def fit_gaussian(spectra, f_ppm, lb=2.6, ub=3.6):
"""
Fit a gaussian function to the difference spectra to be used for estimation
of the GABA peak.
Parameters
----------
spectra : array of shape (n_transients, n_points)
Typically the difference of the on/off spectra in each transient.
f_ppm : array
lb, ub : floats
In ppm, the range over which optimization is bounded
"""
idx = ut.make_idx(f_ppm, lb, ub)
# We are only going to look at the interval between lb and ub
n_points = idx.stop - idx.start
n_params = 5
fit_func = ut.gaussian
# Set the bounds for the optimization
bounds = [
(lb, ub), # peak location
(0, None), # sigma
(0, None), # amp
(None, None), # offset
(None, None) # drift
]
model = np.empty((spectra.shape[0], n_points))
signal = np.empty((spectra.shape[0], n_points))
params = np.empty((spectra.shape[0], n_params))
for ii, xx in enumerate(spectra):
# We fit to the real spectrum:
signal[ii] = np.real(xx[idx])
# Use the signal for a rough estimate of the parameters for
# initialization :
max_idx = np.argmax(signal[ii])
max_sig = np.max(signal[ii])
initial_f0 = f_ppm[idx][max_idx]
half_max_idx = np.argmin(np.abs(signal[ii] - max_sig / 2))
# We estimate sigma as the hwhm:
initial_sigma = np.abs(initial_f0 - f_ppm[idx][half_max_idx])
initial_off = np.min(signal[ii])
initial_drift = 0
initial_amp = max_sig
initial = (initial_f0, initial_sigma, initial_amp, initial_off,
initial_drift)
params[ii], _ = lsq.leastsqbound(mopt.err_func,
initial,
args=(f_ppm[idx], np.real(signal[ii]),
fit_func),
bounds=bounds)
model[ii] = fit_func(f_ppm[idx], *params[ii])
return model, signal, params
def fit_water(self,
line_broadening=5,
zerofill=100,
filt_method=None,
min_ppm=-5.0,
max_ppm=5.0):
"""
"""
# Get the water spectrum as well:
f_hz, w_spectra = ana.get_spectra(self.water_fid,
line_broadening=line_broadening,
zerofill=zerofill,
filt_method=filt_method)
f_ppm = ut.freq_to_ppm(f_hz)
# Averaging across echos:
self.water_spectra = np.mean(w_spectra, 1)
model, signal, params = ana.fit_lorentzian(self.water_spectra,
self.f_ppm,
lb=min_ppm,
ub=max_ppm)
# Store the params:
self.water_model = model
self.water_signal = signal
self.water_params = params
self.water_idx = ut.make_idx(self.f_ppm, min_ppm, max_ppm)
mean_params = stats.nanmean(params, 0)
self.water_auc = self._calc_auc(ut.lorentzian, params, self.water_idx)
def fit_water(self, line_broadening=5, zerofill=100,
filt_method=None, min_ppm=-5.0, max_ppm=5.0):
"""
"""
# Get the water spectrum as well:
f_hz, w_spectra = ana.get_spectra(self.water_fid,
line_broadening=line_broadening,
zerofill=zerofill,
filt_method=filt_method)
f_ppm = ut.freq_to_ppm(f_hz)
# Averaging across echos:
self.water_spectra = np.mean(w_spectra, 1)
model, signal, params = ana.fit_lorentzian(self.water_spectra,
self.f_ppm,
lb=min_ppm,
ub=max_ppm)
# Store the params:
self.water_model = model
self.water_signal = signal
self.water_params = params
self.water_idx = ut.make_idx(self.f_ppm, min_ppm, max_ppm)
mean_params = stats.nanmean(params, 0)
self.water_auc = self._calc_auc(ut.lorentzian, params, self.water_idx)
def fit_two_gaussian(spectra, f_ppm, lb=3.6, ub=3.9):
"""
Fit a gaussian function to the difference spectra
This is useful for estimation of the Glx peak, which tends to have two
peaks.
