def __init__(self):
models = []
for model_def in template.models:
if isinstance(model_def, str):
models.append(mdt.get_model(model_def)())
else:
models.append(mdt.get_model(model_def[0])(model_def[1]))
super().__init__(deepcopy(template.name), models)
def __init__(self, *args):
models = []
for model_def in template.models:
if isinstance(model_def, six.string_types):
models.append(mdt.get_model(model_def)())
else:
models.append(
mdt.get_model(model_def[0])(model_def[1]))
new_args = [deepcopy(template.name), models]
for ind, arg in args:
new_args[ind] = arg
super(AutoCreatedCascadeModel, self).__init__(*new_args)
def prepare_charmed_r1_params(params_cube):
"""Set the primary direction of the Tensor to the first CHARMED Restricted compartment. \
Sorts the diffusivities of the Tensor
"""
param_names = mdt.get_model('CHARMED_r1')().get_free_param_names()
params_cube[..., param_names.index('Tensor.theta')] = params_cube[
..., param_names.index('CHARMEDRestricted0.theta')]
params_cube[..., param_names.index('Tensor.phi')] = params_cube[
..., param_names.index('CHARMEDRestricted0.phi')]
params_cube[:, 1:4] = np.sort(params_cube[:, 1:4], axis=1)[:, ::-1]
def prepare_noddi_params(params_cube):
"""Normalizes the w_ic.w and w_ec.w if > 1, else leaves it as is."""
param_names = mdt.get_model('NODDI')().get_free_param_names()
w_ec_w = params_cube[:, param_names.index('w_ec.w')]
w_ic_w = params_cube[:, param_names.index('w_ic.w')]
summed = w_ec_w + w_ic_w
w_ec_w[summed > 1] = (w_ec_w / summed)[summed > 1]
w_ic_w[summed > 1] = (w_ic_w / summed)[summed > 1]
params_cube[:, param_names.index('w_ec.w')] = w_ec_w
params_cube[:, param_names.index('w_ic.w')] = w_ic_w
def prepare_charmed2_params(params_cube):
"""Make sure the weights sum to 1 and sort the Tensor diffusivities"""
param_names = mdt.get_model('CHARMED_r2')().get_free_param_names()
params_cube[..., param_names.index('Tensor.theta')] = params_cube[
..., param_names.index('CHARMEDRestricted0.theta')]
params_cube[..., param_names.index('Tensor.phi')] = params_cube[
..., param_names.index('CHARMEDRestricted0.phi')]
weights_sum = np.sum(params_cube[:, [7, 11]], axis=1)
indices = weights_sum > 1
params_cube[indices, 7] /= weights_sum[indices]
params_cube[indices, 11] /= weights_sum[indices]
params_cube[:, (7, 11)] = np.sort(params_cube[:, (7, 11)], axis=1)[:, ::-1]
params_cube[:, 1:4] = np.sort(params_cube[:, 1:4], axis=1)[:, ::-1]
def create_simulations(protocol_name, model_name):
output_pjoin = pjoin.create_extended(protocol_name, model_name)
if os.path.exists(output_pjoin()):
return
else:
output_pjoin.make_dirs()
model = mdt.get_model(model_name)(volume_selection=False)
param_names = model.get_free_param_names()
model_config = simulations[model_name]
parameters = create_parameter_combinations(10000, [
param_names.index(name)
for name in model_config['randomize_parameters']
],
model.get_initial_parameters(),
model_config['lower_bounds'],
model_config['upper_bounds'],
seed=0)
if model_config['prepare_params_cube_cb'] is not None:
model_config['prepare_params_cube_cb'](parameters)
simulated_signals = mdt.simulate_signals(
model, mdt.load_protocol(pjoin(protocol_name)), parameters)
mdt.write_nifti(parameters[None, :, None, :],
output_pjoin('original_parameters.nii'))
mdt.write_nifti(simulated_signals[None, :, None, :],
output_pjoin('simulated_signals.nii'))
for snr in noise_snrs:
noisy_signals = mdt.add_rician_noise(simulated_signals,
unweighted_signal_height / snr,
seed=0)
mdt.write_nifti(noisy_signals[None, :, None, :],
output_pjoin('noisy_signals_{}.nii'.format(snr)))
mdt.create_blank_mask(
output_pjoin('noisy_signals_{}.nii'.format(noise_snrs[0])),
output_pjoin('mask.nii'))
def simulate_signals(model_name, protocol, parameters):
"""Generate the signal for the given model for each of the parameters.
