/
dataset.py
executable file
·606 lines (519 loc) · 20.6 KB
/
dataset.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
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
"""
Classes for representing datasets of images and/or coordinates.
"""
import copy
import json
import logging
import nibabel as nib
import numpy as np
import pandas as pd
from .base import NiMAREBase
from .utils import (
dict_to_coordinates,
dict_to_df,
get_masker,
get_template,
listify,
mm2vox,
try_prepend,
validate_df,
validate_images_df,
)
LGR = logging.getLogger(__name__)
class Dataset(NiMAREBase):
"""
Storage container for a coordinate- and/or image-based meta-analytic
dataset/database.
Parameters
----------
source : :obj:`str` or :obj:`dict`
JSON file containing dictionary with database information or the dict()
object
target : :obj:`str`, optional
Desired coordinate space for coordinates. Names follow NIDM convention.
Default is 'mni152_2mm' (MNI space with 2x2x2 voxels).
mask : :obj:`str`, :class:`nibabel.nifti1.Nifti1Image`, \
:class:`nilearn.input_data.NiftiMasker` or similar, or None, optional
Mask(er) to use. If None, uses the target space image, with all
non-zero voxels included in the mask.
Attributes
----------
ids : 1D :class:`numpy.ndarray`
Identifiers
masker : :class:`nilearn.input_data.NiftiMasker` or similar
Masker object defining the space and location of the area of interest
(e.g., 'brain').
space : :obj:`str`
Standard space. Same as ``target`` parameter.
annotations : :class:`pandas.DataFrame`
Labels describing studies
coordinates : :class:`pandas.DataFrame`
Peak coordinates from studies
images : :class:`pandas.DataFrame`
Images from studies
metadata : :class:`pandas.DataFrame`
Metadata describing studies
texts : :class:`pandas.DataFrame`
Texts associated with studies
Notes
-----
Images loaded into a Dataset are assumed to be in the same space.
If images have different resolutions or affines from the Dataset's masker,
then they will be resampled automatically, at the point where they're used,
by :obj:`Dataset.masker`.
"""
_id_cols = ["id", "study_id", "contrast_id"]
def __init__(self, source, target="mni152_2mm", mask=None):
if isinstance(source, str):
with open(source, "r") as f_obj:
data = json.load(f_obj)
elif isinstance(source, dict):
data = source
else:
raise Exception("`source` needs to be a file path or a dictionary")
# Datasets are organized by study, then experiment
# To generate unique IDs, we combine study ID with experiment ID
# build list of ids
id_columns = ["id", "study_id", "contrast_id"]
all_ids = []
for pid in data.keys():
for expid in data[pid]["contrasts"].keys():
id_ = "{0}-{1}".format(pid, expid)
all_ids.append([id_, pid, expid])
id_df = pd.DataFrame(columns=id_columns, data=all_ids)
id_df = id_df.set_index("id", drop=False)
self._ids = id_df.index.values
# Set up Masker
if mask is None:
mask = get_template(target, mask="brain")
self.masker = mask
self.space = target
self.annotations = dict_to_df(id_df, data, key="labels")
self.coordinates = dict_to_coordinates(data, masker=self.masker, space=self.space)
self.images = dict_to_df(id_df, data, key="images")
self.metadata = dict_to_df(id_df, data, key="metadata")
self.texts = dict_to_df(id_df, data, key="text")
self.basepath = None
@property
def ids(self):
"""numpy.ndarray: 1D array of identifiers in Dataset.
The associated setter for this property is private, as ``Dataset.ids`` is immutable.
"""
return self.__ids
@ids.setter
def _ids(self, ids):
ids = np.sort(np.asarray(ids))
assert isinstance(ids, np.ndarray) and ids.ndim == 1
self.__ids = ids
@property
def masker(self):
""":class:`nilearn.input_data.NiftiMasker` or similar: Masker object
defining the space and location of the area of interest (e.g., 'brain').
"""
return self.__masker
@masker.setter
def masker(self, mask):
mask = get_masker(mask)
if hasattr(self, "masker") and not np.array_equal(
self.masker.mask_img.affine, mask.mask_img.affine
):
LGR.info(
"New masker does not match old masker. "
"Space is assumed to be the same, but coordinates will "
"be transformed to new matrix."
