def test_calc_noise():
# Inputs for functions
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 200
tr_number = int(np.floor(duration / tr_duration))
dimensions_tr = np.array([10, 10, 10, tr_number])
# Preset the noise dict
nd_orig = {
'auto_reg_sigma': 0.6,
'drift_sigma': 0.4,
'snr': 30,
'sfnr': 30,
'max_activity': 1000,
'fwhm': 4,
}
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(
onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
# Mask the volume to be the same shape as a brain
mask, template = sim.mask_brain(dimensions_tr, mask_threshold=0.2)
stimfunction_tr = stimfunction[::int(tr_duration * 100)]
noise = sim.generate_noise(
dimensions=dimensions_tr[0:3],
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=nd_orig,
)
# Check that noise_system is being calculated correctly
spatial_sd = 5
temporal_sd = 5
noise_system = sim._generate_noise_system(dimensions_tr, spatial_sd,
temporal_sd)
precision = abs(noise_system[0, 0, 0, :].std() - spatial_sd)
assert precision < spatial_sd, 'noise_system calculated incorrectly'
precision = abs(noise_system[:, :, :, 0].std() - temporal_sd)
assert precision < spatial_sd, 'noise_system calculated incorrectly'
# Calculate the noise
nd_calc = sim.calc_noise(volume=noise, mask=mask)
# How precise are these estimates
precision = abs(nd_calc['snr'] - nd_orig['snr'])
assert precision < nd_orig['snr'], 'snr calculated incorrectly'
precision = abs(nd_calc['sfnr'] - nd_orig['sfnr'])
assert precision < nd_orig['sfnr'], 'sfnr calculated incorrectly'
def test_calc_noise():
# Inputs for functions
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 200
tr_number = int(np.floor(duration / tr_duration))
dimensions_tr = np.array([10, 10, 10, tr_number])
# Preset the noise dict
nd_orig = {'auto_reg_sigma': 0.6,
'drift_sigma': 0.4,
'snr': 30,
'sfnr': 30,
'max_activity': 1000,
'fwhm': 4,
}
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
# Mask the volume to be the same shape as a brain
mask, template = sim.mask_brain(dimensions_tr, mask_threshold=0.2)
stimfunction_tr = stimfunction[::int(tr_duration * 100)]
noise = sim.generate_noise(dimensions=dimensions_tr[0:3],
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=nd_orig,
)
# Check that noise_system is being calculated correctly
spatial_sd = 5
temporal_sd = 5
noise_system = sim._generate_noise_system(dimensions_tr,
spatial_sd,
temporal_sd)
precision = abs(noise_system[0, 0, 0, :].std() - spatial_sd)
assert precision < spatial_sd, 'noise_system calculated incorrectly'
precision = abs(noise_system[:, :, :, 0].std() - temporal_sd)
assert precision < spatial_sd, 'noise_system calculated incorrectly'
# Calculate the noise
nd_calc = sim.calc_noise(volume=noise,
mask=mask)
# How precise are these estimates
precision = abs(nd_calc['snr'] - nd_orig['snr'])
assert precision < nd_orig['snr'], 'snr calculated incorrectly'
precision = abs(nd_calc['sfnr'] - nd_orig['sfnr'])
assert precision < nd_orig['sfnr'], 'sfnr calculated incorrectly'
def test_calc_noise():
# Inputs for functions
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 100
tr_number = int(np.floor(duration / tr_duration))
dimensions_tr = np.array([10, 10, 10, tr_number])
# Preset the noise dict
nd_orig = {
'auto_reg_sigma': 1,
'drift_sigma': 0.5,
'overall': 0.1,
'snr': 30,
'spatial_sigma': 0.15,
'system_sigma': 1,
}
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(
onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
# Mask the volume to be the same shape as a brain
mask = sim.mask_brain(dimensions_tr)
noise = sim.generate_noise(
dimensions=dimensions_tr[0:3],
stimfunction=stimfunction,
tr_duration=tr_duration,
mask=mask,
noise_dict=nd_orig,
)
# Calculate the noise
nd_calc = sim.calc_noise(noise, mask)
assert abs(nd_calc['overall'] - nd_orig['overall']) < 0.1, 'overall ' \
'calculated ' \
'incorrectly'
assert abs(nd_calc['snr'] - nd_orig['snr']) < 10, 'snr calculated ' \
'incorrectly'
assert abs(nd_calc['system_sigma'] - nd_orig['system_sigma']) < 1, \
'snr calculated incorrectly'
def test_calc_noise():
# Inputs for functions
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 100
tr_number = int(np.floor(duration / tr_duration))
dimensions_tr = np.array([10, 10, 10, tr_number])
# Preset the noise dict
nd_orig = {
'auto_reg_sigma': 0.6,
'drift_sigma': 0.4,
'temporal_noise': 5,
'sfnr': 30,
'max_activity': 1000,
'fwhm': 4,
}
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(
onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
# Mask the volume to be the same shape as a brain
mask = sim.mask_brain(dimensions_tr)
stimfunction_tr = stimfunction[::int(tr_duration * 1000)]
noise = sim.generate_noise(
dimensions=dimensions_tr[0:3],
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
mask=mask,
noise_dict=nd_orig,
)
# Calculate the noise
nd_calc = sim.calc_noise(noise, mask)
# How precise are these estimates
precision = abs(nd_calc['temporal_noise'] - nd_orig['temporal_noise'])
assert precision < 1, 'temporal_noise calculated incorrectly'
precision = abs(nd_calc['sfnr'] - nd_orig['sfnr'])
assert precision < 5, 'sfnr calculated incorrectly'
def test_calc_noise():
# Inputs for functions
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 100
tr_number = int(np.floor(duration / tr_duration))
dimensions_tr = np.array([10, 10, 10, tr_number])
# Preset the noise dict
nd_orig = {'auto_reg_sigma': 1,
'drift_sigma': 0.5,
'overall': 0.1,
'snr': 30,
'spatial_sigma': 0.15,
'system_sigma': 1,
}
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
# Mask the volume to be the same shape as a brain
mask = sim.mask_brain(dimensions_tr)
noise = sim.generate_noise(dimensions=dimensions_tr[0:3],
stimfunction=stimfunction,
tr_duration=tr_duration,
mask=mask,
noise_dict=nd_orig,
)
# Calculate the noise
nd_calc = sim.calc_noise(noise, mask)
assert abs(nd_calc['overall'] - nd_orig['overall']) < 0.1, 'overall ' \
'calculated ' \
'incorrectly'
assert abs(nd_calc['snr'] - nd_orig['snr']) < 10, 'snr calculated ' \
'incorrectly'
assert abs(nd_calc['system_sigma'] - nd_orig['system_sigma']) < 1, \
'snr calculated incorrectly'
signal *= mask > 0
# Iterate through the participants and store participants
epochs = []
for participantcounter in range(1, participants + 1):
# Add the epoch cube
epochs += [epoch]
# Save a file name
savename = directory + 'p' + str(participantcounter) + '.nii'
# Create the noise volumes (using the default parameters
noise = sim.generate_noise(dimensions=dimensions,
stimfunction=stimfunction,
tr_duration=tr_duration,
mask=mask,
)
# Combine the signal and the noise
brain = signal + noise
# Save the volume
affine_matrix = np.diag([-1, 1, 1, 1]) # LR gets flipped
brain_nifti = nibabel.Nifti1Image(brain, affine_matrix)
nibabel.save(brain_nifti, savename)
# Save the epochs
np.save(directory + 'epoch_labels.npy', epochs)
# Store the mask
def test_generate_noise():
dimensions = np.array([64, 64, 36]) # What is the size of the brain
feature_size = [2]
feature_type = ['cube']
feature_coordinates = np.array(
[[32, 32, 18]])
signal_magnitude = [30]
# Generate a volume representing the location and quality of the signal
volume_static = sim.generate_signal(dimensions=dimensions,
feature_coordinates=feature_coordinates,
feature_type=feature_type,
feature_size=feature_size,
signal_magnitude=signal_magnitude,
)
# Inputs for generate_stimfunction
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 100
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(onsets=onsets,
event_durations=event_durations,
total_time=duration,
tr_duration=tr_duration,
)
signal_function = sim.double_gamma_hrf(stimfunction=stimfunction,
)
# Convolve the HRF with the stimulus sequence
signal = sim.apply_signal(signal_function=signal_function,
volume_static=volume_static,
)
# Create the noise volumes (using the default parameters)
noise = sim.generate_noise(dimensions=dimensions,
stimfunction=stimfunction,
tr_duration=tr_duration,
)
assert signal.shape == noise.shape, "The dimensions of signal " \
"and noise the same"
Z_noise = sim._generate_noise_temporal(stimfunction, tr_duration, 1)
