Spaces:
Sleeping
Sleeping
felipekitamura
commited on
Commit
•
ef4e685
1
Parent(s):
033e507
Upload 2 files
Browse files- __init__.py +1 -0
- face_deid_ct.py +271 -0
__init__.py
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
from .face_deid_ct import *
|
face_deid_ct.py
ADDED
@@ -0,0 +1,271 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import pydicom
|
3 |
+
import numpy as np
|
4 |
+
import cv2
|
5 |
+
from matplotlib import pyplot as plt
|
6 |
+
import random
|
7 |
+
from tqdm import tqdm
|
8 |
+
import time
|
9 |
+
|
10 |
+
FACE_MAX_VALUE = 50
|
11 |
+
FACE_MIN_VALUE = -125
|
12 |
+
|
13 |
+
AIR_THRESHOLD = -800
|
14 |
+
KERNEL_SIZE = 35
|
15 |
+
|
16 |
+
|
17 |
+
|
18 |
+
def is_dicom(file_path):
|
19 |
+
try:
|
20 |
+
pydicom.dcmread(file_path)
|
21 |
+
return True
|
22 |
+
except Exception:
|
23 |
+
return False
|
24 |
+
|
25 |
+
def get_first_directory(path):
|
26 |
+
# Normalize the path to always use Unix-style path separators
|
27 |
+
normalized_path = path.replace("\\", "/")
|
28 |
+
split_path = normalized_path.split("/")[-1]
|
29 |
+
|
30 |
+
return split_path # Return None if no directories are found
|
31 |
+
|
32 |
+
def list_dicom_directories(root_dir):
|
33 |
+
dicom_dirs = set()
|
34 |
+
|
35 |
+
for root, dirs, files in os.walk(root_dir):
|
36 |
+
for file in files:
|
37 |
+
file_path = os.path.join(root, file)
|
38 |
+
if is_dicom(file_path):
|
39 |
+
dicom_dirs.add(root)
|
40 |
+
break
|
41 |
+
|
42 |
+
return list(dicom_dirs)
|
43 |
+
|
44 |
+
def load_scan(path):
|
45 |
+
slices = [pydicom.read_file(path + '/' + s) for s in os.listdir(path)]
|
46 |
+
slices.sort(key = lambda x: float(x.ImagePositionPatient[2]))
|
47 |
+
try:
|
48 |
+
slice_thickness = np.abs(slices[0].ImagePositionPatient[2] - slices[1].ImagePositionPatient[2])
|
49 |
+
except:
|
50 |
+
slice_thickness = np.abs(slices[0].SliceLocation - slices[1].SliceLocation)
|
51 |
+
|
52 |
+
for s in slices:
|
53 |
+
s.SliceThickness = slice_thickness
|
54 |
+
|
55 |
+
return slices
|
56 |
+
|
57 |
+
def get_pixels_hu(slices):
|
58 |
+
image = np.stack([s.pixel_array for s in slices])
|
59 |
+
# Convert to int16 (from sometimes int16),
|
60 |
+
# should be possible as values should always be low enough (<32k)
|
61 |
+
image = image.astype(np.int16)
|
62 |
+
|
63 |
+
# Set outside-of-scan pixels to 0
|
64 |
+
# The intercept is usually -1024, so air is approximately 0
|
65 |
+
image[image == -2000] = 0
|
66 |
+
|
67 |
+
# Convert to Hounsfield units (HU)
|
68 |
+
for slice_number in range(len(slices)):
|
69 |
+
|
70 |
+
intercept = slices[slice_number].RescaleIntercept
|
71 |
+
slope = slices[slice_number].RescaleSlope
|
72 |
+
|
73 |
+
if slope != 1:
|
74 |
+
image[slice_number] = slope * image[slice_number].astype(np.float64)
|
75 |
+
image[slice_number] = image[slice_number].astype(np.int16)
|
76 |
+
|
77 |
+
image[slice_number] += np.int16(intercept)
|
78 |
+
|
79 |
+
return np.array(image, dtype=np.int16)
|
80 |
+
|
81 |
+
def binarize_volume(volume, air_hu=AIR_THRESHOLD):
|
82 |
+
binary_volume = np.zeros_like(volume, dtype=np.uint8)
|
83 |
+
binary_volume[volume <= air_hu] = 1
|
84 |
+
return binary_volume
|
85 |
+
|
86 |
+
def largest_connected_component(binary_image):
|
87 |
+
# Find all connected components and stats
|
88 |
+
num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(binary_image, connectivity=8)
|
89 |
+
|
90 |
+
# Get the index of the largest component, ignoring the background
|
91 |
+
# The background is considered as a component by connectedComponentsWithStats and it is usually the first component
|
92 |
+
largest_component_index = np.argmax(stats[1:, cv2.CC_STAT_AREA]) + 1
|
93 |
+
|
94 |
+
# Create an image to keep largest component only
|
95 |
+
largest_component_image = np.zeros(labels.shape, dtype=np.uint8)
|
96 |
+
largest_component_image[labels == largest_component_index] = 1
|
97 |
+
|
98 |
+
return largest_component_image
|
99 |
+
|
100 |
+
def get_largest_component_volume(volume):
|
101 |
+
# Initialize an empty array to hold the processed volume
|
102 |
+
processed_volume = np.empty_like(volume, dtype=np.uint8)
|
103 |
+
|
104 |
+
# Iterate over each slice in the volume
|
105 |
+