Parameters
----------
spectra : array of shape (n_transients, n_points)
Typically the difference of the on/off spectra in each transient.
f_ppm : array
lb, ub : floats
In ppm, the range over which optimization is bounded
"""
idx = ut.make_idx(f_ppm, lb, ub)
# We are only going to look at the interval between lb and ub
n_points = idx.stop - idx.start
n_params = 8
fit_func = ut.two_gaussian
# Set the bounds for the optimization
bounds = [
(lb, ub), # peak 1 location
(lb, ub), # peak 2 location
(0, None), # sigma 1
(0, None), # sigma 2
(0, None), # amp 1
(0, None), # amp 2
(None, None), # offset
(None, None), # drift
]
model = np.empty((spectra.shape[0], n_points))
signal = np.empty((spectra.shape[0], n_points))
params = np.empty((spectra.shape[0], n_params))
for ii, xx in enumerate(spectra):
# We fit to the real spectrum:
signal[ii] = np.real(xx[idx])
params[ii] = _do_two_gaussian_fit(f_ppm[idx],
np.real(signal[ii]),
bounds=bounds)
model[ii] = fit_func(f_ppm[idx], *params[ii])
return model, signal, params
def fit_two_lorentzian(spectra, f_ppm, lb=2.6, ub=3.6):
"""
Fit a lorentzian function to the sum spectra to be used for estimation of
the creatine and choline peaks.
Parameters
----------
spectra : array of shape (n_transients, n_points)
Typically the sum of the on/off spectra in each transient.
f_ppm : array
lb, ub : floats
In ppm, the range over which optimization is bounded
"""
# We are only going to look at the interval between lb and ub
idx = ut.make_idx(f_ppm, lb, ub)
n_points = np.abs(idx.stop - idx.start)
n_params = 10 # Lotsa params!
# Set the bounds for the optimization
bounds = [
(lb, ub), #peak1
(lb, ub), #peak2
(0, None), #area1
(0, None), #area2
(0, ub - lb), #hwhm1
(0, ub - lb), #hwhm2
(-np.pi / 2, np.pi / 2), #phase
(-np.pi / 2, np.pi / 2), #phase
(None, None), #offset
(None, None)
] #drift
model = np.empty((spectra.shape[0], n_points))
signal = np.empty((spectra.shape[0], n_points))
params = np.empty((spectra.shape[0], n_params))
for ii, xx in enumerate(spectra):
# We fit to the real spectrum:
signal[ii] = np.real(xx[idx])
params[ii] = _do_two_lorentzian_fit(f_ppm[idx],
np.real(signal[ii]),
bounds=bounds)
model[ii] = ut.two_lorentzian(f_ppm[idx], *params[ii])
return model, signal, params
def fit_two_gaussian(spectra, f_ppm, lb=3.6, ub=3.9):
"""
Fit a gaussian function to the difference spectra
This is useful for estimation of the Glx peak, which tends to have two
peaks.
Parameters
----------
spectra : array of shape (n_transients, n_points)
Typically the difference of the on/off spectra in each transient.
f_ppm : array
lb, ub: floats
In ppm, the range over which optimization is bounded
"""
idx = ut.make_idx(f_ppm, lb, ub)
# We are only going to look at the interval between lb and ub
n_points = idx.stop - idx.start
n_params = 8
fit_func = ut.two_gaussian
# Set the bounds for the optimization
bounds = [(lb,ub), # peak 1 location
(lb,ub), # peak 2 location
(0,None), # sigma 1
(0,None), # sigma 2
(0,None), # amp 1
(0,None), # amp 2
(None, None), # offset
(None, None), # drift
]
model = np.empty((spectra.shape[0], n_points))
signal = np.empty((spectra.shape[0], n_points))
params = np.empty((spectra.shape[0], n_params))
for ii, xx in enumerate(spectra):
# We fit to the real spectrum:
signal[ii] = np.real(xx[idx])
params[ii] = _do_two_gaussian_fit(f_ppm[idx], np.real(signal[ii]),
bounds=bounds)
model[ii] = fit_func(f_ppm[idx], *params[ii])
return model, signal, params
def fit_two_lorentzian(spectra, f_ppm, lb=2.6, ub=3.6):
"""
Fit a lorentzian function to the sum spectra to be used for estimation of
the creatine and choline peaks.