noise = sim._generate_noise_temporal(stimfunction, tr_duration, 0)
assert np.std(Z_noise) < np.std(noise), "Z scoring is not working"
# Combine the signal and the noise
volume = signal + noise
assert np.std(signal) < np.std(noise), "Noise was not created"
noise = sim.generate_noise(dimensions=dimensions,
stimfunction=stimfunction,
tr_duration=tr_duration,
noise_strength=[0, 0, 0]
)
assert np.sum(noise) == 0, "Noise strength could not be manipulated"
assert np.std(noise) == 0, "Noise strength could not be manipulated"
else:
# Load the file name instead
with open(input_noise_dict_name, 'r') as f:
noise_dict = f.read()
print('Loading ' + input_noise_dict_name)
noise_dict = eval(noise_dict)
print('Generating brain for this permutation ')
noise_dict['matched'] = match_noise
brain = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=np.zeros((trs, 1)),
tr_duration=int(tr_duration),
template=template,
mask=mask,
noise_dict=noise_dict,
)
# Save the node brain (if you are given an output name)
if output_name is not None:
print('Saving noise volume')
nii = nibabel.Nifti1Image(brain.astype('int16'), nii.affine)
hdr = nii.header
hdr.set_zooms((dimsize[0], dimsize[1], dimsize[2], dimsize[3]))
nibabel.save(nii, output_name) # Save
# Calculate and save the output dict if a value is supplied
if output_noise_dict_name is not None:
print('Testing noise generation')
def test_generate_noise():
dimensions = np.array([10, 10, 10]) # What is the size of the brain
feature_size = [2]
feature_type = ['cube']
feature_coordinates = np.array([[5, 5, 5]])
signal_magnitude = [1]
# Generate a volume representing the location and quality of the signal
volume = sim.generate_signal(
dimensions=dimensions,
feature_coordinates=feature_coordinates,
feature_type=feature_type,
feature_size=feature_size,
signal_magnitude=signal_magnitude,
)
# Inputs for generate_stimfunction
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 100
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(
onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
signal_function = sim.double_gamma_hrf(
stimfunction=stimfunction,
tr_duration=tr_duration,
)
# Convolve the HRF with the stimulus sequence
signal = sim.apply_signal(
signal_function=signal_function,
volume_static=volume,
)
# Generate the mask of the signal
mask = sim.mask_brain(signal, mask_threshold=0.1)
assert min(mask[mask > 0]) > 0.1, "Mask thresholding did not work"
stimfunction_tr = stimfunction[::int(tr_duration * 1000)]
# Create the noise volumes (using the default parameters)
noise = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
mask=mask,
)
assert signal.shape == noise.shape, "The dimensions of signal and noise " \
"the same"
assert np.std(signal) < np.std(noise), "Noise was not created"
noise = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
mask=mask,
noise_dict={
'temporal_noise': 0,
'sfnr': 10000
},
)
temporal_noise = np.std(noise[mask[:, :, :, 0] > 0], 1).mean()
assert temporal_noise <= 0.1, "Noise strength could not be manipulated"
def test_generate_noise():
dimensions = np.array([10, 10, 10]) # What is the size of the brain
feature_size = [2]
feature_type = ['cube']
feature_coordinates = np.array(
[[5, 5, 5]])
signal_magnitude = [1]
# Generate a volume representing the location and quality of the signal
volume = sim.generate_signal(dimensions=dimensions,
feature_coordinates=feature_coordinates,
feature_type=feature_type,
feature_size=feature_size,
signal_magnitude=signal_magnitude,
)
# Inputs for generate_stimfunction
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 100
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
signal_function = sim.double_gamma_hrf(stimfunction=stimfunction,
tr_duration=tr_duration,
)
# Convolve the HRF with the stimulus sequence
signal = sim.apply_signal(signal_function=signal_function,
volume_static=volume,
)
# Generate the mask of the signal
mask = sim.mask_brain(signal, mask_threshold=0.1)
assert min(mask[mask > 0]) > 0.1, "Mask thresholding did not work"
# Create the noise volumes (using the default parameters)
noise = sim.generate_noise(dimensions=dimensions,
stimfunction=stimfunction,
tr_duration=tr_duration,
mask=mask,
)
assert signal.shape == noise.shape, "The dimensions of signal and noise " \
"the same"
assert np.std(signal) < np.std(noise), "Noise was not created"
noise = sim.generate_noise(dimensions=dimensions,
stimfunction=stimfunction,
tr_duration=tr_duration,
mask=mask,
noise_dict={'overall': 0},
)
assert np.sum(noise) == 0, "Noise strength could not be manipulated"
assert np.std(noise) == 0, "Noise strength could not be manipulated"
signal = signal_A + signal_B
# Combine the stim functions
stimfunction = list(np.add(stimfunction_A, stimfunction_B))
# Generate the mask of the signal
mask = sim.mask_brain(signal)
# Mask the signal to the shape of a brain (attenuates signal according to grey
# matter likelihood)
signal *= mask
# Create the noise volumes (using the default parameters
noise = sim.generate_noise(dimensions=dimensions,
stimfunction=stimfunction,
tr_duration=tr_duration,
mask=mask,
)
# Combine the signal and the noise
brain = signal + noise
# Display the brain
fig = plt.figure()
for tr_counter in list(range(0, brain.shape[3])):
# Get the axis to be plotted
ax = sim.plot_brain(fig,
brain[:, :, :, tr_counter],
mask=mask[:, :, :, tr_counter],
percentile=99.9)
# Generate the mask of the signal
mask, template = sim.mask_brain(signal, mask_threshold=0.2)
# Mask the signal to the shape of a brain (attenuates signal according to grey
# matter likelihood)
signal *= mask.reshape(dimensions[0], dimensions[1], dimensions[2], 1)
# Generate original noise dict for comparison later
orig_noise_dict = sim._noise_dict_update({})
# Create the noise volumes (using the default parameters
noise = sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
mask=mask,
template=template,
noise_dict=orig_noise_dict,
)
# Standardize the signal activity to make it percent signal change
mean_act = (mask * orig_noise_dict['max_activity']).sum() / (mask > 0).sum()
signal = signal * mean_act / 100
# Combine the signal and the noise
brain = signal + noise
# Display the brain
fig = plt.figure()
for tr_counter in list(range(0, brain.shape[3])):
def test_apply_signal():
dimensions = np.array([10, 10, 10]) # What is the size of the brain
feature_size = [2]
feature_type = ['cube']
feature_coordinates = np.array(
[[5, 5, 5]])
signal_magnitude = [30]
# Generate a volume representing the location and quality of the signal
volume = sim.generate_signal(dimensions=dimensions,
feature_coordinates=feature_coordinates,
feature_type=feature_type,
feature_size=feature_size,
signal_magnitude=signal_magnitude,
)
# Inputs for generate_stimfunction
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 100
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
signal_function = sim.convolve_hrf(stimfunction=stimfunction,
tr_duration=tr_duration,
)
# Check that you can compute signal change appropriately
# Preset a bunch of things
stimfunction_tr = stimfunction[::int(tr_duration * 100)]
mask, template = sim.mask_brain(dimensions, mask_self=False)
noise_dict = sim._noise_dict_update({})
noise = sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=noise_dict,
iterations=[0, 0]
)
coords = feature_coordinates[0]
noise_function_a = noise[coords[0], coords[1], coords[2], :]
noise_function_a = noise_function_a.reshape(duration // tr_duration, 1)
noise_function_b = noise[coords[0] + 1, coords[1], coords[2], :]
noise_function_b = noise_function_b.reshape(duration // tr_duration, 1)
# Create the calibrated signal with PSC
method = 'PSC'
sig_a = sim.compute_signal_change(signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
sig_b = sim.compute_signal_change(signal_function,
noise_function_a,
noise_dict,
[1.0],
method,
)
assert sig_b.max() / sig_a.max() == 2, 'PSC modulation failed'
# Create the calibrated signal with SFNR
method = 'SFNR'
sig_a = sim.compute_signal_change(signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = sig_a / (noise_function_a.mean() / noise_dict['sfnr'])
sig_b = sim.compute_signal_change(signal_function,
