def load_prediction_data(path_to_img, channels, x_scale, y_scale, z_scale,
normalize, mu, sig, region_of_interest):
# read image data
img, img_header, img_ext = load_data(path_to_img,
'first_queue',
return_extension=True)
if img is None:
InputError.message = "Invalid image data %s." % (
os.path.basename(path_to_img))
raise InputError()
if img_ext != '.am':
img_header = None
z_shape, y_shape, x_shape = img.shape
# automatic cropping of image to region of interest
if np.any(region_of_interest):
min_z, max_z, min_y, max_y, min_x, max_x = region_of_interest[:]
min_z = min(min_z, z_shape)
min_y = min(min_y, y_shape)
min_x = min(min_x, x_shape)
max_z = min(max_z, z_shape)
max_y = min(max_y, y_shape)
max_x = min(max_x, x_shape)
if max_z - min_z < z_shape:
min_z, max_z = 0, z_shape
if max_y - min_y < y_shape:
min_y, max_y = 0, y_shape
if max_x - min_x < x_shape:
min_x, max_x = 0, x_shape
img = np.copy(img[min_z:max_z, min_y:max_y, min_x:max_x], order='C')
region_of_interest = np.array([
min_z, max_z, min_y, max_y, min_x, max_x, z_shape, y_shape, x_shape
])
z_shape, y_shape, x_shape = max_z - min_z, max_y - min_y, max_x - min_x
# scale image data
img = img.astype(np.float32)
img = img_resize(img, z_scale, y_scale, x_scale)
img -= np.amin(img)
img /= np.amax(img)
if normalize:
mu_tmp, sig_tmp = np.mean(img), np.std(img)
img = (img - mu_tmp) / sig_tmp
img = img * sig + mu
img[img < 0] = 0
img[img > 1] = 1
# compute position data
position = None
if channels == 2:
position = np.empty((z_scale, y_scale, x_scale), dtype=np.float32)
position = compute_position(position, z_scale, y_scale, x_scale)
position = np.sqrt(position)
position /= np.amax(position)
return img, img_header, position, z_shape, y_shape, x_shape, region_of_interest
def create_slices(path_to_data, path_to_label):
try:
if path_to_data:
path_to_slices = path_to_data.replace('images', 'sliceviewer', 1)
if path_to_label:
path_to_label_slices = path_to_label.replace(
'images', 'sliceviewer', 1)
if path_to_data:
# load data
path_to_dir, extension = os.path.splitext(path_to_data)
if extension == '.gz':
path_to_dir, extension = os.path.splitext(path_to_dir)
if extension == '.tar':
img_names = []
for data_type in [
'.tif', '.tiff', '.am', '.hdr', '.mhd', '.mha',
'.nrrd', '.nii', '.nii.gz'
]:
tmp_img_names = glob(path_to_dir + '/**/*' + data_type,
recursive=True)
tmp_img_names = sorted(tmp_img_names)
img_names.extend(tmp_img_names)
raw, _ = load_data(img_names[0], 'create_slices')
zsh, ysh, xsh = raw.shape
m = max(ysh, xsh)
m = max(zsh, m)
scale = float(256) / float(m)
z_scale = int(zsh * scale)
y_scale = int(ysh * scale)
x_scale = int(xsh * scale)
raw = img_resize(raw, z_scale, y_scale, x_scale)
for name in img_names[1:]:
tmp, _ = load_data(name, 'create_slices')
tmp = img_resize(tmp, z_scale, y_scale, x_scale)
raw = np.append(raw, tmp, axis=0)
else:
raw, _ = load_data(path_to_data, 'create_slices')
# increase contrast
raw = img_to_uint8(raw)
raw = contrast(raw)
zsh, ysh, xsh = raw.shape
if not path_to_label:
# create slices for slice viewer
if not os.path.isdir(path_to_slices):
# make directory
os.makedirs(path_to_slices)
os.chmod(path_to_slices, 0o770)
# reduce image size
m = min(ysh, xsh)
