def main(hdf_file: str, data_dir: str):
keys = [
FileTypes.T1, FileTypes.T2, FileTypes.GT, FileTypes.MASK,
FileTypes.AGE, FileTypes.GPA, FileTypes.SEX
]
crawler = pymia_load.FileSystemDataCrawler(data_dir, keys,
DataSetFilePathGenerator(),
DirectoryFilter(), '.mha')
subjects = [
Subject(id_, file_dict) for id_, file_dict in crawler.data.items()
]
if os.path.exists(hdf_file):
os.remove(hdf_file)
with pymia_crt.get_writer(hdf_file) as writer:
callbacks = pymia_crt.get_default_callbacks(writer)
# normalize the images and unsqueeze the labels and mask.
# Unsqueeze is needed due to the convention to have the number of channels as last dimension.
# I.e., here we have the shape 10 x 256 x 256 before the unsqueeze operation and after 10 x 256 x 256 x 1
transform = pymia_tfm.ComposeTransform([
pymia_tfm.IntensityNormalization(loop_axis=3,
entries=('images', )),
pymia_tfm.UnSqueeze(entries=('labels', 'mask'))
])
traverser = pymia_crt.SubjectFileTraverser()
traverser.traverse(subjects,
callback=callbacks,
load=LoadData(),
transform=transform)
def main(hdf_file: str, data_dir: str):
if os.path.exists(hdf_file):
raise RuntimeError(
'Dataset file "{}" does already exist'.format(hdf_file))
# let's create some sample data
np.random.seed(42) # to have same sample data
create_sample_data(data_dir, no_subjects=8)
# collect the data
collector = Collector(data_dir)
subjects = collector.get_subject_files()
for subject in subjects:
print(subject.subject)
# get the values for parametric map normalization
min_, max_ = get_normalization_values(subjects, LoadData())
with pymia_crt.Hdf5Writer(hdf_file) as writer:
callbacks = pymia_crt.get_default_callbacks(writer)
callbacks.callbacks.append(
WriteNormalizationCallback(writer, min_, max_))
transform = pymia_tfm.ComposeTransform([
tfm.MRFMaskedLabelNormalization(min_, max_, data.ID_MASK_FG),
pymia_tfm.IntensityNormalization(loop_axis=4,
entries=(pymia_def.KEY_IMAGES, )),
])
traverser = pymia_crt.SubjectFileTraverser()
traverser.traverse(subjects,
callback=callbacks,
load=LoadData(),
transform=transform,
concat_fn=concat)
def main(hdf_file: str, data_dir: str):
keys = [
FileTypes.T1, FileTypes.T2, FileTypes.GT, FileTypes.MASK,
FileTypes.AGE, FileTypes.GPA, FileTypes.SEX
]
crawler = pymia_load.FileSystemDataCrawler(data_dir, keys,
DataSetFilePathGenerator(),
DirectoryFilter(), '.mha')
subjects = [
Subject(id_, file_dict) for id_, file_dict in crawler.data.items()
]
if os.path.exists(hdf_file):
os.remove(hdf_file)
with pymia_crt.get_writer(hdf_file) as writer:
callbacks = pymia_crt.get_default_callbacks(writer)
# add a transform to normalize the images
transform = pymia_tfm.IntensityNormalization(loop_axis=3,
entries=('images', ))
traverser = pymia_crt.SubjectFileTraverser()
traverser.traverse(subjects,
callback=callbacks,
load=LoadData(),
transform=transform)
def get_params(out_file: str,
in_dir=dirs.BRATS_ORIG_DATA_DIR,
split_file=os.path.join(dirs.SPLITS_DIR,
'split_brats18_100-25-160.json'),
is_train_data=True,
prediction_path=None,
label_fn=None,
add_prediction_fn=None):
if add_prediction_fn is None:
add_prediction_fn = add_predictions
params = BuildParameters(
in_dir=in_dir,
out_file=out_file,
transforms=[tfm.IntensityNormalization(loop_axis=-1)],
split_file=split_file,
is_train_data=is_train_data,
prediction_path=prediction_path,
add_prediction_fn=add_prediction_fn)
if label_fn is not None:
params.transforms.append(
tfm.LambdaTransform(label_fn, entries=('labels', )))
return params
def main(hdf_file: str, data_dir: str, meta: bool):
# collect the files for each subject
subjects = get_subject_files(data_dir)
# remove the "old" dataset if it exists
if os.path.exists(hdf_file):
os.remove(hdf_file)
with crt.get_writer(hdf_file) as writer:
# initialize the callbacks that will actually write the data to the dataset file
callbacks = crt.get_default_callbacks(writer, meta_only=meta)
# add a transform to normalize the images
transform = tfm.IntensityNormalization(loop_axis=3, entries=(defs.KEY_IMAGES, ))
# run through the subject files (loads them, applies transformations, and calls the callback for writing them)
traverser = crt.Traverser()
traverser.traverse(subjects, callback=callbacks, load=LoadData(), transform=transform)
def main(hdf_file: str, data_dir: str):
if os.path.exists(hdf_file):
raise RuntimeError('Dataset file "{}" does exist already'.format(hdf_file))
# we threshold I_Q at probability 0.1
probability_threshold = 0.1
# we use image information extracted from 5^3 neighborhood around each point
spatial_size = 5
# let's create some sample data
np.random.seed(42) # to have same sample data
create_sample_data(data_dir, no_subjects=8)
# collect the data
collector = Collector(data_dir)
subjects = collector.get_subject_files()
for subject in subjects:
print(subject.subject)
print('Total of {} subjects'.format(len(subjects)))
os.makedirs(os.path.dirname(hdf_file), exist_ok=True)
with pymia_crt.get_writer(hdf_file) as writer:
callbacks = pymia_crt.get_default_callbacks(writer)
transform = pymia_tfm.ComposeTransform([
pymia_tfm.LambdaTransform(lambda_fn=lambda np_data: np_data.astype(np.float32),
entries=('images',
data.KEY_IMAGE_INFORMATION,
)),
pymia_tfm.LambdaTransform(loop_axis=1, entries=('images', ), lambda_fn=normalize_unit_cube),
pymia_tfm.IntensityNormalization(loop_axis=-1, entries=(data.KEY_IMAGE_INFORMATION,))
])
traverser = pymia_crt.SubjectFileTraverser()
traverser.traverse(subjects, callback=callbacks, load=LoadData(probability_threshold, spatial_size),
transform=transform, concat_fn=concat)