File size: 29,772 Bytes
fe39cc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
Flexible Multi-Task Head and Loss Function
Adapts to any biomarker configuration without hardcoded assumptions
"""

import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np
from typing import Dict, List, Tuple, Any, Union, Optional
import logging
from config.biomarker_config import FlexibleBiomarkerConfig, TensorLayout
from .single_target_strategies import (
    SingleTargetStrategy, 
    create_feature_extractor,
    get_strategy_from_name
)

logger = logging.getLogger(__name__)


class FlexibleMultiTaskHead(nn.Module):
    """Flexible multi-task head that adapts to biomarker configuration"""
    
    def __init__(
        self,
        input_dim: int,
        biomarker_config: FlexibleBiomarkerConfig,
        dropout: float = 0.1,
        single_target_strategy: Optional[Union[str, SingleTargetStrategy]] = None,
        target_feature_dim: Optional[int] = None
    ):
        super().__init__()
        
        self.biomarker_config = biomarker_config
        self.tensor_layout = biomarker_config.get_tensor_layout()
        
        # Handle single-target strategy
        self.single_target_strategy = None
        self.feature_extractor = None
        self.target_feature_dim = target_feature_dim

        if single_target_strategy is not None:
            if isinstance(single_target_strategy, str):
                self.single_target_strategy = get_strategy_from_name(single_target_strategy)
            else:
                self.single_target_strategy = single_target_strategy

            # Create appropriate feature extractor with target feature dimension
            feature_dim = target_feature_dim if target_feature_dim is not None else input_dim
            self.feature_extractor = create_feature_extractor(
                self.single_target_strategy,
                input_dim,
                feature_dim=feature_dim,
                dropout=dropout
            )

            # Use feature extractor output dimension
            processed_input_dim = self.feature_extractor.output_dim
        else:
            # Default behavior - use original input dimension
            processed_input_dim = input_dim
        
        # Shared feature processing (only if no single-target strategy is used)
        if self.single_target_strategy is None:
            # Use target_feature_dim if provided, otherwise use 512 as default
            output_dim = target_feature_dim if target_feature_dim is not None else 512
            self.shared_layers = nn.Sequential(
                nn.Linear(input_dim, output_dim),
                nn.ReLU(inplace=True),
                nn.Dropout(dropout),
                nn.BatchNorm1d(output_dim)
            )
            processed_input_dim = output_dim
        else:
            # Skip shared layers when using single-target strategy
            self.shared_layers = nn.Identity()
            processed_input_dim = self.feature_extractor.output_dim
        
        # Task-specific heads
        self.task_heads = nn.ModuleDict()
        
        # Binary classification heads (one per biomarker)
        for biomarker in biomarker_config.binary_biomarkers:
            self.task_heads[f"binary_{biomarker.name}"] = nn.Linear(processed_input_dim, 1)
        
        # Multiclass classification heads
        for biomarker in biomarker_config.multiclass_biomarkers:
            num_classes = len(biomarker.classes)
            self.task_heads[f"multiclass_{biomarker.name}"] = nn.Linear(processed_input_dim, num_classes)
        
        # Regression heads (one per biomarker)
        for biomarker in biomarker_config.continuous_biomarkers:
            self.task_heads[f"continuous_{biomarker.name}"] = nn.Linear(processed_input_dim, 1)
    
    def forward(self, x):
        """
        Forward pass
        
        Args:
            x: Input features [batch_size, input_dim] or [batch_size, C, H, W] for spatial features
            
        Returns:
            Concatenated outputs [batch_size, total_output_size]
        """
        # Apply single-target strategy feature extraction if specified
        if self.single_target_strategy is not None and self.feature_extractor is not None:
            shared_features = self.feature_extractor(x)
        else:
            shared_features = self.shared_layers(x)
        
        outputs = []
        
        # Binary outputs
        for biomarker in self.biomarker_config.binary_biomarkers:
            head_name = f"binary_{biomarker.name}"
            binary_out = self.task_heads[head_name](shared_features)  # [B, 1]
            outputs.append(binary_out)
        
        # Multiclass outputs
        for biomarker in self.biomarker_config.multiclass_biomarkers:
            head_name = f"multiclass_{biomarker.name}"
            multiclass_out = self.task_heads[head_name](shared_features)  # [B, num_classes]
            outputs.append(multiclass_out)
        
