BrainMapperUNet.py 20.9 KB
Newer Older
1
2
3
4
"""Brain Mapper U-Net Architecture

Description:

5
6
    This folder contains the Pytorch implementation of the core U-net architecture.
    This arcitecture predicts functional connectivity rsfMRI from structural connectivity information from dMRI.
7

8
9
10
11
12
13
Usage:

    To use this module, import it and instantiate is as you wish:

        from BrainMapperUNet import BrainMapperUNet
        deep_learning_model = BrainMapperUnet(parameters)
14
15
16
17
18
19

"""

import numpy as np
import torch
import torch.nn as nn
20
import utils.modules as modules
21

22

23
24
class BrainMapperUNet3D(nn.Module):
    """Architecture class BrainMapper 3D U-net.
25
26
27
28
29
30
31
32

    This class contains the pytorch implementation of the U-net architecture underpinning the BrainMapper project.

    Args:
        parameters (dict): Contains information relevant parameters
        parameters = {
            'kernel_heigth': 5
            'kernel_width': 5
33
            'kernel_depth': 5
34
35
36
37
38
39
40
41
42
43
            'kernel_classification': 1
            'input_channels': 1
            'output_channels': 64
            'convolution_stride': 1
            'dropout': 0.2
            'pool_kernel_size': 2
            'pool_stride': 2
            'up_mode': 'upconv'
            'number_of_classes': 1
        }
44

45
46
    Returns:
        probability_map (torch.tensor): Output forward passed tensor through the U-net block
47
    """
48

49
    def __init__(self, parameters):
50
        super(BrainMapperUNet3D, self).__init__()
51

52
53
        original_input_channels = parameters['input_channels']
        original_output_channels = parameters['output_channels']
54

55
        self.encoderBlock1 = modules.EncoderBlock3D(parameters)
56
        parameters['input_channels'] = parameters['output_channels']
57
58
59
60
61
62
63
64
        parameters['output_channels'] = parameters['output_channels'] * 2
        self.encoderBlock2 = modules.EncoderBlock3D(parameters)
        parameters['input_channels'] = parameters['output_channels']
        parameters['output_channels'] = parameters['output_channels'] * 2
        self.encoderBlock3 = modules.EncoderBlock3D(parameters)
        parameters['input_channels'] = parameters['output_channels']
        parameters['output_channels'] = parameters['output_channels'] * 2
        self.encoderBlock4 = modules.EncoderBlock3D(parameters)
65

66
67
68
        parameters['input_channels'] = parameters['output_channels']
        parameters['output_channels'] = parameters['output_channels'] * 2
        self.bottleneck = modules.ConvolutionalBlock3D(parameters)
69

70
71
72
73
74
75
76
77
78
79
80
81
        parameters['input_channels'] = parameters['output_channels']
        parameters['output_channels'] = parameters['output_channels'] // 2
        self.decoderBlock1 = modules.DecoderBlock3D(parameters)
        parameters['input_channels'] = parameters['output_channels']
        parameters['output_channels'] = parameters['output_channels'] // 2
        self.decoderBlock2 = modules.DecoderBlock3D(parameters)
        parameters['input_channels'] = parameters['output_channels']
        parameters['output_channels'] = parameters['output_channels'] // 2
        self.decoderBlock3 = modules.DecoderBlock3D(parameters)
        parameters['input_channels'] = parameters['output_channels']
        parameters['output_channels'] = parameters['output_channels'] // 2
        self.decoderBlock4 = modules.DecoderBlock3D(parameters)
82
83

        parameters['input_channels'] = parameters['output_channels']
84
85
86
87
        self.classifier = modules.ClassifierBlock3D(parameters)

        parameters['input_channels'] = original_input_channels
        parameters['output_channels'] = original_output_channels
88
89

    def forward(self, X):
90
        """Forward pass for 3D U-net
91

92
        Function computing the forward pass through the 3D U-Net
93
94
95
        The input to the function is the dMRI map

        Args:
96
            X (torch.tensor): Input dMRI map, shape = (N x C x D x H x W) 
97
98
99

        Returns:
            probability_map (torch.tensor): Output forward passed tensor through the U-net block
100
        """
101

