def create_slices(imagePath, maskPath, cads):
# if os.path.isfile(imagePath.replace('original', SAVE_FOLDER_image)) == False:
img, origin, spacing = load_itk(imagePath)
mask, _, _ = load_itk(maskPath)
# determine the cads in a lung from csv file
imageName = os.path.split(imagePath)[1].replace('.mhd', '')
image_cads = cads[cads['seriesuid'] == imageName]
# calculate resize factor
resize_factor = spacing / RESIZE_SPACING
new_real_shape = img.shape * resize_factor
new_shape = np.round(new_real_shape)
real_resize = new_shape / img.shape
new_spacing = spacing / real_resize
# resize image & resize lung-mask
lung_img = scipy.ndimage.interpolation.zoom(img, real_resize)
lung_mask = scipy.ndimage.interpolation.zoom(mask, real_resize)
# print lung_mask.shape()
# lung_mask to 0-1
lung_mask[lung_mask > 0] = 1
# create nodule mask
nodule_mask = draw_circles(lung_img, image_cads, origin, new_spacing)
# Determine which slices contain nodules
sliceList = []
for z in range(nodule_mask.shape[2]):
if np.sum(nodule_mask[:, :, z]) > 0:
sliceList.append(z)
# save slices
for z in sliceList:
lung_slice = lung_img[:, :, z]
lung_mask_slice = lung_mask[:, :, z]
nodule_mask_slice = nodule_mask[:, :, z]
# padding to 512x512
original_shape = lung_img.shape
lung_slice_512 = np.zeros((512, 512)) - 3000
lung_mask_slice_512 = np.zeros((512, 512))
nodule_mask_slice_512 = np.zeros((512, 512))
offset = (512 - original_shape[1])
upper_offset = np.round(offset / 2)
lower_offset = offset - upper_offset
new_origin = voxel_2_world([-upper_offset, -lower_offset, 0], origin,
new_spacing)
lung_slice_512[upper_offset:-lower_offset,
upper_offset:-lower_offset] = lung_slice
lung_mask_slice_512[upper_offset:-lower_offset,
upper_offset:-lower_offset] = lung_mask_slice
nodule_mask_slice_512[upper_offset:-lower_offset,
upper_offset:-lower_offset] = nodule_mask_slice
# save the lung slice
savePath = imagePath.replace('original_lungs', SAVE_FOLDER_image)
f = gzip.open(savePath.replace('.mhd', '_slice{}.pkl.gz'.format(z)),
'wb')
pickle.dump(lung_slice_512, f, protocol=-1)
pickle.dump(new_spacing, f, protocol=-1)
pickle.dump(new_origin, f, protocol=-1)
f.close()
savePath = imagePath.replace('original_lungs', SAVE_FOLDER_lung_mask)
f = gzip.open(savePath.replace('.mhd', '_slice{}.pkl.gz'.format(z)),
'wb')
pickle.dump(lung_mask_slice_512, f, protocol=-1)
pickle.dump(new_spacing, f, protocol=-1)
pickle.dump(new_origin, f, protocol=-1)
f.close()
savePath = imagePath.replace('original_lungs', SAVE_FOLDER_nodule_mask)
f = gzip.open(savePath.replace('.mhd', '_slice{}.pkl.gz'.format(z)),
'wb')
pickle.dump(nodule_mask_slice_512, f, protocol=-1)
pickle.dump(new_spacing, f, protocol=-1)
pickle.dump(new_origin, f, protocol=-1)
f.close()
def process_image(image_path, candidates, save_dir):
# load image
image, origin, spacing = load_itk(image_path)
# calculate resize factor
resize_factor = spacing / OUTPUT_SPACING
new_real_shape = image.shape * resize_factor
new_shape = np.round(new_real_shape)
real_resize = new_shape / image.shape
new_spacing = spacing / real_resize
# resize image
image = scipy.ndimage.interpolation.zoom(image, real_resize)
# Pad image with offset (OUTPUT_DIM/2) to prevent problems with candidates on edge of image
offset = OUTPUT_DIM / 2
image = np.pad(image, offset, 'constant', constant_values=0)