Parameters
----------
spectra : array of shape (n_transients, n_points)
Typically the sum of the on/off spectra in each transient.
f_ppm : array
lb, ub: floats
In ppm, the range over which optimization is bounded
"""
# We are only going to look at the interval between lb and ub
idx = ut.make_idx(f_ppm, lb, ub)
n_points = np.abs(idx.stop - idx.start)
n_params = 10 # Lotsa params!
# Set the bounds for the optimization
bounds = [(lb,ub), #peak1
(lb,ub), #peak2
(0,None), #area1
(0,None), #area2
(0,ub-lb), #hwhm1
(0,ub-lb), #hwhm2
(-np.pi/2, np.pi/2), #phase
(-np.pi/2, np.pi/2), #phase
(None,None), #offset
(None, None)] #drift
model = np.empty((spectra.shape[0], n_points))
signal = np.empty((spectra.shape[0], n_points))
params = np.empty((spectra.shape[0], n_params))
for ii, xx in enumerate(spectra):
# We fit to the real spectrum:
signal[ii] = np.real(xx[idx])
params[ii] = _do_two_lorentzian_fit(f_ppm[idx], np.real(signal[ii]),
bounds=bounds)
model[ii] = ut.two_lorentzian(f_ppm[idx], *params[ii])
return model, signal, params
def fit_naa(self, reject_outliers=3.0, fit_lb=1.8, fit_ub=2.4,
phase_correct=True):
"""
Fit a Lorentzian function to the NAA peak at ~ 2 ppm. Example of
fitting inverted peak: Foerster et al. 2013, An imbalance between
excitatory and inhibitory neurotransmitters in amyothrophic lateral
sclerosis revealed by use of 3T proton MRS
"""
model, signal, params = ana.fit_lorentzian(self.diff_spectra,
self.f_ppm,
lb=fit_lb,
ub=fit_ub)
# Store the params:
self.naa_model = model
self.naa_signal = signal
self.naa_params = params
self.naa_idx = ut.make_idx(self.f_ppm, fit_lb, fit_ub)
mean_params = stats.nanmean(params, 0)
self.naa_auc = self._calc_auc(ut.lorentzian, params, self.naa_idx)
def fit_naa(self, reject_outliers=3.0, fit_lb=1.8, fit_ub=2.4,
phase_correct=True):
"""
Fit a Lorentzian function to the NAA peak at ~ 2 ppm. Example of
fitting inverted peak: Foerster et al. 2013, An imbalance between
excitatory and inhibitory neurotransmitters in amyothrophic lateral
sclerosis revealed by use of 3T proton MRS
"""
model, signal, params = ana.fit_lorentzian(self.diff_spectra,
self.f_ppm,
lb=fit_lb,
ub=fit_ub)
# Store the params:
self.naa_model = model
self.naa_signal = signal
self.naa_params = params
self.naa_idx = ut.make_idx(self.f_ppm, fit_lb, fit_ub)
mean_params = stats.nanmean(params, 0)
self.naa_auc = self._calc_auc(ut.lorentzian, params, self.naa_idx)
def fit_glx2(self,
reject_outliers=3.0,
fit_lb=3.6,
fit_ub=3.9,
phase_correct=True,
scalefit=False):
"""
Parameters
----------
reject_outliers : float or bool
If set to a float, this is the z score threshold for rejection (on
any of the parameters). If set to False, no outlier rejection
fit_lb, fit_ub : float
What part of the spectrum (in ppm) contains the creatine peak.