        # Regression outputs
        for biomarker in self.biomarker_config.continuous_biomarkers:
            head_name = f"continuous_{biomarker.name}"
            regression_out = self.task_heads[head_name](shared_features)  # [B, 1]
            outputs.append(regression_out)
        
        # Concatenate all outputs
        return torch.cat(outputs, dim=1)


class LinearProbeMultiTaskHead(nn.Module):
    """
    True linear probe head - direct mapping from backbone features to task outputs
    No shared layers, minimal parameters, maximum interpretability
    """
    
    def __init__(
        self,
        input_dim: int,
        biomarker_config: FlexibleBiomarkerConfig,
        dropout: float = 0.0,
        single_target_strategy: Optional[Union[str, SingleTargetStrategy]] = None,
        target_feature_dim: Optional[int] = None
    ):
        super().__init__()
        
        self.biomarker_config = biomarker_config
        self.tensor_layout = biomarker_config.get_tensor_layout()
        
        # Handle single-target strategy
        self.single_target_strategy = None
        self.feature_extractor = None
        self.target_feature_dim = target_feature_dim

        if single_target_strategy is not None:
            if isinstance(single_target_strategy, str):
                self.single_target_strategy = get_strategy_from_name(single_target_strategy)
            else:
                self.single_target_strategy = single_target_strategy

            # Create appropriate feature extractor with target feature dimension
            feature_dim = target_feature_dim if target_feature_dim is not None else input_dim
            self.feature_extractor = create_feature_extractor(
                self.single_target_strategy,
                input_dim,
                feature_dim=feature_dim,
                dropout=0.0  # No dropout for linear probe
            )

            # Use feature extractor output dimension
            processed_input_dim = self.feature_extractor.output_dim
        else:
            # Default behavior - use original input dimension
            processed_input_dim = input_dim
        
        # Optional minimal dropout (usually 0.0 for true linear probe)
        self.dropout = nn.Dropout(dropout) if dropout > 0 else nn.Identity()
        
        # Direct task-specific linear layers (no shared processing)
        self.task_heads = nn.ModuleDict()
        
        # Binary classification heads - direct from backbone
        for biomarker in biomarker_config.binary_biomarkers:
            self.task_heads[f"binary_{biomarker.name}"] = nn.Linear(processed_input_dim, 1)
        
        # Multiclass classification heads - direct from backbone  
        for biomarker in biomarker_config.multiclass_biomarkers:
            num_classes = len(biomarker.classes)
            self.task_heads[f"multiclass_{biomarker.name}"] = nn.Linear(processed_input_dim, num_classes)
        
        # Regression heads - direct from backbone
        for biomarker in biomarker_config.continuous_biomarkers:
            self.task_heads[f"continuous_{biomarker.name}"] = nn.Linear(processed_input_dim, 1)
        
        # Initialize weights for better convergence
        self._initialize_weights()
    
    def _initialize_weights(self):
        """Initialize linear layer weights for stable training"""
        for name, module in self.task_heads.items():
            if isinstance(module, nn.Linear):
                # Xavier/Glorot initialization for linear layers
                nn.init.xavier_uniform_(module.weight)
                nn.init.zeros_(module.bias)
    
    def forward(self, x):
        """
        Direct forward pass - no shared processing
        
        Args:
            x: Backbone features [batch_size, input_dim] or [batch_size, C, H, W] for spatial features
            
        Returns:
            Concatenated outputs [batch_size, total_output_size]
        """
        # Apply single-target strategy feature extraction if specified
        if self.single_target_strategy is not None and self.feature_extractor is not None:
            features = self.feature_extractor(x)
        else:
            # Optional dropout on backbone features (usually disabled)
            features = self.dropout(x)  # [batch_size, input_dim]
        
        outputs = []
        
        # Binary outputs - direct linear transformation
        for biomarker in self.biomarker_config.binary_biomarkers:
            head_name = f"binary_{biomarker.name}"
            binary_out = self.task_heads[head_name](features)  # [B, 1]
            outputs.append(binary_out)
        
        # Multiclass outputs - direct linear transformation
        for biomarker in self.biomarker_config.multiclass_biomarkers:
            head_name = f"multiclass_{biomarker.name}"
            multiclass_out = self.task_heads[head_name](features)  # [B, num_classes]
            outputs.append(multiclass_out)
        
        # Regression outputs - direct linear transformation
        for biomarker in self.biomarker_config.continuous_biomarkers:
            head_name = f"continuous_{biomarker.name}"
            regression_out = self.task_heads[head_name](features)  # [B, 1]
            outputs.append(regression_out)
        