102
103
        Y_encoder_1, Y_np1, _ = self.encoderBlock1.forward(X)
        Y_encoder_2, Y_np2, _ = self.encoderBlock2.forward(
104
            Y_encoder_1)
105
106
107

        del Y_encoder_1

108
        Y_encoder_3, Y_np3, _ = self.encoderBlock3.forward(
109
            Y_encoder_2)
110
111
112

        del Y_encoder_2

113
        Y_encoder_4, Y_np4, _ = self.encoderBlock4.forward(
114
115
            Y_encoder_3)

116
117
        del Y_encoder_3

118
119
        Y_bottleNeck = self.bottleneck.forward(Y_encoder_4)

120
121
        del Y_encoder_4

122
        Y_decoder_1 = self.decoderBlock1.forward(
123
            Y_bottleNeck, Y_np4)
124

125
        del Y_bottleNeck, Y_np4
126

127
        Y_decoder_2 = self.decoderBlock2.forward(
128
            Y_decoder_1, Y_np3)
129

130
        del Y_decoder_1, Y_np3
131

132
        Y_decoder_3 = self.decoderBlock3.forward(
133
            Y_decoder_2, Y_np2)
134

135
        del Y_decoder_2, Y_np2
136

137
        Y_decoder_4 = self.decoderBlock4.forward(
138
            Y_decoder_3, Y_np1)
139

140
        del Y_decoder_3, Y_np1
141

142
143
        probability_map = self.classifier.forward(Y_decoder_4)

144
145
        del Y_decoder_4

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
        return probability_map

    def save(self, path):
        """Model Saver

        Function saving the model with all its parameters to a given path.
        The path must end with a *.model argument.

        Args:
            path (str): Path string
        """

        print("Saving Model... {}".format(path))
        torch.save(self, path)

    @property
    def test_if_cuda(self):
        """Cuda Test

        This function tests if the model parameters are allocated to a CUDA enabled GPU.

        Returns:
            bool: Flag indicating True if the tensor is stored on the GPU and Flase otherwhise
        """

        return next(self.parameters()).is_cuda

    def predict(self, X, device=0):
        """Post-training Output Prediction

        This function predicts the output of the of the U-net post-training

        Args:
            X (torch.tensor): input dMRI volume
            device (int/str): Device type used for training (int - GPU id, str- CPU)

        Returns:
            prediction (ndarray): predicted output after training

        """
        self.eval()  # PyToch module setting network to evaluation mode

        if type(X) is np.ndarray:
            X = torch.tensor(X, requires_grad=False).type(torch.FloatTensor)
        elif type(X) is torch.Tensor and not X.is_cuda:
            X = X.type(torch.FloatTensor).cuda(device, non_blocking=True)

        # .cuda() call transfers the densor from the CPU to the GPU if that is the case.
        # Non-blocking argument lets the caller bypas synchronization when necessary

        with torch.no_grad():  # Causes operations to have no gradients
            output = self.forward(X)

        _, idx = torch.max(output, 1)

        # We retrieve the tensor held by idx (.data), and map it to a cpu as an ndarray
        idx = idx.data.cpu().numpy()

        prediction = np.squeeze(idx)

        del X, output, idx

        return prediction

210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
    def reset_parameters(self):
        """Parameter Initialization

        This function (re)initializes the parameters of the defined network.
        This function is a wrapper for the reset_parameters() function defined for each module. 
        More information can be found here: https://discuss.pytorch.org/t/what-is-the-default-initialization-of-a-conv2d-layer-and-linear-layer/16055 + https://discuss.pytorch.org/t/how-to-reset-model-weights-to-effectively-implement-crossvalidation/53859 
        An alternative (re)initialization method is described here: https://discuss.pytorch.org/t/how-to-reset-variables-values-in-nn-modules/32639 
        """

        print("Initializing network parameters...")

        for _, module in self.named_children():
            for _, submodule in module.named_children():
                if isinstance(submodule, (torch.nn.PReLU, torch.nn.Dropout3d, torch.nn.MaxPool3d)) == False:
                    submodule.reset_parameters()

        print("Initialized network parameters!")

228
229
# DEPRECATED ARCHITECTURES!

230

231
232
class BrainMapperUNet(nn.Module):
    """Architecture class BrainMapper U-net.
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255

    This class contains the pytorch implementation of the U-net architecture underpinning the BrainMapper project.

