-
Notifications
You must be signed in to change notification settings - Fork 1
/
instance_seg_main.py
405 lines (349 loc) · 20.1 KB
/
instance_seg_main.py
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
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
# -*- coding: utf-8 -*-
'''
* @Author: KD Zhang
* @Date: 2023-03-05 14:30:35
'''
import os
import cv2
import time
import glob
import SimpleITK as sitk
import numpy as np
from scipy.ndimage import gaussian_filter
from scripts.roi_selection import roi_selection, roi_selection_preSegs
from scripts.otsu_thres import otsu_seg
from scripts.find_initial_circle import find_initial_circle_func
from scripts.counter_refine import contour_refine_func
from scripts.active_counter_iteration import active_contour_propagation
import argparse
def read_masks(pre_seg):
pre_seg_masks = sorted(glob.glob(os.path.join(pre_seg, '*.png')))
pre_segs = []
for i in range(len(pre_seg_masks)):
pre_seg_mask = cv2.imread(pre_seg_masks[i])
if len(pre_seg_mask.shape) == 3:
pre_seg_mask = pre_seg_mask[:, :, 0]
pre_seg_mask[pre_seg_mask > 0] = 1
pre_segs.append(pre_seg_mask)
return pre_segs
def rois_selection(series, roi_thres, use_column, use_label):
n = series.shape[0]
left_rois, left_ups, left_bottoms = [], [], []
right_rois, right_ups, right_bottoms = [], [], []
for i in range(n):
sample = series[i]
left_roi, left_up, left_bottom, right_roi, right_up, right_bottom = roi_selection(sample, roi_thres, use_column,
use_label)
left_rois.append(left_roi)
left_ups.append(left_up)
left_bottoms.append(left_bottom)
right_rois.append(right_roi)
right_ups.append(right_up)
right_bottoms.append(right_bottom)
return left_rois, left_ups, left_bottoms, right_rois, right_ups, right_bottoms
def rois_selection_preSegs(pre_segs, use_column, use_label):
left_rois, left_ups, left_bottoms = [], [], []
right_rois, right_ups, right_bottoms = [], [], []
for i in range(len(pre_segs)):
pre_seg = pre_segs[i]
left_roi, left_up, left_bottom, right_roi, right_up, right_bottom = roi_selection_preSegs(pre_seg, use_column,
use_label)
left_rois.append(left_roi)
left_ups.append(left_up)
left_bottoms.append(left_bottom)
right_rois.append(right_roi)
right_ups.append(right_up)
right_bottoms.append(right_bottom)
return left_rois, left_ups, left_bottoms, right_rois, right_ups, right_bottoms
def calc_pre_seg_func(ct_series, left_ups, left_bottoms, right_ups, right_bottoms):
n, h, w = ct_series.shape
assert len(left_ups) == n
assert len(left_bottoms) == n
assert len(right_ups) == n
assert len(right_bottoms) == n
left_binaries, right_binaries = [], []
for i in range(n):
ct = ct_series[i]
left_side, right_side = ct[:, 0:w // 2], ct[:, w // 2:]
left_binary = otsu_seg(left_side, left_ups[i], left_bottoms[i])
right_binary = otsu_seg(right_side, right_ups[i], right_bottoms[i])
left_binaries.append(left_binary)
right_binaries.append(right_binary)
return left_binaries, right_binaries
def proj_segs(segs, left_ups, left_bottoms, right_ups, right_bottoms):
series_len = len(segs)
h, w = segs[0].shape
assert len(left_ups) == series_len
assert len(left_bottoms) == series_len
assert len(right_ups) == series_len
assert len(right_bottoms) == series_len
left_binaries, right_binaries = [], []
for i in range(series_len):
left_binary = segs[i][left_ups[i]:left_bottoms[i], 0:w // 2]
right_binary = segs[i][right_ups[i]:right_bottoms[i], w // 2:]
