-
Notifications
You must be signed in to change notification settings - Fork 0
/
generate_figures.py
1251 lines (1021 loc) · 47.8 KB
/
generate_figures.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
from itertools import product
from itkwidgets import view
from mpl_toolkits.axes_grid1.inset_locator import zoomed_inset_axes
from pathlib import Path
from scipy.ndimage.morphology import distance_transform_edt
from skimage import io, morphology, segmentation, util
from skimage.color import gray2rgb
from skimage.draw import ellipse
from skimage.exposure import equalize_hist
from skimage.filters import threshold_multiotsu
from skimage.measure import label
from skimage.restoration import denoise_tv_chambolle
from skimage.segmentation import mark_boundaries
from sklearn.metrics import auc, roc_curve
import fullconvnets.constants as const
import itk
import csv
import matplotlib.pyplot as plt
import matplotlib.patches as mpatches
import numpy as np
import os
import warnings
# Setting up the figures appearance.
plt.rcParams['font.family'] = 'monospace'
plt.rcParams['font.size'] = 30
plt.rcParams['axes.labelsize'] = plt.rcParams['font.size']
plt.rcParams['axes.titlesize'] = 1.2*plt.rcParams['font.size']
plt.rcParams['legend.fontsize'] = 0.6*plt.rcParams['font.size']
plt.rcParams['xtick.labelsize'] = plt.rcParams['font.size']
plt.rcParams['ytick.labelsize'] = plt.rcParams['font.size']
# Defining some helping variables.
OFFSET = -15
SEC_TO_HOURS = 3.6E3
LINE_WIDTH = 7
SCATTER_SIZE = 25
FIGURE_SIZE = (15, 12)
BBOX_TO_ANCHOR = (0.1, 1.01)
ZOOM = 3.5
COLOR_TIRAMISU = '#ffae42'
COLOR_TIRAMISU_3D = '#ff5349'
COLOR_UNET = '#8a2be2'
COLOR_UNET_3D = '#0d98ba'
COLORS = [COLOR_TIRAMISU, COLOR_UNET, COLOR_TIRAMISU_3D, COLOR_UNET_3D]
# Setting patches for the legends.
patch_tiramisu = mpatches.Patch(color=COLOR_TIRAMISU, label='Tiramisu')
patch_tiramisu_3d = mpatches.Patch(color=COLOR_TIRAMISU_3D, label='3D Tiramisu')
patch_unet = mpatches.Patch(color=COLOR_UNET, label='U-net')
patch_unet_3d = mpatches.Patch(color=COLOR_UNET_3D, label='3D U-net')
SAVE_FIG_FORMAT = '.pdf'
# Defining network train and predict parameters.
OUTPUT_TRAIN_BASE = Path('/home/alex/pCloudDrive/data/larson_2019/coefficients/output_train')
OUTPUT_TRAIN_TIRAMISU = OUTPUT_TRAIN_BASE/'output.train_tiramisu-67.txt'
OUTPUT_TRAIN_TIRAMISU_3D = OUTPUT_TRAIN_BASE/'output.train_tiramisu_3d-67.txt'
OUTPUT_TRAIN_UNET = OUTPUT_TRAIN_BASE/'output.train_unet.txt'
OUTPUT_TRAIN_UNET_3D = OUTPUT_TRAIN_BASE/'output.train_unet_3d.txt'
OUTPUT_PREDICT_BASE = Path('/home/alex/pCloudDrive/data/larson_2019/coefficients/output_predict')
OUTPUT_PREDICT_TIRAMISU = OUTPUT_PREDICT_BASE/'output.predict_tiramisu-67.txt'
OUTPUT_PREDICT_TIRAMISU_3D = OUTPUT_PREDICT_BASE/'output.predict_tiramisu_3d-67.txt'
OUTPUT_PREDICT_UNET = OUTPUT_PREDICT_BASE/'output.predict_unet.txt'
OUTPUT_PREDICT_UNET_3D = OUTPUT_PREDICT_BASE/'output.predict_unet_3d.txt'
