def load_slices(data_path, case_list):
img_list = []
pan_list = []
les_list = []
for ID in case_list:
img_path = data_path + '/' + ID + '/ctscan.npy'
pan_path = data_path + '/' + ID + '/pancreas.npy'
pancreas = np.load(pan_path)
pancreas = smoothing(pancreas)
img = np.load(img_path)
img = windowing(img)
img = minmax_normalization(img)
lesion = np.zeros(img.shape)
if ID[:2] != 'NP' and ID[:2] != 'AD':
les_path = data_path + '/' + ID + '/lesion.npy'
lesion = np.load(les_path)
pancreas[np.where(lesion == 1)] = 1
# pancreas is union of pancreas and lesion by hands.
img_pan = img * pancreas
img_list.append(img)
pan_list.append(img_pan)
les_list.append(lesion)
return img_list, pan_list, les_list
def preprocessing_test(self, image, pancreas, lesion):
from skimage import morphology
if self.data_type == 'tcia':
image = image[:, ::-1, :]
pancreas = pancreas[:, ::-1, :]
lesion = lesion[:, ::-1, :]
pancreas = smoothing(pancreas)
lesion = smoothing(lesion)
elif self.data_type == 'msd':
image = image[::-1, ::-1, :]
pancreas = pancreas[::-1, ::-1, :]
lesion = lesion[::-1, ::-1, :]
elif self.data_type == 'ntuh':
image = image[::-1, ::-1, :]
pancreas = pancreas[::-1, ::-1, :]
lesion = lesion[::-1, ::-1, :]
pancreas = morphology.dilation(pancreas, np.ones([3, 3, 1]))
lesion = morphology.dilation(lesion, np.ones([3, 3, 1]))
image = windowing(image)
image = pancreas_normalization(image, pancreas, lesion)
return image, pancreas, lesion
def patch_generator(data_path,
tumor_id,
patch_size=50,
dilation=[5, 5, 5],
add_mask=True,
stride=30,
threshold=0.5,
max_amount=1000):
"""
Usage: Complete process of patch generating
Parameters
----------
data_path (str): The path for the box data
tumor_id (str): The target ID
Example: 'PT1'
patch_size (int): Patch size, default to 50
dilation (int): the dilation of non-lesion region
Returns
-------
List: each content is a 2D patch
List: each content is a label
"""
X = []
Y = []
img_path = data_path + '/' + tumor_id + '/ctscan.npy'
pan_path = data_path + '/' + tumor_id + '/pancreas.npy'
pancreas = np.load(pan_path)
pancreas = smoothing(pancreas)
if tumor_id[:2] == 'PC' or tumor_id[:2] == 'PT':
les_path = data_path + '/' + tumor_id + '/lesion.npy'
lesion = np.load(les_path)
lesion = smoothing(lesion)
else:
lesion = np.zeros(pancreas.shape)
pancreas = morphology.dilation(pancreas, np.ones([3, 3, 1]))
lesion = morphology.dilation(lesion, np.ones([3, 3, 1]))
coords = masked_2D_sampler(lesion,
pancreas,
patch_size,
stride,
threshold,
max_amount=max_amount)
img = np.load(img_path)
img = windowing(img)
img = minmax_normalization(img)
if add_mask:
mask = np.zeros(pancreas.shape)
mask[np.where(lesion == 1)] = 1
mask[np.where(pancreas == 1)] = 1
img = img * mask
for coord in coords:
mask_pancreas = img[coord[1]:coord[4], coord[2]:coord[5], coord[3]]
X.append(mask_pancreas)
Y.append(coord[0])
return X, Y