def __init__( self, scalarName, volume, neighbors=0 ):
super( FyRadiusMapAction, self ).__init__( scalarName, volume )
self._startLabels = {}
self._endLabels = {}
self._neighbors = neighbors
self._paddedImage = ap.pad( self._image, neighbors, 'constant', constant_values=( 0 ) )
def pad_images_and_equalize_sizes(images, padding=(0, 0, 0), mode="reflect", **kwargs):
images_padded = []
max = np.max([im.shape for im in images], axis=0)
for im in images:
pad_width = [(p, p+e) for p, e in zip(padding, (max - im.shape))]
im_padded = pad(im, pad_width, mode=str(mode), **kwargs)
images_padded.append(im_padded)
return np.array(images_padded)
def pad_images_and_equalize_sizes_swapped(images, padding=(0, 0), mode="reflect", **kwargs):
images_padded = []
max = np.max([im.shape for im in images], axis=0)[1:]
for image in images:
im_padded = []
for im in image: # divide by channels
pad_width = [(p, p+e) for p, e in zip(padding, (max - im.shape))]
im_padded.append(pad(im, pad_width, mode=str(mode), **kwargs)[np.newaxis,:,:])
image_padded = np.vstack(tuple(im_padded))
images_padded.append(image_padded)
return np.array(images_padded)
def validateMapping( volumefile, trkfile, radius=0, map_intermediate=True ):
'''
Check if a trk file has correctly mapped scalar values from a volume file.
If radius is > 0 take it into account by looking for the most common value in a sphere
around the original point. This only happens on start and end points so.
If map_intermediate is active, also the points between end points are validated but never
using the radius.
Returns TRUE if everything is fine, FALSE if there were errors.
'''
# load the mapped trk file
s = io.loadTrk( trkfile )
volume = io.readImage( volumefile )
imageHeader = volume.header
image_size = volume.shape[:3]
# grab the tracks
tracks = s[0]
# pad the image with zeros
image = ap.pad( volume, radius, 'constant', constant_values=( 0 ) )
# any errors?
any_errors = False
# incorporate spacing
spacing = imageHeader.get_zooms()[:3]
# .. and loop through them
for t in tracks:
points = t[0] # the points of this fiber track
scalars = t[1] # the mapped scalars
for index, p in enumerate( points ):
current_point = [ int( a / b ) for a, b in zip( [p[0], p[1], p[2]], spacing )]
#print 'ORIG', volume[current_point[0], current_point[1], current_point[2]]
is_first_point = ( index == 0 )
is_last_point = ( index == len( points ) - 1 )
# if this is
if not map_intermediate and not is_first_point and not is_last_point:
real_scalar = 0.0
else:
# here we check for the neighborhood if radius > 0
if radius > 0 and ( is_first_point or is_last_point ):
# neighborhood search!
r = radius
a, b, c = current_point
# crop the image according to the neighborhood look-up
# since we zero-padded the image, we don't need boundary checks here
min_x = a - r
max_x = a + r + 1
min_y = b - r
max_y = b + r + 1
min_z = c - r
max_z = c + r + 1
cropped_image = numpy.asarray( image[min_x + r:max_x + r, min_y + r:max_y + r, min_z + r:max_z + r] )
# create a sphere mask
x, y, z = numpy.ogrid[0:2 * r + 1, 0:2 * r + 1, 0:2 * r + 1]
mask = ( x - r ) ** 2 + ( y - r ) ** 2 + ( z - r ) ** 2 <= r * r # 3d sphere mask
# apply the mask
masked_container = cropped_image[mask]
# throw away all zeros (0)
masked_container = masked_container[numpy.nonzero( masked_container )]
# find the most frequent label in the masked container
from collections import Counter
# by default, we use the original one
mostFrequentLabel = volume[a, b, c]
if len( masked_container ) != 0:
counter = Counter( masked_container )
all_labels = counter.most_common()
best_match_label = counter.most_common( 1 )
original_pos = [i for i, v in enumerate( all_labels ) if v[0] == mostFrequentLabel]
if not original_pos or all_labels[original_pos[0]][1] != best_match_label[0][1]:
# the original label appears less often as the new best_match_label
# in this case, we use the new best matched label
mostFrequentLabel = best_match_label[0][0]
# we don't need an else here since the original label is already set
real_scalar = mostFrequentLabel
else:
# simple mapping without radius incorporation and make sure we are inside the volume
real_scalar = volume[min( current_point[0], image_size[0] - 1 ), min( current_point[1], image_size[1] - 1 ) , min( current_point[2], image_size[2] - 1 )]
if type( scalars ) is types.NoneType:
mapped_scalar = -1
else:
mapped_scalar = scalars[index][0]
# now check if the mapped scalar from the trk file matches the real scalar
compare = ( mapped_scalar == real_scalar )
if compare:
compare = Colors.GREEN + 'OK'
else:
compare = Colors.RED + 'WRONG!!!'
any_errors = True
print Colors.PURPLE + 'Probing ' + Colors.CYAN + str( current_point ) + Colors.PURPLE + ' for scalar.. SHOULD BE: ' + Colors.CYAN + str( real_scalar ) + Colors.PURPLE + ' WAS: ' + Colors.CYAN + str( mapped_scalar ) + Colors.PURPLE + ' ... ' + str( compare ) + Colors._CLEAR
# return TRUE if everything went fine and FALSE if there were errors
return not any_errors