# Individual ROC and AUC samples.
OUTPUT_COMP_BASE = Path('comp_coefficients')
OUTPUT_COMP_TIRAMISU = OUTPUT_COMP_BASE/'tiramisu-67'
OUTPUT_COMP_TIRAMISU_3D = OUTPUT_COMP_BASE/'tiramisu_3d-67'
OUTPUT_COMP_UNET = OUTPUT_COMP_BASE/'unet'
OUTPUT_COMP_UNET_3D = OUTPUT_COMP_BASE/'unet_3d'
ROC_AUC_TIRAMISU = {'232p1_wet': OUTPUT_COMP_TIRAMISU/'rec20160324_055424_232p1_wet_1cm_cont_4097im_1500ms_17keV_13_a.h5-tiramisu_roc_auc.csv',
'232p3_cured': OUTPUT_COMP_TIRAMISU/'rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5-tiramisu_roc_auc.csv',
'232p3_wet': OUTPUT_COMP_TIRAMISU/'rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5-tiramisu_roc_auc.csv',
'244p1_cured': OUTPUT_COMP_TIRAMISU/'rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5-tiramisu_roc_auc.csv',
'244p1_wet': OUTPUT_COMP_TIRAMISU/'rec20160318_223946_244p1_1p5cm_cont__4097im_1500ms_ML17keV_7.h5-tiramisu_roc_auc.csv'}
ROC_AUC_TIRAMISU_3D = {'232p1_wet': OUTPUT_COMP_TIRAMISU_3D/'rec20160324_055424_232p1_wet_1cm_cont_4097im_1500ms_17keV_13_a.h5-tiramisu_3d_roc_auc.csv',
'232p3_cured': OUTPUT_COMP_TIRAMISU_3D/'rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5-tiramisu_3d_roc_auc.csv',
'232p3_wet': OUTPUT_COMP_TIRAMISU_3D/'rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5-tiramisu_3d_roc_auc.csv',
'244p1_cured': OUTPUT_COMP_TIRAMISU_3D/'rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5-tiramisu_3d_roc_auc.csv',
'244p1_wet': OUTPUT_COMP_TIRAMISU_3D/'rec20160318_223946_244p1_1p5cm_cont__4097im_1500ms_ML17keV_7.h5-tiramisu_3d_roc_auc.csv'}
ROC_AUC_UNET = {'232p1_wet': OUTPUT_COMP_UNET/'rec20160324_055424_232p1_wet_1cm_cont_4097im_1500ms_17keV_13_a.h5-unet_roc_auc.csv',
'232p3_cured': OUTPUT_COMP_UNET/'rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5-unet_roc_auc.csv',
'232p3_wet': OUTPUT_COMP_UNET/'rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5-unet_roc_auc.csv',
'244p1_cured': OUTPUT_COMP_UNET/'rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5-unet_roc_auc.csv',
'244p1_wet': OUTPUT_COMP_UNET/'rec20160318_223946_244p1_1p5cm_cont__4097im_1500ms_ML17keV_7.h5-unet_roc_auc.csv'}
ROC_AUC_UNET_3D = {'232p1_wet': OUTPUT_COMP_UNET_3D/'rec20160324_055424_232p1_wet_1cm_cont_4097im_1500ms_17keV_13_a.h5-unet_3d_roc_auc.csv',
'232p3_cured': OUTPUT_COMP_UNET_3D/'rec20160323_093947_232p3_cured_1p5cm_cont_4097im_1500ms_17keV_10.h5-unet_3d_roc_auc.csv',
'232p3_wet': OUTPUT_COMP_UNET_3D/'rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5-unet_3d_roc_auc.csv',
'244p1_cured': OUTPUT_COMP_UNET_3D/'rec20160320_160251_244p1_1p5cm_cont_4097im_1500ms_ML17keV_9.h5-unet_3d_roc_auc.csv',
'244p1_wet': OUTPUT_COMP_UNET_3D/'rec20160318_223946_244p1_1p5cm_cont__4097im_1500ms_ML17keV_7.h5-unet_3d_roc_auc.csv'}
# Ignoring warnings.
warnings.filterwarnings('ignore')
def figure_2():
"""
Figure 2. Exemplifying the methodology using Figure 1 as the input image.
(a) Histogram equalization and TV Chambolle's filtering (parameter:
weight=0.3). (b) Multi Otsu's resulting regions (parameter:
classes=4). Fibers are located within the fourth region (in yellow).
(c) Binary image obtained considering region four in (b) as the region
of interest, and the remaining regions as the background. (d) the
processed region from (c), as shown in Figure 1.
Colormaps: (a, c, d) gray, (b) viridis.
"""
filename = 'figures/.Fig_01-original_larson.tif'
image = io.imread(filename, as_gray=True, plugin=None)
image = util.img_as_float(image)
# Figure 2(a).
image_eq = equalize_hist(image)
image_filt = denoise_tv_chambolle(image_eq, weight=0.6)
plt.figure(figsize=FIGURE_SIZE)
plt.imshow(image_filt, cmap='gray')
plt.savefig('figures/Fig_02a' + SAVE_FIG_FORMAT, bbox_inches='tight')
# Figure 2(b).
thresholds = threshold_multiotsu(image_filt, classes=4)
regions = np.digitize(image_filt, bins=thresholds)
plt.figure(figsize=FIGURE_SIZE)
plt.imshow(regions, cmap='viridis')
plt.savefig('figures/Fig_02b' + SAVE_FIG_FORMAT, bbox_inches='tight')
# Figure 2(c).
img_fibers = morphology.remove_small_objects(regions == 3)
plt.figure(figsize=FIGURE_SIZE)
plt.imshow(img_fibers, cmap='gray')
plt.savefig('figures/Fig_02c' + SAVE_FIG_FORMAT, bbox_inches='tight')
# Figure 2(d).
cut_fibers = _cut_roi(img_fibers)
plt.figure(figsize=FIGURE_SIZE)
plt.imshow(cut_fibers, cmap='gray')
plt.savefig('figures/Fig_02d' + SAVE_FIG_FORMAT, bbox_inches='tight')
# Figure 2(e).
label_fibers, _, _ = segmentation_wusem(cut_fibers,
initial_radius=0,
delta_radius=2,
watershed_line=True)
permutate = np.concatenate((np.zeros(1, dtype='int'),
np.random.permutation(10 ** 4)),
axis=None)
label_random = permutate[label_fibers]
label_mark = mark_boundaries(
plt.cm.nipy_spectral(label_random/label_random.max())[..., :3],
label_img=label_random,
color=(0, 0, 0))
plt.figure(figsize=FIGURE_SIZE)
plt.imshow(label_mark, cmap='gist_stern')
_check_if_folder_exists(folder='./figures')
plt.savefig('figures/Fig_02e' + SAVE_FIG_FORMAT, bbox_inches='tight')
return None
def figure_3():
"""Figure 3. (a) Accuracy and (b) loss through time for each training epoch.