        # Concatenate all outputs
        return torch.cat(outputs, dim=1)  # [batch_size, total_output_size]


class FlexibleMultiTaskLoss(nn.Module):
    """Flexible multi-task loss function that adapts to biomarker configuration"""
    
    def __init__(self, biomarker_config: FlexibleBiomarkerConfig, class_weights: Dict[str, float] = None):
        super().__init__()
        
        self.biomarker_config = biomarker_config
        self.tensor_layout = biomarker_config.get_tensor_layout()
        self.class_weights = class_weights or {}
        
        # Create weighted BCE losses for binary tasks
        self.binary_losses = nn.ModuleDict()
        for biomarker in biomarker_config.binary_biomarkers:
            weight = self.class_weights.get(biomarker.name, 1.0)
            if weight != 1.0:
                pos_weight = torch.tensor([weight], dtype=torch.float32)
                self.binary_losses[biomarker.name] = nn.BCEWithLogitsLoss(pos_weight=pos_weight)
            else:
                self.binary_losses[biomarker.name] = nn.BCEWithLogitsLoss()
        
        # Cross-entropy losses for multiclass tasks
        self.multiclass_losses = nn.ModuleDict()
        for biomarker in biomarker_config.multiclass_biomarkers:
            self.multiclass_losses[biomarker.name] = nn.CrossEntropyLoss()
        
        # MSE losses for regression tasks
        self.regression_losses = nn.ModuleDict()
        for biomarker in biomarker_config.continuous_biomarkers:
            self.regression_losses[biomarker.name] = nn.MSELoss()
    
    def forward(self, predictions: torch.Tensor, targets: torch.Tensor) -> Tuple[torch.Tensor, Dict[str, float]]:
        """
        Calculate multi-task loss
        
        Args:
            predictions: [batch_size, total_output_size]
            targets: [batch_size, total_output_size]
            
        Returns:
            total_loss: Combined loss
            loss_dict: Dictionary with individual loss components
        """
        device = predictions.device
        
        # Move pos_weight tensors to correct device
        for biomarker_name, loss_fn in self.binary_losses.items():
            if hasattr(loss_fn, 'pos_weight') and loss_fn.pos_weight is not None:
                loss_fn.pos_weight = loss_fn.pos_weight.to(device)
        
        total_loss = 0.0
        loss_components = {}
        
        # Binary classification losses
        binary_losses = []
        for biomarker in self.biomarker_config.binary_biomarkers:
            layout = self.tensor_layout[biomarker.name]
            
            pred_slice = predictions[:, layout.start_idx:layout.end_idx].squeeze(-1)  # [B]
            target_slice = targets[:, layout.start_idx:layout.end_idx].squeeze(-1)  # [B]
            
            loss_fn = self.binary_losses[biomarker.name]
            binary_loss = loss_fn(pred_slice, target_slice)
            binary_losses.append(binary_loss)
            
            loss_components[f'binary_{biomarker.name}'] = binary_loss.item()
        
        if binary_losses:
            avg_binary_loss = torch.mean(torch.stack(binary_losses))
            total_loss += avg_binary_loss
            loss_components['avg_binary_loss'] = avg_binary_loss.item()
        
        # Multiclass classification losses
        multiclass_losses = []
        for biomarker in self.biomarker_config.multiclass_biomarkers:
            layout = self.tensor_layout[biomarker.name]
            
            pred_slice = predictions[:, layout.start_idx:layout.end_idx]  # [B, num_classes]
            target_slice = targets[:, layout.start_idx:layout.end_idx]  # [B, num_classes]
            
            # Convert one-hot targets to class indices
            target_indices = torch.argmax(target_slice, dim=1)  # [B]
            
            loss_fn = self.multiclass_losses[biomarker.name]
            multiclass_loss = loss_fn(pred_slice, target_indices)
            multiclass_losses.append(multiclass_loss)
            
            loss_components[f'multiclass_{biomarker.name}'] = multiclass_loss.item()
        
        if multiclass_losses:
            avg_multiclass_loss = torch.mean(torch.stack(multiclass_losses))
            total_loss += avg_multiclass_loss
            loss_components['avg_multiclass_loss'] = avg_multiclass_loss.item()
        