left_binaries.append(left_binary)
right_binaries.append(right_binary)
return left_binaries, right_binaries
def contour_prop(contour, ch, cw, cr, space_x, space_z, ups, bottoms, series, blurred_ct, initial_plane_idx, down_scale, up_scale,
mask_shape, pros_type='left'):
w = blurred_ct.shape[2]
z_cr = int(cr * space_x / space_z)
blurred_ct_samples = blurred_ct[initial_plane_idx - z_cr:initial_plane_idx + z_cr + 1]
ct_samples = series[initial_plane_idx - z_cr:initial_plane_idx + z_cr + 1]
new_ups = ups[initial_plane_idx - z_cr:initial_plane_idx + z_cr + 1]
new_bottoms = bottoms[initial_plane_idx - z_cr:initial_plane_idx + z_cr + 1]
pivot = z_cr
cropped_blurred_ct_samples = []
cropped_ct_samples = []
for i in range(len(ct_samples)):
if pros_type == 'left':
cropped_blurred_ct_sample = blurred_ct_samples[i, new_ups[i]:new_bottoms[i], 0:w // 2]
cropped_ct_sample = ct_samples[i, new_ups[i]:new_bottoms[i], 0:w // 2]
else:
cropped_blurred_ct_sample = blurred_ct_samples[i, new_ups[i]:new_bottoms[i], w // 2:]
cropped_ct_sample = ct_samples[i, new_ups[i]:new_bottoms[i], w // 2:]
cropped_blurred_ct_sample = np.flip(cropped_blurred_ct_sample, axis=1)
cropped_ct_sample = np.flip(cropped_ct_sample, axis=1)
cropped_blurred_ct_samples.append(cropped_blurred_ct_sample)
cropped_ct_samples.append(cropped_ct_sample)
femoral_head_mask_sequences, ch, cw, radius = active_contour_propagation(contour, ch, cw, cr, space_x, space_z, new_ups, new_bottoms,
cropped_ct_samples,
cropped_blurred_ct_samples, pivot,
down_scale, up_scale, mask_shape)
radius = radius.astype(int)
return femoral_head_mask_sequences, pivot, ch, cw, radius
def writeDicom(bones, ct, output, idx):
# https://simpleitk.readthedocs.io/en/master/link_DicomSeriesReadModifyWrite_docs.html
# https://simpleitk.readthedocs.io/en/master/link_DicomImagePrintTags_docs.html#lbl-print-image-meta-data-dictionary
# https://simpleitk.readthedocs.io/en/v1.1.0/Examples/DicomSeriesReadModifyWrite/Documentation.html
# https://blog.csdn.net/weixin_45069929/article/details/108690566
# IMPORTANT: The CT must be series to get accurate meta data !!!
# Transform from numpy to DICOM
bones = bones[np.newaxis, :, :]
filtered_image = sitk.GetImageFromArray(bones)
filtered_image.SetSpacing(ct.GetSpacing())
filtered_image.SetDirection(ct.GetDirection())
# Initialize writter
writer = sitk.ImageFileWriter()
writer.KeepOriginalImageUIDOn()
modification_time = time.strftime('%H%M%S')
modification_date = time.strftime('%Y%m%d')
for k in ct.GetMetaDataKeys():
if k != '0028|1052' and k != '0028|1053':
filtered_image.SetMetaData(k, ct.GetMetaData(k))
filtered_image.SetMetaData('0008|0031', modification_time)
filtered_image.SetMetaData('0008|0021', modification_date)
filtered_image.SetMetaData('0008|0008', 'DRIVED\\SECONDARY')
# Set a unique UID to this image
filtered_image.SetMetaData('0020|000e',
'1.2.840.113619.2.404.3.3233826508' + modification_date + '.1' + modification_time)
writer.SetFileName(os.path.join(output, 'im{}'.format(idx) + '.dcm'))
writer.Execute(filtered_image)
def seg_perform(pre_segs, pivot, femoral_head_masks, ch, cw, radius, process_type='left'):
def find_bottom(width, mask):
h_pos = np.where(mask[:, width] == 1)
bottom_h = np.max(h_pos)
return bottom_h
h, w = pre_segs[0].shape
cr = len(radius) // 2
if process_type == 'left':