if m > 400:
scale = float(400) / float(m)
y_shape = int(ysh * scale)
x_shape = int(xsh * scale)
for k in range(zsh):
tmp = cv2.resize(raw[k], (x_shape, y_shape),
interpolation=cv2.INTER_AREA)
tmp = tmp.astype(np.uint8)
im = Image.fromarray(tmp)
im.save(path_to_slices + '/%s.png' % (k))
else:
for k in range(zsh):
im = Image.fromarray(raw[k])
im.save(path_to_slices + '/%s.png' % (k))
if path_to_label:
# load data
path_to_dir, extension = os.path.splitext(path_to_label)
if extension == '.gz':
path_to_dir, extension = os.path.splitext(path_to_dir)
if extension == '.tar':
img_names = []
for data_type in [
'.tif', '.tiff', '.am', '.hdr', '.mhd', '.mha',
'.nrrd', '.nii', '.nii.gz'
]:
tmp_img_names = glob(path_to_dir + '/**/*' + data_type,
recursive=True)
tmp_img_names = sorted(tmp_img_names)
img_names.extend(tmp_img_names)
# load and scale label data corresponding to img data
mask = np.zeros((0, y_scale, x_scale), dtype=np.uint8)
for name in img_names:
arr, _ = load_data(name, 'create_slices')
arr = arr.astype(np.uint8)
np_unique = np.unique(arr)[1:]
next_mask = np.zeros((z_scale, y_scale, x_scale),
dtype=arr.dtype)
for k in np_unique:
tmp = np.zeros_like(arr)
tmp[arr == k] = 1
tmp = img_resize(tmp, z_scale, y_scale, x_scale)
next_mask[tmp == 1] = k
mask = np.append(mask, next_mask, axis=0)
else:
mask, _ = load_data(path_to_label, 'create_slices')
# img to uint8
mask = color_to_gray(mask)
mask = img_to_uint8(mask)
# create slices for slice viewer
if path_to_data and mask.shape == raw.shape:
if not os.path.isdir(path_to_label_slices):
# make directory
os.makedirs(path_to_label_slices)
os.chmod(path_to_label_slices, 0o770)
# define colors
Color = [(255, 0, 0), (255, 255, 0), (0, 0, 255),
(0, 100, 0), (0, 255, 0), (255, 165, 0),
(139, 0, 0), (255, 20, 147), (255, 105, 180),
(255, 0, 0), (139, 0, 139), (255, 0, 255),
(160, 32, 240), (184, 134, 11), (255, 185, 15),
(255, 215, 0), (0, 191, 255), (16, 78, 139),
(104, 131, 139), (255, 64, 64), (165, 42, 42),
(255, 127, 36), (139, 90, 43), (110, 139, 61),
(0, 255, 127), (255, 127, 80), (139, 10, 80),
(219, 112, 147), (178, 34, 34), (255, 48, 48),
(205, 79, 57), (160, 32, 240), (255, 100, 0)] * 8
labels = np.unique(mask)[1:]
Color = Color[:len(labels)]
labels = labels[:len(labels)]
Color = np.array(Color, dtype=np.uint8)
# reduce image size
m = min(ysh, xsh)
if m > 400:
scale = float(400) / float(m)
ysh = int(ysh * scale)
xsh = int(xsh * scale)
# allocate memory
out = np.empty((ysh, xsh, 3), dtype=np.uint8)
gradient = np.empty((ysh, xsh), dtype=np.uint8)
for k in range(zsh):
# resize slice
if m > 400:
raw_tmp = cv2.resize(raw[k], (xsh, ysh),
interpolation=cv2.INTER_AREA)
mask_tmp = np.zeros((ysh, xsh), dtype=mask.dtype)
for l in labels:
tmp = np.zeros_like(mask[k])
tmp[mask[k] == l] = 1
tmp = cv2.resize(tmp, (xsh, ysh),
interpolation=cv2.INTER_AREA)
mask_tmp[tmp == 1] = l
raw_tmp = raw_tmp.astype(np.uint8)
mask_tmp = mask_tmp.astype(np.uint8)
else:
raw_tmp = raw[k]
mask_tmp = mask[k]
# compute gradient
gradient.fill(0)
tmp = np.abs(mask_tmp[:-1] - mask_tmp[1:])
tmp[tmp > 0] = 1
gradient[:-1] += tmp
gradient[1:] += tmp
tmp = np.abs(mask_tmp[:, :-1] - mask_tmp[:, 1:])
tmp[tmp > 0] = 1