        # Regression losses
        regression_losses = []
        for biomarker in self.biomarker_config.continuous_biomarkers:
            layout = self.tensor_layout[biomarker.name]
            
            pred_slice = predictions[:, layout.start_idx:layout.end_idx].squeeze(-1)  # [B]
            target_slice = targets[:, layout.start_idx:layout.end_idx].squeeze(-1)  # [B]
            
            loss_fn = self.regression_losses[biomarker.name]
            regression_loss = loss_fn(pred_slice, target_slice)
            regression_losses.append(regression_loss)
            
            loss_components[f'regression_{biomarker.name}'] = regression_loss.item()
        
        if regression_losses:
            avg_regression_loss = torch.mean(torch.stack(regression_losses))
            total_loss += avg_regression_loss
            loss_components['avg_regression_loss'] = avg_regression_loss.item()
        
        loss_components['total_loss'] = total_loss.item()
        
        return total_loss, loss_components


class FlexibleMetricsCalculator:
    """Calculate comprehensive metrics for flexible multi-task learning"""
    
    def __init__(self, biomarker_config: FlexibleBiomarkerConfig):
        self.biomarker_config = biomarker_config
        self.tensor_layout = biomarker_config.get_tensor_layout()
        
        # Threshold optimization settings from config
        validation_config = biomarker_config.validation
        self.threshold_optimization = validation_config.get('threshold_optimization', False)
        self.optimization_metric = validation_config.get('optimization_metric', 'f1_score')
        self.per_biomarker_thresholds = validation_config.get('per_biomarker_thresholds', True)
        self.threshold_range = validation_config.get('threshold_search_range', [0.1, 0.9])
        self.threshold_steps = validation_config.get('threshold_search_steps', 81)
        self.fallback_threshold = validation_config.get('fallback_threshold', 0.5)
        
        # Store optimal thresholds per biomarker
        self.optimal_thresholds = {}
    
    def find_optimal_threshold(self, pred_probs: np.ndarray, true_labels: np.ndarray, 
                              metric: str = 'f1_score') -> Tuple[float, float]:
        """
        Find optimal threshold for a single biomarker based on specified metric
        
        Args:
            pred_probs: Predicted probabilities [N]
            true_labels: True binary labels [N]
            metric: Metric to optimize ('f1_score', 'sensitivity', 'specificity', etc.)
            
        Returns:
            best_threshold: Optimal threshold value
            best_score: Best metric score achieved
        """
        from sklearn.metrics import f1_score, precision_score, recall_score
        
        # Create threshold search space
        thresholds = np.linspace(self.threshold_range[0], self.threshold_range[1], self.threshold_steps)
        
        best_threshold = self.fallback_threshold
        best_score = 0.0
        
        for threshold in thresholds:
            pred_labels = (pred_probs > threshold).astype(int)
            
            try:
                if metric == 'f1_score':
                    score = f1_score(true_labels, pred_labels, zero_division=0.0)
                elif metric == 'precision':
                    score = precision_score(true_labels, pred_labels, zero_division=0.0)
                elif metric == 'recall' or metric == 'sensitivity':
                    score = recall_score(true_labels, pred_labels, zero_division=0.0)
                elif metric == 'specificity':
                    # Specificity = TN / (TN + FP)
                    tn = np.sum((pred_labels == 0) & (true_labels == 0))
                    fp = np.sum((pred_labels == 1) & (true_labels == 0))
                    score = tn / (tn + fp + 1e-8)
                elif metric == 'accuracy':
                    score = (pred_labels == true_labels).mean()
                else:
                    # Default to f1_score
                    score = f1_score(true_labels, pred_labels, zero_division=0.0)
                
                if score > best_score:
                    best_score = score
                    best_threshold = threshold
                    
            except (ValueError, ZeroDivisionError):
                continue
        
        return best_threshold, best_score
    
    def optimize_thresholds(self, predictions: torch.Tensor, targets: torch.Tensor) -> Dict[str, float]:
        """
        Find optimal thresholds for all binary biomarkers
        
        Args:
            predictions: Model predictions [batch_size, total_output_size]
            targets: True targets [batch_size, total_output_size]
            
        Returns:
            Dictionary mapping biomarker names to optimal thresholds
        """
        # Convert to numpy
        if isinstance(predictions, torch.Tensor):
            predictions = predictions.detach().cpu().numpy()
        if isinstance(targets, torch.Tensor):
            targets = targets.detach().cpu().numpy()
        
        optimal_thresholds = {}
        
        for biomarker in self.biomarker_config.binary_biomarkers:
            layout = self.tensor_layout[biomarker.name]
            
            pred_logits = predictions[:, layout.start_idx]
            pred_probs = 1 / (1 + np.exp(-pred_logits))  # Sigmoid
            true_labels = targets[:, layout.start_idx].astype(int)
            