noise_function_b,
noise_dict,
[1.0],
method,
)
scaled_b = sig_b / (noise_function_b.mean() / noise_dict['sfnr'])
assert scaled_b.max() / scaled_a.max() == 2, 'SFNR modulation failed'
# Create the calibrated signal with CNR_Amp/Noise-SD
method = 'CNR_Amp/Noise-SD'
sig_a = sim.compute_signal_change(signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = sig_a / noise_function_a.std()
sig_b = sim.compute_signal_change(signal_function,
noise_function_b,
noise_dict,
[1.0],
method,
)
scaled_b = sig_b / noise_function_b.std()
assert scaled_b.max() / scaled_a.max() == 2, 'CNR_Amp modulation failed'
# Create the calibrated signal with CNR_Amp/Noise-Var_dB
method = 'CNR_Amp2/Noise-Var_dB'
sig_a = sim.compute_signal_change(signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = np.log(sig_a.max() / noise_function_a.std())
sig_b = sim.compute_signal_change(signal_function,
noise_function_b,
noise_dict,
[1.0],
method,
)
scaled_b = np.log(sig_b.max() / noise_function_b.std())
assert np.round(scaled_b / scaled_a) == 2, 'CNR_Amp dB modulation failed'
# Create the calibrated signal with CNR_Signal-SD/Noise-SD
method = 'CNR_Signal-SD/Noise-SD'
sig_a = sim.compute_signal_change(signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = sig_a.std() / noise_function_a.std()
sig_b = sim.compute_signal_change(signal_function,
noise_function_a,
noise_dict,
[1.0],
method,
)
scaled_b = sig_b.std() / noise_function_a.std()
assert (scaled_b / scaled_a) == 2, 'CNR signal modulation failed'
# Create the calibrated signal with CNR_Amp/Noise-Var_dB
method = 'CNR_Signal-Var/Noise-Var_dB'
sig_a = sim.compute_signal_change(signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = np.log(sig_a.std() / noise_function_a.std())
sig_b = sim.compute_signal_change(signal_function,
noise_function_b,
noise_dict,
[1.0],
method,
)
scaled_b = np.log(sig_b.std() / noise_function_b.std())
assert np.round(scaled_b / scaled_a) == 2, 'CNR signal dB modulation ' \
'failed'
# Convolve the HRF with the stimulus sequence
signal = sim.apply_signal(signal_function=signal_function,
volume_signal=volume,
)
assert signal.shape == (dimensions[0], dimensions[1], dimensions[2],
duration / tr_duration), "The output is the " \
"wrong size"
signal = sim.apply_signal(signal_function=stimfunction,
volume_signal=volume,
)
assert np.any(signal == signal_magnitude), "The stimfunction is not binary"
# Check that there is an error if the number of signal voxels doesn't
# match the number of non zero brain voxels
with pytest.raises(IndexError):
sig_vox = (volume > 0).sum()
vox_pattern = np.tile(stimfunction, (1, sig_vox - 1))
sim.apply_signal(signal_function=vox_pattern,
volume_signal=volume,
)
def test_generate_noise():
dimensions = np.array([10, 10, 10]) # What is the size of the brain
feature_size = [2]
feature_type = ['cube']
feature_coordinates = np.array(
[[5, 5, 5]])
signal_magnitude = [1]
# Generate a volume representing the location and quality of the signal
volume = sim.generate_signal(dimensions=dimensions,
feature_coordinates=feature_coordinates,
feature_type=feature_type,
feature_size=feature_size,
signal_magnitude=signal_magnitude,
)
# Inputs for generate_stimfunction
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 200
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
signal_function = sim.convolve_hrf(stimfunction=stimfunction,
tr_duration=tr_duration,
)
# Convolve the HRF with the stimulus sequence
signal = sim.apply_signal(signal_function=signal_function,
volume_signal=volume,
)
# Generate the mask of the signal
mask, template = sim.mask_brain(signal,
mask_self=None)
assert min(mask[mask > 0]) > 0.1, "Mask thresholding did not work"
assert len(np.unique(template) > 2), "Template creation did not work"
stimfunction_tr = stimfunction[::int(tr_duration * 100)]
# Create the noise volumes (using the default parameters)
noise = sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
iterations=[1, 0],
)
assert signal.shape == noise.shape, "The dimensions of signal and noise " \
"the same"
noise_high = sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict={'sfnr': 50, 'snr': 25},
iterations=[1, 0],
)
noise_low = sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict={'sfnr': 100, 'snr': 25},
iterations=[1, 0],
)
system_high = np.std(noise_high[mask > 0], 1).mean()
system_low = np.std(noise_low[mask > 0], 1).mean()
assert system_low < system_high, "SFNR noise could not be manipulated"
# Check that you check for the appropriate template values
with pytest.raises(ValueError):
sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template * 2,
mask=mask,
noise_dict={},
)
# Check that iterations does what it should
sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict={},
iterations=[0, 0],
)
sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict={},
iterations=None,
)
# Test drift noise
trs = 1000
period = 100
drift = sim._generate_noise_temporal_drift(trs,
tr_duration,
'sine',
period,
)
# Check that the max frequency is the appropriate frequency
power = abs(np.fft.fft(drift))[1:trs // 2]
freq = np.linspace(1, trs // 2 - 1, trs // 2 - 1) / trs
period_freq = np.where(freq == 1 / (period // tr_duration))
max_freq = np.argmax(power)
assert period_freq == max_freq, 'Max frequency is not where it should be'
# Do the same but now with cosine basis functions, answer should be close
drift = sim._generate_noise_temporal_drift(trs,
tr_duration,
'discrete_cos',
period,
)
# Check that the appropriate frequency is peaky (may not be the max)
power = abs(np.fft.fft(drift))[1:trs // 2]
freq = np.linspace(1, trs // 2 - 1, trs // 2 - 1) / trs
period_freq = np.where(freq == 1 / (period // tr_duration))[0][0]
assert power[period_freq] > power[period_freq + 1], 'Power is low'
assert power[period_freq] > power[period_freq - 1], 'Power is low'
# Check it gives a warning if the duration is too short
drift = sim._generate_noise_temporal_drift(50,
tr_duration,
'discrete_cos',
period,
)
# Test physiological noise (using unrealistic parameters so that it's easy)
timepoints = list(np.linspace(0, (trs - 1) * tr_duration, trs))
resp_freq = 0.2
heart_freq = 1.17
phys = sim._generate_noise_temporal_phys(timepoints,
resp_freq,
heart_freq,
)
# Check that the max frequency is the appropriate frequency
power = abs(np.fft.fft(phys))[1:trs // 2]
freq = np.linspace(1, trs // 2 - 1, trs // 2 - 1) / (trs * tr_duration)
peaks = (power > (power.mean() + power.std())) # Where are the peaks
peak_freqs = freq[peaks]
assert np.any(resp_freq == peak_freqs), 'Resp frequency not found'
assert len(peak_freqs) == 2, 'Two peaks not found'
# Test task noise
sim._generate_noise_temporal_task(stimfunction_tr,
motion_noise='gaussian',
)
sim._generate_noise_temporal_task(stimfunction_tr,
motion_noise='rician',
)
# Test ARMA noise
with pytest.raises(ValueError):
noise_dict = {'fwhm': 4, 'auto_reg_rho': [1], 'ma_rho': [1, 1]}
sim._generate_noise_temporal_autoregression(stimfunction_tr,
noise_dict,
dimensions,
mask,
)
# Generate spatial noise
vol = sim._generate_noise_spatial(np.array([10, 10, 10, trs]))
assert len(vol.shape) == 3, 'Volume was not reshaped to ignore TRs'
# Switch some of the noise types on
noise_dict = dict(physiological_sigma=1, drift_sigma=1, task_sigma=1,
auto_reg_sigma=0)
sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=noise_dict,
iterations=[0, 0],
)
def generate_data(outputDir, user_settings):
"""Generate simulated fMRI data
Use a few parameters that might be relevant for real time analysis
Parameters
----------
outputDir : str
Specify output data dir where the data should be saved
user_settings : dict
A dictionary to specify the parameters used for making data,
specifying the following keys
numTRs - int - Specify the number of time points
multivariate_patterns - bool - Is the difference between conditions
univariate (0) or multivariate (1)
different_ROIs - bool - Are there different ROIs for each condition (
1) or is it in the same ROI (0). If it is the same ROI and you are
using univariate differences, the second condition will have a
smaller evoked response than the other.