pre_segs = [pre_segs[i][:, 0:w // 2] for i in range(len(pre_segs))]
else:
pre_segs = [np.flip(pre_segs[i][:, w // 2:], axis=1) for i in range(len(pre_segs))]
sampled_pre_segs = pre_segs[pivot - cr:pivot + cr + 1]
assert len(sampled_pre_segs) == len(femoral_head_masks)
femoral_segs, acetabulum_segs = [], []
for i in range(len(femoral_head_masks)):
sampled_pre_seg = sampled_pre_segs[i]
femoral_head_mask = femoral_head_masks[i]
r = radius[i]
valid_width = cw - r
for width in range(cw, cw - r - 1, -1):
valid_height = find_bottom(width, femoral_head_mask)
if np.sum(sampled_pre_seg[valid_height:valid_height + r, width]) <= int(r / 3):
valid_width = width
break
femoral_mask = np.zeros((h, w // 2))
femoral_mask[:, 0:valid_width] = 1
femoral_mask[:, valid_width:] = femoral_head_mask[:, valid_width:]
femoral = sampled_pre_seg * femoral_mask
acetabulum = sampled_pre_seg * (1 - femoral_mask)
femoral_segs.append(femoral)
acetabulum_segs.append(acetabulum)
return femoral_segs, acetabulum_segs
def visualize_masks(output, series, femoral_masks, acetabulum_masks, pivot, process_type='left'):
h, w = series[0].shape
femoral_out_dir = os.path.join(output, process_type, 'vis_masks', 'fem')
acetabulum_out_dir = os.path.join(output, process_type, 'vis_masks', 'ace')
if not os.path.exists(femoral_out_dir):
os.makedirs(femoral_out_dir)
if not os.path.exists(acetabulum_out_dir):
os.makedirs(acetabulum_out_dir)
if process_type == 'right':
femoral_masks = [np.flip(femoral_masks[i], axis=1) for i in range(len(femoral_masks))]
acetabulum_masks = [np.flip(acetabulum_masks[i], axis=1) for i in range(len(acetabulum_masks))]
cropped_series = series[:, :, w // 2:]
else:
cropped_series = series[:, :, 0:w // 2]
cr = len(femoral_masks) // 2
counter = 0
for i in range(pivot - cr, pivot + cr + 1):
femoral = cropped_series[i] * femoral_masks[counter]
acetabulum = cropped_series[i] * acetabulum_masks[counter]
scaled_femoral = (np.maximum(femoral, 0) / femoral.max()) * 255.0
scaled_acetabulum = (np.maximum(acetabulum, 0) / acetabulum.max()) * 255.0
counter += 1
cv2.imwrite(os.path.join(femoral_out_dir, '{:05d}.png'.format(i)), scaled_femoral)
cv2.imwrite(os.path.join(acetabulum_out_dir, '{:05d}.png'.format(i)), scaled_acetabulum)
assert counter == len(femoral_masks) and len(femoral_masks) == len(acetabulum_masks)
def save_masks(pivot, output, masks, process_type, bone_type):
assert process_type in ['left', 'right'], f'Invalid process type: {process_type}'
assert bone_type in ['fem', 'ace'], f'Invalid bone type: {bone_type}'
out_dir = os.path.join(output, process_type, 'masks', bone_type)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
if process_type == 'right':
masks = [np.flip(masks[i], axis=1) for i in range(len(masks))]
cr = len(masks) // 2
for i in range(len(masks)):
mask = masks[i]
idx = pivot - cr + i
cv2.imwrite(os.path.join(out_dir, '{:05d}.png'.format(idx)), mask * 255)
def main(args):
ct_series, pre_seg, output = args.ct_series, args.pre_seg, args.output
sigma, radius = args.sigma, args.radius
use_column, use_label = args.use_column, args.use_label
left_initial_plane_idx, right_initial_plane_idx = args.left_initial_plane_idx, args.right_initial_plane_idx
shrink = args.shrink
calc_pre_seg = args.calc_pre_seg
dilate, circle_peaks, percent = args.dilate, args.circle_peaks, args.percent
down_scale, up_scale = args.down_scale, args.up_scale
