def get_statistics_voids(h5_volume_dir, dataset_name, scale=2):
scale = float(scale)
print("Calculating statistics voids")
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
Volume[np.where(Volume != 1)] = 0
Volume = ndi.zoom(Volume, 1.0 / scale)
Volume, num_voids = measure.label(Volume, return_num=True)
Volume[np.where(Volume > 0)] += 1000000
list_voids = fit_all_voids_parallel(Volume)
for el in list_voids:
f = open(h5_volume_dir + "/void_dictionary.txt", "w")
for k in range(len(list_voids)):
el = list_voids[k]
if(el[-1] != -1):
f.write("{},{:.0f},{:.0f},{:.0f},{:.2f}, {:.3f}, {:.4f}, {:.4f}, {:.4f}\n".format(el[0], int(scale) * el[1].item(), int(scale) * el[2].item(), int(scale) * el[3].item(), int(scale) * el[4].item(), int(scale) * int(scale) * int(scale) * el[5], el[6], el[7], el[8]))
f.close()
Volume = ndi.zoom(Volume, 2, order=0)
Volume = Volume[:-1, :-1, :-1]
Volume_fibers = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
Volume_fibers[np.where(Volume_fibers == 1)] = Volume[np.where(Volume_fibers == 1)]
tensors_io.save_volume_h5(Volume_fibers, name=dataset_name + '_labeled_voids', dataset_name=dataset_name + '_labeled_voids', directory=h5_volume_dir)
try:
read_dictionary_voids_volume(h5_volume_dir + "/void_statistics.txt")
except:
print("Volume plotting is not possible. Plot Manually")
def crop_volume(h5_volume_dir, dataset_name, dataset_name2, start=[600, 600, 80], window_size=100):
print("Cropping volume")
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
print(len(np.where(Volume == 1)[0]))
Volume = Volume[start[0]:start[0] + window_size,start[1]:start[1] + window_size, start[2]:start[2] + window_size]
data_volume = tensors_io.read_volume_h5(dataset_name2, dataset_name2, h5_volume_dir)
data_volume = data_volume[start[0]:start[0] + window_size,start[1]:start[1] + window_size,start[2]:start[2] + window_size]
tensors_io.save_volume_h5(Volume, name=dataset_name + "_cropped", dataset_name=dataset_name + "_cropped", directory=h5_volume_dir)
tensors_io.save_volume_h5((data_volume * 256).astype(np.int16), name= "volume_cropped", dataset_name= "volume_cropped", directory=h5_volume_dir)
def upsample_full_volume(h5_volume_dir, dataset_name, scale=2):
print("Upsampling Volume")
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
slices = Volume.shape[-1]
for slc in range(0, slices, 100):
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
Volume = Volume[..., slc: slc + 100]
Volume = ndi.zoom(Volume, scale, order=0)
if(slc == 0):
tensors_io.save_volume_h5(Volume, directory=h5_volume_dir, name=dataset_name + "_full_resolution", dataset_name=dataset_name + "_full_resolution")
else:
tensors_io.append_volume_h5(Volume, directory=h5_volume_dir, name=dataset_name + "_full_resolution", dataset_name=dataset_name + "_full_resolution")
def downsample_volume(h5_volume_dir, dataset_name, scale=2):
scale = float(scale)
print("Downsampling volume")
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
print(Volume.shape)
Volume = ndi.zoom(Volume, 1.0 / scale, order=0)
tensors_io.save_volume_h5(Volume, name=dataset_name + "_small", dataset_name=dataset_name + "_small", directory=h5_volume_dir)
def get_statistics_fibers(h5_volume_dir, fibers_name, scale=2):
print("Downsampling and calculating statistics voids")
scale = float(scale)
V_fibers = tensors_io.read_volume_h5(fibers_name, fibers_name, h5_volume_dir)
V_fibers[np.where(V_fibers == 1)] = 0