gradient[:, :-1] += tmp
gradient[:, 1:] += tmp
# create output slice
for l in range(3):
out[:, :, l] = raw_tmp
# colorize
for j, label in enumerate(labels):
C = Color[j]
tmp = np.logical_and(gradient > 0,
mask_tmp == label)
out[:, :][tmp] = C
# save slice
im = Image.fromarray(out)
im.save(path_to_label_slices + '/%s.png' % (k))
except:
pass
def predict_pre_final(img, path_to_model, x_scale, y_scale, z_scale, z_patch, y_patch, x_patch, \
normalize, mu, sig, channels, stride_size, batch_size):
# img shape
z_shape, y_shape, x_shape = img.shape
# load position data
if channels == 2:
position = np.empty((z_scale, y_scale, x_scale), dtype=np.float32)
position = compute_position(position, z_scale, y_scale, x_scale)
position = np.sqrt(position)
position /= np.amax(position)
# resize img data
img = img.astype(np.float32)
img = img_resize(img, z_scale, y_scale, x_scale)
img -= np.amin(img)
img /= np.amax(img)
if normalize:
mu_tmp, sig_tmp = np.mean(img), np.std(img)
img = (img - mu_tmp) / sig_tmp
img = img * sig + mu
img[img < 0] = 0
img[img > 1] = 1
# img shape
zsh, ysh, xsh = img.shape
# get number of 3D-patches
nb = 0
for k in range(0, zsh - z_patch + 1, stride_size):
for l in range(0, ysh - y_patch + 1, stride_size):
for m in range(0, xsh - x_patch + 1, stride_size):
nb += 1
# allocate memory
x_test = np.empty((nb, z_patch, y_patch, x_patch, channels),
dtype=img.dtype)
# create testing set
nb = 0
for k in range(0, zsh - z_patch + 1, stride_size):
for l in range(0, ysh - y_patch + 1, stride_size):
for m in range(0, xsh - x_patch + 1, stride_size):
x_test[nb, :, :, :, 0] = img[k:k + z_patch, l:l + y_patch,
m:m + x_patch]
if channels == 2:
x_test[nb, :, :, :,
1] = position[k:k + z_patch, l:l + y_patch,
m:m + x_patch]
nb += 1
# reshape testing set
x_test = x_test.reshape(nb, z_patch, y_patch, x_patch, channels)
# create a MirroredStrategy
if os.name == 'nt':
cdo = tf.distribute.HierarchicalCopyAllReduce()
else:
cdo = tf.distribute.NcclAllReduce()
strategy = tf.distribute.MirroredStrategy(cross_device_ops=cdo)
# load model
with strategy.scope():
model = load_model(str(path_to_model))
# predict
tmp = model.predict(x_test, batch_size=batch_size, verbose=0, steps=None)
# create final
final = np.zeros((zsh, ysh, xsh, tmp.shape[4]), dtype=np.float32)
nb = 0
for k in range(0, zsh - z_patch + 1, stride_size):
for l in range(0, ysh - y_patch + 1, stride_size):
for m in range(0, xsh - x_patch + 1, stride_size):
final[k:k + z_patch, l:l + y_patch, m:m + x_patch] += tmp[nb]
nb += 1
# get final
out = np.argmax(final, axis=3)
out = out.astype(np.uint8)
# rescale final to input size
np_unique = np.unique(out)
label = np.zeros((z_shape, y_shape, x_shape), dtype=out.dtype)
for k in np_unique:
tmp = np.zeros_like(out)
tmp[out == k] = 1
tmp = img_resize(tmp, z_shape, y_shape, x_shape)
label[tmp == 1] = k
return label
def predict_semantic_segmentation(img, position, path_to_model, path_to_final,
z_patch, y_patch, x_patch, z_shape, y_shape,
x_shape, compress, header, img_header,
channels, stride_size, allLabels, batch_size,
region_of_interest):
# img shape
zsh, ysh, xsh = img.shape
# list of IDs
list_IDs = []
# get nIds of patches
for k in range(0, zsh - z_patch + 1, stride_size):
