'name {1}. Skip.'.format(data_dir,
os.path.basename(image_name)))
continue
# Read the image
print(' Reading {} ...'.format(image_name))
nim = nib.load(image_name)
image = nim.get_data()
X, Y, Z, T = image.shape
orig_image = image
print(' Segmenting full sequence ...')
start_seg_time = time.time()
# Intensity rescaling
image = rescale_intensity(image, (1, 99))
# Prediction (segmentation)
pred = np.zeros(image.shape)
# Pad the image size to be a factor of 16 so that the
# downsample and upsample procedures in the network will
# result in the same image size at each resolution level.
X2, Y2 = int(math.ceil(X / 16.0)) * 16, int(math.ceil(Y / 16.0)) * 16
x_pre, y_pre = int((X2 - X) / 2), int((Y2 - Y) / 2)
x_post, y_post = (X2 - X) - x_pre, (Y2 - Y) - y_pre
image = np.pad(image,
((x_pre, x_post), (y_pre, y_post), (0, 0), (0, 0)),
'constant')
# Process each time frame
def get_random_batch(filename_list, batch_size, image_size=192, data_augmentation=False,
shift=0.0, rotate=0.0, scale=0.0, intensity=0.0, flip=False):
# Randomly select batch_size images from filename_list
n_file = len(filename_list)
n_selected = 0
images = []
labels = []
while n_selected < batch_size:
rand_index = random.randrange(n_file)
image_name, label_name = filename_list[rand_index]
if os.path.exists(image_name) and os.path.exists(label_name):
print(' Select {0} {1}'.format(image_name, label_name))
# Read image and label
image = nib.load(image_name).get_data()
label = nib.load(label_name).get_data()
# Handle exceptions
if image.shape != label.shape:
print('Error: mismatched size, image.shape = {0}, '
'label.shape = {1}'.format(image.shape, label.shape))
print('Skip {0}, {1}'.format(image_name, label_name))
continue
if image.max() < 1e-6:
print('Error: blank image, image.max = {0}'.format(image.max()))
print('Skip {0} {1}'.format(image_name, label_name))
continue
# Normalise the image size
X, Y, Z = image.shape
cx, cy = int(X / 2), int(Y / 2)
image = crop_image(image, cx, cy, image_size)
label = crop_image(label, cx, cy, image_size)
# Intensity rescaling
image = rescale_intensity(image, (1.0, 99.0))
# Append the image slices to the batch
# Use list for appending, which is much faster than numpy array
for z in range(Z):
images += [image[:, :, z]]
labels += [label[:, :, z]]
# Increase the counter
n_selected += 1
# Convert to a numpy array
images = np.array(images, dtype=np.float32)
labels = np.array(labels, dtype=np.int32)
# Add the channel dimension
# tensorflow by default assumes NHWC format
images = np.expand_dims(images, axis=3)
# Perform data augmentation
if data_augmentation:
images, labels = data_augmenter(images, labels,
shift=shift, rotate=rotate,
scale=scale,
intensity=intensity, flip=flip)
return images, labels