)
coords = self.coordinates
coords[["i", "j", "k"]] = mm2vox(coords[["x", "y", "z"]], mask.mask_img.affine)
self.coordinates = coords
self.__masker = mask
@property
def annotations(self):
""":class:`pandas.DataFrame`: Labels describing studies in the dataset.
Each study/experiment has its own row.
Columns correspond to individual labels (e.g., 'emotion'), and may
be prefixed with a feature group including two underscores
(e.g., 'Neurosynth_TFIDF__emotion').
"""
return self.__annotations
@annotations.setter
def annotations(self, df):
validate_df(df)
self.__annotations = df.sort_values(by="id")
@property
def coordinates(self):
""":class:`pandas.DataFrame`: Coordinates in the dataset.
Each study has one row for each peak.
Columns include ['x', 'y', 'z'] (peak locations in mm),
['i', 'j', 'k'] (peak locations in voxel index based on Dataset's space),
and 'space' (Dataset's space).
"""
return self.__coordinates
@coordinates.setter
def coordinates(self, df):
validate_df(df)
self.__coordinates = df.sort_values(by="id")
@property
def images(self):
""":class:`pandas.DataFrame`: Images in the dataset.
Each image type has its own column (e.g., 'z') with absolute paths to
files and each study has its own row.
Additionally, relative paths to image files are stored in columns with
the suffix '__relative' (e.g., 'z__relative').
Warnings
--------
Images are assumed to be in the same space, although they may have
different resolutions and affines. Images will be resampled as needed
at the point where they are used, via :obj:`Dataset.masker`.
"""
return self.__images
@images.setter
def images(self, df):
validate_df(df)
self.__images = validate_images_df(df).sort_values(by="id")
@property
def metadata(self):
""":class:`pandas.DataFrame`: Metadata describing studies in the dataset.
Each metadata field has its own column (e.g., 'sample_sizes') and each study
has its own row.
"""
return self.__metadata
@metadata.setter
def metadata(self, df):
validate_df(df)
self.__metadata = df.sort_values(by="id")
@property
def texts(self):
""":class:`pandas.DataFrame`: Texts in the dataset.
Each text type has its own column (e.g., 'abstract') and each study
has its own row.
"""
return self.__texts
@texts.setter
def texts(self, df):
validate_df(df)
self.__texts = df.sort_values(by="id")
def slice(self, ids):
"""
Create a new dataset with only requested IDs.
Parameters
----------
ids : array_like
List of study IDs to include in new dataset
Returns
-------
new_dset : :obj:`nimare.dataset.Dataset`
Reduced Dataset containing only requested studies.
"""
new_dset = copy.deepcopy(self)
new_dset._ids = ids
new_dset.annotations = new_dset.annotations.loc[new_dset.annotations["id"].isin(ids)]
new_dset.coordinates = new_dset.coordinates.loc[new_dset.coordinates["id"].isin(ids)]
new_dset.images = new_dset.images.loc[new_dset.images["id"].isin(ids)]
new_dset.metadata = new_dset.metadata.loc[new_dset.metadata["id"].isin(ids)]
new_dset.texts = new_dset.texts.loc[new_dset.texts["id"].isin(ids)]
return new_dset
def update_path(self, new_path):
"""
Update paths to images. Prepends new path to the relative path for
files in Dataset.images.
Parameters
----------
new_path : :obj:`str`
Path to prepend to relative paths of files in Dataset.images.
"""
self.basepath = new_path
df = self.images
relative_path_cols = [c for c in df if c.endswith("__relative")]
for col in relative_path_cols:
abs_col = col.replace("__relative", "")
if abs_col in df.columns:
LGR.info("Overwriting images column {}".format(abs_col))
df[abs_col] = df[col].apply(try_prepend, prefix=new_path)
self.images = df
def copy(self):
"""Create a copy of the Dataset."""
return copy.deepcopy(self)
def get(self, dict_):
"""
Retrieve files and/or metadata from the current Dataset.
Parameters
----------
dict_ : :obj:`dict`
Dictionary specifying images or metadata to collect.
Keys should be variables to be used as keys for results dictionary.
Values should be tuples with two values:
type (e.g., 'image' or 'metadata') and specific field corresponding
to column of type-specific DataFrame (e.g., 'z' or 'sample_sizes').
Returns
-------
results : :obj:`dict`
A dictionary of lists of requested data.