Default (3.5, 4.2)
scalefit : boolean
If this is set to true, attempt is made to tighten the fit to the
peak with a second round of fitting where the fitted curve
is fit with a scale factor. (default false)
"""
if not hasattr(self, 'creatine_params'):
self.fit_creatine()
fit_spectra = self.diff_spectra
# We fit a two-gaussian function to this entire chunk of the spectrum,
# to catch both glx peaks
model, signal, params = ana.fit_two_gaussian(fit_spectra,
self.f_ppm,
lb=fit_lb,
ub=fit_ub)
# Use an array of ones to index everything but the outliers and nans:
ii = np.ones(signal.shape[0], dtype=bool)
# Reject outliers:
if reject_outliers:
model, signal, params, ii = self._outlier_rejection(
params, model, signal, ii)
# We'll keep around a private attribute to tell us which transients
# were good:
self._glx2_transients = np.where(ii)
# Now we separate params of the two glx peaks from each other
# (remember that they both share offset and drift!):
self.glxp1_params = params[:, (0, 2, 4, 6, 7)]
self.glxp2_params = params[:, (1, 3, 5, 6, 7)]
self.glx2_idx = ut.make_idx(self.f_ppm, fit_lb, fit_ub)
# We'll need to generate the model predictions from these parameters,
# because what we're holding in 'model' is for both together:
self.glxp1_model = np.zeros(
(self.glxp1_params.shape[0],
np.abs(self.glx2_idx.stop - self.glx2_idx.start)))
self.glxp2_model = np.zeros(
(self.glxp2_params.shape[0],
np.abs(self.glx2_idx.stop - self.glx2_idx.start)))
for idx in range(self.glxp2_params.shape[0]):
self.glxp2_model[idx] = ut.gaussian(self.f_ppm[self.glx2_idx],
*self.glxp2_params[idx])
self.glxp1_model[idx] = ut.gaussian(self.f_ppm[self.glx2_idx],
*self.glxp1_params[idx])
if scalefit:
combinedmodel = self.glxp2_model + self.glxp1_model
scalefac, scalemodel = ana._do_scale_fit(self.f_ppm[self.glx2_idx],
signal, combinedmodel)
# Reject outliers:
scalemodel, signal, params, ii = self._rm_outlier_by_amp(
params, scalemodel, signal, ii)
self.glx2_model = scalemodel
else:
self.glx2_model = self.glxp1_model + self.glxp2_model
self.glx2_signal = signal
self.glx2_auc = (
self._calc_auc(ut.gaussian, self.glxp2_params, self.glx2_idx) +
self._calc_auc(ut.gaussian, self.glxp1_params, self.glx2_idx))
def _fit_helper(self, fit_spectra, reject_outliers, fit_lb, fit_ub,
fit_func):
"""
This is a helper function for fitting different segments of the spectrum
with Gaussian functions (GLX and GABA).
Parameters
----------
fit_spectra : ndarray
The data to fit
reject_outliers : float or bool
Z score for outlier rejection. If set to `False`, not outlier
rejection.
fit_lb : float
The lower bound of the part of the ppm scale for which the Gaussian
is fit.
fit_ub : float
The upper bound of the part of the scale fit.
fit_func: callable
e.g. `fit_gaussian`
Returns
-------
choose_transients : tuple
Indices into the original data's transients dimension to select
non-outlier transients. If reject_outliers is set to `False`, this is
all the transients
model : ndarray
The model predicition in each transient, based on the fit.
signal : ndarray
The original signal in this part of the difference spectrum.
params : ndarray
The Gaussian parameters in each transient as fit.
this_idx : slice object
A slice into the part of the spectrum that is fit
"""
# fit_idx should already be set from fitting the creatine params:
model, signal, params = fit_func(fit_spectra,
self.f_ppm,
lb=fit_lb,
ub=fit_ub)
# We'll use these indices to reject outliers (or not):
ii = np.ones(signal.shape[0], dtype=bool)
# Reject outliers:
if reject_outliers:
model, signal, params, ii = self._outlier_rejection(
params, model, signal, ii)
choose_transients = np.where(ii)
this_idx = ut.make_idx(self.f_ppm, fit_lb, fit_ub)
return choose_transients, model, signal, params, this_idx
def fit_creatine(self, reject_outliers=3.0, fit_lb=2.7, fit_ub=3.5):
"""
Fit a model to the portion of the summed spectra containing the
creatine and choline signals.