This function only accepts a 2d list of parameters. For a generated parameters cube use function
simulate_signals_param_cube.
Args:
model_name (str): the name of the model we want to generate the values for
protocol (Protocol): the protocol object we use for generating the signals
parameters (ndarray): the 2d matrix with the parameters for every problem instance
Returns:
signal estimates
"""
problem_data = MockDMRIProblemData(protocol, None, None, None)
model = mdt.get_model(model_name)
model.set_problem_data(problem_data)
signal_evaluate = CalculateModelEstimates()
return signal_evaluate.calculate(model, parameters)
def run_model(self):
model = mdt.get_model(self.modelSelection.currentText())
protocol = mdt.load_protocol(self.selectedProtocol.text())
if not model.is_protocol_sufficient(protocol):
msg = ProtocolWarningBox(model.get_protocol_problems(protocol))
msg.exec_()
return
self._run_model_worker.set_args(
model,
mdt.load_problem_data(self.selectedDWI.text(),
self.selectedProtocol.text(),
self.selectedMask.text(),
noise_std=self._optim_options.noise_std),
self.selectedOutputFolder.text(),
recalculate=True,
double_precision=self._optim_options.double_precision,
only_recalculate_last=not self._optim_options.recalculate_all,
optimizer=self._optim_options.get_optimizer(),
save_user_script_info=False)
self._computations_thread.start()
self._run_model_worker.moveToThread(self._computations_thread)
self._run_model_worker.starting.connect(
self._computations_thread.starting)
self._run_model_worker.finished.connect(
self._computations_thread.finished)
self._run_model_worker.starting.connect(
lambda: self.runButton.setEnabled(False))
self._run_model_worker.finished.connect(
lambda: self.runButton.setEnabled(True))
self._run_model_worker.finished.connect(
lambda: self._shared_state.set_output_folder(
self._get_full_model_output_path()))
self._run_model_worker.starting.emit()
def run_model(self):
model = mdt.get_model(self._get_current_model_name())()
if not model.is_input_data_sufficient(
self._input_data_info.build_input_data()):
msg = ProtocolWarningBox(
model.get_input_data_problems(
self._input_data_info.build_input_data()))
msg.exec_()
return
self._run_model_worker.set_args(
model,
self._input_data_info.build_input_data(),
self.selectedOutputFolder.text(),
recalculate=True,
double_precision=self._optim_options.double_precision,
only_recalculate_last=not self._optim_options.recalculate_all,
method=self._optim_options.method)
self._computations_thread.start()
self._run_model_worker.moveToThread(self._computations_thread)
self._run_model_worker.starting.connect(
self._computations_thread.starting)
self._run_model_worker.finished.connect(
self._computations_thread.finished)
self._run_model_worker.starting.connect(
lambda: self.runButton.setEnabled(False))
self._run_model_worker.finished.connect(
lambda: self.runButton.setEnabled(True))
self._run_model_worker.finished.connect(
lambda: self._shared_state.set_output_folder(
self._get_full_model_output_path()))
image_path = split_image_path(self._input_data_info.dwi)
script_basename = os.path.join(
image_path[0], 'scripts', 'fit_model_{}_{}'.format(
self._get_current_model_name().replace('|', '.'),
image_path[1]))
if not os.path.isdir(os.path.join(image_path[0], 'scripts')):
os.makedirs(os.path.join(image_path[0], 'scripts'))
script_info = dict(
optim_options=self._optim_options,
input_data_info=self._input_data_info,
model=self._get_current_model_name(),
output_folder=self.selectedOutputFolder.text(),
recalculate=True,
double_precision=self._optim_options.double_precision,
only_recalculate_last=not self._optim_options.recalculate_all)
self._run_model_worker.finished.connect(
lambda: self._write_python_script_file(script_basename + '.py', **
script_info))
self._run_model_worker.finished.connect(
lambda: self._write_bash_script_file(script_basename + '.sh', **
script_info))
self._run_model_worker.starting.emit()
