-
Notifications
You must be signed in to change notification settings - Fork 15
/
set_images.py
317 lines (257 loc) · 12.2 KB
/
set_images.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
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
import numpy as np
import pandas as pd
from re import sub
from sys import argv, exit
from glob import glob
import os
from os.path import basename, exists
from os import makedirs
import time
from utils import *
np.random.seed(int(time.time()))
def gather_dirs(source_dir, input_str='acq', ext='mnc'):
''' This function takes a source directory and parse it to extract the
information and returns it as lists of strings.
Args:
source_dir (str): the directory where the data is
input_str (str): string that uniquely identifies the input (pet) images in subject directory
ext (str) : string the identifies extension of file (default==nifti)
Returns:
subject_dir (list of str): a list of directories, one per subject
pet_list (list of str): a list of directories, one per pet image
t1_list (list of str): a list of directories, one per t1 image
names (list of str): a list of subject names
'''
input_str = os.path.splitext(input_str)[0]
subject_dirs = glob(source_dir + os.sep + '*', recursive=True)
pet_list = glob(source_dir + os.sep + '**' + os.sep + '*'+input_str+'*.'+ext, recursive=True)
if pet_list == [] :
print("Error -- could not find any files for :")
print(source_dir + os.sep + '**' + os.sep + '*'+input_str+'.'+ext)
exit(1)
names = [basename(f) for f in subject_dirs]
#return(subject_dirs, pet_list, t1_list, names)
return(subject_dirs, pet_list, names)
def print_error_nosubject(source_dir):
''' This function displays an error message if the source directory
is empty, then exits the execution.
Args:
source_dir (str): the directory where the data is (not)
'''
print('No data found in directory :' + source_dir)
print('Are you sure this is the right directory ?')
print('Exiting gracefully')
exit()
def print_error_nosumone(ratio):
''' This function displays an error message if the ratio sum
is not one.
Args:
ratio (list): a list containing the proportions of train/test
subjects. should sum to 1.
'''
print('The sum of ' + str(ratio) + ' is not one.')
print('Check again your test/train ratios')
print('Exiting gracefully')
exit()
#def createdf(name, pet_names, pet, t1, labels, task=False):
def createdf(name, pet_names, pet, labels, task=False):
''' This function creates a dataframe for a given subject.
The dataframe contains information about the files related to
the given subject.
Args:
name (str) : the name of the subject
(example: 'subject-001')
pet_names (list) : a list of the radiotracer used
(example: [raclopride, FDG])
pet (list): a list of (mnc) file names where pet information is stored.
(example: ['../data/sub-02/sub-02_acq-raclopride_pet.mnc',
'../data/sub-02/sub-02_acq-FDG_pet.mnc'] )
t1 (str): a (mnc) file name where the t1 information is stored
(example: '../data/sub-02/sub-02_t1.mnc')
labels (list): a list of str of (mnc) file name where the label is.
(example: ['../data/sub-02/sub-02_labels_brainmask.mnc'])
task (bool or list of str): leave it to False if your data has no task
Returns:
datad (pd.DataFrame): a dataframe that has all these informations
in the right format and number of repetition by subject.
'''
n = len(pet)
datad = {}
datad['subject'] = [name] * n
if pet_names != []: datad['radiotracer'] = pet_names
datad['pet'] = pet
#datad['t1'] = [t1] * n
if task is not False:
datad['label'] = labels
datad['task'] = task
else:
datad['label'] = labels * n
return(datad)
def process(name, source_dir, pet_list, label_str='brainmask', ext='mnc' ):
#def process(name, source_dir, pet_list, t1_list, label_str='brainmask', ext='mnc' ):
''' That function returns a dataframe that has all the information
about a subject. At this point of the code, the category of
a subject (train/test) is still unknown. Given the presence or
absence of 'tasks', the returned dataframe has 5 or 6 columns.
Args:
name(str): the name of the subject
source_dir(str): the directory where the data is
pet_list(list): a list of (mnc) file names where pet
information is stored.
t1_list (list): a list of file names. these files are all mnc t1s.
Returns:
data_subject (pd.DataFrame): a dataframe that has all the information
about a subject. The number of columns depends on the presence of
tasks (ie augmented data).
'''
pet = [f for f in pet_list if name in f]
#t1 = [f for f in t1_list if name in f]
#if not t1 == [] : t1=t1[0]
#else:
# print('Warning: Subject name '+name+' not found in list of t1 images.')
# return(1)
pet_names = [sub('.mnc', '', sub('acq-', '', g))
for f in pet for g in f.split('_') if 'acq' in g]
task_names = [sub('task-', '', g)
for f in pet for g in f.split('_') if 'task' in g]
label_str = os.path.splitext(label_str)[0]
if len(task_names) == 0:
label = glob(source_dir + os.sep + '**' + os.sep + '*'+label_str+'*.'+ext, recursive=True)
data_subject = createdf(name, pet_names, pet, label, task=False)
else :
labels = []
for p, t in zip(pet, task_names):
label_fn = glob(source_dir + os.sep + '**' + os.sep + name + '*' + t + '*'+label_str+'*.'+ext, recursive=True)
if not label_fn == []: label_fn = label_fn[0]
else:
print('Warning: could not find label for ', name, 'with the form:')
print(source_dir + os.sep + '**' + os.sep + name + '*' + t + '*'+label_str+'.'+ext)
return(1)
labels.append(label_fn)
#data_subject = createdf(name, pet_names,pet, t1, labels,task=task_names)
data_subject = createdf(name, pet_names,pet, labels,task=task_names)
return(data_subject)
def create_out(dfd):
''' Function that goes from a dict of df to a single df (successive
concatenation).