for i in range(volume.shape[0]):
|
106 |
+
# Process the slice and store it in the processed volume
|
107 |
+
processed_volume[i] = largest_connected_component(volume[i])
|
108 |
+
|
109 |
+
return processed_volume
|
110 |
+
|
111 |
+
|
112 |
+
|
113 |
+
def dilate_volume(volume, kernel_size=KERNEL_SIZE):
|
114 |
+
# Create the structuring element (kernel) for dilation
|
115 |
+
kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (kernel_size, kernel_size))
|
116 |
+
|
117 |
+
# Initialize an empty array to hold the dilated volume
|
118 |
+
dilated_volume = np.empty_like(volume)
|
119 |
+
|
120 |
+
# Iterate over each slice in the volume
|
121 |
+
for i in range(volume.shape[0]):
|
122 |
+
# Dilate the slice and store it in the dilated volume
|
123 |
+
dilated_volume[i] = cv2.dilate(volume[i].astype(np.uint8), kernel)
|
124 |
+
|
125 |
+
return dilated_volume
|
126 |
+
|
127 |
+
|
128 |
+
def apply_mask_and_get_values(image_volume, mask_volume):
|
129 |
+
# Apply the mask by multiplying the image volume with the mask volume
|
130 |
+
masked_volume = image_volume * mask_volume
|
131 |
+
|
132 |
+
# Get all unique values in the masked volume, excluding zero
|
133 |
+
unique_values = np.unique(masked_volume)
|
134 |
+
unique_values = unique_values[unique_values > FACE_MIN_VALUE]
|
135 |
+
unique_values = unique_values[unique_values < FACE_MAX_VALUE]
|
136 |
+
|
137 |
+
# Convert numpy array to a list
|
138 |
+
unique_values_list = unique_values.tolist()
|
139 |
+
|
140 |
+
return unique_values_list
|
141 |
+
|
142 |
+
|
143 |
+
def apply_random_values_optimized(pixels_hu, dilated_volume, unique_values_list):
|
144 |
+
# Initialize new volume as a copy of the original volume
|
145 |
+
new_volume = np.copy(pixels_hu)
|
146 |
+
|
147 |
+
# Generate random indices
|
148 |
+
random_indices = np.random.choice(len(unique_values_list), size=np.sum(dilated_volume))
|
149 |
+
|
150 |
+
# Select random values from the unique_values_list
|
151 |
+
random_values = np.array(unique_values_list)[random_indices]
|
152 |
+
|
153 |
+
# Apply the random values to the locations where dilated_volume equals 1
|
154 |
+
new_volume[dilated_volume == 1] = random_values
|
155 |
+
|
156 |
+
return new_volume
|
157 |
+
|
158 |
+
def save_new_dicom_files(new_volume, original_dir, out_path, app="_d"):
|
159 |
+
# Create a new directory path by appending "_d" to the original directory
|
160 |
+
if out_path is None:
|
161 |
+
new_dir = original_dir + app
|
162 |
+
else:
|
163 |
+
new_dir = out_path
|
164 |
+
|
165 |
+
# Create the new directory if it doesn't exist
|
166 |
+
if not os.path.exists(new_dir):
|
167 |
+
os.makedirs(new_dir)
|
168 |
+
|
169 |
+
# List all DICOM files in the original directory
|
170 |
+
dicom_files = [os.path.join(original_dir, f) for f in os.listdir(original_dir) if f.endswith('.dcm')]
|
171 |
+
|
172 |
+
# Sort the dicom_files list by SliceLocation
|
173 |
+
dicom_files.sort(key=lambda x: pydicom.dcmread(x).SliceLocation)
|
174 |
+
|
175 |
+
# Loop over each slice of the new volume
|
176 |
+
for i in range(new_volume.shape[0]):
|
177 |
+
# Get the corresponding original DICOM file
|
178 |
+
dicom_file = dicom_files[i]
|
179 |
+
|
180 |
+
# Read the file
|
181 |
+
ds = pydicom.dcmread(dicom_file)
|
182 |
+
|
183 |
+
# Revert the slope and intercept operation on the slice
|
184 |
+
new_slice = (new_volume[i] - ds.RescaleIntercept) / ds.RescaleSlope
|
185 |
+
|
186 |
+
# Update the pixel data with the data from the new slice
|
187 |
+
ds.PixelData = new_slice.astype(np.int16).tobytes()
|
188 |
+
|
189 |
+
# Generate new file name
|
190 |
+
new_file_name = os.path.join(new_dir, f"new_image_{i}.dcm")
|
191 |
+
|
192 |
+
# Save the new DICOM file
|
193 |
+
ds.save_as(new_file_name)
|
194 |
+
|
195 |
+
|
196 |
+
|
197 |
+
def drown_volume(in_path, out_path=None, replacer='face'):
|
198 |
+
"""
|
199 |
+
Processes DICOM files from the provided directory by binarizing, getting the largest connected component,
|
200 |
+
dilating and applying mask. Then applies random values to the dilated volume based on a unique values list
|
201 |
+
obtained from the masked volume (or air value). The results are saved as new DICOM files in a specified directory.