    Args:
        parameters (dict): Contains information relevant parameters
        parameters = {
            'kernel_heigth': 5
            'kernel_width': 5
            'kernel_classification': 1
            'input_channels': 1
            'output_channels': 64
            'convolution_stride': 1
            'dropout': 0.2
            'pool_kernel_size': 2
            'pool_stride': 2
            'up_mode': 'upconv'
            'number_of_classes': 1
        }

    Returns:
        probability_map (torch.tensor): Output forward passed tensor through the U-net block
    """

256
    def __init__(self, parameters):
257
        super(BrainMapperUNet, self).__init__()
258

259
        # TODO: currently, architecture based on QuickNAT - need to adjust parameter values accordingly!
260

261
        self.encoderBlock1 = modules.EncoderBlock(parameters)
262
        parameters['input_channels'] = parameters['output_channels']
263
264
265
        self.encoderBlock2 = modules.EncoderBlock(parameters)
        self.encoderBlock3 = modules.EncoderBlock(parameters)
        self.encoderBlock4 = modules.EncoderBlock(parameters)
266

267
        self.bottleneck = modules.ConvolutionalBlock(parameters)
268

269
270
271
272
273
        parameters['input_channels'] = parameters['output_channels'] * 2.0
        self.decoderBlock1 = modules.DecoderBlock(parameters)
        self.decoderBlock2 = modules.DecoderBlock(parameters)
        self.decoderBlock3 = modules.DecoderBlock(parameters)
        self.decoderBlock4 = modules.DecoderBlock(parameters)
274
275

        parameters['input_channels'] = parameters['output_channels']
276
        self.classifier = modules.ClassifierBlock(parameters)
277
278

    def forward(self, X):
279
        """Forward pass for U-net
280

281
        Function computing the forward pass through the U-Net
282
283
284
        The input to the function is the dMRI map

        Args:
285
            X (torch.tensor): Input dMRI map, shape = (N x C x H x W) 
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

        Returns:
            probability_map (torch.tensor): Output forward passed tensor through the U-net block
        """

        Y_encoder_1, Y_np1, pool_indices1 = self.encoderBlock1.forward(X)
        Y_encoder_2, Y_np2, pool_indices2 = self.encoderBlock2.forward(
            Y_encoder_1)

        del Y_encoder_1

        Y_encoder_3, Y_np3, pool_indices3 = self.encoderBlock3.forward(
            Y_encoder_2)

        del Y_encoder_2

        Y_encoder_4, Y_np4, pool_indices4 = self.encoderBlock4.forward(
            Y_encoder_3)

        del Y_encoder_3

        Y_bottleNeck = self.bottleneck.forward(Y_encoder_4)

        del Y_encoder_4

        Y_decoder_1 = self.decoderBlock1.forward(
            Y_bottleNeck, Y_np4, pool_indices4)

        del Y_bottleNeck, Y_np4, pool_indices4
315

316
317
318
319
320
321
322
323
324
325
        Y_decoder_2 = self.decoderBlock2.forward(
            Y_decoder_1, Y_np3, pool_indices3)

        del Y_decoder_1, Y_np3, pool_indices3

        Y_decoder_3 = self.decoderBlock3.forward(
            Y_decoder_2, Y_np2, pool_indices2)

        del Y_decoder_2, Y_np2, pool_indices2

326
        Y_decoder_4 = self.decoderBlock4.forward(
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
            Y_decoder_3, Y_np1, pool_indices1)

        del Y_decoder_3, Y_np1, pool_indices1

        probability_map = self.classifier.forward(Y_decoder_4)

        del Y_decoder_4

        return probability_map

    def save(self, path):
        """Model Saver

        Function saving the model with all its parameters to a given path.
        The path must end with a *.model argument.

        Args:
            path (str): Path string
        """

        print("Saving Model... {}".format(path))
        torch.save(self, path)

    @property
    def test_if_cuda(self):
        """Cuda Test

        This function tests if the model parameters are allocated to a CUDA enabled GPU.

        Returns:
            bool: Flag indicating True if the tensor is stored on the GPU and Flase otherwhise
        """

        return next(self.parameters()).is_cuda

    def predict(self, X, device=0):
        """Post-training Output Prediction

        This function predicts the output of the of the U-net post-training

        Args:
            X (torch.tensor): input dMRI volume
            device (int/str): Device type used for training (int - GPU id, str- CPU)

        Returns:
            prediction (ndarray): predicted output after training

        """
        self.eval()  # PyToch module setting network to evaluation mode

        if type(X) is np.ndarray:
            X = torch.tensor(X, requires_grad=False).type(torch.FloatTensor)
        elif type(X) is torch.Tensor and not X.is_cuda:
            X = X.type(torch.FloatTensor).cuda(device, non_blocking=True)

