-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_network_liver.py
2236 lines (1866 loc) · 135 KB
/
train_network_liver.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
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
from torch.utils.data import Dataset
import torch.nn.functional as F
import numpy as np
from networks import resnet
from networks import resnext
import time
import os
from os import path
import random
from stl import mesh
import SimpleITK as sitk
import cv2
from datetime import datetime
import argparse
import tools
from networks import RegS2Vnet
import sys
# from pytorch_msssim import ssim, ms_ssim, SSIM, MS_SSIM
from scipy import stats
import xml.etree.ElementTree as ET
import pandas as pd
import itk
from dicttoxml import dicttoxml
from xml.dom.minidom import parseString
from monai.config import print_config
from monai.data import ITKReader, ITKWriter
import matplotlib.pyplot as plt
from monai.transforms import (
EnsureChannelFirstd,
Compose,
LoadImaged,
Spacingd,
CenterSpatialCropd,
MaskIntensityd,
ScaleIntensityd,
MapTransform,
SpatialPadd
)
from monai.losses import (
LocalNormalizedCrossCorrelationLoss
)
from monai.data import DataLoader, Dataset, CacheDataset
from monai.config import KeysCollection
import matplotlib.pyplot as plt
import platform
import math
import json
from utils import loss_functions as loss_F
from utils import util_plot
# import test_network2
# import test_network
################
desc = 'Training registration generator'
parser = argparse.ArgumentParser(description=desc)
parser.add_argument('-t', '--training_mode',
type=str,
help="mode of training with different starting points",
default='scratch')
parser.add_argument('-m', '--model_filename',
type=str,
help="name of the pre-trained mode file",
default='None')
parser.add_argument('-l', '--learning_rate',
type=float,
help='Learning rate',
default=1e-7)
parser.add_argument('-d', '--device_no',
type=int,
choices=[0, 1, 2, 3, 4, 5, 6, 7],
help='GPU device number [0-7]',
default=0)
parser.add_argument('-e', '--epochs',
type=int,
help='number of training epochs',
default=500)
parser.add_argument('-b', '--batch_size',
type=int,
help='number of batch size',
default=2)
parser.add_argument('-n', '--network_type',
type=str,
help='choose different network architectures'
'the size of inputs/outputs are the same'
'could be original, resnext101',
default='mynet')
pretrain_model_str = '0213-092230'
networks3D = ['resnext50', 'resnext101', 'densenet121', 'mynet', 'mynet2', 'p3d',
'autoencoder', 'uda']
use_last_pretrained = False
current_epoch = 0
args = parser.parse_args()
device_no = args.device_no
device = torch.device("cuda:{}".format(device_no))
batch_size = args.batch_size
num_epochs = args.epochs
project_dir = os.getcwd()
output_dir = os.path.join(project_dir, "src/outputs_DeepS2VFF_simplified_nondrop_continue")
isExist = os.path.exists(output_dir)
if not isExist:
os.makedirs(output_dir)
DEEP_MODEL = True
NONDEEP_MODEL = False
RESUME_MODEL = True
TRAINED_MODEL = 5
trained_model_list = {'1': 'FVnet-supervised', '2': 'DeepS2VFF', '3':'DeepRCS2V', '4':"DeepS2VFF_simplified", '5':"DeepS2VFF_simplified_nodrop", '6':"DeepS2VFF_simplified_adddrop"}
# net = 'DeepS2VFF_refine_leakyReLU_localNCC'
net = 'DeepS2VFF_simplified_nondrop'
addNoise = True
print('start device {}'.format(device))
print("network training type: {}".format(net))
print("output directory: {}".format(output_dir))
print("RESUME_MODEL : {}".format(RESUME_MODEL))
print('target training epoches: {}'.format(num_epochs))
print("training batch size: {}".format(batch_size))
print("learning rate: {}".format(args.learning_rate))
print("TRAINED_MODEL : {}".format(trained_model_list[str(TRAINED_MODEL)]))
print("adding noise to data: {}".format(addNoise))
now = datetime.now()
now_str = now.strftime('%m%d_%H%M%S')
print('now_str: {}'.format(now_str))
# create saving file
file = open(os.path.join(output_dir, '{}.txt'.format(now_str)), 'w')
file.close()
def filename_list(dir):
images = []
dir = os.path.expanduser(dir)
print('dir {}'.format(dir))
for filename in os.listdir(dir):
# print(filename)
file_path = path.join(dir, filename)
images.append(file_path)
# print(file_path)
# print(images)
return images
def save_info():
file = open('infos/experiment_diary/{}.txt'.format(now_str), 'a+')
file.write('Time_str: {}\n'.format(now_str))
# file.write('Initial_mode: {}\n'.format(args.init_mode))
file.write('Training_mode: {}\n'.format(args.training_mode))
file.write('Model_filename: {}\n'.format(args.model_filename))
file.write('Device_no: {}\n'.format(args.device_no))
file.write('Epochs: {}\n'.format(args.epochs))
file.write('Network_type: {}\n'.format(args.network_type))
file.write('Learning_rate: {}\n'.format(args.learning_rate))
# file.write('Neighbour_slices: {}\n'.format(args.neighbour_slice))
# file.write('Infomation: {}\n'.format(args.information))
file.write('Best_epoch: 0\n')
file.write('Val_loss: {:.4f}\n'.format(1000))
file.close()
print('Information has been saved!')
def calculate_cascaded_transform(theta_list, device):
cascaded_transform = tools.dof2mat_tensor(input_dof=torch.zeros(1, 6)).type(torch.FloatTensor)
cascaded_transform = cascaded_transform.to(device)
for i in range(len(theta_list)):
trans = tools.dof2mat_tensor(input_dof=theta_list[i]).type(torch.FloatTensor)
trans = trans.to(device)
cascaded_transform =torch.matmul(trans, cascaded_transform)
return cascaded_transform
def update_info(best_epoch, current_epoch, lowest_val_TRE, tv_hist):
# readFile = open(os.path.join(output_dir, '{}.txt'.format(now_str)), "w")
# lines = readFile.readlines()
# readFile.close()
loss_combined_train = tv_hist['train'][-1][0]
loss_image_train = tv_hist['train'][-1][1]
loss_dof_train = tv_hist['train'][-1][2]
loss_combined_val = tv_hist['val'][-1][0]
loss_image_val = tv_hist['val'][-1][1]
loss_dof_val = tv_hist['val'][-1][2]
file = open(os.path.join(output_dir, '{}.txt'.format(now_str)), 'a')
file.write('Best_epoch: {}/{}, Val_loss: {:.4f}, loss_combined_train: {:.4f}, loss_image_train: {:.4f}, loss_dof_train: {:.4f}, loss_combined_val: {:.4f}, loss_image_val: {:.4f}, loss_dof_val: {:.4f}\n'.format(best_epoch, current_epoch, lowest_val_TRE, loss_combined_train, loss_image_train, loss_dof_train, loss_combined_val, loss_image_val, loss_dof_val))
file.close()
print('Info updated in {}!'.format(now_str))
def ConvertITKTransform2NumpyArray(transform):
"""ITK transfrom: transform_from_parent_LPS."""
