def count_num_subjects(self):
"""Count the number of subjects in the experiment."""
if self._radiomix_feature_file:
f = pd.read_excel(self._radiomix_feature_file)
pids = f.values[:, 4]
tocount = pids
elif self._images_train:
tocount = self._images_train[0]
elif self._features_train:
tocount = self._features_train[0]
elif self.images_train:
tocount = self.images_train[0]
elif self.features_train:
tocount = self.features_train[0]
else:
message = 'No features or images given, cannot count number ' +\
' of subjects. Make sure you input at least one of these ' +\
'as source.'
raise WORCValueError(message)
if type(tocount) == dict():
num_subjects = len(list(tocount.keys()))
else:
num_subjects = len(tocount)
self._num_subjects = num_subjects
def find_exampledatadir():
"""Find WORC example data folder."""
# Check if example data directory exists
base_dir = WORCDirectoryDetector().do_detection()
if not base_dir:
raise WORCValueError('WORC installation directory not detected!')
else:
return os.path.join(base_dir, 'exampledata')
def test_combat():
"""Test ComBat feature harmonization."""
# Check if example data directory exists
example_data_dir = th.find_exampledatadir()
# Check if example data required exists
features = glob.glob(
os.path.join(example_data_dir, 'examplefeatures_Patient*.hdf5'))
if len(features) < 7:
message = 'Too few example features for ComBat testing not found! ' +\
'Run the create_example_data script from the WORC exampledata ' +\
'directory!'
raise WORCValueError(message)
elif len(features) > 7:
message = 'Too many example features for ComBat testing not found! ' +\
'Run the create_example_data script from the WORC exampledata ' +\
'directory!'
raise WORCValueError(message)
objectlabels = os.path.join(example_data_dir, 'objectlabels.csv')
# Python
config = os.path.join(example_data_dir, 'ComBatConfig_python.ini')
features_train_out = [
f.replace('examplefeatures_', 'examplefeatures_ComBat_python_')
for f in features
]
# First run synthetic test
# Synthetictest()
# # Run the Combat function: only for training
# ComBat(features_train_in=features,
# labels_train=objectlabels,
# config=config,
# features_train_out=features_train_out)
# # Run the Combat function: now for train + testing
ComBat(features_train_in=features[0:4],
labels_train=objectlabels,
config=config,
features_train_out=features_train_out[0:4],
features_test_in=features[4:],
labels_test=objectlabels,
features_test_out=features_train_out[4:])
def test_iccthreshold():
'''
Test ICC Thresholding statistical testing and computation.
'''
# Check if example data directory exists
example_data_dir = th.find_exampledatadir()
# Check if example data required exists
features = glob.glob(
os.path.join(example_data_dir, 'examplefeatures_Patient*.hdf5'))
if len(features) < 6:
message = 'Too few example features for ICC testing not found!' +\
'Run the create_example_data script from the WORC exampledata ' +\
'directory!'
raise WORCValueError(message)
elif len(features) > 6:
message = 'Too many example features for ICC testing not found!' +\
'Run the create_example_data script from the WORC exampledata ' +\
'directory!'
raise WORCValueError(message)
# Pretend that features are from three observers
features_multi = list()
for i in range(0, 3):
shuffle(features)
features_multi.append(features[:])
features_out = [
i.replace('examplefeatures_', 'examplefeatures_ICC_') for i in features
]
features_out = [features_out, features_out, features_out]
# CSV to save ICC values to
csv_out = os.path.join(example_data_dir, 'ICCValues.csv')
# Run the ICC threshold function: only for training
convert_features_ICC_threshold(features_in=features_multi,
csv_out=csv_out,
features_out=features_out)
# Remove the feature files
for i in glob.glob(
os.path.join(example_data_dir, '*examplefeatures_ICC_*.hdf5')):
os.remove(i)
def find_testdatadir():
"""Find WORC test data folder."""
# Check if example data directory exists
base_dir = WORCDirectoryDetector().do_detection()
if not base_dir:
raise WORCValueError('WORC installation directory not detected!')
else:
testdatadir = os.path.join(base_dir, 'test', 'tmp')
if not os.path.exists(testdatadir):
os.mkdir(testdatadir)
return testdatadir
def ComBatMatlab(dat, batch, command, mod=None, par=1, per_feature='true'):
"""
Run the ComBat Function Matlab script.
par = 0 is non-parametric.