Parameters
----------
reject_outliers : float or bool
If set to a float, this is the z score threshold for rejection (on
any of the parameters). If set to False, no outlier rejection
fit_lb, fit_ub : float
What part of the spectrum (in ppm) contains the creatine peak.
Default (2.7, 3.5)
Note
----
We use upper and lower bounds that are a variation on the bounds
mentioned on the GANNET ISMRM2013 poster [1]_.
[1] RAE Edden et al (2013). Gannet GABA analysis toolkit. ISMRM
conference poster.
"""
# We fit a two-lorentz function to this entire chunk of the spectrum,
# to catch both choline and creatine
model, signal, params = ana.fit_two_lorentzian(self.sum_spectra,
self.f_ppm,
lb=fit_lb,
ub=fit_ub)
# Use an array of ones to index everything but the outliers and nans:
ii = np.ones(signal.shape[0], dtype=bool)
# Reject outliers:
if reject_outliers:
model, signal, params, ii = self._outlier_rejection(
params, model, signal, ii)
# We'll keep around a private attribute to tell us which transients
# were good (this is for both creatine and choline):
self._cr_transients = np.where(ii)
# Now we separate choline and creatine params from each other (remember
# that they both share offset and drift!):
self.choline_params = params[:, (0, 2, 4, 6, 8, 9)]
self.creatine_params = params[:, (1, 3, 5, 7, 8, 9)]
self.cr_idx = ut.make_idx(self.f_ppm, fit_lb, fit_ub)
# We'll need to generate the model predictions from these parameters,
# because what we're holding in 'model' is for both together:
self.choline_model = np.zeros(
(self.creatine_params.shape[0],
np.abs(self.cr_idx.stop - self.cr_idx.start)))
self.creatine_model = np.zeros(
(self.choline_params.shape[0],
np.abs(self.cr_idx.stop - self.cr_idx.start)))
for idx in range(self.creatine_params.shape[0]):
self.creatine_model[idx] = ut.lorentzian(
self.f_ppm[self.cr_idx], *self.creatine_params[idx])
self.choline_model[idx] = ut.lorentzian(self.f_ppm[self.cr_idx],
*self.choline_params[idx])
self.creatine_signal = signal
self.creatine_auc = self._calc_auc(ut.lorentzian, self.creatine_params,
self.cr_idx)
self.choline_auc = self._calc_auc(ut.lorentzian, self.choline_params,
self.cr_idx)
def fit_gaussian(spectra, f_ppm, lb=2.6, ub=3.6):
"""
Fit a gaussian function to the difference spectra to be used for estimation
of the GABA peak.
Parameters
----------
spectra : array of shape (n_transients, n_points)
Typically the difference of the on/off spectra in each transient.
