File size: 12,431 Bytes
71cf8b5 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 | import SimpleITK as sitk
import numpy as np
import scipy.ndimage as ndimage
from scipy.ndimage import distance_transform_edt
from scipy.ndimage import label, find_objects
def split_connected_components(labels, label_value, offset, min_volume=400, top_n=6):
"""
Split the region with the specified label_value into multiple connected components and reassign labels.
Parameters:
labels (np.ndarray): Input label array
label_value (int): The label value to split
offset (int): Offset used to generate new label values
min_volume (int): Minimum volume to retain connected components
top_n (int): Retain the top-n connected components by volume
Returns:
np.ndarray: Relabeled array
"""
# Get a binary mask where the label is equal to label_value
binary_mask = (labels == label_value)
structure = np.array([[[0, 0, 0],
[0, 1, 0],
[0, 0, 0]],
[[0, 1, 0],
[1, 1, 1],
[0, 1, 0]],
[[0, 0, 0],
[0, 1, 0],
[0, 0, 0]]], dtype=int)
# Use scipy.ndimage.label to mark connected components
labeled_array, num_features = label(binary_mask, structure=structure)
# Create new_labels as a copy of the input labels
new_labels = labels.copy()
# Get the volume of all connected components
volumes = [np.sum(labeled_array == i) for i in range(1, num_features + 1)]
# Get indices of the top-n connected components by volume
top_n_indices = np.argsort(volumes)[-top_n:][::-1]
top_n_volumes_labels = [(volumes[i], i + 1) for i in top_n_indices] # Note that component indices start from 1
# Iterate through all connected components in descending order of volume and reassign labels to avoid conflicts
current_label = offset
for volume, i in top_n_volumes_labels:
region_mask = (labeled_array == i)
if volume >= min_volume:
new_labels[region_mask] = current_label
current_label += 1
else:
new_labels[region_mask] = 0
return new_labels
def remove_small_connected_components(prediction, min_volume, label_values):
"""
Remove small connected components and set them as background.
Parameters:
prediction (np.ndarray): Model output predictions
min_volume (int): Minimum volume to retain connected components
label_values (list): List of label values to process
Returns:
np.ndarray: Processed prediction array
"""
new_prediction = prediction.copy()
# Define the connectivity structure for identifying connected components
structure = np.array([[[0, 0, 0],
[0, 1, 0],
[0, 0, 0]],
[[0, 1, 0],
[1, 1, 1],
[0, 1, 0]],
[[0, 0, 0],
[0, 1, 0],
[0, 0, 0]]], dtype=int)
for index, label_value in enumerate(label_values):
print(f"Processing label {label_value}:")
# Get binary mask for the specified label
binary_mask = (prediction == label_value)
minimum = min_volume[index]
labeled_array, num_features = label(binary_mask, structure=structure)
# Get slices of each connected component
slices = find_objects(labeled_array)
retained_sizes = []
removed_sizes = []
# Iterate through each connected component and remove those smaller than the minimum volume
for i, slice_ in enumerate(slices):
region_size = np.sum(labeled_array[slice_] == (i + 1))
if region_size <= minimum:
removed_sizes.append(region_size)
new_prediction[labeled_array == (i + 1)] = 0
else:
retained_sizes.append(region_size)
# Print the sizes of retained and removed regions
if retained_sizes:
print(f" Retained regions sizes: {retained_sizes}")
if removed_sizes:
print(f" Removed regions sizes: {removed_sizes}")
return new_prediction
def calculate_iou(label1, label2):
intersection = np.logical_and(label1, label2).sum()
union = np.logical_or(label1, label2).sum()
return intersection / union
def relabel_connected_components(segmentation):
"""
Handle partial confusion between left and right hip bones in stage0 segmentation.
Parameters:
segmentation (np.ndarray): The segmentation labels from stage0.
Returns:
np.ndarray: Relabeled segmentation with confusion resolved.
"""
# Detect connected components for labels 2 and 3
label_2, num_features_2 = ndimage.label(segmentation == 2)
label_3, num_features_3 = ndimage.label(segmentation == 3)
# Create arrays to store the sizes of the connected components
size_2 = np.bincount(label_2.ravel())
size_3 = np.bincount(label_3.ravel())
# Initialize a new segmentation array for relabeling
new_segmentation = np.copy(segmentation)
# Create a structural element to detect boundaries, suitable for 3D
struct = ndimage.generate_binary_structure(3, 1)
# Iterate over the connected components for label 2
for label in range(1, num_features_2 + 1):
current_region = (label_2 == label)
neighbors = ndimage.binary_dilation(current_region, structure=struct) & (segmentation == 3)
if neighbors.any():
touching_labels_3 = np.unique(label_3[neighbors])
for lbl_3 in touching_labels_3:
if lbl_3 > 0:
if 5 * size_2[label] < size_3[lbl_3]:
print(f"Change: class_2 (size: {size_2[label]}) -> class_3 (size: {size_3[lbl_3]})")
new_segmentation[current_region] = 3
elif 5 * size_3[lbl_3] < size_2[label]:
print(f"Change: class_3 (size: {size_3[lbl_3]}) -> class_2 (size: {size_2[label]})")
new_segmentation[label_3 == lbl_3] = 2
return new_segmentation
def refine_labels(label1, label2, threshold=0.99):
"""
Refine label2 based on reference from label1 if IoU is below the threshold.