event_duration - int - How long, in seconds, is each event
scale_percentage - float - What is the percent signal change
trDuration - float - How many seconds per volume
save_dicom - bool - Save to data as a dicom (1) or numpy (0)
save_realtime - bool - Do you want to save the data in real time (1)
or as fast as possible (0)?
isi - float - What is the time between each event (in seconds)
burn_in - int - How long before the first event (in seconds)
"""
data_dict = default_settings.copy()
data_dict.update(user_settings)
# If the folder doesn't exist then make it
os.system('mkdir -p %s' % outputDir)
logger.info('Load template of average voxel value')
# Get the file names needed for loading in the data
ROI_A_file, ROI_B_file, template_path, noise_dict_file = \
_get_input_names(data_dict)
# Load in the template data (it may already be loaded if doing a test)
if isinstance(template_path, str):
template_nii = nibabel.load(template_path)
template = template_nii.get_data()
else:
template = template_path
dimensions = np.array(template.shape[0:3])
logger.info('Create binary mask and normalize the template range')
mask, template = sim.mask_brain(
volume=template,
mask_self=True,
)
# Write out the mask as a numpy file
outFile = os.path.join(outputDir, 'mask.npy')
np.save(outFile, mask.astype(np.uint8))
# Load the noise dictionary
logger.info('Loading noise parameters')
# If this isn't a string, assume it is a resource stream file
if type(noise_dict_file) is str:
with open(noise_dict_file, 'r') as f:
noise_dict = f.read()
else:
# Read the resource stream object
noise_dict = noise_dict_file.decode()
noise_dict = eval(noise_dict)
noise_dict['matched'] = 0 # Increases processing time
# Add it here for easy access
data_dict['noise_dict'] = noise_dict
logger.info('Generating noise')
temp_stimfunction = np.zeros((data_dict['numTRs'], 1))
noise = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=temp_stimfunction,
tr_duration=int(data_dict['trDuration']),
template=template,
mask=mask,
noise_dict=noise_dict,
)
# Create the stimulus time course of the conditions
total_time = int(data_dict['numTRs'] * data_dict['trDuration'])
onsets_A = []
onsets_B = []
curr_time = data_dict['burn_in']
while curr_time < (total_time - data_dict['event_duration']):
# Flip a coin for each epoch to determine whether it is A or B
if np.random.randint(0, 2) == 1:
onsets_A.append(curr_time)
else:
onsets_B.append(curr_time)
# Increment the current time
curr_time += data_dict['event_duration'] + data_dict['isi']
# How many timepoints per second of the stim function are to be generated?
temporal_res = 1 / data_dict['trDuration']
# Create a time course of events
event_durations = [data_dict['event_duration']]
stimfunc_A = sim.generate_stimfunction(
onsets=onsets_A,
event_durations=event_durations,
total_time=total_time,
temporal_resolution=temporal_res,
)
stimfunc_B = sim.generate_stimfunction(
onsets=onsets_B,
event_durations=event_durations,
total_time=total_time,
temporal_resolution=temporal_res,
)
# Create a labels timecourse
outFile = os.path.join(outputDir, 'labels.npy')
np.save(outFile, (stimfunc_A + (stimfunc_B * 2)))
# How is the signal implemented in the different ROIs
signal_A = _generate_ROIs(ROI_A_file, stimfunc_A, noise,
data_dict['scale_percentage'], data_dict)
if data_dict['different_ROIs'] is True:
signal_B = _generate_ROIs(ROI_B_file, stimfunc_B, noise,
data_dict['scale_percentage'], data_dict)
else:
# Halve the evoked response if these effects are both expected in the
# same ROI
if data_dict['multivariate_pattern'] is False:
signal_B = _generate_ROIs(ROI_A_file, stimfunc_B, noise,
data_dict['scale_percentage'] * 0.5,
data_dict)
else:
signal_B = _generate_ROIs(ROI_A_file, stimfunc_B, noise,
data_dict['scale_percentage'], data_dict)
# Combine the two signal timecourses
signal = signal_A + signal_B
logger.info('Generating TRs in real time')
for idx in range(data_dict['numTRs']):
# Create the brain volume on this TR
brain = noise[:, :, :, idx] + signal[:, :, :, idx]
# Convert file to integers to mimic what you get from MR
brain_int32 = brain.astype(np.int32)
# Store as dicom or nifti?
if data_dict['save_dicom'] is True:
# Save the volume as a DICOM file, with each TR as its own file
output_file = os.path.join(outputDir,
'rt_' + format(idx, '03d') + '.dcm')
_write_dicom(output_file, brain_int32, idx + 1)
else:
# Save the volume as a numpy file, with each TR as its own file
output_file = os.path.join(outputDir,
'rt_' + format(idx, '03d') + '.npy')
np.save(output_file, brain_int32)
logger.info("Generate {}".format(output_file))
# Sleep until next TR
if data_dict['save_realtime'] == 1:
time.sleep(data_dict['trDuration'])
# Generate the mask of the signal
mask = sim.mask_brain(signal)
# Mask the signal to the shape of a brain (attenuates signal according to grey
# matter likelihood)
signal *= mask
# Generate original noise dict for comparison later
orig_noise_dict = sim._noise_dict_update({})
# Create the noise volumes (using the default parameters
noise = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
mask=mask,
noise_dict=orig_noise_dict,
)
# Standardize the signal activity to make it percent signal change
mean_act = (mask * orig_noise_dict['max_activity']).sum() / (mask > 0).sum()
signal = signal * mean_act / 100
# Combine the signal and the noise
brain = signal + noise
# Display the brain
fig = plt.figure()
for tr_counter in list(range(0, brain.shape[3])):
signal_function_B = sim.double_gamma_hrf(stimfunction=stimfunction_B)
# Multiply the HRF timecourse with the signal
signal_A = sim.apply_signal(signal_function=signal_function_A, volume_static=volume_static_A)
signal_B = sim.apply_signal(signal_function=signal_function_B, volume_static=volume_static_B)
# Combine the signals from the two conditions
signal = signal_A + signal_B
# Combine the stim functions
stimfunction = list(np.add(stimfunction_A, stimfunction_B))
# Create the noise volumes (using the default parameters
noise = sim.generate_noise(dimensions=dimensions, stimfunction=stimfunction, tr_duration=tr_duration)
# Combine the signal and the noise
volume = signal + noise
# Mask the volume to be the same shape as a brain
brain = sim.mask_brain(volume)
# Display the brain
fig = plt.figure()
for tr_counter in list(range(0, brain.shape[3])):
# Get the axis to be plotted
ax = sim.plot_brain(fig, brain[:, :, :, tr_counter], percentile=99.9)
# Wait for an input
def test_generate_noise():
dimensions = np.array([10, 10, 10]) # What is the size of the brain
feature_size = [2]
feature_type = ['cube']
feature_coordinates = np.array(
[[5, 5, 5]])
signal_magnitude = [1]
# Generate a volume representing the location and quality of the signal
volume = sim.generate_signal(dimensions=dimensions,
feature_coordinates=feature_coordinates,
feature_type=feature_type,
feature_size=feature_size,
signal_magnitude=signal_magnitude,
)
# Inputs for generate_stimfunction
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 200
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
signal_function = sim.convolve_hrf(stimfunction=stimfunction,
tr_duration=tr_duration,
)
# Convolve the HRF with the stimulus sequence
signal = sim.apply_signal(signal_function=signal_function,
volume_signal=volume,
)
# Generate the mask of the signal
mask, template = sim.mask_brain(signal, mask_threshold=0.1)
assert min(mask[mask > 0]) > 0.1, "Mask thresholding did not work"
assert len(np.unique(template) > 2), "Template creation did not work"
stimfunction_tr = stimfunction[::int(tr_duration * 100)]
# Create the noise volumes (using the default parameters)
noise = sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
)
assert signal.shape == noise.shape, "The dimensions of signal and noise " \