f_ppm : array
lb, ub: floats
In ppm, the range over which optimization is bounded
"""
idx = ut.make_idx(f_ppm, lb, ub)
# We are only going to look at the interval between lb and ub
n_points = idx.stop - idx.start
n_params = 5
fit_func = ut.gaussian
# Set the bounds for the optimization
bounds = [(lb,ub), # peak location
(0,None), # sigma
(0,None), # amp
(None, None), # offset
(None, None) # drift
]
model = np.empty((spectra.shape[0], n_points))
signal = np.empty((spectra.shape[0], n_points))
params = np.empty((spectra.shape[0], n_params))
for ii, xx in enumerate(spectra):
# We fit to the real spectrum:
signal[ii] = np.real(xx[idx])
# Use the signal for a rough estimate of the parameters for
# initialization :
max_idx = np.argmax(signal[ii])
max_sig = np.max(signal[ii])
initial_f0 = f_ppm[idx][max_idx]
half_max_idx = np.argmin(np.abs(signal[ii] - max_sig/2))
# We estimate sigma as the hwhm:
initial_sigma = np.abs(initial_f0 - f_ppm[idx][half_max_idx])
initial_off = np.min(signal[ii])
initial_drift = 0
initial_amp = max_sig
initial = (initial_f0,
initial_sigma,
initial_amp,
initial_off,
initial_drift)
params[ii], _ = lsq.leastsqbound(mopt.err_func,
initial,
args=(f_ppm[idx],
np.real(signal[ii]),
fit_func), bounds=bounds)
model[ii] = fit_func(f_ppm[idx], *params[ii])
return model, signal, params
def fit_glx2(self, reject_outliers=3.0, fit_lb=3.6, fit_ub=3.9,
phase_correct=True, scalefit=False):
"""
Parameters
----------
reject_outliers : float or bool
If set to a float, this is the z score threshold for rejection (on
any of the parameters). If set to False, no outlier rejection
fit_lb, fit_ub : float
What part of the spectrum (in ppm) contains the creatine peak.
Default (3.5, 4.2)
scalefit : boolean
If this is set to true, attempt is made to tighten the fit to the
peak with a second round of fitting where the fitted curve
is fit with a scale factor. (default false)
"""
if not hasattr(self, 'creatine_params'):
self.fit_creatine()
fit_spectra = self.diff_spectra
# We fit a two-gaussian function to this entire chunk of the spectrum,
# to catch both glx peaks
model, signal, params = ana.fit_two_gaussian(fit_spectra,
self.f_ppm,
lb=fit_lb,
ub=fit_ub)
# Use an array of ones to index everything but the outliers and nans:
ii = np.ones(signal.shape[0], dtype=bool)
# Reject outliers:
if reject_outliers:
model, signal, params, ii = self._outlier_rejection(params,
model,
signal,
ii)
# We'll keep around a private attribute to tell us which transients
# were good:
self._glx2_transients = np.where(ii)
# Now we separate params of the two glx peaks from each other
# (remember that they both share offset and drift!):
self.glxp1_params = params[:, (0, 2, 4, 6, 7)]
self.glxp2_params = params[:, (1, 3, 5, 6, 7)]
self.glx2_idx = ut.make_idx(self.f_ppm, fit_lb, fit_ub)
# We'll need to generate the model predictions from these parameters,
# because what we're holding in 'model' is for both together:
self.glxp1_model = np.zeros((self.glxp1_params.shape[0],
np.abs(self.glx2_idx.stop-self.glx2_idx.start)))
self.glxp2_model = np.zeros((self.glxp2_params.shape[0],
np.abs(self.glx2_idx.stop-self.glx2_idx.start)))
for idx in range(self.glxp2_params.shape[0]):
self.glxp2_model[idx] = ut.gaussian(self.f_ppm[self.glx2_idx],
*self.glxp2_params[idx])
self.glxp1_model[idx] = ut.gaussian(self.f_ppm[self.glx2_idx],
*self.glxp1_params[idx])
if scalefit:
combinedmodel = self.glxp2_model + self.glxp1_model
scalefac, scalemodel = ana._do_scale_fit(
self.f_ppm[self.glx2_idx], signal,combinedmodel)
# Reject outliers:
scalemodel, signal, params, ii = self._rm_outlier_by_amp(params,
scalemodel,
signal,
ii)
self.glx2_model = scalemodel
else:
self.glx2_model = self.glxp1_model + self.glxp2_model
self.glx2_signal = signal
self.glx2_auc = (
self._calc_auc(ut.gaussian, self.glxp2_params, self.glx2_idx) +
self._calc_auc(ut.gaussian, self.glxp1_params, self.glx2_idx))
def _fit_helper(self, fit_spectra, reject_outliers, fit_lb, fit_ub,
fit_func):
"""
This is a helper function for fitting different segments of the spectrum
with Gaussian functions (GLX and GABA).