All networks were trained during five epochs, reaching accuracy higher than
0.9 and loss lower than 0.1 on the first training epoch, except for the
two-dimensional U-net. However, 2D U-net is the fastest to finish training,
and reaches the lowest loss between the candidates. We attribute the subtle
loss increase or accuracy decrease on the start of each epoch to the data
augmentation process.
"""
epochs_tiramisu = _split_epochs(filename=OUTPUT_TRAIN_TIRAMISU)
epochs_tiramisu_3d = _split_epochs(filename=OUTPUT_TRAIN_TIRAMISU_3D)
epochs_unet = _split_epochs(filename=OUTPUT_TRAIN_UNET)
epochs_unet_3d = _split_epochs(filename=OUTPUT_TRAIN_UNET_3D)
accuracies_tiramisu, losses_tiramisu, time_tiramisu, _, ends_tiramisu = _return_all_measures(epochs_tiramisu)
accuracies_tiramisu_3d, losses_tiramisu_3d, time_tiramisu_3d, _, ends_tiramisu_3d = _return_all_measures(epochs_tiramisu_3d)
accuracies_unet, losses_unet, time_unet, _, ends_unet = _return_all_measures(epochs_unet)
accuracies_unet_3d, losses_unet_3d, time_unet_3d, _, ends_unet_3d = _return_all_measures(epochs_unet_3d)
# setting patches for the legends.
patch_tiramisu = mpatches.Patch(color=COLOR_TIRAMISU, label='Tiramisu')
patch_tiramisu_3d = mpatches.Patch(color=COLOR_TIRAMISU_3D, label='3D Tiramisu')
patch_unet = mpatches.Patch(color=COLOR_UNET, label='U-net')
patch_unet_3d = mpatches.Patch(color=COLOR_UNET_3D, label='3D U-net')
# Figure 3(a).
fig, ax = plt.subplots(nrows=2, figsize=FIGURE_SIZE)
ax[0] = _plot_individual_measure(time_tiramisu,
accuracies_tiramisu,
ends=ends_tiramisu,
c=COLOR_TIRAMISU,
s=1,
linestyle='--',
ax=ax[0])
ax[0] = _plot_individual_measure(time_unet,
accuracies_unet,
ends=ends_unet,
c=COLOR_UNET,
s=1,
linestyle='--',
ax=ax[0])
ax[0].legend(handles=[patch_tiramisu,
patch_unet,
patch_tiramisu_3d,
patch_unet_3d],
loc='lower left', bbox_to_anchor=BBOX_TO_ANCHOR, ncol=4,
borderaxespad=0, frameon=False)
ax[1] = _plot_individual_measure(time_tiramisu_3d,
accuracies_tiramisu_3d,
ends=ends_tiramisu_3d,
c=COLOR_TIRAMISU_3D,
s=1,
linestyle='--',
ax=ax[1])
ax[1] = _plot_individual_measure(time_unet_3d,
accuracies_unet_3d,
ends=ends_unet_3d,
c=COLOR_UNET_3D,
s=1,
linestyle='--',
ax=ax[1])
fig.text(0.04, 0.5, 'Accuracy', va='center', rotation='vertical')
ax[1].set_xlabel('Time (hours)')
plt.savefig('figures/Fig_03a' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
# Figure 3(b).
fig, ax = plt.subplots(nrows=2, figsize=FIGURE_SIZE)
ax[0] = _plot_individual_measure(time_tiramisu,
losses_tiramisu,
ends=ends_tiramisu,
c=COLOR_TIRAMISU,
s=1,
linestyle='--',
ax=ax[0])
ax[0] = _plot_individual_measure(time_unet,
losses_unet,
ends=ends_unet,
c=COLOR_UNET,
s=1,
linestyle='--',
ax=ax[0])
ax[0].legend(handles=[patch_tiramisu,
patch_unet,
patch_tiramisu_3d,
patch_unet_3d],
loc='lower left', bbox_to_anchor=BBOX_TO_ANCHOR, ncol=4,
borderaxespad=0, frameon=False)
ax[1] = _plot_individual_measure(time_tiramisu_3d,
losses_tiramisu_3d,
ends=ends_tiramisu_3d,
c=COLOR_TIRAMISU_3D,
s=1,
linestyle='--',
ax=ax[1])
ax[1] = _plot_individual_measure(time_unet_3d,
losses_unet_3d,
ends=ends_unet_3d,
c=COLOR_UNET_3D,
s=1,
linestyle='--',
ax=ax[1])
fig.text(0.02, 0.5, 'Loss', va='center', rotation='vertical')
ax[1].set_xlabel('Time (hours)')
_check_if_folder_exists(folder='./figures')
plt.savefig('figures/Fig_03b' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
def figure_4():
"""Figure 4. Accuracy vs. loss on the first epoch. Accuracy surpasses 0.9
and loss is lower than 0.1 for all networks during the first epoch, except
for 2D U-net (loss of 0.23). The large size of the training set and the
similarities in the data are responsible for such numbers. Validation
accuracy and validation loss on the first epoch are represented by diamonds.