Parameters:
label1 (np.ndarray): The reference label.
label2 (np.ndarray): The label to be refined.
threshold (float): IoU threshold for refinement. Default is 0.99.
Returns:
np.ndarray: Refined label.
"""
iou = calculate_iou(label1 > 0, label2 > 0) # Calculate IoU considering only foreground and background
if iou >= threshold:
return label2
print('Refining label...')
fixed_label2 = label2.copy()
# Label the connected components in label2
structure = np.array([[[0, 0, 0],
[0, 1, 0],
[0, 0, 0]],
[[0, 1, 0],
[1, 1, 1],
[0, 1, 0]],
[[0, 0, 0],
[0, 1, 0],
[0, 0, 0]]], dtype=int)
labeled_a, num_features_a = label(label2, structure=structure)
# Iterate over the connected components in label2
for component_id in range(1, num_features_a + 1):
component_mask = (labeled_a == component_id)
if not np.any(component_mask & (label1 > 0)):
# If there is no intersection with label1, set the component to background
fixed_label2[component_mask] = 0
# Foreground areas in label1 that are background in label2
fg_to_bg_mask = (label1 > 0) & (label2 == 0)
# Find the nearest foreground pixel label
if fg_to_bg_mask.any():
distance, indices = distance_transform_edt(fixed_label2 == 0, return_indices=True)
nearest_foreground = label2[tuple(indices)]
fixed_label2[fg_to_bg_mask] = nearest_foreground[fg_to_bg_mask]
return fixed_label2
def process_final_label(segmentation):
"""
Process the final segmentation labels by refining the connected components of specific labels.
Parameters:
segmentation (np.ndarray): The final segmentation labels.
Returns:
np.ndarray: Refined segmentation with certain connected components removed or relabeled.
"""
# Initialize a new segmentation array for relabeling
new_segmentation = np.copy(segmentation)
# Mask out sacrum labels (set to background)
segmentation = np.where((segmentation >= 1) & (segmentation <= 10), 0, segmentation)
# Detect connected components for labels 11 and 21
label_11, num_features_11 = ndimage.label(segmentation == 11)
label_21, num_features_21 = ndimage.label(segmentation == 21)
# Calculate the size of each connected component
size_11 = np.bincount(label_11.ravel())
size_21 = np.bincount(label_21.ravel())
# Find the index of the largest connected component for labels 11 and 21
largest_label_11_index = np.argmax(size_11[1:]) + 1 # Skip index 0 (background)
largest_label_21_index = np.argmax(size_21[1:]) + 1
assert num_features_11 > 0 and num_features_21 > 0, "label 11 and label 21 have no connected components!!"
# Remove the largest connected components from label_11 and label_21 (mark them as background)
label_11[label_11 == largest_label_11_index] = 0 # Mark largest connected component as background
num_features_11 -= 1 # Update the number of connected components
label_21[label_21 == largest_label_21_index] = 0 # Mark largest connected component as background
num_features_21 -= 1 # Update the number of connected components
# Create a structural element for boundary detection, suitable for 3D
struct = ndimage.generate_binary_structure(3, 1)
# Define a function to process connected components for a given label
def process_label(label, segment_label, num_features):
if num_features < 1:
return # Do not process if no connected components remain after removing the largest one
for lbl in range(1, num_features + 1):
current_region = (label == lbl)
neighbors = ndimage.binary_dilation(current_region, structure=struct) & (segmentation != segment_label)
if neighbors.any():
# Find all touching labels, excluding background
touching_labels = np.unique(segmentation[neighbors])
touching_labels = touching_labels[touching_labels != 0] # Exclude background
touching_labels = touching_labels[touching_labels != segment_label] # Exclude current label
if touching_labels.size > 0:
# Calculate the volume of each touching label
touching_label_sizes = {label: np.sum(segmentation == label) for label in touching_labels}
# Find the touching label with the largest volume
max_touching_label = max(touching_label_sizes, key=touching_label_sizes.get)
print(f"Changing segment {lbl} from {segment_label} to {max_touching_label}")
new_segmentation[current_region] = max_touching_label
# Process connected components for label 11
process_label(label_11, 11, num_features_11)
# Process connected components for label 21
process_label(label_21, 21, num_features_21)
return new_segmentation
if __name__ == "__main__":
labels = sitk.ReadImage(
"/home/ypy/Code/PENGWIN-example-algorithm-main/PENGWIN-challenge-packages/preliminary-development-phase-ct/stage1_label_after_remove_101_1.nii.gz")
spacing = labels.GetSpacing()
direction = labels.GetDirection()
origin = labels.GetOrigin()
labels = sitk.GetArrayFromImage(labels)
# label_value = 2
# offset = 22
# new_labels = split_connected_components(labels, label_value, offset)
# new_labels = np.where(new_labels == 1, 21, new_labels)
# new_labels = remove_small_connected_components(labels, 20000, [1, 2, 3])
new_labels = relabel_connected_components(labels)
save_label = sitk.GetImageFromArray(new_labels.astype(np.int8))
save_label.SetSpacing(spacing)
save_label.SetDirection(direction)
save_label.SetOrigin(origin)
sitk.WriteImage(save_label, "stage1_label_after.nii.gz", useCompression=True)
|