Args :
dfd (dict): a dictionary of pd.DataFrame
Returns:
out (pd.DataFrame): a pd.DataFrame that is a concatenation of all the
pd.DataFrames in dfd. It also has an extra 'category' column and
the index is reset.
'''
if len(list(dfd.keys())) == 0 :
print('Error: subject data frame was empty')
exit(1)
out = pd.DataFrame(columns=dfd[list(dfd.keys())[0]].columns)
for k, v in dfd.items():
out = pd.concat([out, v]) #,sort=True)
out["category"] = "unknown"
out.reset_index(inplace=True, drop=True)
return(out)
def attribute_category(out, category, category_class,ratio, verbose=1):
''' This function distributes each subject in a 'train' or 'test' category. The 'train' and 'test'
categories are assigned so as to make sure that all of the different radiotracers are contained
within the 'train' category.
Args:
out (pd.DataFrame): a pd.DataFrame that contains the info of all files
by subject.
ratios (list): a list containing the proportions of train/test
subjects. should sum to 1 and supposedly it has been tested before.
Returns:
out (pd.DataFrame): a pd.DataFrame that contains the info of all files
by subject where the 'category' column has been set to either
train or test depending the result of the random draw.
The value of test or train is the same for a given subject.
'''
nImages=out.shape[0]
n = int(round(nImages * ratio))
i=0
category_classes = pd.Series(out[category_class])
unique_category_classes = np.unique(category_classes)
while True :
for r in unique_category_classes :
unknown_df = out[ (out.category == "unknown") & (category_classes == r) ]
n_unknown = unknown_df.shape[0]
if n_unknown == 0: continue
random_i = np.random.randint(0,n_unknown)
row = out[out.category == "unknown"].iloc[random_i,]
out.loc[ out.index[out.subject == row.subject ], 'category' ]=category
i += out.loc[ out.index[out.subject == row.subject ], 'category' ].shape[0]
if i >= n : break
n_unknown = out[ (out.category == "unknown") & (category_classes == r) ].shape[0]
if i >= n or n_unknown == 0 : break
if verbose > 0 :
print(category, ": expected/real ratio = %3.2f / %3.2f" % (100. * ratio, 100.*out.category.loc[ out.category ==category].shape[0]/nImages ))
def set_valid_samples(images):
'''for each image, identify the number of samples that are valid. some samples should be excluded because they contain
bad or no information
'''
total_slices=0
images['valid_samples']=np.repeat(0, images.shape[0])
images['total_samples']=np.repeat(0, images.shape[0])
for index, row in images.iterrows():
#meera
#as i mentioned in the comments in the utils.py script, it would be good if
#your new version of the safe_h5py_open function just returned an array
minc_pet_f = safe_h5py_open(row.pet, 'r')
pet=np.array(minc_pet_f['minc-2.0/']['image']['0']['image'])
pet = normalize(pet)
#meera
#If a 4D image is used, then the next two lines just sums over the time dimension.
#With minc files the time dimension is the first (i.e., 0) dimension, but I think with
#nifti images it might be the opposite.
time_dimension=0
if len(pet.shape) == 4: pet = np.sum(pet, axis=time_dimension)
images['total_samples'].iloc[index] = pet.shape[0]
valid_slices = 0
for j in range(pet.shape[0]):
if pet[j].sum() != 0 :
valid_slices += 1
images['valid_samples'].iloc[index] = valid_slices
total_slices += valid_slices
return(total_slices)
def set_images(source_dir, ratios, images_fn, input_str='pet', label_str='brainmask', ext='mnc' ):
''' This function takes a source directory, where the data is, and a
ratio list (split test/train).
It returns a pd.DataFrame that links file names to concepts, like
t1 of subject 2 or pet-FDG for subject 15.
This dataframe is exported is csv.
Args:
source_dir (str): the directory where the data is
input_str (str): string used to identify input files
label_str (str): string used to identify label files
ext (str) : string the identifies extension of file (default==nifti)
Returns:
out (pd.DataFrame): a dataframe that synthesises the information
of the source_dir.
'''
# 1 - gathering information (parsing the source directory)
subject_dirs, pet_list, names = gather_dirs(source_dir, input_str, ext )
# 2 - checking for potential errors
if len(names) == 0:
print_error_nosubject(source_dir)
if sum(ratios) > 1.:
print_error_nosumone(ratios)
# 3 - creating an empty directory of dataframes, then filling it.
dfd = {}
for name in names:
data_subject = process(name, source_dir, pet_list, label_str, ext)
#data_subject = process(name, source_dir, pet_list, t1_list, label_str, ext)
if not data_subject == 1: dfd[name] = pd.DataFrame(data_subject) # formerly subject_df
# 4 - concatenation of the dict of df to a single df
out = create_out(dfd)
# 5 - attributing a train/validate/test category for all subject
if "radiotracer" in out.columns : category_class="radiotracer"
else : category_class="task"
attribute_category(out, 'train',category_class, ratios[0])
attribute_category(out, 'validate',category_class, ratios[1])
out.category.loc[ out.category=="unknown" ] = "test"
#5.5 Set the number of valid samples per image (some samples exluded because they contain no information)
set_valid_samples(out)
# 6 - export and return
out.to_csv(images_fn, index=False)
return out
if __name__ == '__main__':
set_images(argv[1], [0.7, 0.3])