-
Notifications
You must be signed in to change notification settings - Fork 9
/
nyul_normalization.py
119 lines (79 loc) · 3.76 KB
/
nyul_normalization.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
import numpy as np
import matplotlib as mpl
mpl.use('Agg')
import matplotlib.pyplot as plt
data_dir = '/data1/users/adoyle/'
def normalize(image, mask, target_image, target_mask):
img_shape = image.shape
n_landmarks = 10
valid_orig_image = image[mask]
valid_target_image = target_image[target_mask]
valid_orig_flat = []
for x in range(img_shape[0]):
for y in range(img_shape[1]):
for z in range(img_shape[2]):
if mask[x, y, z] == 1:
valid_orig_flat.append(image[x, y, z])
hist_original = np.histogram(valid_orig_flat, bins=256)
fig, (ax1, ax2, ax3) = plt.subplots(1, 3, sharey=True)
ax1.bar(hist_original[1][:-1], hist_original[0])
ax1.set_title('original histogram')
plt.savefig(data_dir + 'orig_hist.png')
orig_tails = np.percentile(valid_orig_image, (2, 98))
deciles = np.linspace(2, 98, n_landmarks, endpoint=True)
orig_landmarks = np.percentile(valid_orig_image, tuple(deciles))
print('original deciles:', deciles )
print('original landmarks:', orig_landmarks)
target_tails = np.percentile(valid_target_image, (2, 98))
target_landmarks = np.percentile(valid_target_image, tuple(deciles))
hist_target = np.histogram(valid_target_image, bins=256)
ax2.bar(hist_target[1][:-1], hist_target[0])
ax2.set_title('target histogram')
rescaled_image = np.copy(image)
rescaled = []
for i in range(n_landmarks-1):
low_orig, high_orig = orig_landmarks[i], orig_landmarks[i+1]
low_target, high_target = target_landmarks[i], target_landmarks[i+1]
b_orig = low_orig
a_orig = high_orig - low_orig
b_target = low_target
a_target = high_target - low_target
b_transform = b_target - b_orig
a_transform = a_target - a_orig
for x in range(img_shape[0]):
for y in range(img_shape[1]):
for z in range(img_shape[2]):
if mask[x, y, z] == 1:
pixel = image[x, y, z]
if pixel >= low_orig and pixel < high_orig:
rescaled_image[x, y, z] = b_transform + pixel*a_transform
rescaled.append(b_transform + pixel*a_transform)
hist_rescaled = np.histogram(rescaled, bins=256)
ax3.bar(hist_rescaled[1][:-1], hist_rescaled[0])
ax3.set_title('result histogram')
plt.tight_layout()
plt.savefig(data_dir + 'histograms.png')
return rescaled_image
if __name__ == '__main__':
import nibabel as nib
from nibabel.processing import resample_from_to
target_size = (192, 256, 192)
orig = nib.load(data_dir + '/deep_abide/resampled/50002.mnc').get_data()
atlas = nib.load(data_dir + 'mni_icbm152_t1_tal_nlin_asym_09a.mnc')
target = nib.load(data_dir + '/ds030/sub-10225.nii.gz')
target = resample_from_to(target, atlas)
target = target.get_data()
mask = nib.load(data_dir + 'mni_icbm152_t1_tal_nlin_asym_09a_mask.mnc').get_data()
from make_datasets import resize_image_with_crop_or_pad, normalise_zero_one
orig = normalise_zero_one(resize_image_with_crop_or_pad(orig, target_size, mode='constant'))
target = normalise_zero_one(resize_image_with_crop_or_pad(target, target_size, mode='constant'))
mask = np.asarray(resize_image_with_crop_or_pad(mask, target_size, mode='constant'), dtype='bool')
target_mask = np.copy(mask)
returned_image = normalize(orig, mask, target, target_mask)
print('shapes:', orig.shape, target.shape, mask.shape, returned_image.shape)
fig, (ax1, ax2, ax3, ax4) = plt.subplots(1, 4)
ax1.imshow(orig[96, :, :])
ax2.imshow(target[96, :, :])
ax3.imshow(mask[96, :, :])
ax4.imshow(returned_image[96, :, :])
plt.savefig(data_dir + 'nyul_results.png')