"the same"
assert np.std(signal) < np.std(noise), "Noise was not created"
noise = sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict={'sfnr': 10000, 'snr': 10000},
)
system_noise = np.std(noise[mask > 0], 1).mean()
assert system_noise <= 0.1, "Noise strength could not be manipulated"
def generate_data(cfgFile):
cfg = loadConfigFile(cfgFile)
frame = inspect.currentframe()
moduleFile = typing.cast(str, frame.f_code.co_filename) # type: ignore
moduleDir = os.path.dirname(moduleFile)
cfgDate = parser.parse(cfg.session.date).strftime("%Y%m%d")
dataDir = os.path.join(
cfg.session.dataDir, "subject{}/day{}".format(cfg.session.subjectNum,
cfg.session.subjectDay))
imgDir = os.path.join(
cfg.session.imgDir, "{}.{}.{}".format(cfgDate, cfg.session.subjectName,
cfg.session.subjectName))
if os.path.exists(dataDir) and os.path.exists(imgDir):
print(
"output data and imgage directory already exist, skippig data generation"
)
return
runPatterns = [
'patternsdesign_1_20180101T000000.mat',
'patternsdesign_2_20180101T000000.mat',
'patternsdesign_3_20180101T000000.mat'
]
template_filename = os.path.join(moduleDir, 'sub_template.nii.gz')
noise_dict_filename = os.path.join(moduleDir, 'sub_noise_dict.txt')
roiA_filename = os.path.join(moduleDir, 'ROI_A.nii.gz')
roiB_filename = os.path.join(moduleDir, 'ROI_B.nii.gz')
output_file_pattern = '001_0000{}_000{}.mat'
if not os.path.exists(imgDir):
os.makedirs(imgDir)
if not os.path.exists(dataDir):
os.makedirs(dataDir)
print('Load data')
template_nii = nibabel.load(template_filename)
template = template_nii.get_data()
# dimsize = template_nii.header.get_zooms()
roiA_nii = nibabel.load(roiA_filename)
roiB_nii = nibabel.load(roiB_filename)
roiA = roiA_nii.get_data()
roiB = roiB_nii.get_data()
dimensions = np.array(template.shape[0:3]) # What is the size of the brain
print('Create mask')
# Generate the continuous mask from the voxels
mask, template = sim.mask_brain(
volume=template,
mask_self=True,
)
# Write out the mask as matlab
mask_uint8 = mask.astype(np.uint8)
maskfilename = os.path.join(
dataDir, 'mask_{}_{}.mat'.format(cfg.session.subjectNum,
cfg.session.subjectDay))
sio.savemat(maskfilename, {'mask': mask_uint8})
# Load the noise dictionary
with open(noise_dict_filename, 'r') as f:
noise_dict = f.read()
print('Loading ' + noise_dict_filename)
noise_dict = eval(noise_dict)
noise_dict['matched'] = 0
runNum = 1
scanNum = 0
for patfile in runPatterns:
fullPatfile = os.path.join(moduleDir, patfile)
# make dataDir run directory
runDir = os.path.join(dataDir, "run{}".format(runNum))
if not os.path.exists(runDir):
os.makedirs(runDir)
shutil.copy(fullPatfile, runDir)
runNum += 1
pat = sio.loadmat(fullPatfile)
scanNum += 1
# shifted labels are in regressor field
shiftedLabels = pat['patterns']['regressor'][0][0]
# non-shifted labels are in attCateg field and whether stimulus applied in the stim field
nsLabels = pat['patterns']['attCateg'][0][0] * pat['patterns']['stim'][
0][0]
labels_A = (nsLabels == 1).astype(int)
labels_B = (nsLabels == 2).astype(int)
# trialType = pat['patterns']['type'][0][0]
tr_duration = pat['TR'][0][0]
disdaqs = pat['disdaqs'][0][0]
begTrOffset = disdaqs // tr_duration
nTRs = pat['nTRs'][0][0]
# nTestTRs = np.count_nonzero(trialType == 2)
# Preset some of the parameters
total_trs = nTRs + begTrOffset # How many time points are there?
print('Generating data')
start = time.time()
noiseVols = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=np.zeros((total_trs, 1)),
tr_duration=int(tr_duration),
template=template,
mask=mask,
noise_dict=noise_dict,
)
print("Time: generate noise vols {} sec".format(time.time() - start))
nVoxelsA = int(roiA.sum())
nVoxelsB = int(roiB.sum())
# Multiply each pattern by each voxel time course
weights_A = np.tile(labels_A.reshape(-1, 1), nVoxelsA)
weights_B = np.tile(labels_B.reshape(-1, 1), nVoxelsB)
print('Creating signal time course')
signal_func_A = sim.convolve_hrf(
stimfunction=weights_A,
tr_duration=tr_duration,
temporal_resolution=(1 / tr_duration),
scale_function=1,
)
signal_func_B = sim.convolve_hrf(
stimfunction=weights_B,
tr_duration=tr_duration,
temporal_resolution=(1 / tr_duration),
scale_function=1,
)
max_activity = noise_dict['max_activity']
signal_change = 10 # .01 * max_activity
signal_func_A *= signal_change
signal_func_B *= signal_change
# Combine the signal time course with the signal volume
print('Creating signal volumes')
signal_A = sim.apply_signal(
signal_func_A,
roiA,
)
signal_B = sim.apply_signal(
signal_func_B,
roiB,
)
# Combine the two signal timecourses
signal = signal_A + signal_B
# testTrId = 0
numVols = noiseVols.shape[3]
for idx in range(numVols):
start = time.time()
brain = noiseVols[:, :, :, idx]
if idx >= begTrOffset:
# some initial scans are skipped as only instructions and not stimulus are shown
signalIdx = idx - begTrOffset
brain += signal[:, :, :, signalIdx]
# TODO: how to create a varying combined percentage of A and B signals
# if trialType[0][idx] == 1:
# # training TR, so create pure A or B signal
# if labels_A[idx] != 0:
# brain = brain + roiA
# elif labels_B[idx] != 0:
# brain = brain + roiB
# elif trialType[0][idx] == 2:
# # testing TR, so create a mixture of A and B signal
# testTrId += 1
# testPercent = testTrId / nTestTRs
# brain = brain + testPercent * roiA + (1-testPercent) * roiB
# Save the volume as a matlab file
filenum = idx + 1
filename = output_file_pattern.format(
str(scanNum).zfill(2),
str(filenum).zfill(3))
outputfile = os.path.join(imgDir, filename)
brain_float32 = brain.astype(np.float32)
sio.savemat(outputfile, {'vol': brain_float32})
print("Time: generate vol {}: {} sec".format(
filenum,
time.time() - start))
def test_calc_noise():
# Inputs for functions
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 200
temporal_res = 100
tr_number = int(np.floor(duration / tr_duration))
dimensions_tr = np.array([10, 10, 10, tr_number])
# Preset the noise dict
nd_orig = sim._noise_dict_update({})
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(onsets=onsets,
event_durations=event_durations,
total_time=duration,
temporal_resolution=temporal_res,
)
# Mask the volume to be the same shape as a brain
mask, template = sim.mask_brain(dimensions_tr, mask_self=None)
stimfunction_tr = stimfunction[::int(tr_duration * temporal_res)]
nd_orig['matched'] = 0
noise = sim.generate_noise(dimensions=dimensions_tr[0:3],
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=nd_orig,
)
# Check the spatial noise match
nd_orig['matched'] = 1
noise_matched = sim.generate_noise(dimensions=dimensions_tr[0:3],
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=nd_orig,
iterations=[50, 0]
)
# Calculate the noise parameters from this newly generated volume
nd_new = sim.calc_noise(noise, mask, template)
nd_matched = sim.calc_noise(noise_matched, mask, template)
# Check the values are reasonable"
assert nd_new['snr'] > 0, 'snr out of range'
assert nd_new['sfnr'] > 0, 'sfnr out of range'
assert nd_new['auto_reg_rho'][0] > 0, 'ar out of range'
# Check that the dilation increases SNR
no_dilation_snr = sim._calc_snr(noise_matched,
mask,
dilation=0,
reference_tr=tr_duration,
)
assert nd_new['snr'] > no_dilation_snr, "Dilation did not increase SNR"
# Check that template size is in bounds
with pytest.raises(ValueError):
sim.calc_noise(noise, mask, template * 2)
# Check that Mask is set is checked
with pytest.raises(ValueError):
sim.calc_noise(noise, None, template)
# Check that it can deal with missing noise parameters
temp_nd = sim.calc_noise(noise, mask, template, noise_dict={})
assert temp_nd['voxel_size'][0] == 1, 'Default voxel size not set'
temp_nd = sim.calc_noise(noise, mask, template, noise_dict=None)