Examples
--------
>>> dset.get({'z_maps': ('image', 'z'), 'sample_sizes': ('metadata', 'sample_sizes')})
"""
results = {}
results["id"] = self.ids
keep_idx = np.arange(len(self.ids), dtype=int)
for k in dict_:
vals = dict_[k]
if vals[0] == "image":
temp = self.get_images(imtype=vals[1])
elif vals[0] == "metadata":
temp = self.get_metadata(field=vals[1])
elif vals[0] == "coordinates":
temp = [self.coordinates.loc[self.coordinates["id"] == id_] for id_ in self.ids]
else:
raise ValueError('Input "{}" not understood.'.format(vals[0]))
results[k] = temp
temp_keep_idx = np.where([t is not None for t in temp])[0]
keep_idx = np.intersect1d(keep_idx, temp_keep_idx)
# reduce
if len(keep_idx) != len(self.ids):
LGR.info("Retaining {0}/{1} studies".format(len(keep_idx), len(self.ids)))
for k in results:
results[k] = [results[k][i] for i in keep_idx]
if dict_.get(k, [None])[0] == "coordinates":
results[k] = pd.concat(results[k])
return results
def get_labels(self, ids=None):
"""
Extract list of labels for which studies in Dataset have annotations.
Parameters
----------
ids : :obj:`list`, optional
A list of IDs in the Dataset for which to find labels. Default is
None, in which case all labels are returned.
Returns
-------
labels : :obj:`list`
List of labels for which there are annotations in the Dataset.
"""
if not isinstance(ids, list) and ids is not None:
ids = listify(ids)
result = [c for c in self.annotations.columns if c not in self._id_cols]
if ids is not None:
temp_annotations = self.annotations.loc[self.annotations["id"].isin(ids)]
res = temp_annotations[result].any(axis=0)
result = res.loc[res].index.tolist()
return result
def get_texts(self, ids=None, text_type=None):
"""
Extract list of texts of a given type for selected IDs.
Parameters
----------
ids : :obj:`list`, optional
A list of IDs in the Dataset for which to find texts. Default is
None, in which case all texts of requested type are returned.
text_type : :obj:`str`, optional
Type of text to extract. Corresponds to column name in
Dataset.texts DataFrame. Default is None.
Returns
-------
texts : :obj:`list`
List of texts of requested type for selected IDs.
"""
# Rename variables
value = text_type
df = self.texts
return_first = False
if isinstance(ids, str) and value is not None:
return_first = True
ids = listify(ids)
available_types = [c for c in df.columns if c not in self._id_cols]
if (value is not None) and (value not in available_types):
raise ValueError(
'Text type "{0}" not found.\n'
"Available types: "
"{1}".format(value, ", ".join(available_types))
)
if value is not None:
if ids is not None:
result = df[value].loc[df["id"].isin(ids)].tolist()
else:
result = df[value].tolist()
else:
if ids is not None:
result = {v: df[v].loc[df["id"].isin(ids)].tolist() for v in available_types}
result = {k: v for k, v in result.items() if any(v)}
else:
result = {v: df[v].tolist() for v in available_types}
result = list(result.keys())
if return_first:
return result[0]
else:
return result
return result
def get_metadata(self, ids=None, field=None):
"""
Get metadata from Dataset.
Parameters
----------
ids : :obj:`list`, optional
A list of IDs in the Dataset for which to find texts. Default is
None, in which case all texts of requested type are returned.
field : :obj:`str`, optional
Metadata field to extract. Corresponds to column name in
Dataset.metadata DataFrame. Default is None.
Returns
-------
metadata : :obj:`list`
List of values of requested type for selected IDs.
"""
# Rename variables
value = field
df = self.metadata
return_first = False
if isinstance(ids, str) and value is not None:
return_first = True
ids = listify(ids)
available_types = [c for c in df.columns if c not in self._id_cols]
if (value is not None) and (value not in available_types):
raise ValueError(
'Metadata field "{0}" not found.\n'
"Available fields: "
"{1}".format(field, ", ".join(available_types))
)
if value is not None:
if ids is not None:
result = df[value].loc[df["id"].isin(ids)].tolist()
else:
result = df[value].tolist()
else:
if ids is not None:
result = {v: df[v].loc[df["id"].isin(ids)].tolist() for v in available_types}
result = {k: v for k, v in result.items() if any(v)}
else:
result = {v: df[v].tolist() for v in available_types}
result = list(result.keys())
if return_first:
return result[0]
else:
return result
def get_images(self, ids=None, imtype=None):
"""
Get images of a certain type for a subset of studies in the dataset.