"""
epochs_tiramisu = _split_epochs(filename=OUTPUT_TRAIN_TIRAMISU)
epochs_tiramisu_3d = _split_epochs(filename=OUTPUT_TRAIN_TIRAMISU_3D)
epochs_unet = _split_epochs(filename=OUTPUT_TRAIN_UNET)
epochs_unet_3d = _split_epochs(filename=OUTPUT_TRAIN_UNET_3D)
accuracies_tiramisu, losses_tiramisu, _, validation_tiramisu, _ = _return_all_measures(epochs_tiramisu, concatenate=False)
accuracies_unet, losses_unet, _, validation_unet, _ = _return_all_measures(epochs_unet, concatenate=False)
accuracies_tiramisu_3d, losses_tiramisu_3d, _, validation_tiramisu_3d, _ = _return_all_measures(epochs_tiramisu_3d, concatenate=False)
accuracies_unet_3d, losses_unet_3d, _, validation_unet_3d, _ = _return_all_measures(epochs_unet_3d, concatenate=False)
# Figure 4.
_, ax = plt.subplots(figsize=FIGURE_SIZE)
ax_ins = zoomed_inset_axes(ax, zoom=ZOOM-1.5, loc='upper left')
ax = _plot_accuracy_loss(accuracies_tiramisu[0], losses_tiramisu[0],
validation_tiramisu[0], s=1, c=COLOR_TIRAMISU,
ax=ax, ax_ins=ax_ins)
ax = _plot_accuracy_loss(accuracies_unet[0], losses_unet[0],
validation_unet[0], s=1, c=COLOR_UNET, ax=ax,
ax_ins=ax_ins)
ax = _plot_accuracy_loss(accuracies_tiramisu_3d[0], losses_tiramisu_3d[0],
validation_tiramisu_3d[0][::-1], s=1, # acc/loss inverted here!
c=COLOR_TIRAMISU_3D, ax=ax, ax_ins=ax_ins)
ax = _plot_accuracy_loss(accuracies_unet_3d[0], losses_unet_3d[0],
validation_unet_3d[0], s=1, c=COLOR_UNET_3D, ax=ax,
ax_ins=ax_ins)
ax.legend(handles=[patch_tiramisu,
patch_unet,
patch_tiramisu_3d,
patch_unet_3d],
loc='lower left', bbox_to_anchor=BBOX_TO_ANCHOR, ncol=4,
borderaxespad=0, frameon=False)
ax.set_xlabel('Accuracy')
ax.set_ylabel('Loss')
_check_if_folder_exists(folder='./figures')
plt.savefig('figures/Fig_04' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
return None
def figure_5():
"""Figure 5. Mean and standard deviation for prediction times for each sample.
The processing time results are similar to the training ones: 2D U-net and
2D Tiramisu are the fastest architectures to process a sample, predicting in
a whole sample in one hour, in average. 3D Tiramisu, being the slowest, takes
in average more than a day to process one sample.
"""
time_tiramisu = _split_predictions(filename=OUTPUT_PREDICT_TIRAMISU)
time_unet = _split_predictions(filename=OUTPUT_PREDICT_UNET)
time_tiramisu_3d = _split_predictions(filename=OUTPUT_PREDICT_TIRAMISU_3D)
time_unet_3d = _split_predictions(filename=OUTPUT_PREDICT_UNET_3D)
sum_tiramisu = _sum_duration_predictions(time_tiramisu)
sum_unet = _sum_duration_predictions(time_unet)
sum_tiramisu_3d = _sum_duration_predictions(time_tiramisu_3d)
sum_unet_3d = _sum_duration_predictions(time_unet_3d)
width = 0.4 # the width of the bars
# Figure 5.
sum_networks = [sum_tiramisu,
sum_unet,
sum_tiramisu_3d,
sum_unet_3d]
bars_range = np.arange(len(sum_networks))
x_ticks = bars_range
x_labels = ['Tiramisu', 'U-net', '3D Tiramisu', '3D U-net']
fig, ax = plt.subplots(figsize=(16, 10))
ax.set_xticks(x_ticks)
ax.set_xticklabels(x_labels)
# ax.set_yscale('log')
for idx in bars_range:
aux_mean = sum_networks[idx].mean() / 3600
aux_std = sum_networks[idx].std() / 3600
# adjusting significant figures
aux_std = np.round(aux_std+np.modf(aux_mean)[0], decimals=2)
aux_mean = int(aux_mean)
ax.bar(idx, aux_mean, width, yerr=aux_std, label=x_labels,
color=COLORS[idx])
if idx in (bars_range[-1], bars_range[-2]):
ax.annotate(f'{aux_mean}±{aux_std}',
xy=(idx - width / 1.5, aux_mean),
ha='center', va='bottom')
else:
ax.annotate(f'{aux_mean}±{aux_std}',
xy=(idx + width / 1.5, aux_mean),
ha='center', va='bottom')
ax.set_ylabel('Time (hours)')
# Checking if the folder 'figures' exists.
_check_if_folder_exists(folder='./figures')
plt.savefig('figures/Fig_05' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
return None
def figure_6():
""" Figure 6. Receiver operating characteristic (ROC) and area under curve
(AUC) from the comparison between the prediction for each network and the
segmentation made available for five samples by Larson et al. (2019). ROC
curves were calculated to all slices in a dataset; their mean areas and
standard deviation intervals are presented. AUC is larger than 98% in all
comparisons, showing that our predictions are accurate when compared with
Larson et al. semi-supervised method. The 2D versions of U-net and Tiramisu
perform better when compared to their 3D alternatives.