assert temp_nd['voxel_size'][0] == 1, 'Default voxel size not set'
# Check that the fitting worked
snr_diff = abs(nd_orig['snr'] - nd_new['snr'])
snr_diff_match = abs(nd_orig['snr'] - nd_matched['snr'])
assert snr_diff > snr_diff_match, 'snr fit incorrectly'
# Test that you can generate rician and exponential noise
sim._generate_noise_system(dimensions_tr,
1,
1,
spatial_noise_type='exponential',
temporal_noise_type='rician',
)
# Check the temporal noise match
nd_orig['matched'] = 1
noise_matched = sim.generate_noise(dimensions=dimensions_tr[0:3],
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=nd_orig,
iterations=[0, 50]
)
nd_matched = sim.calc_noise(noise_matched, mask, template)
sfnr_diff = abs(nd_orig['sfnr'] - nd_new['sfnr'])
sfnr_diff_match = abs(nd_orig['sfnr'] - nd_matched['sfnr'])
assert sfnr_diff > sfnr_diff_match, 'sfnr fit incorrectly'
ar1_diff = abs(nd_orig['auto_reg_rho'][0] - nd_new['auto_reg_rho'][0])
ar1_diff_match = abs(nd_orig['auto_reg_rho'][0] - nd_matched[
'auto_reg_rho'][0])
assert ar1_diff > ar1_diff_match, 'AR1 fit incorrectly'
# Check that you can calculate ARMA for a single voxel
vox = noise[5, 5, 5, :]
arma = sim._calc_ARMA_noise(vox,
None,
sample_num=2,
)
assert len(arma) == 2, "Two outputs not given by ARMA"
stimfunction_tr = stimfunction[::int(tr_duration * 1000)]
# Iterate through the participants and store participants
epochs = []
for participantcounter in range(1, participants + 1):
# Add the epoch cube
epochs += [epoch]
# Save a file name
savename = directory + 'p' + str(participantcounter) + '.nii'
# Create the noise volumes (using the default parameters
noise = sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
)
# Combine the signal and the noise
brain = signal + noise
# Save the volume
affine_matrix = np.diag([-1, 1, 1, 1]) # LR gets flipped
brain_nifti = nibabel.Nifti1Image(brain, affine_matrix)
nibabel.save(brain_nifti, savename)
# Save the epochs
np.save(directory + 'epoch_labels.npy', epochs)
# Store the mask
plt.xlim((0, 200))
plt.ylim((-1, 5))
plt.savefig(signal_func_save)
# Convert the stim func into a binary vector of dim 1
stimfunc_binary = np.mean(np.abs(stimfunc_all)>0, 1) > 0
stimfunc_binary = stimfunc_binary[::int(tr_duration * temporal_res)]
# Bound, can happen if the duration is not rounded to a TR
stimfunc_binary = stimfunc_binary[0:signal_func.shape[0]]
# Create the noise volumes (using the default parameters)
noise = sim.generate_noise(dimensions=dimensions[0:3],
stimfunction_tr=stimfunc_binary,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=noise_dict,
)
# Change the type of noise
noise = noise.astype('double')
# Create a noise function (same voxels for signal function as for noise)
noise_function = noise[idxs[:, 0], idxs[:, 1], idxs[:, 2], :].T
# Compute the signal magnitude for the data
signal_func_scaled = sim.compute_signal_change(signal_function=signal_func,
noise_function=noise_function,
noise_dict=noise_dict,
magnitude=[
def test_apply_signal():
dimensions = np.array([10, 10, 10]) # What is the size of the brain
feature_size = [2]
feature_type = ['cube']
feature_coordinates = np.array([[5, 5, 5]])
signal_magnitude = [30]
# Generate a volume representing the location and quality of the signal
volume = sim.generate_signal(
dimensions=dimensions,
feature_coordinates=feature_coordinates,
feature_type=feature_type,
feature_size=feature_size,
signal_magnitude=signal_magnitude,
)
# Inputs for generate_stimfunction
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 100
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(
onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
signal_function = sim.convolve_hrf(
stimfunction=stimfunction,
tr_duration=tr_duration,
)
# Check that you can compute signal change appropriately
# Preset a bunch of things
stimfunction_tr = stimfunction[::int(tr_duration * 100)]
mask, template = sim.mask_brain(dimensions, mask_self=False)
noise_dict = sim._noise_dict_update({})
noise = sim.generate_noise(dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=noise_dict,
iterations=[0, 0])
coords = feature_coordinates[0]
noise_function_a = noise[coords[0], coords[1], coords[2], :]
noise_function_a = noise_function_a.reshape(duration // tr_duration, 1)
noise_function_b = noise[coords[0] + 1, coords[1], coords[2], :]
noise_function_b = noise_function_b.reshape(duration // tr_duration, 1)
# Create the calibrated signal with PSC
method = 'PSC'
sig_a = sim.compute_signal_change(
signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
sig_b = sim.compute_signal_change(
signal_function,
noise_function_a,
noise_dict,
[1.0],
method,
)
assert sig_b.max() / sig_a.max() == 2, 'PSC modulation failed'
# Create the calibrated signal with SFNR
method = 'SFNR'
sig_a = sim.compute_signal_change(
signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = sig_a / (noise_function_a.mean() / noise_dict['sfnr'])
sig_b = sim.compute_signal_change(
signal_function,
noise_function_b,
noise_dict,
[1.0],
method,
)
scaled_b = sig_b / (noise_function_b.mean() / noise_dict['sfnr'])
assert scaled_b.max() / scaled_a.max() == 2, 'SFNR modulation failed'
# Create the calibrated signal with CNR_Amp/Noise-SD
method = 'CNR_Amp/Noise-SD'
sig_a = sim.compute_signal_change(
signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = sig_a / noise_function_a.std()
sig_b = sim.compute_signal_change(
signal_function,
noise_function_b,
noise_dict,
[1.0],
method,
)
scaled_b = sig_b / noise_function_b.std()
assert scaled_b.max() / scaled_a.max() == 2, 'CNR_Amp modulation failed'
# Create the calibrated signal with CNR_Amp/Noise-Var_dB
method = 'CNR_Amp2/Noise-Var_dB'
sig_a = sim.compute_signal_change(
signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = np.log(sig_a.max() / noise_function_a.std())
sig_b = sim.compute_signal_change(
signal_function,
noise_function_b,
noise_dict,
[1.0],
method,
)
scaled_b = np.log(sig_b.max() / noise_function_b.std())
assert np.round(scaled_b / scaled_a) == 2, 'CNR_Amp dB modulation failed'
# Create the calibrated signal with CNR_Signal-SD/Noise-SD
method = 'CNR_Signal-SD/Noise-SD'
sig_a = sim.compute_signal_change(
signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = sig_a.std() / noise_function_a.std()
sig_b = sim.compute_signal_change(
signal_function,
noise_function_a,
noise_dict,
[1.0],
method,
)
scaled_b = sig_b.std() / noise_function_a.std()
assert (scaled_b / scaled_a) == 2, 'CNR signal modulation failed'
# Create the calibrated signal with CNR_Amp/Noise-Var_dB
method = 'CNR_Signal-Var/Noise-Var_dB'
sig_a = sim.compute_signal_change(
signal_function,
noise_function_a,
noise_dict,
[0.5],
method,
)
scaled_a = np.log(sig_a.std() / noise_function_a.std())
sig_b = sim.compute_signal_change(
signal_function,
noise_function_b,
noise_dict,
[1.0],
method,
)
scaled_b = np.log(sig_b.std() / noise_function_b.std())
assert np.round(scaled_b / scaled_a) == 2, 'CNR signal dB modulation ' \
'failed'
# Convolve the HRF with the stimulus sequence
signal = sim.apply_signal(
signal_function=signal_function,
volume_signal=volume,
)
assert signal.shape == (dimensions[0], dimensions[1], dimensions[2],
duration / tr_duration), "The output is the " \
"wrong size"
signal = sim.apply_signal(
signal_function=stimfunction,
volume_signal=volume,
)
assert np.any(signal == signal_magnitude), "The stimfunction is not binary"
# Check that there is an error if the number of signal voxels doesn't
# match the number of non zero brain voxels
with pytest.raises(IndexError):
sig_vox = (volume > 0).sum()
vox_pattern = np.tile(stimfunction, (1, sig_vox - 1))
sim.apply_signal(
signal_function=vox_pattern,
volume_signal=volume,
)
def generate_data(inputDir, outputDir, data_dict):