|
202 |
+
|
203 |
+
Parameters:
|
204 |
+
in_path (str): The path to the directory containing the input DICOM files.
|
205 |
+
out_path (str, optional): The path to the directory where the output DICOM files will be saved.
|
206 |
+
If not provided, the output files will be saved in the input directory appended by "_d".
|
207 |
+
replacer (str, optional): Indicates what kind of pixels are going to be replaced. Default is 'face'.
|
208 |
+
'face': replaces air and face with random values that are found in the skin and subcutaneous fat.
|
209 |
+
'air': replaces air and face with -1000 HU.
|
210 |
+
int: replaces air and face with int HU.
|
211 |
+
|
212 |
+
Returns:
|
213 |
+
None. The function saves new DICOM files and prints the total elapsed time of the operation.
|
214 |
+
"""
|
215 |
+
start_time = time.time()
|
216 |
+
|
217 |
+
if out_path is None:
|
218 |
+
out_path = '_d'
|
219 |
+
|
220 |
+
|
221 |
+
dirs = list_dicom_directories(in_path)
|
222 |
+
|
223 |
+
for _d in tqdm(dirs):
|
224 |
+
|
225 |
+
with tqdm(total=8, desc="Processing DICOM Files", ncols=80) as pbar:
|
226 |
+
# Load the DICOM files
|
227 |
+
slices = load_scan(_d)
|
228 |
+
pbar.update()
|
229 |
+
|
230 |
+
# Get the pixel values and convert them to Hounsfield Units (HU)
|
231 |
+
pixels_hu = get_pixels_hu(slices)
|
232 |
+
pbar.update()
|
233 |
+
|
234 |
+
# Apply the binarization function on the HU volume
|
235 |
+
binarized_volume = binarize_volume(pixels_hu)
|
236 |
+
pbar.update()
|
237 |
+
|
238 |
+
# Get the largest connected component from the binarized volume
|
239 |
+
processed_volume = get_largest_component_volume(binarized_volume)
|
240 |
+
pbar.update()
|
241 |
+
|
242 |
+
# Dilate the processed volume
|
243 |
+
dilated_volume = dilate_volume(processed_volume)
|
244 |
+
pbar.update()
|
245 |
+
if replacer == 'face':
|
246 |
+
# Apply the mask to the original volume and get unique values list
|
247 |
+
unique_values_list = apply_mask_and_get_values(pixels_hu, dilated_volume - processed_volume)
|
248 |
+
elif replacer == 'air':
|
249 |
+
unique_values_list = [0]
|
250 |
+
else:
|
251 |
+
try:
|
252 |
+
replacer = int(replacer)
|
253 |
+
unique_values_list = [replacer]
|
254 |
+
except:
|
255 |
+
print('replacer must be either air, face, or an integer number in Hounsfield units, but ' + str(replacer) + ' was provided.')
|
256 |
+
print('replacing with face')
|
257 |
+
unique_values_list = apply_mask_and_get_values(pixels_hu, dilated_volume - processed_volume)
|
258 |
+
|
259 |
+
pbar.update()
|
260 |
+
|
261 |
+
# Apply random values to the dilated volume based on the unique values list
|
262 |
+
new_volume = apply_random_values_optimized(pixels_hu, dilated_volume, unique_values_list)
|
263 |
+
pbar.update()
|
264 |
+
|
265 |
+
# Save the new DICOM files
|
266 |
+
out_path_n = _d.replace(get_first_directory(_d), get_first_directory(_d) + out_path)
|
267 |
+
save_new_dicom_files(new_volume, _d, out_path_n)
|
268 |
+
pbar.update()
|
269 |
+
|
270 |
+
elapsed_time = time.time() - start_time
|
271 |
+
print(f"Total elapsed time for 1 study: {elapsed_time} seconds")
|