# Make a indixlist of the candidates of the image
image_name = os.path.split(image_path)[1].replace('.mhd', '')
indices = candidates[candidates['image_name'] == image_name].index
# loop through the candidates within this image
for i in indices:
# get row data and nodule voxel coords
row = candidates.iloc[i]
world_coords = np.array([row.x, row.y, row.z])
# add offset to voxel coords to cope with padding
coords = np.floor(world_2_voxel(world_coords, origin,
new_spacing)) + offset
label = row.label
# Create xy, xz, yz
xy_slice = np.transpose(image[coords[0] - offset:coords[0] + offset,
coords[1] - offset:coords[1] + offset,
coords[2]])
xz_slice = np.rot90(image[coords[0] - offset:coords[0] + offset,
coords[1],
coords[2] - offset:coords[2] + offset])
yz_slice = np.rot90(image[coords[0],
coords[1] - offset:coords[1] + offset,
coords[2] - offset:coords[2] + offset])
# UNCOMMENT THESE LINES IF YOU WANT TO MANUALLY COMPARE IMAGES WITH MEVISLAB
# test_coords means coords you need to look up in mevislab
# test_coords = world_2_voxel(world_coords,origin,spacing)
# print 'x:',test_coords[0],'y:',test_coords[1],'z:',test_coords[2]
# plt.imshow(xy_slice)
# plt.gray()
# plt.figure(2)
# plt.imshow(xz_slice)
# plt.gray()
# plt.figure(3)
# plt.imshow(yz_slice)
# plt.gray()
# plt.show()
# Create output
output = np.zeros([3, OUTPUT_DIM, OUTPUT_DIM])
output[0, :, :] = xy_slice
output[1, :, :] = xz_slice
output[2, :, :] = yz_slice
# Determine save_path based on label and indices (need +2 due to starting with zero + header in csv)
if label:
save_path = save_dir.format('True') + '/{}.pkl.gz'.format(i + 2)
else:
save_path = save_dir.format('False') + '/{}.pkl.gz'.format(i + 2)
# save with gzip/pickle
with gzip.open(save_path, 'wb') as f:
pickle.dump(output, f, protocol=-1)
def process_image(image_path, candidates, save_dir):
#load image
image, origin, spacing = load_itk(image_path)
#calculate resize factor
resize_factor = spacing / OUTPUT_SPACING
new_real_shape = image.shape * resize_factor
new_shape = np.round(new_real_shape)
real_resize = new_shape / image.shape
new_spacing = spacing / real_resize
#resize image
image = scipy.ndimage.interpolation.zoom(image, real_resize)
#Pad image with offset (OUTPUT_DIM/2) to prevent problems with candidates on edge of image
offset = OUTPUT_DIM/2
image = np.pad(image,offset,'constant',constant_values=0)
#Make a indixlist of the candidates of the image
image_name = os.path.split(image_path)[1].replace('.mhd','')
indices = candidates[candidates['image_name'] == image_name].index
#loop through the candidates within this image
for i in indices:
#get row data and nodule voxel coords
row = candidates.iloc[i]
world_coords = np.array([row.x, row.y, row.z])
# add offset to voxel coords to cope with padding
coords = np.floor(world_2_voxel(world_coords,origin,new_spacing)) + offset
label = row.label
# Create xy, xz, yz
xy_slice = np.transpose(image[coords[0]-offset:coords[0]+offset,coords[1]-offset:coords[1]+offset,coords[2]])
xz_slice = np.rot90(image[coords[0]-offset:coords[0]+offset,coords[1],coords[2]-offset:coords[2]+offset])
yz_slice = np.rot90(image[coords[0],coords[1]-offset:coords[1]+offset,coords[2]-offset:coords[2]+offset])
# UNCOMMENT THESE LINES IF YOU WANT TO MANUALLY COMPARE IMAGES WITH MEVISLAB
# test_coords means coords you need to look up in mevislab
#test_coords = world_2_voxel(world_coords,origin,spacing)