# Generate simulated fMRI data with a few parameters that might be
# relevant for real time analysis
# inputDir - Specify input data dir where the parameters for fmrisim are
# outputDir - Specify output data dir where the data should be saved
# data_dict contains:
# numTRs - Specify the number of time points
# multivariate_patterns - Is the difference between conditions
# univariate (0) or multivariate (1)
# different_ROIs - Are there different ROIs for each condition (1) or
# is it in the same ROI (0). If it is the same ROI and you are using
# univariate differences, the second condition will have a smaller evoked
# response than the other.
# event_duration - How long, in seconds, is each event
# scale_percentage - What is the percent signal change
# trDuration - How many seconds per volume
# save_dicom - Do you want to save data as a dicom (1) or numpy (0)
# save_realtime - Do you want to save the data in real time (1) or as
# fast as possible (0)?
# isi - What is the time between each event (in seconds)
# burn_in - How long before the first event (in seconds)
# If the folder doesn't exist then make it
if os.path.isdir(outputDir) is False:
os.makedirs(outputDir, exist_ok=True)
print('Load template of average voxel value')
templateFile = os.path.join(inputDir, 'sub_template.nii.gz')
template_nii = nibabel.load(templateFile)
template = template_nii.get_data()
dimensions = np.array(template.shape[0:3])
print('Create binary mask and normalize the template range')
mask, template = sim.mask_brain(
volume=template,
mask_self=True,
)
# Write out the mask as a numpy file
outFile = os.path.join(outputDir, 'mask.npy')
np.save(outFile, mask.astype(np.uint8))
# Load the noise dictionary
print('Loading noise parameters')
noiseFile = os.path.join(inputDir, 'sub_noise_dict.txt')
with open(noiseFile, 'r') as f:
noise_dict = f.read()
noise_dict = eval(noise_dict)
noise_dict['matched'] = 0 # Increases processing time
# Add it here for easy access
data_dict['noise_dict'] = data_dict
print('Generating noise')
temp_stimfunction = np.zeros((data_dict['numTRs'], 1))
noise = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=temp_stimfunction,
tr_duration=int(data_dict['trDuration']),
template=template,
mask=mask,
noise_dict=noise_dict,
)
# Create the stimulus time course of the conditions
total_time = int(data_dict['numTRs'] * data_dict['trDuration'])
onsets_A = []
onsets_B = []
curr_time = data_dict['burn_in']
while curr_time < (total_time - data_dict['event_duration']):
# Flip a coin for each epoch to determine whether it is A or B
if np.random.randint(0, 2) == 1:
onsets_A.append(curr_time)
else:
onsets_B.append(curr_time)
# Increment the current time
curr_time += data_dict['event_duration'] + data_dict['isi']
# How many timepoints per second of the stim function are to be generated?
temporal_res = 1 / data_dict['trDuration']
# Create a time course of events
event_durations = [data_dict['event_duration']]
stimfunc_A = sim.generate_stimfunction(
onsets=onsets_A,
event_durations=event_durations,
total_time=total_time,
temporal_resolution=temporal_res,
)
stimfunc_B = sim.generate_stimfunction(
onsets=onsets_B,
event_durations=event_durations,
total_time=total_time,
temporal_resolution=temporal_res,
)
# Create a labels timecourse
outFile = os.path.join(outputDir, 'labels.npy')
np.save(outFile, (stimfunc_A + (stimfunc_B * 2)))
roiA_file = os.path.join(inputDir, 'ROI_A.nii.gz')
roiB_file = os.path.join(inputDir, 'ROI_B.nii.gz')
# How is the signal implemented in the different ROIs
signal_A = generate_ROIs(roiA_file, stimfunc_A, noise,
data_dict['scale_percentage'], data_dict)
if data_dict['different_ROIs'] is True:
signal_B = generate_ROIs(roiB_file, stimfunc_B, noise,
data_dict['scale_percentage'], data_dict)
else:
# Halve the evoked response if these effects are both expected in the same ROI
if data_dict['multivariate_pattern'] is False:
signal_B = generate_ROIs(roiA_file, stimfunc_B, noise,
data_dict['scale_percentage'] * 0.5,
data_dict)
else:
signal_B = generate_ROIs(roiA_file, stimfunc_B, noise,
data_dict['scale_percentage'], data_dict)
# Combine the two signal timecourses
signal = signal_A + signal_B
print('Generating TRs in real time')
for idx in range(data_dict['numTRs']):
# Create the brain volume on this TR
brain = noise[:, :, :, idx] + signal[:, :, :, idx]
# Convert file to integers to mimic what you get from MR
brain_int32 = brain.astype(np.int32)
# Store as dicom or nifti?
if data_dict['save_dicom'] is True:
# Save the volume as a DICOM file, with each TR as its own file
output_file = os.path.join(outputDir,
'rt_' + format(idx, '03d') + '.dcm')
write_dicom(output_file, brain_int32, idx + 1)
else:
# Save the volume as a numpy file, with each TR as its own file
output_file = os.path.join(outputDir,
'rt_' + format(idx, '03d') + '.npy')
np.save(output_file, brain_int32)
print("Generate {}".format(output_file))
# Sleep until next TR
if data_dict['save_realtime'] == 1:
time.sleep(data_dict['trDuration'])
def test_generate_noise():
dimensions = np.array([10, 10, 10]) # What is the size of the brain
feature_size = [2]
feature_type = ['cube']
feature_coordinates = np.array([[5, 5, 5]])
signal_magnitude = [1]
# Generate a volume representing the location and quality of the signal
volume = sim.generate_signal(
dimensions=dimensions,
feature_coordinates=feature_coordinates,
feature_type=feature_type,
feature_size=feature_size,
signal_magnitude=signal_magnitude,
)
# Inputs for generate_stimfunction
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 200
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(
onsets=onsets,
event_durations=event_durations,
total_time=duration,
)
signal_function = sim.convolve_hrf(
stimfunction=stimfunction,
tr_duration=tr_duration,
)
# Convolve the HRF with the stimulus sequence
signal = sim.apply_signal(
signal_function=signal_function,
volume_signal=volume,
)
# Generate the mask of the signal
mask, template = sim.mask_brain(signal, mask_self=None)
assert min(mask[mask > 0]) > 0.1, "Mask thresholding did not work"
assert len(np.unique(template) > 2), "Template creation did not work"
stimfunction_tr = stimfunction[::int(tr_duration * 100)]
# Create the noise volumes (using the default parameters)
noise = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
iterations=[1, 0],
)
assert signal.shape == noise.shape, "The dimensions of signal and noise " \
"the same"
noise_high = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict={
'sfnr': 50,
'snr': 25
},
iterations=[1, 0],
)
noise_low = sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict={
'sfnr': 100,
'snr': 25
},
iterations=[1, 0],
)
system_high = np.std(noise_high[mask > 0], 1).mean()
system_low = np.std(noise_low[mask > 0], 1).mean()
assert system_low < system_high, "SFNR noise could not be manipulated"
# Check that you check for the appropriate template values
with pytest.raises(ValueError):
sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template * 2,
mask=mask,
noise_dict={},
)
# Check that iterations does what it should
sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict={},
iterations=[0, 0],
)
sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict={},
iterations=None,
)
# Test drift noise
trs = 1000
period = 100
drift = sim._generate_noise_temporal_drift(
trs,
tr_duration,
'sine',
period,
)
# Check that the max frequency is the appropriate frequency
power = abs(np.fft.fft(drift))[1:trs // 2]
freq = np.linspace(1, trs // 2 - 1, trs // 2 - 1) / trs
period_freq = np.where(freq == 1 / (period // tr_duration))
max_freq = np.argmax(power)