        # .cuda() call transfers the densor from the CPU to the GPU if that is the case.
        # Non-blocking argument lets the caller bypas synchronization when necessary

        with torch.no_grad():  # Causes operations to have no gradients
            output = self.forward(X)

        _, idx = torch.max(output, 1)

        # We retrieve the tensor held by idx (.data), and map it to a cpu as an ndarray
        idx = idx.data.cpu().numpy()

        prediction = np.squeeze(idx)

        del X, output, idx

        return prediction

399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
    def reset_parameters(self):
        """Parameter Initialization

        This function (re)initializes the parameters of the defined network.
        This function is a wrapper for the reset_parameters() function defined for each module. 
        More information can be found here: https://discuss.pytorch.org/t/what-is-the-default-initialization-of-a-conv2d-layer-and-linear-layer/16055 + https://discuss.pytorch.org/t/how-to-reset-model-weights-to-effectively-implement-crossvalidation/53859 
        An alternative (re)initialization method is described here: https://discuss.pytorch.org/t/how-to-reset-variables-values-in-nn-modules/32639 
        """

        print("Initializing network parameters...")

        for _, module in self.named_children():
            for _, submodule in module.named_children():
                if isinstance(submodule, (torch.nn.PReLU, torch.nn.Dropout3d, torch.nn.MaxPool3d)) == False:
                    submodule.reset_parameters()

        print("Initialized network parameters!")
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
class BrainMapperUNet3D_Simple(nn.Module):
    """Architecture class BrainMapper 3D U-net.

    This class contains the pytorch implementation of the U-net architecture underpinning the BrainMapper project.

    Args:
        parameters (dict): Contains information relevant parameters
        parameters = {
            'kernel_heigth': 5
            'kernel_width': 5
            'kernel_depth': 5
            'kernel_classification': 1
            'input_channels': 1
            'output_channels': 64
            'convolution_stride': 1
            'dropout': 0.2
            'pool_kernel_size': 2
            'pool_stride': 2
            'up_mode': 'upconv'
            'number_of_classes': 1
        }

    Returns:
        probability_map (torch.tensor): Output forward passed tensor through the U-net block
    """

443
    def __init__(self, parameters):
444
        super(BrainMapperUNet3D_Simple, self).__init__()
445
446
447
448
449
450
451
452
453
454
455

        # TODO: currently, architecture based on QuickNAT - need to adjust parameter values accordingly!

        self.encoderBlock1 = modules.EncoderBlock3D(parameters)
        parameters['input_channels'] = parameters['output_channels']
        self.encoderBlock2 = modules.EncoderBlock3D(parameters)
        self.encoderBlock3 = modules.EncoderBlock3D(parameters)
        self.encoderBlock4 = modules.EncoderBlock3D(parameters)

        self.bottleneck = modules.ConvolutionalBlock3D(parameters)

456
        parameters['input_channels'] = parameters['output_channels'] * 2
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
        self.decoderBlock1 = modules.DecoderBlock3D(parameters)
        self.decoderBlock2 = modules.DecoderBlock3D(parameters)
        self.decoderBlock3 = modules.DecoderBlock3D(parameters)
        self.decoderBlock4 = modules.DecoderBlock3D(parameters)

        parameters['input_channels'] = parameters['output_channels']
        self.classifier = modules.ClassifierBlock3D(parameters)

    def forward(self, X):
        """Forward pass for 3D U-net

        Function computing the forward pass through the 3D U-Net
        The input to the function is the dMRI map

        Args:
            X (torch.tensor): Input dMRI map, shape = (N x C x D x H x W) 

        Returns:
            probability_map (torch.tensor): Output forward passed tensor through the U-net block
        """

478
        Y_encoder_1, Y_np1, pool_indices1 = self.encoderBlock1.forward(X)
479
480
        Y_encoder_2, Y_np2, pool_indices2 = self.encoderBlock2.forward(
            Y_encoder_1)
481
482
483

        del Y_encoder_1

484
485
        Y_encoder_3, Y_np3, pool_indices3 = self.encoderBlock3.forward(
            Y_encoder_2)
486
487
488

        del Y_encoder_2

489
490
        Y_encoder_4, Y_np4, pool_indices4 = self.encoderBlock4.forward(
            Y_encoder_3)
491

492
493
        del Y_encoder_3

494
495
        Y_bottleNeck = self.bottleneck.forward(Y_encoder_4)