for l in range(0, ysh - y_patch + 1, stride_size):
for m in range(0, xsh - x_patch + 1, stride_size):
list_IDs.append(k * ysh * xsh + l * xsh + m)
# make length of list divisible by batch size
rest = batch_size - (len(list_IDs) % batch_size)
list_IDs = list_IDs + list_IDs[:rest]
# parameters
params = {
'dim': (z_patch, y_patch, x_patch),
'dim_img': (zsh, ysh, xsh),
'batch_size': batch_size,
'n_channels': channels
}
# data generator
predict_generator = PredictDataGenerator(img, position, list_IDs, **params)
# create a MirroredStrategy
if os.name == 'nt':
cdo = tf.distribute.HierarchicalCopyAllReduce()
else:
cdo = tf.distribute.NcclAllReduce()
strategy = tf.distribute.MirroredStrategy(cross_device_ops=cdo)
# load model
with strategy.scope():
model = load_model(str(path_to_model))
# predict
probabilities = model.predict(predict_generator, verbose=0, steps=None)
# create final
final = np.zeros((zsh, ysh, xsh, probabilities.shape[4]), dtype=np.float32)
nb = 0
for k in range(0, zsh - z_patch + 1, stride_size):
for l in range(0, ysh - y_patch + 1, stride_size):
for m in range(0, xsh - x_patch + 1, stride_size):
final[k:k + z_patch, l:l + y_patch,
m:m + x_patch] += probabilities[nb]
nb += 1
# get final
out = np.argmax(final, axis=3)
out = out.astype(np.uint8)
# rescale final to input size
np_unique = np.unique(out)
label = np.zeros((z_shape, y_shape, x_shape), dtype=out.dtype)
for k in np_unique:
tmp = np.zeros_like(out)
tmp[out == k] = 1
tmp = img_resize(tmp, z_shape, y_shape, x_shape)
label[tmp == 1] = k
# revert automatic cropping
if np.any(region_of_interest):
min_z, max_z, min_y, max_y, min_x, max_x, z_shape, y_shape, x_shape = region_of_interest[:]
tmp = np.zeros((z_shape, y_shape, x_shape), dtype=out.dtype)
tmp[min_z:max_z, min_y:max_y, min_x:max_x] = label
label = np.copy(tmp)
# save final
label = label.astype(np.uint8)
label = get_labels(label, allLabels)
if header is not None:
header = get_image_dimensions(header, label)
if img_header is not None:
try:
header = get_physical_size(header, img_header)
except:
pass
save_data(path_to_final, label, header=header, compress=compress)
def load_training_data(normalize, img_list, label_list, channels, x_scale,
y_scale, z_scale, crop_data):
# get filenames
img_names, label_names = [], []
for img_name, label_name in zip(img_list, label_list):
img_dir, img_ext = os.path.splitext(img_name)
if img_ext == '.gz':
img_dir, img_ext = os.path.splitext(img_dir)
label_dir, label_ext = os.path.splitext(label_name)
if label_ext == '.gz':
label_dir, label_ext = os.path.splitext(label_dir)
if img_ext == '.tar' and label_ext == '.tar':
for data_type in [
'.am', '.tif', '.tiff', '.hdr', '.mhd', '.mha', '.nrrd',
'.nii', '.nii.gz'
]:
tmp_img_names = glob(img_dir + '/**/*' + data_type,
recursive=True)
tmp_label_names = glob(label_dir + '/**/*' + data_type,
recursive=True)
tmp_img_names = sorted(tmp_img_names)
tmp_label_names = sorted(tmp_label_names)
img_names.extend(tmp_img_names)
label_names.extend(tmp_label_names)
if len(img_names) == 0:
InputError.message = "Invalid image TAR file."
raise InputError()
if len(label_names) == 0:
InputError.message = "Invalid label TAR file."