assert period_freq == max_freq, 'Max frequency is not where it should be'
# Do the same but now with cosine basis functions, answer should be close
drift = sim._generate_noise_temporal_drift(
trs,
tr_duration,
'discrete_cos',
period,
)
# Check that the appropriate frequency is peaky (may not be the max)
power = abs(np.fft.fft(drift))[1:trs // 2]
freq = np.linspace(1, trs // 2 - 1, trs // 2 - 1) / trs
period_freq = np.where(freq == 1 / (period // tr_duration))[0][0]
assert power[period_freq] > power[period_freq + 1], 'Power is low'
assert power[period_freq] > power[period_freq - 1], 'Power is low'
# Check it runs fine
drift = sim._generate_noise_temporal_drift(
50,
tr_duration,
'discrete_cos',
period,
)
# Check it runs fine
drift = sim._generate_noise_temporal_drift(
300,
tr_duration,
'cos_power_drop',
period,
)
# Check that when the TR is greater than the period it errors
with pytest.raises(ValueError):
sim._generate_noise_temporal_drift(30, 10, 'cos_power_drop', 5)
# Test physiological noise (using unrealistic parameters so that it's easy)
timepoints = list(np.linspace(0, (trs - 1) * tr_duration, trs))
resp_freq = 0.2
heart_freq = 1.17
phys = sim._generate_noise_temporal_phys(
timepoints,
resp_freq,
heart_freq,
)
# Check that the max frequency is the appropriate frequency
power = abs(np.fft.fft(phys))[1:trs // 2]
freq = np.linspace(1, trs // 2 - 1, trs // 2 - 1) / (trs * tr_duration)
peaks = (power > (power.mean() + power.std())) # Where are the peaks
peak_freqs = freq[peaks]
assert np.any(resp_freq == peak_freqs), 'Resp frequency not found'
assert len(peak_freqs) == 2, 'Two peaks not found'
# Test task noise
sim._generate_noise_temporal_task(
stimfunction_tr,
motion_noise='gaussian',
)
sim._generate_noise_temporal_task(
stimfunction_tr,
motion_noise='rician',
)
# Test ARMA noise
with pytest.raises(ValueError):
noise_dict = {'fwhm': 4, 'auto_reg_rho': [1], 'ma_rho': [1, 1]}
sim._generate_noise_temporal_autoregression(
stimfunction_tr,
noise_dict,
dimensions,
mask,
)
# Generate spatial noise
vol = sim._generate_noise_spatial(np.array([10, 10, 10, trs]))
assert len(vol.shape) == 3, 'Volume was not reshaped to ignore TRs'
# Switch some of the noise types on
noise_dict = dict(physiological_sigma=1,
drift_sigma=1,
task_sigma=1,
auto_reg_sigma=0)
sim.generate_noise(
dimensions=dimensions,
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=noise_dict,
iterations=[0, 0],
)
def test_generate_noise():
dimensions = np.array([64, 64, 36]) # What is the size of the brain
feature_size = [2]
feature_type = ['cube']
feature_coordinates = np.array([[32, 32, 18]])
signal_magnitude = [30]
# Generate a volume representing the location and quality of the signal
volume_static = sim.generate_signal(
dimensions=dimensions,
feature_coordinates=feature_coordinates,
feature_type=feature_type,
feature_size=feature_size,
signal_magnitude=signal_magnitude,
)
# Inputs for generate_stimfunction
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 100
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(
onsets=onsets,
event_durations=event_durations,
total_time=duration,
tr_duration=tr_duration,
)
signal_function = sim.double_gamma_hrf(stimfunction=stimfunction, )
# Convolve the HRF with the stimulus sequence
signal = sim.apply_signal(
signal_function=signal_function,
volume_static=volume_static,
)
# Create the noise volumes (using the default parameters)
noise = sim.generate_noise(
dimensions=dimensions,
stimfunction=stimfunction,
tr_duration=tr_duration,
)
assert signal.shape == noise.shape, "The dimensions of signal " \
"and noise the same"
Z_noise = sim._generate_noise_temporal(stimfunction, tr_duration, 1)
noise = sim._generate_noise_temporal(stimfunction, tr_duration, 0)
assert np.std(Z_noise) < np.std(noise), "Z scoring is not working"
# Combine the signal and the noise
volume = signal + noise
assert np.std(signal) < np.std(noise), "Noise was not created"
noise = sim.generate_noise(dimensions=dimensions,
stimfunction=stimfunction,
tr_duration=tr_duration,
noise_strength=[0, 0, 0])
assert np.sum(noise) == 0, "Noise strength could not be manipulated"
assert np.std(noise) == 0, "Noise strength could not be manipulated"
def test_calc_noise():
# Inputs for functions
onsets = [10, 30, 50, 70, 90]
event_durations = [6]
tr_duration = 2
duration = 200
temporal_res = 100
tr_number = int(np.floor(duration / tr_duration))
dimensions_tr = np.array([10, 10, 10, tr_number])
# Preset the noise dict
nd_orig = sim._noise_dict_update({})
# Create the time course for the signal to be generated
stimfunction = sim.generate_stimfunction(
onsets=onsets,
event_durations=event_durations,
total_time=duration,
temporal_resolution=temporal_res,
)
# Mask the volume to be the same shape as a brain
mask, template = sim.mask_brain(dimensions_tr, mask_self=None)
stimfunction_tr = stimfunction[::int(tr_duration * temporal_res)]
nd_orig['matched'] = 0
noise = sim.generate_noise(
dimensions=dimensions_tr[0:3],
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=nd_orig,
)
# Check the spatial noise match
nd_orig['matched'] = 1
noise_matched = sim.generate_noise(dimensions=dimensions_tr[0:3],
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=nd_orig,
iterations=[50, 0])
# Calculate the noise parameters from this newly generated volume
nd_new = sim.calc_noise(noise, mask, template)
nd_matched = sim.calc_noise(noise_matched, mask, template)
# Check the values are reasonable"
assert nd_new['snr'] > 0, 'snr out of range'
assert nd_new['sfnr'] > 0, 'sfnr out of range'
assert nd_new['auto_reg_rho'][0] > 0, 'ar out of range'
# Check that the dilation increases SNR
no_dilation_snr = sim._calc_snr(
noise_matched,
mask,
dilation=0,
reference_tr=tr_duration,
)
assert nd_new['snr'] > no_dilation_snr, "Dilation did not increase SNR"
# Check that template size is in bounds
with pytest.raises(ValueError):
sim.calc_noise(noise, mask, template * 2)
# Check that Mask is set is checked
with pytest.raises(ValueError):
sim.calc_noise(noise, None, template)
# Check that it can deal with missing noise parameters
temp_nd = sim.calc_noise(noise, mask, template, noise_dict={})
assert temp_nd['voxel_size'][0] == 1, 'Default voxel size not set'
temp_nd = sim.calc_noise(noise, mask, template, noise_dict=None)
assert temp_nd['voxel_size'][0] == 1, 'Default voxel size not set'
# Check that the fitting worked
snr_diff = abs(nd_orig['snr'] - nd_new['snr'])
snr_diff_match = abs(nd_orig['snr'] - nd_matched['snr'])
assert snr_diff > snr_diff_match, 'snr fit incorrectly'
# Test that you can generate rician and exponential noise
sim._generate_noise_system(
dimensions_tr,
1,
1,
spatial_noise_type='exponential',
temporal_noise_type='rician',
)
# Check the temporal noise match
nd_orig['matched'] = 1
noise_matched = sim.generate_noise(dimensions=dimensions_tr[0:3],
stimfunction_tr=stimfunction_tr,
tr_duration=tr_duration,
template=template,
mask=mask,
noise_dict=nd_orig,
iterations=[0, 50])
nd_matched = sim.calc_noise(noise_matched, mask, template)
sfnr_diff = abs(nd_orig['sfnr'] - nd_new['sfnr'])
sfnr_diff_match = abs(nd_orig['sfnr'] - nd_matched['sfnr'])
assert sfnr_diff > sfnr_diff_match, 'sfnr fit incorrectly'
ar1_diff = abs(nd_orig['auto_reg_rho'][0] - nd_new['auto_reg_rho'][0])
ar1_diff_match = abs(nd_orig['auto_reg_rho'][0] -
nd_matched['auto_reg_rho'][0])
assert ar1_diff > ar1_diff_match, 'AR1 fit incorrectly'
# Check that you can calculate ARMA for a single voxel
vox = noise[5, 5, 5, :]
arma = sim._calc_ARMA_noise(
vox,
None,
sample_num=2,
)
assert len(arma) == 2, "Two outputs not given by ARMA"
