-
Notifications
You must be signed in to change notification settings - Fork 0
/
NURAD_Local.py
203 lines (166 loc) · 7.25 KB
/
NURAD_Local.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
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
from sys import platform as sys_pf
# nilearn library for 3d visualizations: https://nilearn.github.io/plotting/index.html#adding-overlays-edges-contours-contour-fillings-markers-scale-bar
# if sys_pf == 'darwin':
# import matplotlib
# matplotlib.use("TkAgg")
import os
import numpy as np
from matplotlib import pyplot as plt
import matplotlib
import SimpleITK as sitk
from scipy import ndimage
import random
import math
import tensorflow as tf
# from tensorflow.keras.models import Sequential
# from tensorflow.keras.layers import Concatenate, Add, Average, Input, Dense, Flatten, BatchNormalization, Activation, LeakyReLU, Reshape
# from tensorflow.keras.layers import Dense, Dropout, Activation, Flatten, Conv2D
# from tensorflow.keras import backend as K
import tensorflow.keras.callbacks as callbacks
import tensorflow.keras.optimizers as optimizers
from sklearn import preprocessing as prepro
from sklearn.model_selection import train_test_split
# from keras.models import *
from tensorflow.keras.layers import *
from tensorflow.keras import Model
# from keras.optimizers import *
from tensorflow.keras.callbacks import ModelCheckpoint, LearningRateScheduler
# from keras import backend as keras
image_shape = (256, 256, 176)
# Read in transformed images
def read_file(filename):
reader = sitk.ImageFileReader()
reader.SetImageIO("NiftiImageIO")
reader.SetFileName(filename)
image = reader.Execute()
# img1 = sitk.ReadImage(folder + item) # alternative way to pull in image
# convert image into np array & perform fft
img = sitk.GetArrayFromImage(image)
# Transpose the image so the first axis is Anterior-Posterior
return img
# Read in non-transformed images
def read_original(filename):
reader = sitk.ImageFileReader()
reader.SetImageIO("NiftiImageIO")
reader.SetFileName(filename)
image = reader.Execute()
# img1 = sitk.ReadImage(folder + item) # alternative way to pull in image
# convert image into np array & perform fft
img = sitk.GetArrayFromImage(image)
# Transpose the image so the first axis is Anterior-Posterior
img = np.transpose(img, (2, 1, 0))
return img
def visualize(orig, new):
plt.figure(figsize=(20, 20))
plt.subplot(121), plt.imshow(orig, cmap='gray')
plt.title('Original'), plt.xticks([]), plt.yticks([])
plt.subplot(122), plt.imshow(new, cmap='gray')
plt.title('New'), plt.xticks([]), plt.yticks([])
def visualize3(orig, blur, predict):
plt.figure(figsize=(30, 30))
plt.subplot(131), plt.imshow(orig, cmap='gray')
plt.title('Original'), plt.xticks([]), plt.yticks([])
plt.subplot(132), plt.imshow(blur, cmap='gray')
plt.title('Blurred'), plt.xticks([]), plt.yticks([])
plt.subplot(133), plt.imshow(predict, cmap='gray')
plt.title('Model Output'), plt.xticks([]), plt.yticks([])
def visualize_model(model, ind, X, y, save=False, fn=''):
x_input = X[ind]
out = model.predict(np.array(x_input[np.newaxis, ...]))
