medfuse-omar/fusion_model.py at main · oe2015/medfuse-omar · GitHub

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

import torch.nn as nn
import torchvision
import torch
import numpy as np

from torch.nn.functional import kl_div, softmax, log_softmax
from .loss import RankingLoss, CosineLoss, KLDivLoss
import torch.nn.functional as F

class Fusion(nn.Module):
    def __init__(self, args, ehr_model, cxr_model):

        super(Fusion, self).__init__()
        self.args = args
        self.ehr_model = ehr_model
        self.cxr_model = cxr_model


        target_classes = self.args.num_classes
        lstm_in = self.ehr_model.feats_dim
        lstm_out = self.cxr_model.feats_dim
        projection_in = self.cxr_model.feats_dim


        if self.args.labels_set == 'radiology':
            target_classes = self.args.vision_num_classes
            lstm_in = self.cxr_model.feats_dim
            projection_in = self.ehr_model.feats_dim

        # import pdb; pdb.set_trace()
        self.projection = nn.Linear(projection_in, lstm_in)
        feats_dim = 2 * self.ehr_model.feats_dim
        # feats_dim = self.ehr_model.feats_dim + self.cxr_model.feats_dim

        self.fused_cls = nn.Sequential(
            nn.Linear(feats_dim, self.args.num_classes),
            nn.Sigmoid()
        )

        self.align_loss = CosineLoss()
        self.kl_loss = KLDivLoss()

        self.lstm_fused_cls =  nn.Sequential(
            nn.Linear(lstm_out, target_classes),
            nn.Sigmoid()
        )

        self.lstm_fusion_layer = nn.LSTM(
            lstm_in, lstm_out,
            batch_first=True,
            dropout = 0.0)


    def forward_uni_cxr(self, x, seq_lengths=None, img=None ):
        cxr_preds, _ , feats = self.cxr_model(img)
        return {
            'uni_cxr': cxr_preds,
            'cxr_feats': feats
            }
    #
    def forward(self, x, seq_lengths=None, img=None, pairs=None ):
        if self.args.fusion_type == 'uni_cxr':
            return self.forward_uni_cxr(x, seq_lengths=seq_lengths, img=img)
        elif self.args.fusion_type in ['joint',  'early', 'late_avg', 'unified']:
            return self.forward_fused(x, seq_lengths=seq_lengths, img=img, pairs=pairs )
        elif self.args.fusion_type == 'uni_ehr':
            return self.forward_uni_ehr(x, seq_lengths=seq_lengths, img=img)
        elif self.args.fusion_type == 'lstm':
            return self.forward_lstm_fused(x, seq_lengths=seq_lengths, img=img, pairs=pairs )

        elif self.args.fusion_type == 'uni_ehr_lstm':
            return self.forward_lstm_ehr(x, seq_lengths=seq_lengths, img=img, pairs=pairs )

    def forward_uni_ehr(self, x, seq_lengths=None, img=None ):
        ehr_preds , feats = self.ehr_model(x, seq_lengths)
        return {
            'uni_ehr': ehr_preds,
            'ehr_feats': feats
            }

    def forward_fused(self, x, seq_lengths=None, img=None, pairs=None ):

        ehr_preds , ehr_feats = self.ehr_model(x, seq_lengths)
        cxr_preds, _ , cxr_feats = self.cxr_model(img)
        projected = self.projection(cxr_feats)

        # loss = self.align_loss(projected, ehr_feats)

        feats = torch.cat([ehr_feats, projected], dim=1)
        fused_preds = self.fused_cls(feats)

        # late_avg = (cxr_preds + ehr_preds)/2
        return {
            'early': fused_preds,
            'joint': fused_preds,
            # 'late_avg': late_avg,
            # 'align_loss': loss,
            'ehr_feats': ehr_feats,
            'cxr_feats': projected,
            'unified': fused_preds
            }

    def forward_lstm_fused(self, x, seq_lengths=None, img=None, pairs=None ):
        if self.args.labels_set == 'radiology':
            _ , ehr_feats = self.ehr_model(x, seq_lengths)

            _, _ , cxr_feats = self.cxr_model(img)

            feats = cxr_feats[:,None,:]

            ehr_feats = self.projection(ehr_feats)

            ehr_feats[list(~np.array(pairs))] = 0
            feats = torch.cat([feats, ehr_feats[:,None,:]], dim=1)
        else:

            _ , ehr_feats = self.ehr_model(x, seq_lengths)
            # if

            _, _ , cxr_feats = self.cxr_model(img)
            cxr_feats = self.projection(cxr_feats)

            cxr_feats[list(~np.array(pairs))] = 0
            if len(ehr_feats.shape) == 1:
                # print(ehr_feats.shape, cxr_feats.shape)
                # import pdb; pdb.set_trace()
                feats = ehr_feats[None,None,:]
                feats = torch.cat([feats, cxr_feats[:,None,:]], dim=1)
            else:
                feats = ehr_feats[:,None,:]
                feats = torch.cat([feats, cxr_feats[:,None,:]], dim=1)
        seq_lengths = np.array([1] * len(seq_lengths))
        seq_lengths[pairs] = 2

        feats = torch.nn.utils.rnn.pack_padded_sequence(feats, seq_lengths, batch_first=True, enforce_sorted=False)

        x, (ht, _) = self.lstm_fusion_layer(feats)

        out = ht.squeeze()

        fused_preds = self.lstm_fused_cls(out)

        return {
            'lstm': fused_preds,
            'ehr_feats': ehr_feats,
            'cxr_feats': cxr_feats,
        }

    def forward_lstm_ehr(self, x, seq_lengths=None, img=None, pairs=None ):
        _ , ehr_feats = self.ehr_model(x, seq_lengths)
        feats = ehr_feats[:,None,:]


        seq_lengths = np.array([1] * len(seq_lengths))

        feats = torch.nn.utils.rnn.pack_padded_sequence(feats, seq_lengths, batch_first=True, enforce_sorted=False)

        x, (ht, _) = self.lstm_fusion_layer(feats)

        out = ht.squeeze()

        fused_preds = self.lstm_fused_cls(out)

        return {
            'uni_ehr_lstm': fused_preds,
        }