"""
# Figure 6 (a).
SAMPLE_232p1_wet = [ROC_AUC_TIRAMISU['232p1_wet'],
ROC_AUC_UNET['232p1_wet'],
ROC_AUC_TIRAMISU_3D['232p1_wet'],
ROC_AUC_UNET_3D['232p1_wet']]
_, ax = plt.subplots(figsize=FIGURE_SIZE)
ax_ins = zoomed_inset_axes(ax, zoom=ZOOM, loc='lower right')
area_under_curve = []
for idx, filename in enumerate(SAMPLE_232p1_wet):
fp_rate, tp_rate, auc_mean = _read_csv_roc_auc(filename)
area_under_curve.append(auc_mean)
_plot_roc_and_auc(fp_rate, tp_rate, c=COLORS[idx],
linestyle='-', ax=ax, ax_ins=ax_ins)
ax = _add_auc_legend(area_under_curve, ax=ax)
plt.savefig('figures/Fig_06a' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
# Figure 6 (b).
SAMPLE_232p3_cured = [ROC_AUC_TIRAMISU['232p3_cured'],
ROC_AUC_UNET['232p3_cured'],
ROC_AUC_TIRAMISU_3D['232p3_cured'],
ROC_AUC_UNET_3D['232p3_cured']]
_, ax = plt.subplots(figsize=FIGURE_SIZE)
ax_ins = zoomed_inset_axes(ax, zoom=ZOOM, loc='lower right')
area_under_curve = []
for idx, filename in enumerate(SAMPLE_232p3_cured):
fp_rate, tp_rate, auc_mean = _read_csv_roc_auc(filename)
area_under_curve.append(auc_mean)
_plot_roc_and_auc(fp_rate, tp_rate, c=COLORS[idx],
linestyle='-', ax=ax, ax_ins=ax_ins)
ax = _add_auc_legend(area_under_curve, ax=ax)
plt.savefig('figures/Fig_06b' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
# Figure 6 (c).
SAMPLE_232p3_wet = [ROC_AUC_TIRAMISU['232p3_wet'],
ROC_AUC_UNET['232p3_wet'],
ROC_AUC_TIRAMISU_3D['232p3_wet'],
ROC_AUC_UNET_3D['232p3_wet']]
_, ax = plt.subplots(figsize=FIGURE_SIZE)
ax_ins = zoomed_inset_axes(ax, zoom=ZOOM, loc='lower right')
area_under_curve = []
for idx, filename in enumerate(SAMPLE_232p3_wet):
fp_rate, tp_rate, auc_mean = _read_csv_roc_auc(filename)
area_under_curve.append(auc_mean)
_plot_roc_and_auc(fp_rate, tp_rate, c=COLORS[idx],
linestyle='-', ax=ax, ax_ins=ax_ins)
ax = _add_auc_legend(area_under_curve, ax=ax)
plt.savefig('figures/Fig_06c' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
# Figure 6 (d).
SAMPLE_244p1_cured = [ROC_AUC_TIRAMISU['244p1_cured'],
ROC_AUC_UNET['244p1_cured'],
ROC_AUC_TIRAMISU_3D['244p1_cured'],
ROC_AUC_UNET_3D['244p1_cured']]
_, ax = plt.subplots(figsize=FIGURE_SIZE)
ax_ins = zoomed_inset_axes(ax, zoom=ZOOM, loc='lower right')
area_under_curve = []
for idx, filename in enumerate(SAMPLE_244p1_cured):
fp_rate, tp_rate, auc_mean = _read_csv_roc_auc(filename)
area_under_curve.append(auc_mean)
_plot_roc_and_auc(fp_rate, tp_rate, c=COLORS[idx],
linestyle='-', ax=ax, ax_ins=ax_ins)
ax = _add_auc_legend(area_under_curve, ax=ax)
plt.savefig('figures/Fig_06d' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
# Figure 6 (e).
SAMPLE_244p1_wet = [ROC_AUC_TIRAMISU['244p1_wet'],
ROC_AUC_UNET['244p1_wet'],
ROC_AUC_TIRAMISU_3D['244p1_wet'],
ROC_AUC_UNET_3D['244p1_wet']]
_, ax = plt.subplots(figsize=FIGURE_SIZE)
ax_ins = zoomed_inset_axes(ax, zoom=ZOOM, loc='lower right')
area_under_curve = []
for idx, filename in enumerate(SAMPLE_244p1_wet):
fp_rate, tp_rate, auc_mean = _read_csv_roc_auc(filename)
area_under_curve.append(auc_mean)
_plot_roc_and_auc(fp_rate, tp_rate, c=COLORS[idx],
linestyle='-', ax=ax, ax_ins=ax_ins)
ax = _add_auc_legend(area_under_curve, ax=ax)
_check_if_folder_exists(folder='./figures')
plt.savefig('figures/Fig_06e' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
return None
def figure_8():
"""
Notes
-----
Colors from Bang Wong's color-blind friendly colormap. Available at:
https://www.nature.com/articles/nmeth.1618
Wong's map acquired from David Nichols page. Available at:
https://davidmathlogic.com/colorblind/.
