def _collect(self):
self.subject_files.clear()
img_dir, label_dir = self.get_img_and_label_dirs()
assert os.path.exists(img_dir)
assert os.path.exists(label_dir)
files_by_id = {}
for file_path in glob.glob(img_dir + '/*') + glob.glob(label_dir +
'/*'):
base_name = os.path.basename(file_path)
id_ = base_name[:12]
if base_name.endswith('_superpixels.png'):
files_by_id.setdefault(id_, {})['superpixel'] = file_path
elif base_name.endswith('_segmentation.png'):
files_by_id.setdefault(id_, {})['gt'] = file_path
elif base_name.endswith('.jpg'):
files_by_id.setdefault(id_, {})['image'] = file_path
for id_, files in files_by_id.items():
assert len(files) == 3, 'id "{}" has not 3 entries'.format(id_)
params = {
'images': {
'image': files['image']
},
'labels': {
'gt': files['gt']
}
}
if self.with_super_pixels:
params['misc'] = {'superpixel': files['superpixel']}
sf = data.SubjectFile(id_, **params)
self.subject_files.append(sf)
def _collect(self):
self.subject_files.clear()
flair_paths = glob.glob(self.root_dir + '/**/*_flair.nii.gz',
recursive=True)
t1_paths = glob.glob(self.root_dir + '/**/*_t1.nii.gz', recursive=True)
t2_paths = glob.glob(self.root_dir + '/**/*_t2.nii.gz', recursive=True)
t1c_paths = glob.glob(self.root_dir + '/**/*_t1ce.nii.gz',
recursive=True)
label_paths = glob.glob(self.root_dir + '/**/*_seg.nii.gz',
recursive=True)
flair_paths.sort()
t1_paths.sort()
t2_paths.sort()
t1c_paths.sort()
label_paths.sort()
if not (len(flair_paths) == len(t1_paths) == len(t2_paths) ==
len(t1c_paths)):
raise ValueError(
'all sequences must have same amount of files in the dataset')
has_gt = len(label_paths) > 0
if has_gt and len(flair_paths) != len(label_paths):
raise ValueError(
'label must have same amount of files as other sequences')
for subject_index in range(len(flair_paths)):
subject_dir = os.path.dirname(flair_paths[subject_index])
identifier = os.path.basename(subject_dir)
if self.crop_brats_prefix:
identifier = identifier[len('BratsXX_'):]
if self.with_grade:
grade = os.path.basename(os.path.dirname(subject_dir))
identifier = '{}_{}'.format(identifier, grade)
image_files = {
'flair': flair_paths[subject_index],
't1': t1_paths[subject_index],
't2': t2_paths[subject_index],
't1c': t1c_paths[subject_index]
}
label_files = {}
if has_gt:
label_files['gt'] = label_paths[subject_index]
sf = data.SubjectFile(identifier,
images=image_files,
labels=label_files)
self.subject_files.append(sf)
def _collect(self):
self.subject_files.clear()
subject_dirs = glob.glob(os.path.join(self.root_dir, '*'))
subject_dirs = list(
filter(
lambda path: os.path.basename(path).lower().startswith(
'subject') and os.path.isdir(path), subject_dirs))
subject_dirs.sort(key=lambda path: os.path.basename(path))
# for each subject
for subject_dir in subject_dirs:
subject = os.path.basename(subject_dir)
images = {
data.FileTypes.Data.name: os.path.join(subject_dir,
'MRFreal.mha')
}
labels = {
data.FileTypes.T1H2Omap.name:
os.path.join(subject_dir, 'T1H2O.mha'),
data.FileTypes.FFmap.name:
os.path.join(subject_dir, 'FF.mha'),
data.FileTypes.B1map.name:
os.path.join(subject_dir, 'B1.mha')
}
mask_fg = {
data.FileTypes.ForegroundTissueMask.name:
os.path.join(subject_dir, 'MASK_FG.mha')
}
mask_t1h2o = {
data.FileTypes.T1H2OTissueMask.name:
os.path.join(subject_dir, 'MASK_FG.mha')
}
sf = pymia_data.SubjectFile(subject,
images=images,
labels=labels,
mask_fg=mask_fg,
mask_t1h2o=mask_t1h2o)
self.subject_files.append(sf)
def _collect(self):
self.subject_files.clear()
subject_dirs = glob.glob(os.path.join(self.root_dir, '*'))
subject_dirs = list(filter(lambda path: os.path.basename(path).lower().startswith('subject')
and os.path.isdir(path),
subject_dirs))
subject_dirs.sort(key=lambda path: os.path.basename(path))
for subject_dir in subject_dirs:
subject = os.path.basename(subject_dir)
# we generate an entry for the coordinates of the points in our clouds
# note that the "---" is an ugly hack to be able to pass two paths
images = {data.FileTypes.COORDINATE.name:
os.path.join(subject_dir, '{}_PROBABILITY.mha'.format(subject)) +
'---' +
os.path.join(subject_dir, '{}_GROUND_TRUTH.mha'.format(subject))
}
# we create an entry for the labels of each point
labels = {data.FileTypes.LABEL.name:
os.path.join(subject_dir, '{}_GROUND_TRUTH.mha'.format(subject))}
indices = {data.FileTypes.INDICES.name:
os.path.join(subject_dir, '{}_PROBABILITY.mha'.format(subject))}
image_information = {data.FileTypes.IMAGE_INFORMATION.name:
os.path.join(subject_dir, '{}_PROBABILITY.mha'.format(subject))}
# we also save the ground truth in image format for easier evaluation
gt = {data.FileTypes.GTM.name:
os.path.join(subject_dir, '{}_GROUND_TRUTH.mha'.format(subject))}
sf = pymia_data.SubjectFile(subject, images=images, labels=labels,
indices=indices,
image_information=image_information,
gt=gt
)
self.subject_files.append(sf)
def _collect(self):
self.subject_files.clear()
files_by_id = {}
for post_fix in self.post_fixes:
post_fix_paths = glob.glob(self.prediction_path +
'/**/*_{}.nii.gz'.format(post_fix),
recursive=True)
for path_ in post_fix_paths:
id_ = os.path.basename(
path_)[:-len('_{}.nii.gz'.format(post_fix))]
files_by_id.setdefault(id_, {})[post_fix] = path_
for id_, files in files_by_id.items():
assert set(files.keys()) == set(self.post_fixes), \
'id "{}" has not all required entries "({})"'.format(id_, list(self.post_fixes))
categories = {}
for post_fix, category in self.post_fix_to_category.items():
categories.setdefault(category, {})[post_fix] = files[post_fix]
sf = data.SubjectFile(id_, **categories)
self.subject_files.append(sf)