visualize3(y[ind][..., 0], x_input[..., 0], out[0, ..., 0])
if save:
plt.savefig(fn)
def generate_data_3d(t_filenames, o_filenames):
# Takes in the files of the transformed images and the originals to create the pairs
# Also sets the min and max slice numbers to take in
x_data = []
y_data = []
for i in range(len(t_filenames)):
print('Generating data for: ' + t_filenames[i])
x_mri = read_file(t_filenames[i])
y_mri = read_original(o_filenames[i])
x_data.append(
(x_mri - np.amin(x_mri)) / (np.amax(x_mri) - np.amin(x_mri))) # normalizes the intensity to between 0 and 1
y_data.append((y_mri - np.amin(y_mri)) / (np.amax(y_mri) - np.amin(y_mri)))
# Shape of x & y will be [# mris, 256, 256, 176, 1]
x_data = np.array(x_data)[..., np.newaxis]
y_data = np.array(y_data)[..., np.newaxis]
return x_data, y_data
# Simple 3d unet with mse loss
def conv_layer(filts, dim):
# abstracted a single conv layer out since the parameters outside of dimension were kept the same
return Conv3D(filts, dim, activation='relu', padding='same', kernel_initializer='he_normal')
# u_net model
def unet_model(lr=1e-4, input_size=(256, 256, 176, 1), dropout_level=0.1):
inputs = Input(input_size)
conv1 = conv_layer(64, 3)(inputs)
conv1 = conv_layer(64, 3)(conv1)
drop1 = Dropout(dropout_level)(conv1)
pool1 = MaxPooling3D(pool_size=(2, 2, 2))(drop1)
conv2 = conv_layer(128, 3)(pool1)
conv2 = conv_layer(128, 3)(conv2)
drop2 = Dropout(dropout_level)(conv2)
pool2 = MaxPooling3D(pool_size=(2, 2, 2))(drop2)
conv3 = conv_layer(256, 3)(pool2)
conv3 = conv_layer(256, 3)(conv3)
drop3 = Dropout(dropout_level)(conv3)
pool3 = MaxPooling3D(pool_size=(2, 2, 2))(drop3)
conv4 = conv_layer(512, 3)(pool3)
conv4 = conv_layer(512, 3)(conv4)
drop4 = Dropout(dropout_level)(conv4)
pool4 = MaxPooling3D(pool_size=(2, 2, 2))(drop4)
conv5 = conv_layer(1024, 3)(pool4)
conv5 = conv_layer(1024, 3)(conv5)
drop5 = Dropout(dropout_level)(conv5)
# Decoder
up6 = conv_layer(512, 2)(UpSampling3D(size=(2, 2, 2))(drop5))
merge6 = concatenate([drop4, up6], axis=4)
drop6 = Dropout(dropout_level)(merge6)
conv6 = conv_layer(512, 3)(drop6)
conv6 = conv_layer(512, 3)(conv6)
up7 = conv_layer(256, 2)(UpSampling3D(size=(2, 2, 2))(conv6))
merge7 = concatenate([conv3, up7], axis=4)
drop7 = Dropout(dropout_level)(merge7)
conv7 = conv_layer(256, 3)(drop7)
conv7 = conv_layer(256, 3)(conv7)
up8 = conv_layer(128, 2)(UpSampling3D(size=(2, 2, 2))(conv7))
merge8 = concatenate([conv2, up8], axis=4)
drop8 = Dropout(dropout_level)(merge8)
conv8 = conv_layer(128, 3)(drop8)
conv8 = conv_layer(128, 3)(conv8)
up9 = conv_layer(64, 2)(UpSampling3D(size=(2, 2, 2))(conv8))
merge9 = concatenate([conv1, up9], axis=4)
drop9 = Dropout(dropout_level)(merge9)
conv9 = conv_layer(64, 3)(drop9)
conv9 = conv_layer(64, 3)(conv9)
conv9 = conv_layer(2, 3)(conv9)
conv10 = Conv3D(1, 1, activation='linear')(conv9)
model = Model(inputs, conv10)
model.compile(loss="mean_squared_error", optimizer="adam")
model.summary()
return model
# generate 3d training data
trans_folder = '../transforms/'
orig_folder = '../originals/'
X_filenames = []
y_filenames = []
for item in os.listdir(trans_folder):
if item.endswith(".nii"):
X_filenames.append(trans_folder + item)
y_filenames.append(orig_folder + item[:-12] + '.nii')
# X, y = generate_data(['gdrive/My Drive/NU_Rad/transforms/M02_motion5_trans.txt.gz'], ['gdrive/My Drive/NU_Rad/mris/M02.nii'])
# limiting # of files to train on:
lim = 1
X_train, y_train = generate_data_3d(X_filenames[:lim], y_filenames[:lim])
print('Data Generated')
print(np.shape(X_train))
# Checkpoint
check_path = '../models/3d_unet_mse_model'
checkpoint = ModelCheckpoint(check_path, monitor='val_loss', verbose=1, save_best_only=True, mode='min')
callbacks_list = [checkpoint]
# sess = tf.compat.v1.Session(config=tf.compat.v1.ConfigProto(log_device_placement=True))
gpus = tf.config.experimental.list_physical_devices('GPU')
for gpu in gpus:
tf.config.experimental.set_memory_growth(gpu, True)
# model = unet_model()
# model.fit(X_train, y_train, batch_size=1, epochs=3, use_multiprocessing=True, callbacks=callbacks_list, verbose=1)