"""
# choosing test sample and network.
sample = const.SAMPLE_232p3_wet
network_folder = const.FOLDER_PRED_UNET
# we will return a 10 x 10 matthews matrix; each for a crop
matthews_coefs = np.ones((10, 10))
worst_indexes = np.zeros((10, 10))
# a variable to obtain inlay data.
inlay_data = []
# reading input data.
is_registered = sample['registered_path'] is not None
data_pred, data_gs = _pred_and_goldstd(sample,
folder_prediction=network_folder,
is_registered=is_registered,
is_binary=True)
data_pred = data_pred[slice(*sample['segmentation_interval'])]
# comp_color starts as gray (background).
comp_color = np.ones(
(*data_pred[0].shape, 3)
) * (np.asarray((238, 238, 238)) / 255)
for idx, (img_pred, img_gs) in enumerate(zip(data_pred, data_gs)):
# crop images in 100 (256, 256) pieces.
crop_pred = util.view_as_blocks(img_pred,
block_shape=(256, 256))
crop_gs = util.view_as_blocks(img_gs,
block_shape=(256, 256))
for i, _ in enumerate(crop_pred):
for j, _ in enumerate(crop_pred[i]):
# calculate the Matthews coefficient for each crop.
aux_conf = _confusion_matrix(crop_gs[i, j],
crop_pred[i, j])
aux_matthews = _measure_matthews(aux_conf)
# if smaller than previously, save results.
# restricting aux_matthews > 0.1 due to errors in all-TN regions
if (0.1 < aux_matthews < matthews_coefs[i, j]):
matthews_coefs[i, j] = aux_matthews
worst_indexes[i, j] = idx
aux_comp = _comparison_color(crop_gs[i, j], crop_pred[i, j])
comp_color[i*256:(i+1)*256, j*256:(j+1)*256] = aux_comp
# grab inlay data from crops we want to highlight.
for i, j in [(2, 2), (8, 7)]:
inlay_data.append(comp_color[i*256:(i+1)*256, j*256:(j+1)*256])
# Figure 8(a).
plt.figure(figsize=FIGURE_SIZE)
plt.imshow(comp_color)
for idx in np.arange(start=0, stop=2560, step=256): # according to image
plt.axvline(idx, color='white')
plt.axhline(idx, color='white')
matthews_coefs = np.round(matthews_coefs * 100, decimals=2)
for i, j in product(range(10), repeat=2):
facecolor, textcolor = _label_color(matthews_coefs[j, i])
plt.text(x=i*256 + 30, y=j*256 + 50,
s=str(matthews_coefs[j, i]),
fontsize=8,
color=textcolor,
bbox=dict(facecolor=facecolor, alpha=0.9))
_check_if_folder_exists(folder='./figures')
plt.savefig('figures/Fig_08a' + SAVE_FIG_FORMAT, bbox_inches='tight')
plt.close()
# Figures 8(b, c).
indexes = {0: 'b', 1: 'c'}
for idx in indexes.keys():
plt.figure(figsize=FIGURE_SIZE)
plt.imshow(inlay_data[idx])
_check_if_folder_exists(folder='./figures')
plt.savefig(f'figures/Fig_08{indexes[idx]}' + SAVE_FIG_FORMAT,
bbox_inches='tight')
plt.close()
return None
def figure_9():
"""
Notes
-----
You should run this code inside a Jupyter Notebook.
"""
# reading all images from the sample.
pattern = 'unet/rec20160318_191511_232p3_2cm_cont__4097im_1500ms_ML17keV_6.h5/predict/*.png'
images = io.ImageCollection(pattern)
images = images.concatenate()[:, ::2, ::2]
# slicing the outer part to show inside fibers.
images = images[:, :1000, :]
images = itk.GetImageFromArray(images)
view(images, label_image_blend=0, gradient_opacity=0, shadow=False,
interpolation=False)
return None
def _label_color(coef):
"""
"""
colors = {100: '#ffffff', 90: '#fff5f0', 80: '#fee0d2', 70: '#fcbba1',
60: '#fc9272', 50: '#fb6a4a', 40: '#ef3b2c', 30: '#cb181d',
20: '#a50f15', 10: '#67000d', 0: '#000000'}
percentage = np.asarray(list(colors.keys()))
min_perc = min(percentage[percentage - coef >= 0])
if min_perc > 50:
textcolor = 'black'
else:
textcolor = 'white'
return colors[min_perc], textcolor
def _comparison_color(data_gs, data_pred):
"""
"""
# comp_color starts as gray (background).
comparison = np.ones(
(*data_gs.shape, 3)
) * (np.asarray((238, 238, 238)) / 255)
# calculating true positives, false positives, and false negatives.
true_pos = (data_gs & data_pred)
false_pos = (~data_gs & data_pred)
false_neg = (data_gs & ~data_pred)
# defining colors for the plot.
color_fp = np.asarray((213, 94, 0)) / 255 # vermillion
color_tp = np.asarray((0, 158, 115)) / 255 # bluish green
color_fn = np.asarray((230, 159, 0)) / 255 # orange
# determining where comparison will take TP, FP, and FN values.
np.copyto(comparison, color_fp, where=false_pos[..., np.newaxis])
np.copyto(comparison, color_tp, where=true_pos[..., np.newaxis])
np.copyto(comparison, color_fn, where=false_neg[..., np.newaxis])
return comparison
def _confusion_matrix(data_gs, data_pred):
"""Compares reference and test data to generate a confusion matrix.
Parameters
----------
data_gs : ndarray
Reference binary data (ground truth).
data_pred : ndarray
Test binary data.