#print 'x:',test_coords[0],'y:',test_coords[1],'z:',test_coords[2]
#plt.imshow(xy_slice)
#plt.gray()
#plt.figure(2)
#plt.imshow(xz_slice)
#plt.gray()
#plt.figure(3)
#plt.imshow(yz_slice)
#plt.gray()
#plt.show()
# Create output
output = np.zeros([3,OUTPUT_DIM,OUTPUT_DIM])
output[0,:,:] = xy_slice
output[1,:,:] = xz_slice
output[2,:,:] = yz_slice
# Determine save_path based on label and indices (need +2 due to starting with zero + header in csv)
if label:
save_path = save_dir.format('True') + '/{}.pkl.gz'.format(i+2)
else:
save_path = save_dir.format('False') + '/{}.pkl.gz'.format(i+2)
# save with gzip/pickle
with gzip.open(save_path,'wb') as f:
pickle.dump(output,f,protocol=-1)
def create_slices(imagePath, maskPath, cads):
#if os.path.isfile(imagePath.replace('original',SAVE_FOLDER_image)) == False:
img, origin, spacing = load_itk(imagePath)
mask, _, _ = load_itk(maskPath)
#determine the cads in a lung from csv file
imageName = os.path.split(imagePath)[1].replace('.mhd','')
image_cads = cads[cads['seriesuid'] == imageName]
#calculate resize factor
resize_factor = spacing / RESIZE_SPACING
new_real_shape = img.shape * resize_factor
new_shape = np.round(new_real_shape)
real_resize = new_shape / img.shape
new_spacing = spacing / real_resize
#resize image & resize lung-mask
lung_img = scipy.ndimage.interpolation.zoom(img, real_resize)
lung_mask = scipy.ndimage.interpolation.zoom(mask, real_resize)
#print lung_mask.shape()
#lung_mask to 0-1
lung_mask[lung_mask >0] = 1
#create nodule mask
nodule_mask = draw_circles(lung_img,image_cads,origin,new_spacing)
#Determine which slices contain nodules
sliceList = []
for z in range(nodule_mask.shape[2]):
if np.sum(nodule_mask[:,:,z]) > 0:
sliceList.append(z)
#save slices
for z in sliceList:
lung_slice = lung_img[:,:,z]
lung_mask_slice = lung_mask[:,:,z]
nodule_mask_slice = nodule_mask[:,:,z]
#padding to 512x512
original_shape = lung_img.shape
lung_slice_512 = np.zeros((512,512)) - 3000
lung_mask_slice_512 = np.zeros((512,512))
nodule_mask_slice_512 = np.zeros((512,512))
offset = (512 - original_shape[1])
upper_offset = np.round(offset/2)
lower_offset = offset - upper_offset
new_origin = voxel_2_world([-upper_offset,-lower_offset,0],origin,new_spacing)
lung_slice_512[upper_offset:-lower_offset,upper_offset:-lower_offset] = lung_slice
lung_mask_slice_512[upper_offset:-lower_offset,upper_offset:-lower_offset] = lung_mask_slice
nodule_mask_slice_512[upper_offset:-lower_offset,upper_offset:-lower_offset] = nodule_mask_slice
#save the lung slice
savePath = imagePath.replace('original_lungs',SAVE_FOLDER_image)
file = gzip.open(savePath.replace('.mhd','_slice{}.pkl.gz'.format(z)),'wb')
pickle.dump(lung_slice_512,file,protocol=-1)
pickle.dump(new_spacing,file, protocol=-1)
pickle.dump(new_origin,file, protocol=-1)
file.close()
savePath = imagePath.replace('original_lungs',SAVE_FOLDER_lung_mask)
file = gzip.open(savePath.replace('.mhd','_slice{}.pkl.gz'.format(z)),'wb')
pickle.dump(lung_mask_slice_512,file,protocol=-1)
pickle.dump(new_spacing,file, protocol=-1)
pickle.dump(new_origin,file, protocol=-1)
file.close()
savePath = imagePath.replace('original_lungs',SAVE_FOLDER_nodule_mask)
file = gzip.open(savePath.replace('.mhd','_slice{}.pkl.gz'.format(z)),'wb')
pickle.dump(nodule_mask_slice_512,file,protocol=-1)
pickle.dump(new_spacing,file, protocol=-1)
pickle.dump(new_origin,file, protocol=-1)
file.close()