rotation = transform.GetMatrix()
translation = transform.GetTranslation()
center = transform.GetCenter()
rotation_array = np.array([[rotation[0], rotation[1], rotation[2]], [rotation[3], rotation[4], rotation[5]], [rotation[6], rotation[7], rotation[8]]])
offset = -rotation_array@center + center + translation
transform_array = np.array([[rotation[0], rotation[1], rotation[2], offset[0]], [rotation[3], rotation[4], rotation[5], offset[1]], [rotation[6], rotation[7], rotation[8], offset[2]], [0, 0, 0, 1]])
return transform_array
def ConvertNumpyArray2ITKTransform(transform_array):
"""ITK transform: transfrom_from_parent_LPS"""
dimension = 3
transform = sitk.AffineTransform(dimension)
rotation = [transform_array[0][0], transform_array[0][1], transform_array[0][2], transform_array[1][0], transform_array[1][1], transform_array[1][2], transform_array[2][0], transform_array[2][1], transform_array[2][2]]
translation = [transform_array[0][3], transform_array[1][3], transform_array[2][3]]
transform.SetMatrix(rotation)
transform.SetTranslation(translation)
return transform
def CreateLookupTable(cases_metadata, project_dir, phase = 'train', save_flag = True):
project_src_dir = os.path.join(project_dir, "src")
project_data_dir = os.path.join(project_dir, "data")
if phase == 'train':
dataset_dict_fileNAME = os.path.join(project_src_dir, "training_dataset_dict.xml")
volume_dict_fileNAME = os.path.join(project_src_dir, "training_volume_dict.xml")
elif phase == 'val':
dataset_dict_fileNAME = os.path.join(project_src_dir, "validation_dataset_dict.xml")
volume_dict_fileNAME = os.path.join(project_src_dir, "validation_volume_dict.xml")
elif phase == 'test':
dataset_dict_fileNAME = os.path.join(project_src_dir, "test_dataset_dict.xml")
volume_dict_fileNAME = os.path.join(project_src_dir, "test_volume_dict.xml")
else:
dataset_dict_fileNAME = os.path.join(project_src_dir, phase + "_dataset_dict.xml")
volume_dict_fileNAME = os.path.join(project_src_dir, phase + "_volume_dict.xml")
if platform.system() == "Linux":
dataset_dict_fileNAME = '/'.join(dataset_dict_fileNAME.split('\\'))
volume_dict_fileNAME = '/'.join(volume_dict_fileNAME.split('\\'))
num_of_cases = len(cases_metadata) # number of volumes
alldataset_LUT = []
volume_LUT = []
for case_index in range(0, num_of_cases):
path = os.path.join(project_data_dir, cases_metadata[case_index].text)
if platform.system() == 'Linux':
path = '/'.join(path.split('\\'))
case_tree = ET.parse(path)
case_root = case_tree.getroot()
# get the volume name (moving image)
moving_image_metadata = case_root.find('moving_image')
moving_image_fileNAME = os.path.join(project_data_dir, moving_image_metadata.find('directory').text, moving_image_metadata.find('name_raw_US_volume').text)
if platform.system() == 'Linux':
moving_image_fileNAME = '/'.join(moving_image_fileNAME.split('\\'))
# get the mask of volume (moving image)
moving_image_mask_metadata = case_root.find('moving_image_mask')
moving_image_mask_fileNAME = os.path.join(project_data_dir, moving_image_mask_metadata.find('directory').text, moving_image_mask_metadata.find('name_rawdata').text)
if platform.system() == 'Linux':
moving_image_mask_fileNAME = '/'.join(moving_image_mask_fileNAME.split('\\'))
# get the fixed image
fixed_images_metadata = case_root.find('fixed_image')
rangeMin = int(fixed_images_metadata.find('rangeMin').text)
rangeMax = int(fixed_images_metadata.find('rangeMax').text)
# get the mask of fixed image
fixed_image_mask_metadata = case_root.find('fixed_image_mask')
fixed_image_mask_fileNAME = os.path.join(project_data_dir, fixed_image_mask_metadata.find('directory').text, fixed_image_mask_metadata.find('name').text)
if platform.system() == 'Linux':
fixed_image_mask_fileNAME = '/'.join(fixed_image_mask_fileNAME.split('\\'))
volume_case_dict = {'volume_ID': case_index,'volume_name': moving_image_fileNAME, "volume_mask_name": moving_image_mask_fileNAME}
volume_LUT.append(volume_case_dict)
# get the transform (3DUS to CT/MRI)
tfm_3DUS_CT_metadata = case_root.find('transform_3DUS_CT')
tfm_3DUS_CT_fileNAME = os.path.join(project_data_dir, tfm_3DUS_CT_metadata.find('directory').text, tfm_3DUS_CT_metadata.find('name').text)
if platform.system() == 'Linux':
tfm_3DUS_CT_fileNAME = '/'.join(tfm_3DUS_CT_fileNAME.split('\\'))
# get the frame flip flag
US_image_setting_metadata = case_root.find('US_image_setting')
frame_flip_flag = US_image_setting_metadata.find("flip").text
# get the transform metadata (slice to volume registration)
tfm_RegS2V_metadata = case_root.find('slice_to_volume_registration')
for frame_index in range(rangeMin, rangeMax + 1, 2):
num_letters_frame_index = len(str(frame_index))
if num_letters_frame_index == 1:
full_frame_index = '000' + str(frame_index)
elif num_letters_frame_index == 2:
full_frame_index = '00' + str(frame_index)
elif num_letters_frame_index == 3:
full_frame_index = '0' + str(frame_index)
elif num_letters_frame_index == 4:
full_frame_index = str(frame_index)
else :
print("checking out the maximum length of the filename!")
# get the fixed image filename
fixed_image_filename = "Image_" + full_frame_index + ".mha"
fixed_image_fileNAME = os.path.join(project_data_dir, fixed_images_metadata.find('directory').text, fixed_image_filename)
if platform.system() == 'Linux':
fixed_image_fileNAME = '/'.join(fixed_image_fileNAME.split('\\'))
initial_frame_name = "initial_"+ full_frame_index + ".tfm"
correction_frame_name = "correction_" + full_frame_index+ ".tfm"
tfm_RegS2V_initial_fileNAME = os.path.join(project_data_dir, tfm_RegS2V_metadata.find('US_directory').text, tfm_RegS2V_metadata.find('initial_folder').text, initial_frame_name)
tfm_RegS2V_correction_fileNAME = os.path.join(project_data_dir, tfm_RegS2V_metadata.find('US_directory').text, tfm_RegS2V_metadata.find('correction_folder').text, correction_frame_name)
if platform.system() == 'Linux':
tfm_RegS2V_initial_fileNAME = '/'.join(tfm_RegS2V_initial_fileNAME.split('\\'))
tfm_RegS2V_correction_fileNAME = '/'.join(tfm_RegS2V_correction_fileNAME.split('\\'))
if frame_index == rangeMin:
# fixed_image_fileNAME_pre = fixed_image_fileNAME
tfm_RegS2V_correction_fileNAME_pre = tfm_RegS2V_correction_fileNAME
else:
num_letters_frame_index_pre = len(str(frame_index-1))
if num_letters_frame_index_pre == 1:
full_frame_index_pre = '000' + str(frame_index-1)
elif num_letters_frame_index_pre == 2:
full_frame_index_pre = '00' + str(frame_index-1)
elif num_letters_frame_index_pre == 3:
full_frame_index_pre = '0' + str(frame_index-1)
elif num_letters_frame_index_pre == 4:
full_frame_index_pre = str(frame_index-1)
else :
print("checking out the maximum length of the filename!")