496
497
        del Y_encoder_4

498
499
        Y_decoder_1 = self.decoderBlock1.forward(
            Y_bottleNeck, Y_np4, pool_indices4)
500
501

        del Y_bottleNeck, Y_np4, pool_indices4
502

503
504
        Y_decoder_2 = self.decoderBlock2.forward(
            Y_decoder_1, Y_np3, pool_indices3)
505
506
507

        del Y_decoder_1, Y_np3, pool_indices3

508
509
        Y_decoder_3 = self.decoderBlock3.forward(
            Y_decoder_2, Y_np2, pool_indices2)
510
511
512

        del Y_decoder_2, Y_np2, pool_indices2

513
        Y_decoder_4 = self.decoderBlock4.forward(
514
            Y_decoder_3, Y_np1, pool_indices1)
515

516
517
518
519
520
521
        del Y_decoder_3, Y_np1, pool_indices1

        probability_map = self.classifier.forward(Y_decoder_4)

        del Y_decoder_4

522
        probability_map = self.classifier.forward(Y_decoder_4)
523

524
        return probability_map
525

526
    def save(self, path):
Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
527
        """Model Saver
528

Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
529
530
        Function saving the model with all its parameters to a given path.
        The path must end with a *.model argument.
531

Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
532
533
534
        Args:
            path (str): Path string
        """
535

Andrei-Claudiu Roibu's avatar
Andrei-Claudiu Roibu committed
536
537
        print("Saving Model... {}".format(path))
        torch.save(self, path)
538
539
540
541
542
543
544
545
546
547
548
549

    @property
    def test_if_cuda(self):
        """Cuda Test

        This function tests if the model parameters are allocated to a CUDA enabled GPU.

        Returns:
            bool: Flag indicating True if the tensor is stored on the GPU and Flase otherwhise
        """

        return next(self.parameters()).is_cuda
550
551

    def predict(self, X, device=0):
552
        """Post-training Output Prediction
553

554
555
556
557
        This function predicts the output of the of the U-net post-training

        Args:
            X (torch.tensor): input dMRI volume
558
            device (int/str): Device type used for training (int - GPU id, str- CPU)
559
560
561
562

        Returns:
            prediction (ndarray): predicted output after training

563
        """
564
        self.eval()  # PyToch module setting network to evaluation mode
565
566
567
568
569
570
571
572
573

        if type(X) is np.ndarray:
            X = torch.tensor(X, requires_grad=False).type(torch.FloatTensor)
        elif type(X) is torch.Tensor and not X.is_cuda:
            X = X.type(torch.FloatTensor).cuda(device, non_blocking=True)

        # .cuda() call transfers the densor from the CPU to the GPU if that is the case.
        # Non-blocking argument lets the caller bypas synchronization when necessary

574
        with torch.no_grad():  # Causes operations to have no gradients
575
576
577
578
            output = self.forward(X)

        _, idx = torch.max(output, 1)

579
580
581
        # We retrieve the tensor held by idx (.data), and map it to a cpu as an ndarray
        idx = idx.data.cpu().numpy()

582
583
584
585
586
587
        prediction = np.squeeze(idx)

        del X, output, idx

        return prediction

588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
    def reset_parameters(self):
        """Parameter Initialization

        This function (re)initializes the parameters of the defined network.
        This function is a wrapper for the reset_parameters() function defined for each module. 
        More information can be found here: https://discuss.pytorch.org/t/what-is-the-default-initialization-of-a-conv2d-layer-and-linear-layer/16055 + https://discuss.pytorch.org/t/how-to-reset-model-weights-to-effectively-implement-crossvalidation/53859 
        An alternative (re)initialization method is described here: https://discuss.pytorch.org/t/how-to-reset-variables-values-in-nn-modules/32639 
        """

        print("Initializing network parameters...")

        for _, module in self.named_children():
            for _, submodule in module.named_children():
                if isinstance(submodule, (torch.nn.PReLU, torch.nn.Dropout3d, torch.nn.MaxPool3d)) == False:
                    submodule.reset_parameters()

        print("Initialized network parameters!")

606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
# if __name__ == '__main__':

#     # For debugging - To be deleted later! TODO

#     parameters = {
#         'kernel_heigth': 5,
#         'kernel_width': 5,
#         'kernel_depth': 5,
#         'kernel_classification': 1,
#         'input_channels': 1,
#         'output_channels': 64,
#         'convolution_stride': 1,
#         'dropout': 0.2,
#         'pool_kernel_size': 2,
#         'pool_stride': 2,
#         'up_mode': 'upconv',
#         'number_of_classes': 1
#     }

#     network = BrainMapperUNet3D(parameters)