if not os.path.exists(output):
os.makedirs(output)
# Step 1: Read and blur the CT sequences
reader = sitk.ImageSeriesReader()
dicom_names = reader.GetGDCMSeriesFileNames(ct_series)
dicom_names = dicom_names[::-1]
reader.SetFileNames(dicom_names)
reader.MetaDataDictionaryArrayUpdateOn()
reader.LoadPrivateTagsOn()
images = reader.Execute()
space_x, space_y, space_z = images.GetSpacing()
assert space_x == space_y, f'Invalid space x {space_x} and space y {space_y}'
series = sitk.GetArrayFromImage(images)
blurred_ct = gaussian_filter(series, sigma=sigma, radius=radius) # blur OK
n, h, w = series.shape
# Step 2: Get the pre-segmentation results
pre_segs = read_masks(pre_seg)
# Step 3: Get the ROIs from the CTs
left_rois, left_ups, left_bottoms, right_rois, right_ups, right_bottoms = rois_selection_preSegs(pre_segs.copy(),
use_column,
use_label)
# Step 4: Get the coarse segmentation in the ROIs
if calc_pre_seg:
# Result: the seg mask of the femoral head is smaller
left_segs, right_segs = calc_pre_seg_func(blurred_ct, left_ups, left_bottoms, right_ups, right_bottoms)
else:
# Result: the seg mask of the femoral head is larger
# Consistent improvement
left_segs, right_segs = proj_segs(pre_segs, left_ups, left_bottoms, right_ups, right_bottoms)
mask_shape = (h, w // 2)
# Step 5: Determine the initial planes of the left and the right sides
if left_initial_plane_idx > 0:
left_initial_plane = left_segs[left_initial_plane_idx]
# Step 6: Find the coarse anatomic circle in the femoral head
left_circle_rr, left_circle_cc, left_ch, left_cw, left_radius = find_initial_circle_func(left_initial_plane,
shrink=args.shrink,
vis=True)
# Step 7: Refine the coarse anatomy circle
initial_left_up = left_ups[left_initial_plane_idx]
initial_left_bottom = left_bottoms[left_initial_plane_idx]
left_initial_array = series[left_initial_plane_idx, initial_left_up:initial_left_bottom, 0:w // 2]
left_initial_blurred_array = blurred_ct[left_initial_plane_idx, initial_left_up:initial_left_bottom, 0:w // 2]
left_contour, left_refined_ch, left_refined_cw, left_refined_cr = contour_refine_func(
left_initial_blurred_array,
left_initial_array, left_ch,
left_cw, left_radius, shrink,
dilate, percent, vis=True,
circle_peaks=circle_peaks)
# Step 8: Propagate the refined contour to the related slices that contain femoral head
left_femoral_heads, left_pivot, left_ch, left_cw, left_radius = contour_prop(left_contour, left_refined_ch,
left_refined_cw, left_refined_cr,
space_x, space_z, left_ups,
left_bottoms, series,
blurred_ct, left_initial_plane_idx,
down_scale, up_scale, mask_shape)
# Step 9: Impose the contours to the ct series
left_femoral_segs, left_acetabulum_segs = seg_perform(pre_segs, left_initial_plane_idx, left_femoral_heads,
left_ch,
left_cw, left_radius, process_type='left')
# Visualize for debug
visualize_masks(output, series, left_femoral_segs, left_acetabulum_segs, left_initial_plane_idx,
process_type='left')
# Save masks
save_masks(left_initial_plane_idx, output, left_femoral_segs, 'left', 'fem')
save_masks(left_initial_plane_idx, output, left_acetabulum_segs, 'left', 'ace')
if right_initial_plane_idx > 0:
right_initial_plane = right_segs[right_initial_plane_idx]
# Horizontal inverse the right side (all processing are based on left side)
hor_inv_right_initial_plane = np.flip(right_initial_plane, axis=1)