            # Skip if all labels are the same (no positive or negative examples)
            if len(np.unique(true_labels)) < 2:
                optimal_thresholds[biomarker.name] = self.fallback_threshold
                continue
            
            # Find optimal threshold
            best_threshold, best_score = self.find_optimal_threshold(
                pred_probs, true_labels, self.optimization_metric
            )
            
            optimal_thresholds[biomarker.name] = best_threshold
            
            # Log the optimization result
            logger.info(
                "  %s: threshold=%.3f, %s=%.3f",
                biomarker.name,
                best_threshold,
                self.optimization_metric,
                best_score,
            )
        
        return optimal_thresholds
    
    def update_optimal_thresholds(self, predictions: torch.Tensor, targets: torch.Tensor):
        """Update optimal thresholds based on current predictions and targets"""
        if self.threshold_optimization:
            logger.info("Optimizing thresholds...")
            self.optimal_thresholds = self.optimize_thresholds(predictions, targets)
        else:
            # Use fallback threshold for all biomarkers
            for biomarker in self.biomarker_config.binary_biomarkers:
                self.optimal_thresholds[biomarker.name] = self.fallback_threshold
    
    def calculate_binary_metrics(self, predictions: torch.Tensor, targets: torch.Tensor, 
                                use_optimal_thresholds: bool = True) -> Dict[str, Dict[str, float]]:
        """Calculate metrics for binary classification tasks"""
        metrics = {}
        
        # Convert to numpy
        if isinstance(predictions, torch.Tensor):
            predictions = predictions.detach().cpu().numpy()
        if isinstance(targets, torch.Tensor):
            targets = targets.detach().cpu().numpy()
        
        for biomarker in self.biomarker_config.binary_biomarkers:
            layout = self.tensor_layout[biomarker.name]
            
            pred_logits = predictions[:, layout.start_idx]
            pred_probs = 1 / (1 + np.exp(-pred_logits))  # Sigmoid
            true_labels = targets[:, layout.start_idx].astype(int)
            
            # AUROC (threshold-independent)
            try:
                from sklearn.metrics import roc_auc_score
                auroc = roc_auc_score(true_labels, pred_probs)
            except (ValueError, ImportError):
                auroc = 0.0
            
            # Get threshold for this biomarker
            if use_optimal_thresholds and biomarker.name in self.optimal_thresholds:
                threshold = self.optimal_thresholds[biomarker.name]
            else:
                threshold = self.fallback_threshold
            
            # Predictions with threshold
            pred_labels = (pred_probs > threshold).astype(int)
            
            # Basic metrics
            accuracy = (pred_labels == true_labels).mean()
            
            # Confusion matrix components
            true_positives = np.sum((pred_labels == 1) & (true_labels == 1))
            true_negatives = np.sum((pred_labels == 0) & (true_labels == 0))
            false_positives = np.sum((pred_labels == 1) & (true_labels == 0))
            false_negatives = np.sum((pred_labels == 0) & (true_labels == 1))
            
            # Sensitivity and Specificity
            sensitivity = true_positives / (true_positives + false_negatives) if (true_positives + false_negatives) > 0 else 0.0
            specificity = true_negatives / (true_negatives + false_positives) if (true_negatives + false_positives) > 0 else 0.0
            
            # F1 Score
            try:
                from sklearn.metrics import f1_score
                f1 = f1_score(true_labels, pred_labels, zero_division=0.0)
            except ImportError:
                precision = true_positives / (true_positives + false_positives) if (true_positives + false_positives) > 0 else 0.0
                recall = sensitivity
                f1 = 2 * (precision * recall) / (precision + recall + 1e-8)
            
            metrics[biomarker.name] = {
                'auroc': float(auroc),
                'accuracy': float(accuracy),
                'sensitivity': float(sensitivity),
                'specificity': float(specificity),
                'f1': float(f1),
                'threshold': float(threshold),  # Include threshold used
                'true_positives': int(true_positives),
                'true_negatives': int(true_negatives),
                'false_positives': int(false_positives),
                'false_negatives': int(false_negatives)
            }
        
        return metrics
    
    def calculate_multiclass_metrics(self, predictions: torch.Tensor, targets: torch.Tensor) -> Dict[str, Dict[str, Any]]:
        """Calculate metrics for multiclass classification tasks"""
        metrics = {}
        
