo_from_specific_data(data, margin)
def _interactivity_end(self, igc):
"""
This is called after processing step. All data are rescaled to original
resolution.
"""
logger.debug('_interactivity_end()')
self.__resize_to_orig(igc.seeds)
self.organ_interactivity_counter = igc.interactivity_counter
logger.debug("org inter counter " +
str(self.organ_interactivity_counter))
logger.debug('nonzero segm ' + str(np.nonzero(self.segmentation)))
# if False:
if False:
# TODO dodělat postprocessing PV
import segmentation
outputSegmentation = segmentation.vesselSegmentation( # noqa
self.data3d,
self.segmentation,
threshold=-1,
inputSigma=0.15,
dilationIterations=10,
nObj=1,
biggestObjects=False,
seeds=(self.segmentation > 0).astype(np.int8),
useSeedsOfCompactObjects=True,
interactivity=True,
binaryClosingIterations=2,
binaryOpeningIterations=0)
self._segmentation_postprocessing()
# rint 'autocrop', self.autocrop
if self.autocrop is True:
# rint
# mport pdb; pdb.set_trace()
tmpcrinfo = qmisc.crinfo_from_specific_data(
self.segmentation,
self.autocrop_margin)
self.crop(tmpcrinfo)
if self.texture_analysis not in (None, False):
import texture_analysis
# doplnit nějaký kód, parametry atd
# elf.orig_scale_segmentation =
# texture_analysis.segmentation(self.data3d,
# self.orig_scale_segmentation, params = self.texture_analysis)
self.segmentation = texture_analysis.segmentation(
self.data3d,
self.segmentation,
self.voxelsize_mm
)
# set label number
# !! pomaly!!!
# @TODO make faster
self.segmentation[self.segmentation == 1] = self.output_label
#
logger.debug('self.slab')
logger.debug(str(self.slab))
self.processing_time = (
datetime.datetime.now() - self.time_start).total_seconds()
logger.debug('processing_time = ' + str(self.processing_time))
def _interactivity_end(self, igc):
logger.debug('_interactivity_end()')
# @TODO remove old code in except part
try:
# rint 'pred vyjimkou'
# aise Exception ('test without skimage')
# rint 'za vyjimkou'
import skimage
import skimage.transform
# Now we need reshape seeds and segmentation to original size
segm_orig_scale = skimage.transform.resize(
self.segmentation, self.data3d.shape, order=0)
seeds = skimage.transform.resize(
igc.seeds, self.data3d.shape, order=0)
self.segmentation = segm_orig_scale
self.seeds = seeds
except:
segm_orig_scale = scipy.ndimage.zoom(
self.segmentation,
1.0 / self.zoom,
mode='nearest',
order=0
).astype(np.int8)
seeds = scipy.ndimage.zoom(
igc.seeds,
1.0 / self.zoom,
mode='nearest',
order=0
)
self.organ_interactivity_counter = igc.interactivity_counter
logger.debug("org inter counter " +
str(self.organ_interactivity_counter))
# @TODO odstranit hack pro oříznutí na stejnou velikost
# v podstatě je to vyřešeno, ale nechalo by se to dělat elegantněji v zoom
# tam je bohužel patrně bug
# rint 'd3d ', self.data3d.shape
# rint 's orig scale shape ', segm_orig_scale.shape
shp = [
np.min([segm_orig_scale.shape[0], self.data3d.shape[0]]),
np.min([segm_orig_scale.shape[1], self.data3d.shape[1]]),
np.min([segm_orig_scale.shape[2], self.data3d.shape[2]]),
]
# elf.data3d = self.data3d[0:shp[0], 0:shp[1], 0:shp[2]]
# mport ipdb; ipdb.set_trace() # BREAKPOINT
self.segmentation = np.zeros(self.data3d.shape, dtype=np.int8)
self.segmentation[
0:shp[0],
0:shp[1],
0:shp[2]] = segm_orig_scale[0:shp[0], 0:shp[1], 0:shp[2]]
del segm_orig_scale
self.seeds[
0:shp[0],