Parameters
----------
ids : :obj:`list`, optional
A list of IDs in the Dataset for which to find texts. Default is
None, in which case all texts of requested type are returned.
imtype : :obj:`str`, optional
Type of image to extract. Corresponds to column name in
Dataset.images DataFrame. Default is None.
Returns
-------
images : :obj:`list`
List of images of requested type for selected IDs.
"""
# Rename variables
value = imtype
df = self.images
return_first = False
if isinstance(ids, str) and value is not None:
return_first = True
ids = listify(ids)
metadata_fields = ["space"]
available_types = [c for c in df.columns if c not in self._id_cols]
available_types = [c for c in available_types if not c.endswith("__relative")]
available_types = [c for c in available_types if c not in metadata_fields]
if (value is not None) and (value not in available_types):
raise ValueError(
'Image type "{0}" not found.\n'
"Available types: "
"{1}".format(value, ", ".join(available_types))
)
if value is not None:
if ids is not None:
result = self.images[value].loc[self.images["id"].isin(ids)].tolist()
else:
result = self.images[value].tolist()
else:
if ids is not None:
result = {v: df[v].loc[df["id"].isin(ids)].tolist() for v in available_types}
result = {k: v for k, v in result.items() if any(v)}
else:
result = {v: df[v].tolist() for v in available_types}
result = list(result.keys())
if return_first:
return result[0]
else:
return result
def get_studies_by_label(self, labels=None, label_threshold=0.5):
"""
Extract list of studies with a given label.
Parameters
----------
labels : :obj:`list`, optional
List of labels to use to search Dataset. If a contrast has all of
the labels above the threshold, it will be returned.
Default is None.
label_threshold : :obj:`float`, optional
Default is 0.5.
Returns
-------
found_ids : :obj:`list`
A list of IDs from the Dataset found by the search criteria.
"""
if isinstance(labels, str):
labels = [labels]
elif labels is None:
# For now, labels are all we can search by.
return self.ids
elif not isinstance(labels, list):
raise ValueError('Argument "labels" cannot be {0}'.format(type(labels)))
found_labels = [label for label in labels if label in self.annotations.columns]
temp_annotations = self.annotations[self._id_cols + found_labels]
found_rows = (temp_annotations[found_labels] >= label_threshold).all(axis=1)
if any(found_rows):
found_ids = temp_annotations.loc[found_rows, "id"].tolist()
else:
found_ids = []
return found_ids
def get_studies_by_mask(self, mask):
"""
Extract list of studies with at least one coordinate in mask.
Parameters
----------
mask : img_like
Mask across which to search for coordinates.
Returns
-------
found_ids : :obj:`list`
A list of IDs from the Dataset with at least one focus in the mask.
"""
from scipy.spatial.distance import cdist
if isinstance(mask, str):
mask = nib.load(mask)
dset_mask = self.masker.mask_img
if not np.array_equal(dset_mask.affine, mask.affine):
from nilearn.image import resample_to_img
mask = resample_to_img(mask, dset_mask, interpolation="nearest")
mask_ijk = np.vstack(np.where(mask.get_fdata())).T
distances = cdist(mask_ijk, self.coordinates[["i", "j", "k"]].values)
distances = np.any(distances == 0, axis=0)
found_ids = list(self.coordinates.loc[distances, "id"].unique())
return found_ids
def get_studies_by_coordinate(self, xyz, r=20):
"""
Extract list of studies with at least one focus within radius r of
requested coordinates.
Parameters
----------
xyz : (X x 3) array_like
List of coordinates against which to find studies.
r : :obj:`float`, optional
Radius (in mm) within which to find studies. Default is 20mm.
Returns
-------
found_ids : :obj:`list`
A list of IDs from the Dataset with at least one focus within
radius r of requested coordinates.
"""
from scipy.spatial.distance import cdist
xyz = np.array(xyz)
assert xyz.shape[1] == 3 and xyz.ndim == 2
distances = cdist(xyz, self.coordinates[["x", "y", "z"]].values)
distances = np.any(distances <= r, axis=0)
found_ids = list(self.coordinates.loc[distances, "id"].unique())
return found_ids