Parameters
----------
fit_spectra : ndarray
The data to fit
reject_outliers : float or bool
Z score for outlier rejection. If set to `False`, not outlier
rejection.
fit_lb : float
The lower bound of the part of the ppm scale for which the Gaussian
is fit.
fit_ub : float
The upper bound of the part of the scale fit.
fit_func: callable
e.g. `fit_gaussian`
Returns
-------
choose_transients : tuple
Indices into the original data's transients dimension to select
non-outlier transients. If reject_outliers is set to `False`, this is
all the transients
model : ndarray
The model predicition in each transient, based on the fit.
signal : ndarray
The original signal in this part of the difference spectrum.
params : ndarray
The Gaussian parameters in each transient as fit.
this_idx : slice object
A slice into the part of the spectrum that is fit
"""
# fit_idx should already be set from fitting the creatine params:
model, signal, params = fit_func(fit_spectra,
self.f_ppm,
lb=fit_lb,
ub=fit_ub)
# We'll use these indices to reject outliers (or not):
ii = np.ones(signal.shape[0], dtype=bool)
# Reject outliers:
if reject_outliers:
model, signal, params, ii = self._outlier_rejection(params,
model,
signal,
ii)
choose_transients = np.where(ii)
this_idx = ut.make_idx(self.f_ppm, fit_lb, fit_ub)
return choose_transients, model, signal, params, this_idx
def fit_creatine(self, reject_outliers=3.0, fit_lb=2.7, fit_ub=3.5):
"""
Fit a model to the portion of the summed spectra containing the
creatine and choline signals.
Parameters
----------
reject_outliers : float or bool
If set to a float, this is the z score threshold for rejection (on
any of the parameters). If set to False, no outlier rejection
fit_lb, fit_ub : float
What part of the spectrum (in ppm) contains the creatine peak.
Default (2.7, 3.5)
Note
----
We use upper and lower bounds that are a variation on the bounds
mentioned on the GANNET ISMRM2013 poster [1]_.
[1] RAE Edden et al (2013). Gannet GABA analysis toolkit. ISMRM
conference poster.
"""
# We fit a two-lorentz function to this entire chunk of the spectrum,
# to catch both choline and creatine
model, signal, params = ana.fit_two_lorentzian(self.sum_spectra,
self.f_ppm,
lb=fit_lb,
ub=fit_ub)
# Use an array of ones to index everything but the outliers and nans:
ii = np.ones(signal.shape[0], dtype=bool)
# Reject outliers:
if reject_outliers:
model, signal, params, ii = self._outlier_rejection(params,
model,
signal,
ii)
# We'll keep around a private attribute to tell us which transients
# were good (this is for both creatine and choline):
self._cr_transients = np.where(ii)
# Now we separate choline and creatine params from each other (remember
# that they both share offset and drift!):
self.choline_params = params[:, (0,2,4,6,8,9)]
self.creatine_params = params[:, (1,3,5,7,8,9)]
self.cr_idx = ut.make_idx(self.f_ppm, fit_lb, fit_ub)
# We'll need to generate the model predictions from these parameters,
# because what we're holding in 'model' is for both together:
self.choline_model = np.zeros((self.creatine_params.shape[0],
np.abs(self.cr_idx.stop-self.cr_idx.start)))
self.creatine_model = np.zeros((self.choline_params.shape[0],
np.abs(self.cr_idx.stop-self.cr_idx.start)))
for idx in range(self.creatine_params.shape[0]):
self.creatine_model[idx] = ut.lorentzian(self.f_ppm[self.cr_idx],
*self.creatine_params[idx])
self.choline_model[idx] = ut.lorentzian(self.f_ppm[self.cr_idx],
*self.choline_params[idx])
self.creatine_signal = signal
self.creatine_auc = self._calc_auc(ut.lorentzian,
self.creatine_params,
self.cr_idx)
self.choline_auc = self._calc_auc(ut.lorentzian,
self.choline_params,
self.cr_idx)