"""
# Mod: default argument is empty list
if mod is None:
mod = []
# TODO: Add check whether matlab executable is found
# Save the features in a .mat MatLab Compatible format
# NOTE: Should change this_folder to a proper temporary directory
this_folder = os.path.dirname(os.path.realpath(__file__))
tempdir = tempfile.gettempdir()
tempfile_in = os.path.join(tempdir, 'combat_input.mat')
tempfile_out = os.path.join(tempdir, 'combat_output.mat')
ComBatFolder = os.path.join(os.path.dirname(this_folder), 'external',
'ComBatHarmonization', 'Matlab', 'scripts')
dict = {
'output': tempfile_out,
'ComBatFolder': ComBatFolder,
'datvar': dat,
'batchvar': batch,
'modvar': mod,
'parvar': par,
'per_feature': per_feature
}
sio.savemat(tempfile_in, dict)
# Make sure there is no tempfile out from the previous run
if os.path.exists(tempfile_out):
os.remove(tempfile_out)
# Run ComBat
currentdir = os.getcwd()
if platform == "linux" or platform == "linux2":
commandseparator = ' ; '
elif platform == "win32":
commandseparator = ' & '
# BIGR Cluster: /cm/shared/apps/matlab/R2015b/bin/matlab
regcommand = ('cd "' + this_folder + '"' + commandseparator + '"' +
command + '" -nodesktop -nosplash -nojvm -r "combatmatlab(' +
"'" + str(tempfile_in) + "'" + ')"' + commandseparator +
'cd "' + currentdir + '"')
print(f'Executing ComBat in Matlab through command: {regcommand}.')
proc = subprocess.Popen(
regcommand,
shell=True,
stdin=subprocess.PIPE,
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
)
proc.wait()
stdout_value, stderr_value = proc.communicate()
# BUG: Waiting does not work, just wait for output to arrive, either with
# the actual output or an error message
succes = False
while succes is False:
if os.path.exists(tempfile_out):
try:
mat_dict = sio.loadmat(tempfile_out)
try:
data_harmonized = mat_dict['data_harmonized']
succes = True
except KeyError:
try:
message = mat_dict['message']
raise WORCValueError(
f'Error in Matlab ComBat execution: {message}.')
except KeyError:
pass
except (sio.matlab.miobase.MatReadError, ValueError):
pass
# Check if expected output file exists
if not os.path.exists(tempfile_out):
raise WORCValueError(
f'Error in Matlab ComBat execution: command: {regcommand}, stdout: {stdout_value}, stderr: {stderr_value}'
)
# Read the output from ComBat
mat_dict = sio.loadmat(tempfile_out)
data_harmonized = mat_dict['data_harmonized']
data_harmonized = np.transpose(data_harmonized)
# Remove temporary files
os.remove(tempfile_out)
os.remove(tempfile_in)
return data_harmonized
def ComBatPython(dat,
batch,
mod=None,
par=1,
eb=1,
per_feature=False,
plotting=False):
"""
Run the ComBat Function python script.
par = 0 is non-parametric.
"""
# convert inputs to neuroCombat format.
covars = dict()
categorical_cols = list()
covars['batch'] = batch
if mod is not None:
for i_mod in range(mod.shape[1]):
label = f'mod_{i_mod}'
covars[label] = [m for m in mod[:, i_mod]]
categorical_cols.append(label)
covars = pd.DataFrame(covars)
batch_col = 'batch'
if par == 0:
parametric = False
elif par == 1:
parametric = True
else:
raise WORCValueError(f'Par should be 0 or 1, now {par}.')
if eb == 0:
eb = False
elif eb == 1:
eb = True
else:
raise WORCValueError(f'eb should be 0 or 1, now {eb}.')
if per_feature == 0:
per_feature = False
elif per_feature == 1:
per_feature = True
else:
raise WORCValueError(
f'per_feature should be 0 or 1, now {per_feature}.')