# fixed_image_filename_pre = "Image_" + full_frame_index_pre + ".mha"
# fixed_image_fileNAME_pre = os.path.join(fixed_images_metadata.find('directory').text, fixed_image_filename_pre)
correction_frame_name = "correction_" + full_frame_index_pre+ ".tfm"
tfm_RegS2V_correction_fileNAME_pre = os.path.join(project_data_dir, tfm_RegS2V_metadata.find('US_directory').text, tfm_RegS2V_metadata.find('correction_folder').text, correction_frame_name)
if platform.system() == 'Linux':
tfm_RegS2V_correction_fileNAME_pre = '/'.join(tfm_RegS2V_correction_fileNAME_pre.split('\\'))
# case_frame_pair = [case_index, frame_index, frame_index-1]
# case_dict = {'volume_ID': case_index, 'volume_name': moving_image_fileNAME, "volume_mask_name": moving_image_mask_fileNAME,
# "frame_name":fixed_image_fileNAME, "frame_name_pre": fixed_image_fileNAME_pre, 'frame_mask_name': fixed_image_mask_fileNAME,
# "tfm_3DUS_CT_fileNAME": tfm_3DUS_CT_fileNAME, "tfm_RegS2V_initial_fileNAME": tfm_RegS2V_initial_fileNAME, "tfm_RegS2V_correction_fileNAME": tfm_RegS2V_correction_fileNAME,
# "tfm_RegS2V_correction_fileNAME_pre": tfm_RegS2V_correction_fileNAME_pre}
case_dict = {'volume_ID': case_index,
"frame_name":fixed_image_fileNAME,'frame_mask_name': fixed_image_mask_fileNAME,
"tfm_RegS2V": [tfm_3DUS_CT_fileNAME, tfm_RegS2V_initial_fileNAME, tfm_RegS2V_correction_fileNAME, tfm_RegS2V_correction_fileNAME_pre],
"tfm_gt_diff_mat": None, "tfm_gt_diff_dof": None,
"tfm_RegS2V_initial_mat": None, "tfm_RegS2V_gt_mat": None, "frame_flip_flag": frame_flip_flag}
alldataset_LUT.append(case_dict)
data_LUT_np = np.array(alldataset_LUT)
# print("Lookuptable: ", data_LUT_np)
# print("size of lut:", data_LUT_np.shape)
if save_flag:
# dataframe = pd.DataFrame(data_LUT_np, columns=['volume case ID', 'Transform ID (gt)', 'Transform ID (initial)'])
# dataframe.to_csv(r"E:\PROGRAM\Project_PhD\Registration\Deepcode\FVR-Net\dataset_LUT.csv")
alldataset_xml = dicttoxml(alldataset_LUT, custom_root="all_cases")
dom = parseString(alldataset_xml)
dom.writexml( open(dataset_dict_fileNAME, 'w'),
indent="\t",
addindent="\t",
newl='\n')
volume_dict_xml = dicttoxml(volume_LUT, custom_root="volume_cases")
dom = parseString(volume_dict_xml)
dom.writexml( open(volume_dict_fileNAME, 'w'),
indent="\t",
addindent="\t",
newl='\n')
# print(dom.toprettyxml())
return alldataset_LUT, volume_LUT
def CreateLookupTable_new(cases_metadata, project_dir, phase = 'train', save_flag = True, augmented = True, create_augment_data=False):
project_src_dir = os.path.join(project_dir, "src")
project_data_dir = os.path.join(project_dir, "data")
if phase == 'train':
if augmented:
dataset_dict_fileNAME = os.path.join(project_src_dir, "training_dataset_dict_aug.xml")
volume_dict_fileNAME = os.path.join(project_src_dir, "training_volume_dict_aug.xml")
else:
dataset_dict_fileNAME = os.path.join(project_src_dir, "training_dataset_dict.xml")
volume_dict_fileNAME = os.path.join(project_src_dir, "training_volume_dict.xml")
elif phase == 'val':
dataset_dict_fileNAME = os.path.join(project_src_dir, "validation_dataset_dict.xml")
volume_dict_fileNAME = os.path.join(project_src_dir, "validation_volume_dict.xml")
elif phase == 'test':
dataset_dict_fileNAME = os.path.join(project_src_dir, "test_dataset_dict.xml")
volume_dict_fileNAME = os.path.join(project_src_dir, "test_volume_dict.xml")
else:
dataset_dict_fileNAME = os.path.join(project_src_dir, phase + "_dataset_dict.xml")
volume_dict_fileNAME = os.path.join(project_src_dir, phase + "_volume_dict.xml")
if platform.system() == "Linux":
dataset_dict_fileNAME = '/'.join(dataset_dict_fileNAME.split('\\'))
volume_dict_fileNAME = '/'.join(volume_dict_fileNAME.split('\\'))
num_of_cases = len(cases_metadata) # number of volumes
alldataset_LUT = []
volume_LUT = []
for case_index in range(0, num_of_cases):
path = os.path.join(project_data_dir, cases_metadata[case_index].text)
if platform.system() == 'Linux':
path = '/'.join(path.split('\\'))
case_tree = ET.parse(path)
case_root = case_tree.getroot()
# get the volume name (moving image)
moving_image_metadata = case_root.find('moving_image')
moving_image_fileNAME = os.path.join(project_data_dir, moving_image_metadata.find('directory').text, moving_image_metadata.find('name_raw_US_volume').text)
if platform.system() == 'Linux':
moving_image_fileNAME = '/'.join(moving_image_fileNAME.split('\\'))
# get the mask of volume (moving image)
moving_image_mask_metadata = case_root.find('moving_image_mask')
moving_image_mask_fileNAME = os.path.join(project_data_dir, moving_image_mask_metadata.find('directory').text, moving_image_mask_metadata.find('name_rawdata').text)
if platform.system() == 'Linux':
moving_image_mask_fileNAME = '/'.join(moving_image_mask_fileNAME.split('\\'))
# get the fixed image
fixed_images_metadata = case_root.find('fixed_image')
rangeMin = int(fixed_images_metadata.find('rangeMin').text)
rangeMax = int(fixed_images_metadata.find('rangeMax').text)
# get the mask of fixed image
fixed_image_mask_metadata = case_root.find('fixed_image_mask')
fixed_image_mask_fileNAME = os.path.join(project_data_dir, fixed_image_mask_metadata.find('directory').text, fixed_image_mask_metadata.find('name').text)
if platform.system() == 'Linux':
fixed_image_mask_fileNAME = '/'.join(fixed_image_mask_fileNAME.split('\\'))
volume_case_dict = {'volume_ID': case_index,'volume_name': moving_image_fileNAME, "volume_mask_name": moving_image_mask_fileNAME}
volume_LUT.append(volume_case_dict)
# get the transform (3DUS to CT/MRI)
tfm_3DUS_CT_metadata = case_root.find('transform_3DUS_CT')
tfm_3DUS_CT_fileNAME = os.path.join(project_data_dir, tfm_3DUS_CT_metadata.find('directory').text, tfm_3DUS_CT_metadata.find('name').text)
if platform.system() == 'Linux':
tfm_3DUS_CT_fileNAME = '/'.join(tfm_3DUS_CT_fileNAME.split('\\'))
# get the frame flip flag
US_image_setting_metadata = case_root.find('US_image_setting')
frame_flip_flag = US_image_setting_metadata.find("flip").text
# get the transform metadata (slice to volume registration)
tfm_RegS2V_metadata = case_root.find('slice_to_volume_registration')
if augmented:
if create_augment_data:
step = 1 # when not loading data, this is for creating the dataset
else:
step = 2 # this number may change
else:
step = 1
for frame_index in range(rangeMin, rangeMax + 1, step): # add step
num_letters_frame_index = len(str(frame_index))
if num_letters_frame_index == 1:
full_frame_index = '000' + str(frame_index)
elif num_letters_frame_index == 2:
full_frame_index = '00' + str(frame_index)
elif num_letters_frame_index == 3:
full_frame_index = '0' + str(frame_index)
elif num_letters_frame_index == 4:
full_frame_index = str(frame_index)
else :
print("checking out the maximum length of the filename!")