raise InputError()
else:
img_names.append(img_name)
label_names.append(label_name)
# load first label
region_of_interest = None
a, header, extension = load_data(label_names[0], 'first_queue', True)
if a is None:
InputError.message = "Invalid label data %s." % (os.path.basename(
label_names[0]))
raise InputError()
if crop_data:
region_of_interest = np.zeros(6)
argmin_z, argmax_z, argmin_y, argmax_y, argmin_x, argmax_x = predict_blocksize(
a)
a = np.copy(a[argmin_z:argmax_z, argmin_y:argmax_y, argmin_x:argmax_x],
order='C')
region_of_interest += [
argmin_z, argmax_z, argmin_y, argmax_y, argmin_x, argmax_x
]
a = a.astype(np.uint8)
np_unique = np.unique(a)
label = np.zeros((z_scale, y_scale, x_scale), dtype=a.dtype)
for k in np_unique:
tmp = np.zeros_like(a)
tmp[a == k] = 1
tmp = img_resize(tmp, z_scale, y_scale, x_scale)
label[tmp == 1] = k
# load first img
img, _ = load_data(img_names[0], 'first_queue')
if img is None:
InputError.message = "Invalid image data %s." % (os.path.basename(
img_names[0]))
raise InputError()
if crop_data:
img = np.copy(img[argmin_z:argmax_z, argmin_y:argmax_y,
argmin_x:argmax_x],
order='C')
img = img.astype(np.float32)
img = img_resize(img, z_scale, y_scale, x_scale)
img -= np.amin(img)
img /= np.amax(img)
mu, sig = np.mean(img), np.std(img)
for img_name, label_name in zip(img_names[1:], label_names[1:]):
# append label
a, _ = load_data(label_name, 'first_queue')
if a is None:
InputError.message = "Invalid label data %s." % (
os.path.basename(name))
raise InputError()
if crop_data:
argmin_z, argmax_z, argmin_y, argmax_y, argmin_x, argmax_x = predict_blocksize(
a)
a = np.copy(a[argmin_z:argmax_z, argmin_y:argmax_y,
argmin_x:argmax_x],
order='C')
region_of_interest += [
argmin_z, argmax_z, argmin_y, argmax_y, argmin_x, argmax_x
]
a = a.astype(np.uint8)
np_unique = np.unique(a)
next_label = np.zeros((z_scale, y_scale, x_scale), dtype=a.dtype)
for k in np_unique:
tmp = np.zeros_like(a)
tmp[a == k] = 1
tmp = img_resize(tmp, z_scale, y_scale, x_scale)
next_label[tmp == 1] = k
label = np.append(label, next_label, axis=0)
# append image
a, _ = load_data(img_name, 'first_queue')
if a is None:
InputError.message = "Invalid image data %s." % (
os.path.basename(name))
raise InputError()
if crop_data:
a = np.copy(a[argmin_z:argmax_z, argmin_y:argmax_y,
argmin_x:argmax_x],
order='C')
a = a.astype(np.float32)
a = img_resize(a, z_scale, y_scale, x_scale)
a -= np.amin(a)
a /= np.amax(a)
if normalize:
mu_tmp, sig_tmp = np.mean(a), np.std(a)
a = (a - mu_tmp) / sig_tmp
a = a * sig + mu
img = np.append(img, a, axis=0)
# automatic cropping
if crop_data:
region_of_interest /= float(len(img_names))
region_of_interest = np.round(region_of_interest)
region_of_interest[region_of_interest < 0] = 0
region_of_interest = region_of_interest.astype(int)
# scale image data to [0,1]
img[img < 0] = 0
img[img > 1] = 1
# compute position data
position = None
if channels == 2:
position = np.empty((z_scale, y_scale, x_scale), dtype=np.float32)
position = compute_position(position, z_scale, y_scale, x_scale)
position = np.sqrt(position)
position /= np.amax(position)
for k in range(len(img_names[1:])):
a = np.copy(position)
position = np.append(position, a, axis=0)
# labels must be in ascending order
allLabels = np.unique(label)
for k, l in enumerate(allLabels):
label[label == l] = k
return img, label, position, allLabels, mu, sig, header, extension, region_of_interest
def load_training_data(normalize, img_list, label_list, channels, x_scale, y_scale, z_scale,
crop_data, x_puffer=25, y_puffer=25, z_puffer=25):
# get filenames
img_names, label_names = [], []
for img_name, label_name in zip(img_list, label_list):
# check for tarball
img_dir, img_ext = os.path.splitext(img_name)
if img_ext == '.gz':
img_dir, img_ext = os.path.splitext(img_dir)
label_dir, label_ext = os.path.splitext(label_name)
if label_ext == '.gz':
label_dir, label_ext = os.path.splitext(label_dir)
if (img_ext == '.tar' and label_ext == '.tar') or (os.path.isdir(img_name) and os.path.isdir(label_name)):
# extract files if necessary
if img_ext == '.tar' and not os.path.exists(img_dir):
tar = tarfile.open(img_name)
tar.extractall(path=img_dir)
tar.close()
if label_ext == '.tar' and not os.path.exists(label_dir):
tar = tarfile.open(label_name)
tar.extractall(path=label_dir)
tar.close()
for data_type in ['.am','.tif','.tiff','.hdr','.mhd','.mha','.nrrd','.nii','.nii.gz']:
tmp_img_names = glob(img_dir+'/**/*'+data_type, recursive=True)
tmp_label_names = glob(label_dir+'/**/*'+data_type, recursive=True)
tmp_img_names = sorted(tmp_img_names)
tmp_label_names = sorted(tmp_label_names)
img_names.extend(tmp_img_names)
label_names.extend(tmp_label_names)
if len(img_names)==0:
InputError.message = "Invalid image TAR file."