# execute ComBat
if not per_feature:
data_harmonized = neuroCombat(dat=dat,
covars=covars,
batch_col=batch_col,
categorical_cols=categorical_cols,
eb=eb,
parametric=parametric)
elif per_feature:
print('\t Executing ComBat per feature.')
data_harmonized = np.zeros(dat.shape)
# Shape: (features, samples)
for i in range(dat.shape[0]):
if eb:
# Copy feature + random noise
random_feature = np.random.rand(dat[i, :].shape[0])
feat_temp = np.asarray([dat[i, :], dat[i, :] + random_feature])
else:
# Just use the single feature
feat_temp = np.asarray([dat[i, :]])
feat_temp = neuroCombat(dat=feat_temp,
covars=covars,
batch_col=batch_col,
categorical_cols=categorical_cols,
eb=eb,
parametric=parametric)
data_harmonized[i, :] = feat_temp[0, :]
if plotting:
feat1 = dat[i, :]
feat1_harm = data_harmonized[i, :]
print(len(feat1))
feat1_b1 = [f for f, b in zip(feat1, batch[0]) if b == 1.0]
feat1_b2 = [f for f, b in zip(feat1, batch[0]) if b == 2.0]
print(len(feat1_b1))
print(len(feat1_b2))
feat1_harm_b1 = [
f for f, b in zip(feat1_harm, batch[0]) if b == 1.0
]
feat1_harm_b2 = [
f for f, b in zip(feat1_harm, batch[0]) if b == 2.0
]
plt.figure()
ax = plt.subplot(2, 1, 1)
ax.scatter(np.ones((len(feat1_b1))), feat1_b1, color='red')
ax.scatter(np.ones((len(feat1_b2))) + 1,
feat1_b2,
color='blue')
plt.title('Before Combat')
ax = plt.subplot(2, 1, 2)
ax.scatter(np.ones((len(feat1_b1))),
feat1_harm_b1,
color='red')
ax.scatter(np.ones((len(feat1_b2))) + 1,
feat1_harm_b2,
color='blue')
plt.title('After Combat')
plt.show()
else:
raise WORCValueError(
f'per_feature should be False or True, now {per_feature}.')
return data_harmonized
def test_combat_fastr():
"""Test ComBat feature harmonization."""
# Check if example data directory exists
example_data_dir = th.find_exampledatadir()
# Check if example data required exists
features = glob.glob(
os.path.join(example_data_dir, 'examplefeatures_Patient*.hdf5'))
if len(features) < 6:
message = 'Too few example features for ComBat testing not found!' +\
'Run the create_example_data script from the WORC exampledata ' +\
'directory!'
raise WORCValueError(message)
elif len(features) > 6:
message = 'Too many example features for ComBat testing not found!' +\
'Run the create_example_data script from the WORC exampledata ' +\
'directory!'
raise WORCValueError(message)
objectlabels = os.path.join(example_data_dir, 'objectlabels.csv')
# Python
config = os.path.join(example_data_dir, 'ComBatConfig_python.ini')
# Create the fastr network
experiment = fastr.create_network('test_ComBat')
source_features = experiment.create_source('HDF5',
id='features_in',
node_group='features')
source_labels = experiment.create_source('PatientInfoFile',
id='labels',
node_group='pctrain')
source_config = experiment.create_source('ParameterFile',
id='config',
node_group='conf')
sink_features = experiment.create_sink('HDF5', id='features_out')
node_combat = experiment.create_node(
'combat/ComBat:1.0',
tool_version='1.0',
id='ComBat',
)
link_combat_1 = experiment.create_link(source_config.output,
node_combat.inputs['config'])
link_combat_2 = experiment.create_link(
source_labels.output, node_combat.inputs['patientclass_train'])
link_combat_1.collapse = 'conf'
link_combat_2.collapse = 'pctrain'
# Mimic using two feature toolboxes
links_Combat1_train = node_combat.inputs['features_train'][
'MR_0'] << source_features.output
links_Combat1_train.collapse = 'features'
links_Combat2_train = node_combat.inputs['features_train'][
'MR_1'] << source_features.output
links_Combat2_train.collapse = 'features'
links_Combat_out_train = sink_features.input << node_combat.outputs[
'features_train_out']
links_Combat_out_train.collapse = 'ComBat'
# Provide source and sink data
source_data = dict()
source_data['features_in'] = features
source_data['labels'] = objectlabels
source_data['config'] = config
sink_data = dict()
sink_data[
'features_out'] = "vfs://output/test_ComBat/ComBat/features_ComBat_{{sample_id}}_{{cardinality}}{{ext}}"
# Execute
experiment.execute(source_data,
sink_data,
execution_plugin='LinearExecution')
# Remove the feature files
for i in glob.glob(os.path.join(example_data_dir,
'*features_ComBat*.hdf5')):
os.remove(i)