WhiteMatterHyperintensities/twodunet.py at master · pabloduque0/WhiteMatterHyperintensities · 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
169
170
171
172
173
174
175
176
177
178
179
180
181
from keras import models
from keras import layers
from contextlib import redirect_stdout
import os
from keras.callbacks import ModelCheckpoint, TensorBoard
from sklearn.model_selection import train_test_split
from keras.models import load_model
from keras.optimizers import Adam, SGD
from metrics import dice_coef, dice_coef_loss, weighted_crossentropy, predicted_count, ground_truth_count, ground_truth_sum, predicted_sum
from keras.losses import binary_crossentropy
import cv2
import numpy as np


class TwoDUnet():

    def __init__(self, model_path=None, img_shape=None):

        if model_path is None:
            if img_shape is None:
                raise Exception('If no model path is provided img shape is a mandatory argument.')
            model = self.create_model(img_shape)
        else:
            model = load_model(model_path)

        self.model = model


    def create_model(self, img_shape):

        concat_axis = 3

        inputs = layers.Input(shape=img_shape)
        conv1 = layers.Conv2D(64, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(inputs)
        conv2 = layers.Conv2D(64, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv1)
        maxpool1 = layers.MaxPool2D(pool_size=(2, 2))(conv2)

        conv3 = layers.Conv2D(128, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(maxpool1)
        conv4 = layers.Conv2D(128, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv3)
        maxpool2 = layers.MaxPool2D(pool_size=(2, 2))(conv4)

        conv5 = layers.Conv2D(256, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(maxpool2)
        conv6 = layers.Conv2D(256, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv5)
        maxpool3 = layers.MaxPool2D(pool_size=(2, 2))(conv6)

        conv7 = layers.Conv2D(512, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(maxpool3)
        conv8 = layers.Conv2D(512, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv7)
        maxpool4 = layers.MaxPool2D(pool_size=(2, 2))(conv8)

        conv9 = layers.Conv2D(1024, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(maxpool4)
        conv10 = layers.Conv2D(1024, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv9)

        up_conv10 = layers.UpSampling2D(size=(2, 2))(conv10)
        up_samp1 = layers.concatenate([conv8, up_conv10], axis=concat_axis)
        conv11 = layers.Conv2D(512, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(up_samp1)

        conv12 = layers.Conv2D(512, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv11)
        conv13 = layers.Conv2D(512, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv12)

        up_conv13 = layers.UpSampling2D(size=(2, 2))(conv13)
        up_samp2 = layers.concatenate([conv6, up_conv13], axis=concat_axis)
        conv14 = layers.Conv2D(256, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(up_samp2)

        conv15 = layers.Conv2D(256, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv14)
        conv16 = layers.Conv2D(256, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv15)

        up_conv16 = layers.UpSampling2D(size=(2, 2))(conv16)
        up_samp3 = layers.concatenate([conv4, up_conv16], axis=concat_axis)
        conv17 = layers.Conv2D(128, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(up_samp3)

        conv18 = layers.Conv2D(128, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv17)
        conv19 = layers.Conv2D(128, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv18)

        up_conv19 = layers.UpSampling2D(size=(2, 2))(conv19)
        up_samp1 = layers.concatenate([conv2, up_conv19], axis=concat_axis)
        conv20 = layers.Conv2D(64, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(up_samp1)

        conv21 = layers.Conv2D(64, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv20)
        conv22 = layers.Conv2D(64, kernel_size=5, padding='same', kernel_initializer='he_normal', activation='relu')(conv21)

        conv23 = layers.Conv2D(1, kernel_size=1, padding='same', kernel_initializer='he_normal', activation='sigmoid')(conv22)

        model = models.Model(inputs=inputs, outputs=conv23)

        model.compile(optimizer=Adam(lr=0.000001), loss=dice_coef_loss, metrics=[dice_coef, binary_crossentropy, weighted_crossentropy,
                                                                                   predicted_count, predicted_sum, ground_truth_count, ground_truth_sum])

        model.summary()

        return model


    def save_specs(self, specs_path, fit_specs):

        with open(specs_path, 'w') as file:
            with redirect_stdout(file):
                self.model.summary()

        fit_specs_file = specs_path[:-4] + 'fit_specs.txt'

        with open(fit_specs_file, 'w') as fit_file:
            for key, value in fit_specs.items():
                fit_file.write(key + ': ' + str(value) + '\n')

    def create_folders(self, training_name, base_path):

        model_path = base_path + "/models/" + training_name
        if not os.path.exists(model_path):
            os.makedirs(model_path)

        v = 0
        weights_path = model_path + "/model_0.hdf5"
        if os.path.exists(weights_path):
            try:
                v = int(weights_path.split("_")[-1].replace(".hdf5", "")) + 1
            except ValueError:
                v = 1
            weights_path = model_path + "/model_{}.hdf5".format(v)

        log_path = base_path + "/logs/" + training_name + '/'
        if not os.path.exists(log_path):
            os.makedirs(log_path)

        specs_path = log_path + "/specs_{}.txt".format(v)

        return {"log_path": log_path, "weights_path": weights_path,
                "specs_path": specs_path}

    def train(self, X, y, test_size, training_name, base_path, epochs=10, batch_size=32):

        paths = self.create_folders(training_name, base_path)

        checkpointer = ModelCheckpoint(filepath=paths["weights_path"],
                                       save_best_only=True,
                                       verbose=1)

        tensorboard_callback = TensorBoard(log_dir=paths["log_path"],
                                           batch_size=batch_size,
                                           write_graph=False,
                                           write_grads=False,
                                           write_images=False,
                                           embeddings_freq=0,
                                           embeddings_layer_names=None,
                                           embeddings_metadata=None)

        X_train, X_test, y_train, y_test = train_test_split(X,
                                                            y,
                                                            test_size=test_size)

        fit_specs = {
            'epochs': epochs,
            'batch_size': batch_size,
            'test_size': test_size

        }
        self.save_specs(paths['specs_path'], fit_specs)


        self.model.fit(X_train, y_train,
                       batch_size=batch_size,
                       callbacks=[checkpointer, tensorboard_callback],
                       epochs=epochs,
                       validation_data=(X_test, y_test),
                       verbose=1)


    def predict_and_save(self, data, labels, output_path, batch_size=1):

        if not output_path.endswith('/'):
            output_path += '/'

        if not os.path.exists(output_path):
            os.makedirs(output_path)

        predictions = self.model.predict(data, batch_size=batch_size, verbose=1)

        for index, (pred, original, label) in enumerate(zip(predictions, data, labels)):
            print(len(np.flatnonzero(pred)))
            cv2.imwrite(output_path + 'original_' + str(index) + '.png', original * 255)
            cv2.imwrite(output_path + 'prediction_' + str(index) + '.png', pred * 255)
            cv2.imwrite(output_path + 'label_' + str(index) + '.png', label * 255)