forked from multimodallearning/hierarchical-dense-ssl
-
Notifications
You must be signed in to change notification settings - Fork 0
/
preprocess_pretraining_data.py
136 lines (107 loc) · 4.82 KB
/
preprocess_pretraining_data.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
import os
import amid
import random
import pathlib
import subprocess
import numpy as np
import pandas as pd
import nibabel as nib
import reorient_nii as ornt
from matplotlib import pyplot as plt
AMOS_DIR = '/path/to/amos_raw_data_dir/'
FLARE_DIR = '/path/to/flare_raw_data_dir/'
NAKO_DIR = '/path/to/nako_raw_data_dir/'
C3D_TOOL = '/path/to/c3d_tool'
OUTPUT_DIR = '/path/to/output_dir/'
dataset = 'nako' # can be amos, flare, nako
spacing_x = 1.5 # 1.0 for ct 1.5 for mri
spacing_y = 1.5 # 1.0 for ct 1.5 for mri
spacing_z = 1.5 # 2.0 for ct 1.5 for mri
percentile_min = 1 # lower percentile to calculate lower bound for clipping 'mri' images, will be used only if modality == 'mri'
percentile_max = 99.99 # upper percentile to calculate upper bound for clipping 'mri' images, will be used only if modality == 'mri'
vol_clip_min = -175 # -175 is a lower bound for CT abdomen window, will be used only if modality == 'ct'
vol_clip_max = 250 # 250 is an upper bound for CT abdomen window, will be used only if modality == 'ct'
C3D_TASK_STRING_IMG = \
f"-interpolation Linear " \
f"-resample-mm {spacing_x}x{spacing_y}x{spacing_z}mm " \
if dataset == 'amos':
data = amid.AMOS(root=AMOS_DIR)
ids = data.ids[:500]
elif dataset == 'flare':
data = amid.FLARE2022(root=FLARE_DIR)
ids = [data_id for data_id in data.ids if data_id.startswith('TU')]
elif dataset == 'nako':
ids = list(pathlib.Path(NAKO_DIR).rglob('*2_3D_GRE_TRA_W*.nii.gz'))
output_dir_images = os.path.join(OUTPUT_DIR, dataset, 'images')
output_dir_visualizations = os.path.join(OUTPUT_DIR, dataset, 'visualizations')
output_file_analisys = os.path.join(OUTPUT_DIR, dataset, 'data_analisys.csv')
os.makedirs(output_dir_images, exist_ok=True)
os.makedirs(output_dir_visualizations, exist_ok=True)
image_names = []
shape_x, shape_y, shape_z = [], [], []
spacings_x, spacings_y, spacings_z = [], [], []
min_val, max_val = [], []
orientation = []
for idx, data_id in enumerate(ids):
if dataset == 'flare' or dataset == 'amos':
image_names.append(data_id)
output_path = pathlib.Path(output_dir_images).joinpath(data_id + '.nii.gz')
nib.save(nib.Nifti1Image(data.image(data_id), affine=data.affine(data_id)), output_path)
image_path = output_path
elif dataset == 'nako':
image_names.append(data_id.name)
image_path = data_id
output_path = pathlib.Path(output_dir_images).joinpath(data_id.name)
args = C3D_TOOL + " " + str(image_path) + " " + C3D_TASK_STRING_IMG + "-o " + str(output_path)
subprocess.run(args, shell=True)
nib_image = nib.load(output_path)
if ornt.get_orientation(nib_image) != 'RAS':
nib_image = ornt.reorient(nib_image, 'RAS')
orientation.append(ornt.get_orientation(nib_image))
np_image = nib_image.get_fdata()
header = nib_image.header
shape_x.append(np_image.shape[0])
shape_y.append(np_image.shape[1])
shape_z.append(np_image.shape[2])
spacing = header.get_zooms()
spacings_x.append(np.round(spacing[0], 1))
spacings_y.append(np.round(spacing[1], 1))
spacings_z.append(np.round(spacing[2], 1))
if dataset == 'flare' or dataset == 'amos':
np_image = np.clip(np_image, vol_clip_min, vol_clip_max)
np_image = (np_image - vol_clip_min) / (vol_clip_max - vol_clip_min)
nib.save(nib.Nifti1Image(np_image.astype(np.float32), affine=nib_image.affine), output_path)
elif dataset == 'nako':
vol_clip_min = np.percentile(np_image, percentile_min)
vol_clip_max = np.percentile(np_image, percentile_max)
np_image = np.clip(np_image, vol_clip_min, vol_clip_max)
np_image = (np_image - vol_clip_min) / (vol_clip_max - vol_clip_min)
nib.save(nib.Nifti1Image(np_image.astype(np.float32), affine=nib_image.affine), output_path)
min_val.append(np.min(np_image))
max_val.append(np.max(np_image))
if ((idx + 1) % 10) == 0:
print(f'{idx + 1} images processed')
df = pd.DataFrame(data={
'path': image_names,
'shape_x': shape_x,
'shape_y': shape_y,
'shape_z': shape_z,
'spacing_x': spacings_x,
'spacing_y': spacings_y,
'spacing_z': spacings_z,
'min_val': min_val,
'max_val': max_val,
'orientation': orientation
})
df.to_csv(output_file_analisys)
images_paths = sorted(list(pathlib.Path(output_dir_images).glob('*.nii.gz')))
random_sampled_images_paths = random.choices(images_paths, k=20)
for image_path in random_sampled_images_paths:
np_image = nib.load(image_path).get_fdata()
center_slice = np_image.shape[2] // 2
for image_slice in range(center_slice - 2, center_slice + 3):
plt.imshow(np_image[:, :, image_slice], cmap='gray')
plt.colorbar()
output_path = pathlib.Path(output_dir_visualizations).joinpath(image_path.name[:-7] + '_z' + str(image_slice) + '.png')
plt.savefig(output_path)
plt.close()