forked from armiro/COVID-CXNet
/
transfer_learning.py
153 lines (122 loc) · 6.1 KB
/
transfer_learning.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
import glob, numpy as np, matplotlib.pyplot as plt, cv2.cv2 as cv2, seaborn as sns
import keras.callbacks as cb
from keras.applications import ResNet50, DenseNet121, NASNetMobile, Xception
from keras.preprocessing.image import load_img, img_to_array, ImageDataGenerator
from keras.layers import Dense, Flatten, GlobalAveragePooling2D, Dropout
from keras import Model
from sklearn.model_selection import train_test_split
from sklearn.metrics import confusion_matrix, classification_report, roc_curve, auc
from BEASF import BEASF
def data_preparation(path, image_size):
normal_images = list()
for img_name in glob.glob(pathname=path + '/normal/*'):
img = load_img(path=img_name, color_mode='grayscale')
img = img_to_array(img=img, data_format='channels_last')
normal_images.append(img)
normal_images = np.array(normal_images)
print('number of normal chest xrays:', len(normal_images))
covid_images = list()
for img_name in glob.glob(pathname=path + '/covid19/*'):
img = load_img(path=img_name, color_mode='grayscale')
img = img_to_array(img=img, data_format='channels_last')
covid_images.append(img)
covid_images = np.array(covid_images)
print('number of covid chest xrays:', len(covid_images))
normal_labels = [0 for _ in range(len(normal_images))]
covid_labels = [1 for _ in range(len(covid_images))]
X = np.concatenate((covid_images, normal_images))
y = np.array(covid_labels + normal_labels)
X = np.array([cv2.resize(image, dsize=image_size, interpolation=cv2.INTER_CUBIC) for image in X])
X = np.array([np.expand_dims(a=image, axis=-1) for image in X])
X = X.astype(dtype=np.uint8)
# apply image enhancements and concat with the original image
X_beasf = np.array([BEASF(image=image, gamma=1.5) for image in X])
X_clahe = np.array([cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8)).apply(image) for image in X])
X_clahe = np.array([np.expand_dims(a=image, axis=-1) for image in X_clahe])
X = np.concatenate((X, X_beasf, X_clahe), axis=-1)
X = np.array([X[idx] / 255. for idx in range(len(X))])
print('number of total dataset images:', len(X))
print('number of total dataset labels:', len(y))
print('dataset shape:', X.shape)
rnd_idx = np.random.choice(a=len(X), size=None)
plt.imshow(X=X[rnd_idx].squeeze(), cmap='gray')
plt.axis('off')
plt.title(label='a random image from the dataset')
plt.show()
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=18)
return X_train, X_test, y_train, y_test
"""image data preparation"""
X_train, X_test, y_train, y_test = data_preparation(path='./chest_xray_images/', image_size=(96, 96))
"""data augmentation using keras"""
augmenter = ImageDataGenerator(rotation_range=90, horizontal_flip=True, vertical_flip=True, rescale=None)
"""model callbacks"""
checkpoint = cb.ModelCheckpoint(filepath='./checkpoints/densenet121/eps={epoch:03d}_valLoss={val_loss:.4f}.hdf5',
monitor='val_loss', verbose=1, save_best_only=True, mode='auto')
cb_list = [checkpoint]
"""create the classifier from pretrained models"""
base_model = DenseNet121(include_top=False, weights='imagenet', input_shape=(96, 96, 3))
base_model_out = base_model.output
# flatten = Flatten(name='flatten_layer')(base_model_out)
pooling = GlobalAveragePooling2D(name='pooling_layer')(base_model_out)
fc1 = Dense(units=10, activation='sigmoid', name='fc1_layer')(pooling)
drop1 = Dropout(rate=0.2, name='dropout1_layer')(fc1)
fc2 = Dense(units=20, activation='sigmoid', name='fc2_layer')(drop1)
drop2 = Dropout(rate=0.2, name='dropout2_layer')(fc2)
out = Dense(units=1, activation='sigmoid', name='output_layer')(drop2)
classifier = Model(inputs=base_model.input, outputs=out)
for layer in classifier.layers[:-6]:
layer.trainable = False
classifier.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
classifier.summary()
print('number of pretrained network layers:', len(classifier.layers))
"""fine-tuning"""
fine_tuning = classifier.fit(augmenter.flow(x=X_train, y=y_train, batch_size=32), callbacks=cb_list, epochs=10,
verbose=1, validation_data=(X_test, y_test))
fig = plt.figure(figsize=(16, 8))
plt.subplot(1, 2, 1)
plt.plot(fine_tuning.history['loss'], color='r', label='training_loss')
plt.plot(fine_tuning.history['val_loss'], color='g', label='validation_loss')
plt.legend()
plt.subplot(1, 2, 2)
plt.plot(fine_tuning.history['accuracy'], color='r', label='training_accuracy')
plt.plot(fine_tuning.history['val_accuracy'], color='g', label='validation_accuracy')
plt.legend()
plt.show()
# fig.savefig('./checkpoints/densenet121/fine_tuning.png')
"""save the model to a json file"""
# model_json = classifier.to_json()
# with open("./checkpoints/densenet121/densenet121.json", "w") as json_file:
# json_file.write(model_json)
"""classification reports"""
y_pred = classifier.predict(X_test)
print('number of test-set images:', len(y_test))
print(y_test)
y_pred = np.round(np.reshape(a=y_pred, newshape=(1, -1)), decimals=2)[0]
print(y_pred)
y_pred_rnd = np.round(np.reshape(a=y_pred, newshape=(1, -1)))[0]
cm = confusion_matrix(y_true=y_test, y_pred=y_pred_rnd)
print('confusion matrix:')
print(cm)
print('test-set accuracy:', (cm[0][0] + cm[1][1])/np.sum(cm))
print('classification report:')
print(classification_report(y_true=y_test, y_pred=y_pred_rnd,
target_names=['normal', 'covid']))
fig1 = plt.figure()
sns.heatmap(data=cm, cmap='Blues', annot=True, annot_kws={'size': 14}, fmt='d',
vmin=0, vmax=len(y_test)/2.)
plt.title('annotated heatmap for confusion matrix')
plt.show()
# fig1.savefig('./checkpoints/densenet121/cm_heatmap.png')
fpr, tpr, _ = roc_curve(y_true=y_test, y_score=y_pred, pos_label=None)
roc_auc = auc(x=fpr, y=tpr)
fig2 = plt.figure()
plt.plot(fpr, tpr, 'b', label='AUC = %0.4f' % roc_auc)
plt.title('Receiver Operating Characteristic')
plt.legend()
plt.plot([0, 1], [0, 1], 'r--')
plt.xlim([0, 1])
plt.ylim([0, 1])
plt.ylabel('True Positive Rate')
plt.xlabel('False Positive Rate')
plt.show()
# fig2.savefig('./checkpoints/densenet121/roc.png')