        # Convert to numpy
        if isinstance(predictions, torch.Tensor):
            predictions = predictions.detach().cpu().numpy()
        if isinstance(targets, torch.Tensor):
            targets = targets.detach().cpu().numpy()
        
        for biomarker in self.biomarker_config.multiclass_biomarkers:
            layout = self.tensor_layout[biomarker.name]
            
            pred_logits = predictions[:, layout.start_idx:layout.end_idx]
            # Numerically stable softmax
            max_logits = np.max(pred_logits, axis=1, keepdims=True)
            exp_logits = np.exp(pred_logits - max_logits)
            pred_probs = exp_logits / (np.sum(exp_logits, axis=1, keepdims=True) + 1e-12)
            target_one_hot = targets[:, layout.start_idx:layout.end_idx]
            
            # Get predicted and true classes
            pred_classes = np.argmax(pred_probs, axis=1)
            true_classes = np.argmax(target_one_hot, axis=1)
            
            # Overall accuracy
            accuracy = (pred_classes == true_classes).mean()
            
            # Multi-class AUROC
            try:
                from sklearn.metrics import roc_auc_score
                auroc = roc_auc_score(target_one_hot, pred_probs, multi_class='ovr', average='macro')
            except (ValueError, ImportError):
                auroc = 0.0
            
            metrics[biomarker.name] = {
                'accuracy': float(accuracy),
                'auroc': float(auroc)
            }
        
        return metrics
    
    def calculate_regression_metrics(self, predictions: torch.Tensor, targets: torch.Tensor) -> Dict[str, Dict[str, float]]:
        """Calculate metrics for regression tasks"""
        metrics = {}
        
        # Convert to numpy
        if isinstance(predictions, torch.Tensor):
            predictions = predictions.detach().cpu().numpy()
        if isinstance(targets, torch.Tensor):
            targets = targets.detach().cpu().numpy()
        
        for biomarker in self.biomarker_config.continuous_biomarkers:
            layout = self.tensor_layout[biomarker.name]
            
            pred_values = predictions[:, layout.start_idx]
            true_values = targets[:, layout.start_idx]
            
            # Denormalize if needed
            if biomarker.normalization == "min_max":
                pred_denorm = pred_values * (biomarker.max_value - biomarker.min_value) + biomarker.min_value
                true_denorm = true_values * (biomarker.max_value - biomarker.min_value) + biomarker.min_value
            else:
                pred_denorm = pred_values
                true_denorm = true_values
            
            # Calculate metrics
            mse = np.mean((pred_denorm - true_denorm) ** 2)
            mae = np.mean(np.abs(pred_denorm - true_denorm))
            
            # R² score
            ss_res = np.sum((true_denorm - pred_denorm) ** 2)
            ss_tot = np.sum((true_denorm - np.mean(true_denorm)) ** 2)
            r2 = 1 - (ss_res / (ss_tot + 1e-8))
            
            metrics[biomarker.name] = {
                'mse': float(mse),
                'mae': float(mae),
                'r2': float(r2)
            }
        
        return metrics
    
    def calculate_all_metrics(self, predictions: torch.Tensor, targets: torch.Tensor) -> Dict[str, Any]:
        """Calculate all metrics"""
        all_metrics = {}
        
        # Binary metrics
        if self.biomarker_config.binary_biomarkers:
            binary_metrics = self.calculate_binary_metrics(predictions, targets)
            all_metrics.update(binary_metrics)
        
        # Multiclass metrics
        if self.biomarker_config.multiclass_biomarkers:
            multiclass_metrics = self.calculate_multiclass_metrics(predictions, targets)
            all_metrics.update(multiclass_metrics)
        
        # Regression metrics
        if self.biomarker_config.continuous_biomarkers:
            regression_metrics = self.calculate_regression_metrics(predictions, targets)
            all_metrics.update(regression_metrics)
        
        # Calculate both average and median AUROC for comprehensive monitoring
        auroc_values = []
        for biomarker_name, metrics in all_metrics.items():
            if isinstance(metrics, dict) and 'auroc' in metrics:
                auroc_values.append(metrics['auroc'])
        
        if auroc_values:
            all_metrics['average_auroc'] = float(np.mean(auroc_values))
            all_metrics['median_auroc'] = float(np.median(auroc_values))
        else:
            all_metrics['average_auroc'] = 0.0
            all_metrics['median_auroc'] = 0.0
        
        return all_metrics