V_fibers = ndi.zoom(V_fibers, 1.0 / scale, order=0)
list_fibers = fit_all_fibers_parallel(V_fibers)
f = open("dict_fibers.txt","w")
for k in range(len(list_fibers)):
el = list_fibers[k]
f.write("{},{:.0f},{:.0f},{:.0f},{:.2f},{:.2f},{:.0f},{:.0f}\n".format(el[0], scale * el[1][0], scale * el[1][1], scale * el[1][2], scale * el[2], scale * el[3], el[4], el[5]))
f.close()
def find_neighboring_fibers(h5_volume_dir, dataset_name, window_size=50):
print("Cropping volume")
neighbor_dictionary = defaultdict(set)
neighbor_overlap = []
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
rows, cols, slices = Volume.shape
print(Volume.shape)
start_x = []
start_y = []
start_z = []
sz = 0
while(sz + window_size < rows):
start_x.append(sz)
sz = sz + window_size
sz = 0
while(sz + window_size < cols):
start_y.append(sz)
sz = sz + window_size
sz = 0
while(sz + window_size < slices):
start_z.append(sz)
sz = sz + window_size
num_partitions = len(start_z) * len(start_y) * len(start_x)
counter = 0
print("Starting Fitting")
for x in start_x:
for y in start_y:
for z in start_z:
print("Partition {} done out of {}".format(counter, num_partitions))
temp_vol = Volume[x:x + window_size, y:y + window_size, z:z + window_size]
find_n(temp_vol, neighbor_dictionary, neighbor_overlap)
# np.save("neighbor_dictionary", neighbor_dictionary)
counter += 1
f = open( "./neighbor_elements2.txt", "w")
for k in neighbor_dictionary.keys():
el = neighbor_dictionary[k]
f.write("{},".format(k))
for it in el:
f.write("{},".format(it))
f.write("\n")
f.close()
def fit_long_fibers(h5_volume_dir, dataset_name):
print("Fitting Very Long Fibers")
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
complete_fiber_set = set()
edge_fiber_set = set()
with open("fiber_dictionary_CORRECTED_EDGE.txt") as f:
for line in f.readlines():
edge_fiber_set.add(int(line))
print(edge_fiber_set)
fiber_dictionary = dict()
Volume = torch.from_numpy(Volume)
list_fibers = get_fiber_properties_post_processing(Volume, offset=[0, 0, 0], complete_fibers=complete_fiber_set, edge_fiber_set={}, fibers_to_label=edge_fiber_set)
for f_id in list_fibers.keys():
fiber_dictionary[f_id] = list_fibers[f_id]
f = open(h5_volume_dir + "/fiber_dictionary_CORRECTED_EDGES.txt", "w")
for k in fiber_dictionary.keys():
el = fiber_dictionary[k]
f.write("{},{:.0f},{:.0f},{:.0f},{:.2f},{:.2f},{:.0f},{:.0f},{:.2f}\n".format(el[0], el[1], el[2], el[3], el[4], el[5], el[6], el[7], el[8]))
f.close()
def crop_at_a_specific_fiber_debug(h5_volume_dir, dataset_name, fiber_id=2):
print("Cropping volume")
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
coords = np.where(Volume == fiber_id)
centers = [int(coords[i].mean()) for i in range(3)]
maxs = [int(coords[i].max()) for i in range(3)]
mins = [int(coords[i].min()) for i in range(3)]
distances = [maxs[i] - mins[i] for i in range(3)]
window_size = [1.2 * distances[i] for i in range(3)]
min_coords_xyz = [int(max(0, mins[i])) for i in range(3)]
max_coords_xyz = [int(min(Volume.shape[i], maxs[i])) for i in range(3)]
print("Window Size: {}".format(window_size))
print(fiber_id, centers)
Volume = Volume[min_coords_xyz[0]:max_coords_xyz[0], min_coords_xyz[1]:max_coords_xyz[1], min_coords_xyz[2]:max_coords_xyz[2]]
Volume_wrong = np.zeros(Volume.shape)
Volume_small = np.zeros(Volume.shape)
print("Fitting")
list_fibers = fit_all_fibers_parallel(Volume)
for k in list_fibers:
if(k[0] == fiber_id):
fiber_selected = k
if(k[-1] > 200):
Volume_wrong[np.where(Volume == k[0])] = k[0]