# get the fixed image filename
fixed_image_filename = "Image_" + full_frame_index + ".mha"
fixed_image_fileNAME = os.path.join(project_data_dir, fixed_images_metadata.find('directory').text, fixed_image_filename)
if platform.system() == 'Linux':
fixed_image_fileNAME = '/'.join(fixed_image_fileNAME.split('\\'))
initial_frame_name = "initial_"+ full_frame_index + ".tfm"
correction_frame_name = "correction_" + full_frame_index+ ".tfm"
tfm_RegS2V_initial_fileNAME = os.path.join(project_data_dir, tfm_RegS2V_metadata.find('US_directory').text, tfm_RegS2V_metadata.find('initial_folder').text, initial_frame_name)
tfm_RegS2V_correction_fileNAME = os.path.join(project_data_dir, tfm_RegS2V_metadata.find('US_directory').text, tfm_RegS2V_metadata.find('correction_folder').text, correction_frame_name)
if platform.system() == 'Linux':
tfm_RegS2V_initial_fileNAME = '/'.join(tfm_RegS2V_initial_fileNAME.split('\\'))
tfm_RegS2V_correction_fileNAME = '/'.join(tfm_RegS2V_correction_fileNAME.split('\\'))
if frame_index == rangeMin:
# fixed_image_fileNAME_pre = fixed_image_fileNAME
tfm_RegS2V_correction_fileNAME_pre = tfm_RegS2V_correction_fileNAME
else:
num_letters_frame_index_pre = len(str(frame_index-1))
if num_letters_frame_index_pre == 1:
full_frame_index_pre = '000' + str(frame_index-1)
elif num_letters_frame_index_pre == 2:
full_frame_index_pre = '00' + str(frame_index-1)
elif num_letters_frame_index_pre == 3:
full_frame_index_pre = '0' + str(frame_index-1)
elif num_letters_frame_index_pre == 4:
full_frame_index_pre = str(frame_index-1)
else :
print("checking out the maximum length of the filename!")
# fixed_image_filename_pre = "Image_" + full_frame_index_pre + ".mha"
# fixed_image_fileNAME_pre = os.path.join(fixed_images_metadata.find('directory').text, fixed_image_filename_pre)
correction_frame_name = "correction_" + full_frame_index_pre+ ".tfm"
tfm_RegS2V_correction_fileNAME_pre = os.path.join(project_data_dir, tfm_RegS2V_metadata.find('US_directory').text, tfm_RegS2V_metadata.find('correction_folder').text, correction_frame_name)
if platform.system() == 'Linux':
tfm_RegS2V_correction_fileNAME_pre = '/'.join(tfm_RegS2V_correction_fileNAME_pre.split('\\'))
# case_frame_pair = [case_index, frame_index, frame_index-1]
# case_dict = {'volume_ID': case_index, 'volume_name': moving_image_fileNAME, "volume_mask_name": moving_image_mask_fileNAME,
# "frame_name":fixed_image_fileNAME, "frame_name_pre": fixed_image_fileNAME_pre, 'frame_mask_name': fixed_image_mask_fileNAME,
# "tfm_3DUS_CT_fileNAME": tfm_3DUS_CT_fileNAME, "tfm_RegS2V_initial_fileNAME": tfm_RegS2V_initial_fileNAME, "tfm_RegS2V_correction_fileNAME": tfm_RegS2V_correction_fileNAME,
# "tfm_RegS2V_correction_fileNAME_pre": tfm_RegS2V_correction_fileNAME_pre}
case_dict = {'volume_ID': case_index,
"frame_name":fixed_image_fileNAME,'frame_mask_name': fixed_image_mask_fileNAME,
"tfm_RegS2V": [tfm_3DUS_CT_fileNAME, tfm_RegS2V_initial_fileNAME, tfm_RegS2V_correction_fileNAME, tfm_RegS2V_correction_fileNAME_pre],
"tfm_gt_diff_mat": None, "tfm_gt_diff_dof": None,
"tfm_RegS2V_initial_mat": None, "tfm_RegS2V_gt_mat": None, "frame_flip_flag": frame_flip_flag, "augment_flag": 'False'}
if augmented:
if phase == "train":
if case_index < 15:
case_dict['augment_flag'] = 'False'
else:
case_dict['augment_flag'] = 'True'
alldataset_LUT.append(case_dict)
data_LUT_np = np.array(alldataset_LUT)
# print("Lookuptable: ", data_LUT_np)
# print("size of lut:", data_LUT_np.shape)
if save_flag:
# dataframe = pd.DataFrame(data_LUT_np, columns=['volume case ID', 'Transform ID (gt)', 'Transform ID (initial)'])
# dataframe.to_csv(r"E:\PROGRAM\Project_PhD\Registration\Deepcode\FVR-Net\dataset_LUT.csv")
alldataset_xml = dicttoxml(alldataset_LUT, custom_root="all_cases")
dom = parseString(alldataset_xml)
dom.writexml( open(dataset_dict_fileNAME, 'w'),
indent="\t",
addindent="\t",
newl='\n')
volume_dict_xml = dicttoxml(volume_LUT, custom_root="volume_cases")
dom = parseString(volume_dict_xml)
dom.writexml( open(volume_dict_fileNAME, 'w'),
indent="\t",
addindent="\t",
newl='\n')
# print(dom.toprettyxml())
return alldataset_LUT, volume_LUT
class LoadRegistrationTransformd(MapTransform):
def __init__(self, keys: KeysCollection, allow_missing_keys: bool = False, scale = 1, volume_size = None, loading_data = False, step_size = 2) -> None:
super().__init__(keys, allow_missing_keys)
self.keys = keys
self.scale = scale
self.volume_size = volume_size #"W*H*D"
self.loading_data = loading_data
self.step_size = step_size
def __call__(self, data):
"""Keys tfm_3DUS_CT_fileNAME: ["tfm_3DUS_CT_fileNAME", "tfm_RegS2V_initial_fileNAME", "tfm_RegS2V_correction_fileNAME", "tfm_RegS2V_correction_fileNAME_pre"]"""
for key in self.keys:
tfm_3DUS_CT = sitk.ReadTransform(data[key][0]) # transform (from parent) LPS
# print(tfm_3DUS_CT)
tfm_3DUS_CT_np = ConvertITKTransform2NumpyArray(tfm_3DUS_CT) # from parent
# print("tfm_3DUS_CT", np.linalg.inv(tfm_3DUS_CT_np))
tfm_RegS2V_initial = sitk.ReadTransform(data[key][1]) # transform (from parent) LPS
# print(tfm_RegS2V_initial)
tfm_RegS2V_initial_np = ConvertITKTransform2NumpyArray(tfm_RegS2V_initial) # from parent
# print("tfm_RegS2V_initial", np.linalg.inv(tfm_RegS2V_initial_np))
tfm_RegS2V_correction = sitk.ReadTransform(data[key][2]) # transform (from parent) LPS
tfm_RegS2V_correction_np = ConvertITKTransform2NumpyArray(tfm_RegS2V_correction) # from parent
# print("tfm_RegS2V_correction", np.linalg.inv(tfm_RegS2V_correction_np))
"""add fake scaling and translation, this is for training. since the networks don't support to identify the spacing of the data array, we rescaled the data to spacing 1"""
T_scale = np.array([[self.scale, 0, 0, 0], [0, self.scale, 0, 0], [0, 0, self.scale, 0], [0, 0, 0, 1]]) # LPS to parent
T_scale_inv = np.linalg.inv(T_scale)
frame_fileNAME = data["frame_name"]
# print(frame_fileNAME)
reader = sitk.ImageFileReader()