Returns
-------
conf_matrix : array
Matrix containing the number of true positives, false positives,
false negatives, and true negatives.
Notes
-----
The values true positive, false positive, false negative, and false
positive are events obtained in the comparison between data_gs and
data_pred:
data_gs: True False
data_pred:
True True positive | False positive
----------------------------------
False False negative | True negative
References
----------
.. [1] Fawcett T. (2006) "An Introduction to ROC Analysis." Pattern
Recognition Letters, 27 (8): 861-874, :DOI:`10.1016/j.patrec.2005.10.010`
.. [2] Google Developers. "Machine Learning Crash Course with TensorFlow
APIs: Classification: True vs. False and Positive vs. Negative."
Available at:
https://developers.google.com/machine-learning/crash-course/classification/true-false-positive-negative
.. [3] Wikipedia. "Confusion matrix." Available at:
https://en.wikipedia.org/wiki/Confusion_matrix
"""
true_pos = (data_gs & data_pred).sum() / data_gs.size
false_pos = (~data_gs & data_pred).sum() / data_gs.size
false_neg = (data_gs & ~data_pred).sum() / data_gs.size
true_neg = (~data_gs & ~data_pred).sum() / data_gs.size
return np.array([[true_pos, false_pos], [false_neg, true_neg]])
def _measure_matthews(conf_matrix):
"""Calculate the Matthews correlation coefficient for a confusion matrix.
Parameters
----------
conf_matrix : array
Matrix containing the number of true positives, false positives,
false_negatives, and true negatives.
Returns
-------
coef_matthews : float
Matthews correlation index for the input confusion matrix.
Notes
-----
Matthews correlation coefficient tends to be more informative than other
confusion matrix measures, such as matthews coefficient and accuracy, when
evaluating binary classification problems: it considers the balance ratios
of true positives, true negatives, false positives, and false negatives.
References
----------
.. [1] Matthews B. W. (1975) "Comparison of the predicted and observed
secondary structure of T4 phage lysozyme." Biochimica et Biophysica
Acta (BBA) - Protein Structure, 405 (2): 442-451,
:DOI:`10.1016/0005-2795(75)90109-9`
.. [2] Chicco D. (2017) "Ten quick tips for machine learning in computational
biology." BioData Mining, 10 (35): 35, :DOI:`10.1186/s13040-017-0155-3`
.. [3] Wikipedia. "Matthews correlation coefficient." Available at:
https://en.wikipedia.org/wiki/Matthews_correlation_coefficient
Examples
--------
>>> from skimage.measures import confusion_matrix
>>> conf_matrix = confusion_matrix(data_true, data_test)
>>> coef_matthews = measure_matthews(conf_matrix)
"""
tr_pos, fl_pos, fl_neg, tr_neg = conf_matrix.ravel()
coef_matthews = (tr_pos * tr_neg - fl_pos * fl_neg) / \
np.sqrt((tr_pos + fl_pos) * (tr_pos + fl_neg) *
(tr_neg + fl_pos) * (tr_neg + fl_neg))
return coef_matthews
def _pred_and_goldstd(sample, folder_prediction, is_registered=False,
is_binary=True):
"""
"""
aux_folder = sample['folder']
if is_registered:
aux_folder += '_REG'
aux_subfolder = const.SUBFOLDER_GOLDSTD_REG
else:
aux_subfolder = const.SUBFOLDER_GOLDSTD
folder_pred = os.path.join(folder_prediction,
aux_folder,
const.SUBFOLDER_PRED,
f'*{const.EXT_PRED}')
folder_goldstd = os.path.join(const.FOLDER_GOLDSTD,
sample['folder'],
aux_subfolder,
f'*{const.EXT_GOLDSTD}')
data_prediction = io.ImageCollection(load_pattern=folder_pred,
load_func=imread_prediction,
is_binary=is_binary)
data_goldstd = io.ImageCollection(load_pattern=folder_goldstd,
load_func=imread_goldstd)
return data_prediction, data_goldstd
def imread_prediction(image, is_binary=True):
"""Auxiliary function intended to be used with skimage.io.ImageCollection.
Returns a binary prediction image — True when image > 0.5, False
when image <= 0.5.
"""
if is_binary:
return util.img_as_float(io.imread(image)) > 0.5
else:
return util.img_as_float(io.imread(image))
def imread_goldstd(image):
"""Auxiliary function intended to be used with skimage.io.ImageCollection.
Helps to read and process Larson et al's gold standard images.
"""
data = io.imread(image)
if np.unique(data).size > 2:
data = data == 217
return util.img_as_bool(data)
def _add_auc_legend(area_under_curve, ax=None):
"""
"""
if ax is None:
_, ax = plt.subplots(figsize=FIGURE_SIZE)