print(k)
if(k[-1] == -1):
Volume_small[np.where(Volume == k[0])] = k[0]
print(fiber_selected)
tensors_io.save_volume_h5(Volume, name=dataset_name + "_cropped_single", dataset_name=dataset_name + "_cropped_single", directory=h5_volume_dir)
tensors_io.save_volume_h5(Volume_wrong, name=dataset_name + "_wrong_fibers", dataset_name=dataset_name + "_wrong_fibers", directory=h5_volume_dir)
tensors_io.save_volume_h5(Volume_small, name=dataset_name + "_small_fibers", dataset_name=dataset_name + "_wrong_fibers", directory=h5_volume_dir)
def crop_at_a_specific_fiber(h5_volume_dir, dataset_name, fiber_id=2):
print("Cropping volume")
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
coords = np.where(Volume == fiber_id)
centers = [int(coords[i].mean()) for i in range(3)]
maxs = [int(coords[i].max()) for i in range(3)]
mins = [int(coords[i].min()) for i in range(3)]
distances = [maxs[i] - mins[i] for i in range(3)]
window_size = [distances[i] for i in range(3)]
min_coords_xyz = [int(max(0, mins[i] - 20)) for i in range(3)]
max_coords_xyz = [int(min(Volume.shape[i], maxs[i] + 20)) for i in range(3)]
Volume = torch.from_numpy(Volume)
properties = get_fiber_properties_post_processing(Volume, offset=[0, 0, 0], complete_fibers={}, fibers_to_label={fiber_id})
Volume = Volume.numpy()
print(properties)
print("Window Size: {}".format(window_size))
print(fiber_id, centers)
Volume = Volume[min_coords_xyz[0]:max_coords_xyz[0], min_coords_xyz[1]:max_coords_xyz[1], min_coords_xyz[2]:max_coords_xyz[2]]
# idx = np.where(Volume == fiber_id)
# Volume = Volume * 0
# Volume[idx] = 1000000
tensors_io.save_volume_h5(Volume, name=dataset_name + "_cropped_single", dataset_name=dataset_name + "_cropped_single", directory=h5_volume_dir)
def get_whole_volume_statistics(h5_volume_dir, dataset_name, window_size=500, device=None):
print("Getting Fibers Statistics")
if(device is None):
device = torch.device("cuda:0")
fiber_dictionary = {}
edge_fiber_set = set()
complete_fiber_set = set()
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
rows, cols, slices = Volume.shape
print(Volume.shape)
start_x = []
start_y = []
start_z = []
sz = 0
while(sz + window_size < rows):
start_x.append(sz)
sz = sz + window_size / 2
sz = 0
while(sz + window_size < cols):
start_y.append(sz)
sz = sz + window_size / 2
sz = 0
while(sz + window_size < slices):
start_z.append(sz)
sz = sz + window_size / 2
num_partitions = len(start_z) * len(start_y) * len(start_x)
counter = 0
print("Starting Fitting")
for x in start_x:
for y in start_y:
for z in start_z:
print("Partition {} done out of {}".format(counter, num_partitions))
temp_vol = Volume[x:x + window_size, y:y + window_size, z:z + window_size]
temp_vol = torch.from_numpy(temp_vol).to(device)
list_fibers = get_fiber_properties_post_processing(temp_vol, [x, y, z], complete_fibers=complete_fiber_set, edge_fiber_set=edge_fiber_set, vol_dim=Volume.shape)
for f_id in list_fibers.keys():
fiber_is_complete = 1 - list_fibers[f_id][-1]
if(fiber_is_complete):
fiber_dictionary[f_id] = list_fibers[f_id]
complete_fiber_set.add(f_id)
if(f_id in edge_fiber_set):
edge_fiber_set.remove(f_id)
else:
if(f_id not in complete_fiber_set):
edge_fiber_set.add(f_id)
fiber_dictionary[f_id] = list_fibers[f_id]
Volume[x:x + window_size, y:y + window_size, z:z + window_size] = temp_vol.cpu().numpy()
print("Found Complete Fibers: {}. Edge Fibers {} ".format(len(complete_fiber_set), len(edge_fiber_set)))