reader.SetFileName(frame_fileNAME)
reader.LoadPrivateTagsOn()
reader.ReadImageInformation()
spacing = reader.GetSpacing()
size = reader.GetSize()
T_translate = np.array([[1, 0, 0, 0], [0, 1, 0, -spacing[1]*size[1]*0.5], [0, 0, 1, 0], [0, 0, 0, 1]]) # LPS to parent
# print(T_translate)
"""original transform_ITK (ground truth)"""
tfm_RegS2V_gt_np_from = tfm_3DUS_CT_np@tfm_RegS2V_initial_np@tfm_RegS2V_correction_np # from parent
tfm_RegS2V_gt_np_to = np.linalg.inv(tfm_RegS2V_gt_np_from) # (ground truth) to parent
# print("gt transform: {}".format(tfm_RegS2V_gt_np_to))
# print("T_translate: {}".format(T_scale_inv))
"""transform_ITK to fake the pytorch transform (rescaled and recentered the orignal one)"""
tfm_RegS2V_gt_np_to = T_scale@T_translate@tfm_RegS2V_gt_np_to@T_scale_inv # this is the affine_transform_ITK
# print("tfm_RegS2v_gt_np: {}".format(tfm_RegS2V_gt_np_to))
"""get transform_pytorch"""
tfm_RegS2V_gt_pytorch = transform_conversion_ITK_to_pytorch(tfm_RegS2V_gt_np_to, self.volume_size) # this is the affine_transform_pytorch
tfm_RegS2V_gt_mat_tensor = torch.from_numpy(tfm_RegS2V_gt_pytorch).type(torch.FloatTensor)
"""original transform_ITK (initial one)"""
tfm_RegS2V_correction_pre = sitk.ReadTransform(data[key][3]) # transform (from parent) LPS
tfm_RegS2V_correction_np_pre = ConvertITKTransform2NumpyArray(tfm_RegS2V_correction_pre) # from parent
tfm_RegS2V_initial_np_from = tfm_3DUS_CT_np@tfm_RegS2V_initial_np@tfm_RegS2V_correction_np_pre # from parent
tfm_RegS2V_initial_np_to = np.linalg.inv(tfm_RegS2V_initial_np_from) # (initial transform) to parent
# print("initial transform:", tfm_RegS2V_initial_np_to)
# addNoise = True
addAugment = data['augment_flag']
# addAugment = 'True'
# print("addAgument: ", addAugment)
if addAugment == 'True':
# print("addAgument: ", addAugment)
global augmentation_dof_data
global loading_data_index
if self.loading_data:
# print("frame_name: ", data[key][3])
transform_dof_noise = augmentation_dof_data[loading_data_index]
loading_data_index = loading_data_index+ self.step_size
else:
"""adding noise translation: N(mean = 0, std = 3), rotation: N(mean = 0, std = 3)"""
translation_dof_noise = np.array([np.random.uniform(-10, 10), np.random.uniform(-10, 10), np.random.uniform(-5, 5)])
rotation_dof_noise = np.array([np.random.uniform(-5, 5), np.random.uniform(-5, 5), np.random.uniform(-10, 10)])
transform_dof_noise = np.concatenate((translation_dof_noise, rotation_dof_noise), axis =0)
transform_dof_noise = np.expand_dims(transform_dof_noise, axis=0)
augmentation_dof_data = np.append(augmentation_dof_data, transform_dof_noise, axis=0)
transform_dof_noise = np.squeeze(transform_dof_noise, axis = 0)
# print("transform_dof_noise:{}".format(transform_dof_noise.shape))
# print("transform_dof_noise:{}".format(transform_dof_noise))
# print("augmentation_dof_data: {}".format(augmentation_dof_data))
transform_noise_np = tools.dof2mat_np(transform_dof_noise) # 4 by 4 matrix
# print("transform_noise_np:{}".format(transform_noise_np))
"""transform_ITK to fake the pytorch transform (rescaled and recentered the orignal one)"""
tfm_RegS2V_initial_np_to = T_scale@transform_noise_np@T_translate@tfm_RegS2V_initial_np_to@T_scale_inv # this is the affine_transform_ITK
else:
"""transform_ITK to fake the pytorch transform (rescaled and recentered the orignal one)"""
tfm_RegS2V_initial_np_to = T_scale@T_translate@tfm_RegS2V_initial_np_to@T_scale_inv # this is the affine_transform_ITK
# if addNoise:
# """adding noise translation: N(mean = 0, std = 3), rotation: N(mean = 0, std = 3)"""
# translation_dof_noise = np.random.normal(0, 3, (3))
# rotation_dof_noise = np.random.normal(0, 1.5, (3))
# transform_dof_noise = np.concatenate((translation_dof_noise, rotation_dof_noise), axis =0)
# # print("transform_dof_noise:{}".format(transform_dof_noise))
# transform_noise_np = tools.dof2mat_np(transform_dof_noise) # 4 by 4 matrix
# # print("transform_noise_np:{}".format(transform_noise_np))
# """transform_ITK to fake the pytorch transform (rescaled and recentered the orignal one)"""
# tfm_RegS2V_initial_np_to = T_scale@transform_noise_np@T_translate@tfm_RegS2V_initial_np_to@T_scale_inv # this is the affine_transform_ITK
# else:
# """transform_ITK to fake the pytorch transform (rescaled and recentered the orignal one)"""
# tfm_RegS2V_initial_np_to = T_scale@T_translate@tfm_RegS2V_initial_np_to@T_scale_inv # this is the affine_transform_ITK
# """transform_ITK to fake the pytorch transform (rescaled and recentered the orignal one)"""
# tfm_RegS2V_initial_np_to = T_scale@T_translate@tfm_RegS2V_initial_np_to@T_scale_inv # this is the affine_transform_ITK
"""get transform_pytorch"""
tfm_RegS2V_initial_pytorch = transform_conversion_ITK_to_pytorch(tfm_RegS2V_initial_np_to, self.volume_size) # this is the affine_transform_pytorch
tfm_RegS2V_initial_mat_tensor = torch.from_numpy(tfm_RegS2V_initial_pytorch).type(torch.FloatTensor)
"""caculate the different between the initial one and ground truth, which is for training"""
# # """this is based on the transform_ITK, we decided to use the transform_pytorch to train our model, see below"""
# tfm_gt_diff_mat = [email protected](tfm_RegS2V_initial_np_to)
# tfm_gt_diff_mat_tensor = torch.from_numpy(tfm_gt_diff_mat)
# tfm_gt_diff_dof = tools.mat2dof_np(input_mat=tfm_gt_diff_mat)
# tfm_gt_diff_dof_tensor = torch.from_numpy(tfm_gt_diff_dof[:6])
# # tfm_RegS2V_initial_mat_tensor = torch.from_numpy(tfm_RegS2V_initial_np_to)
"""Note: converting to ITK transform needs to be inversed"""
tfm_gt_diff_mat_normalized = [email protected](tfm_RegS2V_initial_pytorch) #
# tfm_gt_diff_mat_normalized = np.linalg.inv(tfm_RegS2V_initial_pytorch) @ tfm_RegS2V_gt_pytorch# need to be inversed for normalized transform
tfm_gt_diff_mat_tensor_normalized = torch.from_numpy(tfm_gt_diff_mat_normalized).type(torch.FloatTensor)