# Setting patches for the legends.
_ = np.round(np.asarray(area_under_curve[0]) * 100, decimals=2)
aux_label = f'Tiramisu (AUC$_\mu$: {_}%)'
leg_auc_tiramisu = mpatches.Patch(color=COLOR_TIRAMISU,
label=aux_label)
_ = np.round(np.asarray(area_under_curve[1]) * 100, decimals=2)
aux_label = f'U-net (AUC$_\mu$: {_}%)'
leg_auc_unet = mpatches.Patch(color=COLOR_UNET,
label=aux_label)
_ = np.round(np.asarray(area_under_curve[2]) * 100, decimals=2)
aux_label = f'3D Tiramisu (AUC$_\mu$: {_}%)'
leg_auc_tiramisu_3d = mpatches.Patch(color=COLOR_TIRAMISU_3D,
label=aux_label)
_ = np.round(np.asarray(area_under_curve[3]) * 100, decimals=2)
aux_label = f'3D U-net (AUC$_\mu$: {_}%)'
leg_auc_unet_3d = mpatches.Patch(color=COLOR_UNET_3D,
label=aux_label)
ax.legend(handles=[leg_auc_tiramisu,
leg_auc_unet,
leg_auc_tiramisu_3d,
leg_auc_unet_3d],
loc='lower left', bbox_to_anchor=BBOX_TO_ANCHOR, ncol=2,
borderaxespad=0, frameon=False)
return ax
def _check_if_folder_exists(folder: str) -> None:
"""Auxiliary function. Check if folder exists and create it if necessary."""
if not os.path.isdir(folder):
os.mkdir(folder)
return None
def segmentation_wusem(image, str_el='disk', initial_radius=10,
delta_radius=5, watershed_line=False):
"""Separates regions on a binary input image using successive
erosions as markers for the watershed algorithm. The algorithm stops
when the erosion image does not have objects anymore.
Parameters
----------
image : (N, M) ndarray
Binary input image.
str_el : string, optional
Structuring element used to erode the input image. Accepts the
strings 'diamond', 'disk' and 'square'. Default is 'disk'.
initial_radius : int, optional
Initial radius of the structuring element to be used in the
erosion. Default is 10.
delta_radius : int, optional
Delta radius used in the iterations:
* Iteration #1: radius = initial_radius + delta_radius
* Iteration #2: radius = initial_radius + 2 * delta_radius,
and so on. Default is 5.
Returns
-------
img_labels : (N, M) ndarray
Labeled image presenting the regions segmented from the input
image.
num_objects : int
Number of objects in the input image.
last_radius : int
Radius size of the last structuring element used on the erosion.
References
----------
.. [1] F.M. Schaller et al. "Tomographic analysis of jammed ellipsoid
packings", in: AIP Conference Proceedings, 2013, 1542: 377-380. DOI:
10.1063/1.4811946.
Examples
--------
>>> from skimage.data import binary_blobs
>>> image = binary_blobs(length=512, seed=0)
>>> img_labels, num_objects, _ = segmentation_wusem(image,
str_el='disk',
initial_radius=10,
delta_radius=3)
"""
rows, cols = image.shape
img_labels = np.zeros((rows, cols))
curr_radius = initial_radius
distance = distance_transform_edt(image)
while True:
aux_se = {
'diamond': morphology.diamond(curr_radius),
'disk': morphology.disk(curr_radius),
'square': morphology.square(curr_radius)
}
str_el = aux_se.get('disk', morphology.disk(curr_radius))
erod_aux = morphology.binary_erosion(image, selem=str_el)
if erod_aux.min() == erod_aux.max():
last_step = curr_radius
break
markers = label(erod_aux)
if watershed_line:
curr_labels = segmentation.watershed(-distance,
markers,
mask=image,
watershed_line=True)
img_labels += curr_labels
else:
curr_labels = segmentation.watershed(-distance,
markers,
mask=image)
img_labels += curr_labels
# preparing for another loop.
curr_radius += delta_radius
# reordering labels.
img_labels = label(img_labels)
# removing small labels.
img_labels, num_objects = label(morphology.remove_small_objects(img_labels),
return_num=True)
return img_labels, num_objects, last_step
def _cut_roi(image, crop_roi=False):
"""
"""
rows, cols = image.shape
rows_c = rows // 2 - 50
cols_c = cols // 2 - 10
radius_a = 900
radius_b = 935
rr, cc = ellipse(rows_c, cols_c, radius_a, radius_b, rotation=np.pi/4)
mask = image < 0
mask[rr, cc] = True
image *= mask
if crop_roi:
image = image[rows_c-radius_a:rows_c+radius_a,
cols_c-radius_b:cols_c+radius_b]
return image
def _plot_accuracy_loss(x, y, validation, c, s=1, ax=None, ax_ins=None):
"""
"""
if ax is None:
_, ax = plt.subplots(figsize=FIGURE_SIZE)
if ax_ins is None:
ax_ins = zoomed_inset_axes(ax, zoom=2.5, loc=2)
ax.scatter(x, y, s=s, c=c, rasterized=True)
ax.scatter(validation[0], validation[1], c=c, s=s*50, marker='D',
edgecolors='w', rasterized=True)
ax_ins.scatter(x, y, s=s*2, c=c, rasterized=True)
ax_ins.scatter(validation[0], validation[1], c=c, s=s*100, marker='D',
edgecolors='w', rasterized=True)
# defining plot limits.
ax.set_xlim([0, 1])
ax.set_ylim([0, 1])
ax.set_yticks([0.0, 0.2, 0.4, 0.6, 0.8])
ax.invert_yaxis()
# defining inset limits.
ax_ins.set_xlim([0.8, 1])
ax_ins.set_ylim([0, 0.2])
ax_ins.set_xticks([1])
ax_ins.set_yticks([0.1])
ax_ins.yaxis.tick_right()
ax_ins.invert_yaxis()
# adding y=1-x.
limits = [
np.min([ax.get_xlim(), ax.get_ylim()]), # min of both axes