counter = counter + 1
f = open(h5_volume_dir + "/fiber_dictionary.txt", "w")
for k in fiber_dictionary.keys():
el = fiber_dictionary[k]
f.write("{},{:.0f},{:.0f},{:.0f},{:.2f},{:.2f},{:.0f},{:.0f},{:.2f}\n".format(el[0], el[1], el[2], el[3], el[4], el[5], el[6], el[7], el[8]))
f.close()
del temp_vol
torch.cuda.empty_cache()
torch.cuda.empty_cache()
print("Saving Partition")
tensors_io.save_volume_h5(Volume, name=dataset_name, dataset_name=dataset_name, directory=h5_volume_dir)
print("Fitting Very Long Fibers")
Volume = torch.from_numpy(Volume)
list_fibers = get_fiber_properties_post_processing(Volume, offset=[0, 0, 0], complete_fibers=complete_fiber_set, edge_fiber_set={}, fibers_to_label=edge_fiber_set)
for f_id in list_fibers.keys():
fiber_dictionary[f_id] = list_fibers[f_id]
f = open(h5_volume_dir + "/fiber_dictionary.txt", "w")
for k in fiber_dictionary.keys():
el = fiber_dictionary[k]
f.write("{},{:.0f},{:.0f},{:.0f},{:.2f},{:.2f},{:.0f},{:.0f},{:.2f}\n".format(el[0], el[1], el[2], el[3], el[4], el[5], el[6], el[7], el[8]))
f.close()
def crop_at_a_specific_fiber_neighbors(h5_volume_dir, dataset_name, neighbors_dir):
print("Cropping volume")
neighbors = read_neighbor_dictionary(neighbors_dir)
fiber_dict = read_fiber_dictionary('h5_statistics/fiber_dictionary.txt')
void_dictionary = read_fiber_dictionary('h5_statistics/void_dictionary2.txt')
for k in neighbors.keys():
if k > 800:
fiber_id = k
list_of_neighbors = neighbors[k]
break
Volume = tensors_io.read_volume_h5(dataset_name, dataset_name, h5_volume_dir)
coords = np.where(Volume == fiber_id)
centers = [int(coords[i].mean()) for i in range(3)]
maxs = [int(coords[i].max()) for i in range(3)]
mins = [int(coords[i].min()) for i in range(3)]
distances = [maxs[i] - mins[i] for i in range(3)]
window_size = [distances[i] for i in range(3)]
min_coords_xyz = [int(max(0, mins[i] - 20)) for i in range(3)]
max_coords_xyz = [int(min(Volume.shape[i], maxs[i] + 20)) for i in range(3)]
Volume = Volume[min_coords_xyz[0]:max_coords_xyz[0], min_coords_xyz[1]:max_coords_xyz[1], min_coords_xyz[2]:max_coords_xyz[2]]
Volume2 = np.zeros(Volume.shape)
list_of_neighbors.append(fiber_id)
void_list = []
fiber_list = []
for el in list_of_neighbors:
idx = np.where(Volume == el)
if(el == fiber_id):
Volume2[idx] = 1
elif(el > 1000000):
Volume2[idx] = 2
print("Void properties")
if(el in void_dictionary.keys()):
void_list.append(void_dictionary[el])
else:
Volume2[idx] = 3
fiber_list.append(fiber_dict[el])
print("Fiber properties")
print(fiber_dict[fiber_id])
'''
for i in fiber_list:
print(i)
print("Viods")
for i in void_list:
print(i)
'''
fiber_list = np.array(fiber_list)
void_list = np.array(void_list)
lengths = fiber_list[:, 5]
thetas = fiber_list[:, 6]
phis = fiber_list[:, 7]
for i in lengths:
print(i, end=",")
print("")
for i in thetas:
print(i, end=",")
print("")
for i in phis:
print(i, end=",")
print("")
print(lengths.mean())
print(thetas.mean())
print(phis.mean())
print(void_list[:, 5].mean())
direction = np.array([void_list[:, 6].mean(), void_list[:, 7].mean(), void_list[:, 8].mean()])
Txy = np.arctan2(direction[1], direction[0]) * 180 / np.pi
if(Txy < 0):
Txy = 180 + Txy
Tz = np.arccos(np.dot(direction, np.array([0, 0, 1])) / np.linalg.norm(direction, 2)) * 180 / np.pi
print(Txy)
print(Tz)
tensors_io.save_volume_h5(Volume2, name=dataset_name + "_cropped_single", dataset_name=dataset_name + "_cropped_single", directory=h5_volume_dir)