tfm_gt_diff_dof_rad_normalized = tools.mat2dof_tensor(tfm_gt_diff_mat_tensor_normalized, degree = 'rad')
# dof_deg = tools.mat2dof_tensor(tfm_gt_diff_mat_tensor_normalized, degree = 'deg')
# matrix = tools.dof2mat_tensor_normalized(dof_rad)
#################test########################
# print("mat_tensor: ", tfm_gt_diff_mat_tensor_normalized)
# print("mat_tensor * ini: ", tfm_gt_diff_mat_tensor_normalized @ tfm_RegS2V_initial_mat_tensor)
# print("gt_tensor: ", tfm_RegS2V_gt_mat_tensor)
# tfm_gt_diff_mat = tfm_RegS2V_gt_np_to @ np.linalg.inv(tfm_RegS2V_initial_np_to)
# tfm_gt_diff_mat_tensor = torch.from_numpy(tfm_gt_diff_mat)
# # print("diff_gt_mat (1): {}".format(tfm_gt_diff_mat_tensor))
# # vol_size = [200, 160, 120]
# correction_transform_ITK_LPS = transform_conversion_pytorch_to_tfm_ITK_LPS(tfm_gt_diff_mat_tensor_normalized, self.volume_size)
# dof_est_unnormalized = tools.mat2dof_tensor(correction_transform_ITK_LPS, degree = 'rad')
# print("dof_est_unnormalized(1): {}".format(dof_est_unnormalized))
# print("dof_est_unnormalized(2): {}".format(tools.mat2dof_tensor(tfm_gt_diff_mat_tensor, degree = 'rad')))
# # print("tfm_gt_diff_mat:{}".format(tfm_gt_diff_mat))
# tfm_gt_diff_dof_normalized = tools.mat2dof_np(input_mat=tfm_gt_diff_mat_normalized)
# tfm_gt_diff_dof_tensor_normalized = torch.from_numpy(tfm_gt_diff_dof_normalized[:6])
# print("tfm_gt_diff_dof_tensor_normalized: ", tfm_gt_diff_dof_tensor_normalized)
# print("dof_est_unnormalized:{}".format(dof_est_unnormalized))
# print("tfm_gt_diff_dof_tensor_normalized:{}".format(tfm_gt_diff_dof_tensor_normalized))
# print("diff_gt_mat(2): {}".format(correction_transform_ITK_LPS))
data["tfm_gt_diff_mat"] = tfm_gt_diff_mat_tensor_normalized # difference normlized by affine_grid
data["tfm_gt_diff_dof"] = tfm_gt_diff_dof_rad_normalized # difference normlized by affine_grid
data["tfm_RegS2V_initial_mat"] = tfm_RegS2V_initial_mat_tensor # difference normlized by affine_grid
data["tfm_RegS2V_gt_mat"] = tfm_RegS2V_gt_mat_tensor # difference normlized by affine_grid
return data
def transform_conversion_pytorch_to_ITK(affine_transform_pytorch_initial, affine_transform_pytorch_correction, volume_size):
#note: affine_transform_ITK (to parent)
""" to calucate the affine_transorm_ITK, when you get the output from our network"""
"""our network is to cacluate the correction transform, assuming we already have a good iniital transform (affine_transform_ITK_initial)"""
affine_transform_pytorch = affine_transform_pytorch_correction@affine_transform_pytorch_initial
T_normalized = np.array([[2/volume_size[0], 0, 0, 0], [0, 2/volume_size[1], 0, 0], [0, 0, 2/volume_size[2], 0], [0, 0, 0, 1]]) # LPS to parent
affine_transform_ITK = np.linalg.inv(T_normalized) @ affine_transform_pytorch @ T_normalized
return affine_transform_ITK
def transform_conversion_pytorch_to_original_ITK(affine_transform_pytorch_initial, affine_transform_pytorch_correction, volume_size, frame_spacing, frame_size, scale):
# TODO: need to test the correctness
#note: affine_transform_ITK (to parent); affine_transform_original_ITK (to parent)
""" to calucate the affine_transorm_ITK, when you get the output from our network"""
"""our network is to cacluate the correction transform, assuming we already have a good iniital transform (affine_transform_ITK_initial)"""
affine_transform_pytorch = affine_transform_pytorch_correction@affine_transform_pytorch_initial
T_normalized = np.array([[2/volume_size[0], 0, 0, 0], [0, 2/volume_size[1], 0, 0], [0, 0, 2/volume_size[2], 0], [0, 0, 0, 1]]) # LPS to parent
affine_transform_ITK = np.linalg.inv(T_normalized) @ affine_transform_pytorch @ T_normalized
T_scale = np.array([[scale, 0, 0, 0], [0, scale, 0, 0], [0, 0, scale, 0], [0, 0, 0, 1]]) # LPS to parent
T_scale_inv = np.linalg.inv(T_scale)
T_translate = np.array([[1, 0, 0, 0], [0, 1, 0, -frame_spacing[1]*frame_size[1]*0.5], [0, 0, 1, 0], [0, 0, 0, 1]]) # LPS to parent
T_translate_inv = np.linalg.inv(T_translate)
#tfm_RegS2V_gt_np_to = T_scale@T_translate@tfm_RegS2V_gt_np_to@T_scale_inv
affine_transform_original_ITK = T_translate_inv@T_scale_inv@affine_transform_ITK@T_scale
return affine_transform_original_ITK
def transform_conversion_dof_normalized_to_ITK(dof_normalized, vol_size, device):
mat_normalized = tools.dof2mat_tensor_normalized(dof_normalized)
mat_normalized = mat_normalized.to(device)
transform_ITK_LPS = transform_conversion_pytorch_to_tfm_ITK_LPS(mat_normalized, vol_size, device=device)
dof_ITK = tools.mat2dof_tensor(transform_ITK_LPS, degree = 'deg')
dof_ITK = dof_ITK.to(device)
return dof_ITK
def transform_conversion_ITK_to_pytorch(affine_transform_ITK, volume_size):
# note: affine_transform_ITK (to parent)
"""to achieve the conversion of transform from the ITK transformation to pytorch transformation"""
"""For the input and output objects, both of them are rescaled to spacing (1mm, 1mm, 1mm) and recentered the volume, which is to fake the pytorch transform"""
"""For the pytorch, the input and output are the torches, which dont have the spacing information, so we can treat them as 1."""
"""To understand how the transform_ITK and transform_pytorch can work well and know the relationship, we need to bring anther package (Scipy) to help us"""
"""The conversion workflow is: transform_ITK -> transform_scipy -> transform_pytorch"""
"""For the scipy, the input and outout dont have the spacing information, which are the same as the pytorch transform. Differently, the input and output are represented as [1, 2,..., N_w] by [1, 2, ..., N_h] by [1, 2, ..., N_d].
And the transform_scipy is applied around the first voxel [0, 0, 0] by default. If the volume center is set as the (0, 0, 0) in spatial domain, we need to translate the transform_ITK to the center of volume first T_translate,
then apply the transform_ITK, after that, we need to translate back the object."""