raise InputError()
if len(label_names)==0:
InputError.message = "Invalid label TAR file."
raise InputError()
else:
img_names.append(img_name)
label_names.append(label_name)
# load first label
a, header, extension = load_data(label_names[0], 'first_queue', True)
if a is None:
InputError.message = "Invalid label data %s." %(os.path.basename(label_names[0]))
raise InputError()
if crop_data:
argmin_z,argmax_z,argmin_y,argmax_y,argmin_x,argmax_x = predict_blocksize(a, x_puffer, y_puffer, z_puffer)
a = np.copy(a[argmin_z:argmax_z,argmin_y:argmax_y,argmin_x:argmax_x], order='C')
a = a.astype(np.uint8)
np_unique = np.unique(a)
label = np.zeros((z_scale, y_scale, x_scale), dtype=a.dtype)
for k in np_unique:
tmp = np.zeros_like(a)
tmp[a==k] = 1
tmp = img_resize(tmp, z_scale, y_scale, x_scale)
label[tmp==1] = k
# load first img
img, _ = load_data(img_names[0], 'first_queue')
if img is None:
InputError.message = "Invalid image data %s." %(os.path.basename(img_names[0]))
raise InputError()
if crop_data:
img = np.copy(img[argmin_z:argmax_z,argmin_y:argmax_y,argmin_x:argmax_x], order='C')
img = img.astype(np.float32)
img = img_resize(img, z_scale, y_scale, x_scale)
img -= np.amin(img)
img /= np.amax(img)
mu, sig = np.mean(img), np.std(img)
for img_name, label_name in zip(img_names[1:], label_names[1:]):
# append label
a, _ = load_data(label_name, 'first_queue')
if a is None:
InputError.message = "Invalid label data %s." %(os.path.basename(name))
raise InputError()
if crop_data:
argmin_z,argmax_z,argmin_y,argmax_y,argmin_x,argmax_x = predict_blocksize(a, x_puffer, y_puffer, z_puffer)
a = np.copy(a[argmin_z:argmax_z,argmin_y:argmax_y,argmin_x:argmax_x], order='C')
a = a.astype(np.uint8)
np_unique = np.unique(a)
next_label = np.zeros((z_scale, y_scale, x_scale), dtype=a.dtype)
for k in np_unique:
tmp = np.zeros_like(a)
tmp[a==k] = 1
tmp = img_resize(tmp, z_scale, y_scale, x_scale)
next_label[tmp==1] = k
label = np.append(label, next_label, axis=0)
# append image
a, _ = load_data(img_name, 'first_queue')
if a is None:
InputError.message = "Invalid image data %s." %(os.path.basename(name))
raise InputError()
if crop_data:
a = np.copy(a[argmin_z:argmax_z,argmin_y:argmax_y,argmin_x:argmax_x], order='C')
a = a.astype(np.float32)
a = img_resize(a, z_scale, y_scale, x_scale)
a -= np.amin(a)
a /= np.amax(a)
if normalize:
mu_tmp, sig_tmp = np.mean(a), np.std(a)
a = (a - mu_tmp) / sig_tmp
a = a * sig + mu
img = np.append(img, a, axis=0)
# scale image data to [0,1]
img[img<0] = 0
img[img>1] = 1
# compute position data
position = None
if channels == 2:
position = np.empty((z_scale, y_scale, x_scale), dtype=np.float32)
position = compute_position(position, z_scale, y_scale, x_scale)
position = np.sqrt(position)
position /= np.amax(position)
for k in range(len(img_names[1:])):
a = np.copy(position)
position = np.append(position, a, axis=0)
# labels must be in ascending order
allLabels, counts = np.unique(label, return_counts=True)
for k, l in enumerate(allLabels):
label[label==l] = k
# configuration data
configuration_data = np.array([channels, x_scale, y_scale, z_scale, normalize, mu, sig])
return img, label, position, allLabels, configuration_data, header, extension, len(img_names), counts