# Step 6: Find the coarse anatomic circle in the femoral head
right_circle_rr, right_circle_cc, right_ch, right_cw, right_radius = find_initial_circle_func(
hor_inv_right_initial_plane, shrink=args.shrink)
# Step 7: Refine the coarse anatomy circle
initial_right_up = right_ups[right_initial_plane_idx]
initial_right_bottom = right_bottoms[right_initial_plane_idx]
right_initial_array = series[right_initial_plane_idx, initial_right_up:initial_right_bottom, w // 2:]
right_initial_blurred_array = blurred_ct[right_initial_plane_idx, initial_right_up:initial_right_bottom, w // 2:]
hor_inv_right_initial_array = np.flip(right_initial_array, axis=1)
hor_inv_right_initial_blurred_array = np.flip(right_initial_blurred_array, axis=1)
# Pass debug
right_contour, right_refined_ch, right_refined_cw, right_refined_cr = contour_refine_func(
hor_inv_right_initial_blurred_array, hor_inv_right_initial_array, right_ch, right_cw, right_radius, shrink,
dilate, circle_peaks)
# Step 8: Propagate the refined contour to the related slices that contain femoral head
mask_shape = (h, w // 2)
right_femoral_heads, right_pivot, right_ch, right_cw, right_radius = contour_prop(right_contour, right_refined_ch,
right_refined_cw,
right_refined_cr, space_x,
space_z, right_ups,
right_bottoms, series, blurred_ct,
right_initial_plane_idx,
down_scale, up_scale, mask_shape,
pros_type='right')
# Step 9: Impose the contours to the ct series
right_femoral_segs, right_acetabulum_segs = seg_perform(pre_segs, right_initial_plane_idx, right_femoral_heads,
right_ch, right_cw, right_radius, process_type='right')
# Visualize for debug
visualize_masks(output, series, right_femoral_segs, right_acetabulum_segs, right_initial_plane_idx,
process_type='right')
# Save masks
save_masks(right_initial_plane_idx, output, right_femoral_segs, 'right', 'fem')
save_masks(right_initial_plane_idx, output, right_acetabulum_segs, 'right', 'ace')
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--ct_series', type=str, default='../../../bones/data/third_dicom/DICOM/PA104/ST0/SE2/')
parser.add_argument('--pre_seg', type=str, default='../debug_rets/pa104_st0_se2_preSegs')
parser.add_argument('--output', type=str, default='../debug_rets/pa104_st0_se2_debug_getSpace')
# Blur args
parser.add_argument('--sigma', type=float, default=1.5)
parser.add_argument('--radius', type=int, default=4)
# Roi selection args
parser.add_argument('--use_column', type=int, default=0)
parser.add_argument('--use_label', type=int, default=0)
# Get pre segmentation results manually?
parser.add_argument('--calc_pre_seg', type=int, default=0)
# Initial plane of the left and right sides
parser.add_argument('--left_initial_plane_idx', type=int, default=-1) # 84, 103 for this sequence
parser.add_argument('--right_initial_plane_idx', type=int, default=-1)
# shrink parameter to find the best atonomy circle in the femoral head
parser.add_argument('--shrink', type=float, default=1.7)
# dilate and circle_peaks for the contour refinement
parser.add_argument('--dilate', type=float, default=1.5)
parser.add_argument('--circle_peaks', type=int, default=5)
parser.add_argument('--percent', type=int, default=92)
# down scale and up scale for determing the search range of hough transform
parser.add_argument('--down_scale', type=float, default=0.8)
parser.add_argument('--up_scale', type=float, default=1.2)
args = parser.parse_args()
main(args)