"""For the pytorch, the transformation is similiar as the transform_scipy. Differently, the input and ouput need to be normalized first. That means that instead of representing by [1, 2,..., N_w] by [1, 2, ..., N_h] by [1, 2, ..., N_d],
the input and output are represented by [-1,,..., 0, ..., 1] by [-1,,..., 0, ..., 1] by [-1,,..., 0, ..., 1]. For the numpy array input, the grid_sample will recoginize them as
[-N_w/2, ..., -2, -1, 0, 1, 2, ..., N_w/2] by [-N_h/2, ..., -2, -1, 0, 1, 2, ..., N_h/2] by [-N_d/2, ..., -2, -1, 0, 1, 2, ..., N_d/2]"""
# print(image_array_volume)
# affine_transform_ITK = data_2DUS[21]["tfm_RegS2V_gt_mat"].numpy()
# affine_transform = np.array([[np.cos(45*np.pi/180), -np.sin(45*np.pi/180), 0, 0], [np.sin(45*np.pi/180), np.cos(45*np.pi/180), 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]]) # LPS to parent
affine_transform_inverse = np.linalg.inv(affine_transform_ITK) # for the grid_sampling, the transform is inversed
"""transform_scipy -> transform_pytorch"""
# T_normalized = np.array([[2/volume_size[0], 0, 0, -1], [0, 2/volume_size[1], 0, -1], [0, 0, 2/volume_size[2], -1], [0, 0, 0, 1]]) # LPS to parent
T_normalized = np.array([[2/volume_size[0], 0, 0, 0], [0, 2/volume_size[1], 0, 0], [0, 0, 2/volume_size[2], 0], [0, 0, 0, 1]]) # LPS to parent
affine_transform_pytorch = T_normalized @ affine_transform_inverse @ np.linalg.inv(T_normalized)
return affine_transform_pytorch # 4 by 4
def transform_conversion_ITK_torch_to_pytorch(affine_transform_ITK_torch, volume_size):
# note: affine_transform_ITK (to parent)
"""to achieve the conversion of transform from the ITK transformation to pytorch transformation"""
"""For the input and output objects, both of them are rescaled to spacing (1mm, 1mm, 1mm) and recentered the volume, which is to fake the pytorch transform"""
"""For the pytorch, the input and output are the tensors, which don't have the spacing information, so we can treat them as 1."""
"""To understand how the transform_ITK and transform_pytorch can work well and know the relationship, we need to bring anther package (Scipy) to help us"""
"""The conversion workflow is: transform_ITK -> transform_scipy -> transform_pytorch"""
"""For the scipy, the input and outout dont have the spacing information neither, which are the same as the pytorch transform. Differently, the input and output are represented as [1, 2,..., N_w] by [1, 2, ..., N_h] by [1, 2, ..., N_d].
And the transform_scipy is applied around the first voxel [0, 0, 0] by default. If the volume center is set as the (0, 0, 0) in spatial domain, we need to translate the transform_ITK to the center of volume first T_translate,
then apply the transform_ITK, after that, we need to translate back the object."""
"""For the pytorch, the transformation is similiar as the transform_scipy. Differently, the input and ouput need to be normalized first. That means that instead of representing by [1, 2,..., N_w] by [1, 2, ..., N_h] by [1, 2, ..., N_d],
the input and output are represented by [-1,,..., 0, ..., 1] by [-1,,..., 0, ..., 1] by [-1,,..., 0, ..., 1]. For the numpy array input, the grid_sample will recoginize them as
[-N_w/2, ..., -2, -1, 0, 1, 2, ..., N_w/2] by [-N_h/2, ..., -2, -1, 0, 1, 2, ..., N_h/2] by [-N_d/2, ..., -2, -1, 0, 1, 2, ..., N_d/2]"""
# print(image_array_volume)
# affine_transform_ITK = data_2DUS[21]["tfm_RegS2V_gt_mat"].numpy()
# affine_transform = np.array([[np.cos(45*np.pi/180), -np.sin(45*np.pi/180), 0, 0], [np.sin(45*np.pi/180), np.cos(45*np.pi/180), 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]]) # LPS to parent
affine_transform_inverse = torch.linalg.inv(affine_transform_ITK_torch)
# affine_transform_inverse = np.linalg.inv(affine_transform_ITK) # for the grid_sampling, the transform is inversed
"""transform_scipy -> transform_pytorch"""
# T_normalized = np.array([[2/volume_size[0], 0, 0, -1], [0, 2/volume_size[1], 0, -1], [0, 0, 2/volume_size[2], -1], [0, 0, 0, 1]]) # LPS to parent
# T_normalized = np.array([[2/volume_size[0], 0, 0, 0], [0, 2/volume_size[1], 0, 0], [0, 0, 2/volume_size[2], 0], [0, 0, 0, 1]]) # LPS to parent
T_normalized = torch.tensor([[2/volume_size[0], 0.0, 0.0, 0.0], [0.0, 2/volume_size[1], 0.0, 0.0], [0.0, 0.0, 2/volume_size[2], 0.0], [0.0, 0.0, 0.0, 1.0]]).type(torch.FloatTensor)
# affine_transform_pytorch = T_normalized @ affine_transform_inverse @ np.linalg.inv(T_normalized)
affine_transform_pytorch = T_normalized @ affine_transform_inverse @ torch.linalg.inv(T_normalized)
return affine_transform_pytorch # 4 by 4
def transform_conversion_pytorch_to_tfm_ITK_LPS(tranform_normalized, volume_size, device = "cpu"):
T_normalized = torch.tensor([[2/volume_size[0], 0.0, 0.0, 0.0], [0.0, 2/volume_size[1], 0.0, 0.0], [0.0, 0.0, 2/volume_size[2], 0.0], [0.0, 0.0, 0.0, 1.0]]).type(torch.FloatTensor)
T_normalized = T_normalized.to(device)
affine_transform_inverse = torch.linalg.inv(T_normalized) @ tranform_normalized @ T_normalized
affine_transform_ITK_LPS = torch.linalg.inv(affine_transform_inverse)
return affine_transform_ITK_LPS
def transform_conversion_ITK_to_scipy(affine_transform_ITK, volume_size):
#note: affine_transform_ITK (to parent)
"""to achieve the conversion of transform from the ITK transformation to pytorch transformation"""
"""For the input and output objects, both of them are rescaled to spacing (1mm, 1mm, 1mm) and recentered the volume, which is to fake the pytorch transform"""
"""For the pytorch, the input and output are the torches, which dont have the spacing information, so we can treat them as 1."""
"""To understand how the transform_ITK and transform_pytorch can work well and know the relationship, we need to bring anther package (Scipy) to help us"""
"""The conversion workflow is: transform_ITK -> transform_scipy -> transform_pytorch"""
"""For the scipy, the input and outout dont have the spacing information, which are the same as the pytorch transform. Differently, the input and output are represented as [1, 2,..., N_w] by [1, 2, ..., N_h] by [1, 2, ..., N_d].
And the transform_scipy is applied around the first voxel [0, 0, 0] by default. If the volume center is set as the (0, 0, 0) in spatial domain, we need to translate the transform_ITK to the center of volume first T_translate,
then apply the transform_ITK, after that, we need to translate back the object."""