0:shp[1],
0:shp[2]] = seeds[0:shp[0], 0:shp[1], 0:shp[2]]
if self.segmentation_smoothing:
self.segm_smoothing(self.smoothing_mm)
# rint 'autocrop', self.autocrop
if self.autocrop is True:
# rint
# mport pdb; pdb.set_trace()
tmpcrinfo = qmisc.crinfo_from_specific_data(
self.segmentation,
self.autocrop_margin)
self.crop(tmpcrinfo)
# seg = self
# rint 'ms d3d ', oseg.data3d.shape
# rint 'ms seg ', oseg.segmentation.shape
# rint 'crinfo ', oseg.crinfo
# elf.segmentation = qmisc.crop(self.segmentation, tmpcrinfo)
# elf.data3d = qmisc.crop(self.data3d, tmpcrinfo)
# elf.crinfo = qmisc.combinecrinfo(self.crinfo, tmpcrinfo)
if self.texture_analysis not in (None, False):
import texture_analysis
# doplnit nějaký kód, parametry atd
# elf.orig_scale_segmentation =
# texture_analysis.segmentation(self.data3d,
# self.orig_scale_segmentation, params = self.texture_analysis)
self.segmentation = texture_analysis.segmentation(
self.data3d,
self.segmentation,
self.voxelsize_mm
)
# set label number
# !! pomaly!!!
# @TODO make faster
self.segmentation[self.segmentation == 1] = self.output_label
#
self.processing_time = time.time() - self.time_start
def resection_portal_vein_new(data,
interactivity=False,
seeds=None,
label=10,
vein=6,
**kwargs):
"""
New function for portal vein segmentation
:param data:
:param interactivity:
:param seeds:
:param kwargs:
:return:
"""
# ed = sed3.sed3(a)
# ed.show()
# from PyQt4 import QtGui
# from PyQt4.QtGui import QApplication, QMainWindow, QVBoxLayout, QHBoxLayout, QLabel, QPushButton, QFrame, \
# QFont, QPixmap, QFileDialog
#
# window = QtGui.QWidget()
# mainLayout = QVBoxLayout()
# window.setLayout(mainLayout)
# mainLayout.addWidget(sed3.sed3qtWidget(data['data3d'], contour=data['segmentation']))
# zachovani puvodnich dat
segmentation = data["segmentation"]
data3d = data["data3d"]
# data pouze se segmentacemi
segm = ((data["segmentation"] == label) * label +
(data["segmentation"] == vein) * vein)
# ufiknutí segmentace
crinfo = qmisc.crinfo_from_specific_data(segm, [0])
data["segmentation"] = qmisc.crop(segm, crinfo)
data["data3d"] = qmisc.crop(data3d, crinfo)
# @TODO zde nahradit střeve čímkoliv smysluplnějším
if interactivity:
print("Select cut")
# seeds = cut_editor_old(data)
seeds = cut_editor_old(data)
elif seeds is None:
logger.error('seeds is None and interactivity is False')
return None
lab, cut = split_vessel(data, seeds)
segm, dist1, dist2 = split_organ_by_two_vessels(data, lab)
# jatra rozdeleny na 3 kusy
a = morphology.label(segm, background=0)
### podmínka nefunguje
if 3 in a: # zda se v segmentaci objevuje 3. cast
print "slape :) :) :P"
a_index = velikosti(segm)
print a_index
i = nejnizsi(a_index[0], a_index[1], a_index[2])
segm = ((a == i) * (segm == 1).astype('int8') + (a != i) *
(segm == 2).astype('int8') + (segm != 0).astype('int8'))
# TODO split this function from visualization
data = virtual_resection_visualization(data,
segm,
dist1,
dist2,
cut,
interactivity=interactivity)
# vrácení původních dat a spojení s upravenými daty
data["data3d"] = data3d
data["segmentation"] = qmisc.uncrop(
data["segmentation"], crinfo,
(len(segmentation), len(segmentation[0]), len(segmentation[0])))
#segmentation = segmentation == vein
data["segmentation"] = (
data["segmentation"] +
(segmentation != label) * segmentation) - (segmentation == vein) * vein
return data