"""For the pytorch, the transformation is similiar as the transform_scipy. Differently, the input and ouput need to be normalized first. That means that instead of representing by [1, 2,..., N_w] by [1, 2, ..., N_h] by [1, 2, ..., N_d],
the input and output are represented by [-1,,..., 0, ..., 1] by [-1,,..., 0, ..., 1] by [-1,,..., 0, ..., 1]. Therefore, based on transform_scipy, we need to renormalized the input first, then apply transform_scipy, after that re-normalized back the ojects"""
# print(image_array_volume)
# affine_transform_ITK = data_2DUS[21]["tfm_RegS2V_gt_mat"].numpy()
# affine_transform = np.array([[np.cos(45*np.pi/180), -np.sin(45*np.pi/180), 0, 0], [np.sin(45*np.pi/180), np.cos(45*np.pi/180), 0, 0], [0, 0, 1, 0], [0, 0, 0, 1]]) # LPS to parent
"""transform_ITK -> transform_scipy"""
T_translate = np.array([[1, 0, 0, -volume_size[0]*0.5], [0, 1, 0, -volume_size[1]*0.5], [0, 0, 1, -volume_size[2]*0.5], [0, 0, 0, 1]]) # LPS to parent
affine_transform_scipy = np.linalg.inv(T_translate)@affine_transform_ITK@T_translate
return affine_transform_scipy # 4 by 4
def train_model(model, training_dataset_frame, training_dataset_volume, validation_dataset_frame, validation_dateset_volume, num_cases):
since = time.time()
loss_mse = nn.MSELoss()
# loss_localNCC = LocalNormalizedCrossCorrelationLoss(spatial_dims=2, kernel_size= 13, kernel_type="rectangular", reduction="mean")
# loss_localNCC = loss_F.LocalNCC(device = device, kernel_size =(71, 71), stride=(1, 1), padding="valid", win_eps= 100)
kernel_size = 91
loss_localNCC = loss_F.LocalNCC_new(device = device, kernel_size =(kernel_size, kernel_size), stride=(2, 2), padding="valid", win_eps= 0.98)
if args.training_mode == 'finetune':
# overwrite the learning rate for finetune
lr = 5e-6
print('Learning rate is overwritten to be {}'.format(lr))
else:
lr = args.learning_rate
print('Learning rate = {}'.format(lr))
optimizer = optim.Adam(model.parameters(), lr=lr)
scheduler = lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.8)
# num_epochs=25
lowest_loss = 2000
best_ep = 0
tv_hist = {'train': [], 'val': []}
for epoch in range(num_epochs):
# global current_epoch
# current_epoch = epoch + 1
# Each epoch has a training and validation phase
for phase in ['train', 'val']:
print('Network is in {}...'.format(phase))
running_loss = 0.0
running_localNCC = 0.0
running_dof = 0.0
if phase == 'train':
# scheduler.step()
model.train()
nth_batch = 0
for i, batch in enumerate(training_dataset_frame):
"""create the batch volume tensor. This way could significantly reduce the computation when preprocessing"""
volume_size = training_dataset_volume[0]['volume_name'].shape
actual_batch_size = len(batch["volume_ID"])
# vol_tensor = torch.zeros(actual_batch_size, volume_size[0], volume_size[1], volume_size[2], volume_size[3])
"""converting from N*C*W*H*D to N*C*D*H*W"""
vol_tensor = torch.zeros(actual_batch_size, volume_size[0], volume_size[3], volume_size[2], volume_size[1])
for i, volume_id in enumerate(batch["volume_ID"]):
# print("volume_id: ", volume_id)
vol_tensor[i, :, :, :, :] = torch.permute(training_dataset_volume[volume_id]['volume_name'].type(torch.FloatTensor), (0, 3, 2, 1))
frame_tensor = torch.permute(batch["frame_name"].type(torch.FloatTensor), (0, 1, 4, 3, 2))
mat_tensor = batch["tfm_gt_diff_mat"].type(torch.FloatTensor)
dof_tensor = batch["tfm_gt_diff_dof"].type(torch.FloatTensor)
# print('vol_tensor {}'.format(vol_tensor.shape))
# print('frame_tensor {}'.format(frame_tensor.shape))
# print('mat_tensor {}'.format(mat_tensor.shape))
# print('dof_tensor {}'.format(dof_tensor.shape))
# sys.exit()
vol_tensor = vol_tensor.to(device)
frame_tensor = frame_tensor.to(device)
mat_tensor = mat_tensor.to(device)
dof_tensor = dof_tensor.to(device)
mat_tensor.require_grad = True
dof_tensor.require_grad = True
frame_tensor_gt = torch.zeros((frame_tensor.shape))
for i, frame_flip_flag in enumerate(batch["frame_flip_flag"]):
if frame_flip_flag == "True":
frame_tensor_gt[i, :,:, :, :] = torch.flip(frame_tensor[i, :, :,:, :], [3])
else:
frame_tensor_gt[i, :, :, :,:] = frame_tensor[i,:, :, :, :]
frame_tensor_gt = frame_tensor_gt.type(torch.FloatTensor).to(device)
# forward
# track history if only in train
with torch.set_grad_enabled(phase == 'train'):
optimizer.zero_grad()
vol_resampled, dof_estimated = model(vol=vol_tensor, frame=frame_tensor_gt, initial_transform = batch['tfm_RegS2V_initial_mat'] ,device=device, training=True) # shape batch_size*6
# print('vol_resampled {}'.format(vol_resampled.shape))
# print('dof_estimated {}'.format(dof_estimated))
# """save the vol_resampled"""
# vol_resampled_= torch.permute(vol_resampled.data, (0, 1, 4, 3, 2))
# vol_resampled_squeezed = vol_resampled_[0, :, :, :, :].squeeze()
# print('vol_resampled_squeezed shape {}'.format(vol_resampled_squeezed.shape))
# writer = ITKWriter(output_dtype=np.float32, affine_lps_to_ras= False)
# writer.set_data_array(vol_resampled_squeezed, channel_dim=None)
# writer.set_metadata(training_dataset_volume[volume_id]['volume_name'].meta, resample=False)
# output_filename = r"E:\PROGRAM\Project_PhD\Registration\Deepcode\test\Pre-Ablation_01_test.mha"
# writer.write(output_filename)
# sys.exit()
"""rotation loss (deg)"""
rotation_loss = loss_mse(dof_estimated[:, 3:], dof_tensor[:, 3:])
"""translation loss (mm)"""
translation_loss = loss_mse(dof_estimated[:, :3], dof_tensor[:, :3])
"""image intensity-based loss (localNCC)"""
frame_estimated = vol_resampled[:,:, int(volume_size[3]*0.5), :, :].to(device)
# print("frame_estimated shape: ", frame_estimated.shape)
frame_tensor_gt_4d = frame_tensor_gt.squeeze(2)
# # print("frame_tensor shape: ", frame_tensor.shape)
# frame_tensor_gt = torch.zeros((frame_tensor_4d.shape))
# for i, frame_flip_flag in enumerate(batch["frame_flip_flag"]):
# if frame_flip_flag == "True":
# frame_tensor_gt[i, :, :, :] = torch.flip(frame_tensor_4d[i, :, :, :], [2])
# else:
# frame_tensor_gt[i, :, :, :] = frame_tensor_4d[i, :, :, :]
# frame_tensor_gt = frame_tensor_gt.type(torch.FloatTensor).to(device)
# print("frame_tensor_gt shape: ", frame_tensor_gt.shape)
##############################################################
# """visualize image (tested)"""
# frame_gt_np = torch.Tensor.numpy(frame_tensor.detach().cpu())
# frame_est_np = torch.Tensor.numpy(frame_estimated.detach().cpu())
# plt.figure("visualize", (12,8))
# plt.subplot(1,2,1)
# plt.title("original image")
# plt.imshow(frame_gt_np[0,0,:,:], cmap="gray")
# plt.subplot(1,2,2)
# plt.title("resampled image")
# plt.imshow(frame_est_np[0,0,:,:], cmap="gray")
# plt.show()
# sys.exit()
image_localNCC_loss, ROI = loss_localNCC(frame_estimated, frame_tensor_gt_4d)
# print("image_localNCC_loss: ", image_localNCC_loss)
# print("image_localNCC_loss device: ", image_localNCC_loss.device)
# coefficients for loss functinos
alpha = 1.0
beta = 1.0