def fa(self):
r"""
Fractional anisotropy (FA) calculated from cached eigenvalues.
Returns
---------
fa : array (V, 1)
Calculated FA. Note: range is 0 <= FA <= 1.
Notes
--------
FA is calculated with the following equation:
.. math::
FA = \sqrt{\frac{1}{2}\frac{(\lambda_1-\lambda_2)^2+(\lambda_1-
\lambda_3)^2+(\lambda_2-lambda_3)^2}{\lambda_1^2+
\lambda_2^2+\lambda_3^2} }
"""
evals, wrap = _makearray(self.model_params[..., :3])
ev1 = evals[..., 0]
ev2 = evals[..., 1]
ev3 = evals[..., 2]
fa = np.sqrt(0.5 * ((ev1 - ev2)**2 + (ev2 - ev3)**2 + (ev3 - ev1)**2)
/ (ev1*ev1 + ev2*ev2 + ev3*ev3))
fa = wrap(np.asarray(fa))
return _filled(fa)
def tensor_eig_from_lo_tri(B, data):
"""Calculates parameters for creating a Tensor instance
Calculates tensor parameters from the six unique tensor elements. This
function can be passed to the Tensor class as a fit_method for creating a
Tensor instance from tensors stored in a nifti file.
Parameters:
-----------
B :
not currently used
data : array_like (..., 6)
diffusion tensors elements stored in lower triangular order
Returns
-------
dti_params
Eigen values and vectors, used by the Tensor class to create an
instance
"""
data, wrap = _makearray(data)
data_flat = data.reshape((-1, data.shape[-1]))
dti_params = np.empty((len(data_flat), 4, 3))
for ii in xrange(len(data_flat)):
tensor = from_lower_triangular(data_flat[ii])
eigvals, eigvecs = decompose_tensor(tensor)
dti_params[ii, 0] = eigvals
dti_params[ii, 1:] = eigvecs
dti_params.shape = data.shape[:-1] + (12,)
dti_params = wrap(dti_params)
return dti_params
def tensor_eig_from_lo_tri(B, data):
"""Calculates parameters for creating a Tensor instance
Calculates tensor parameters from the six unique tensor elements. This
function can be passed to the Tensor class as a fit_method for creating a
Tensor instance from tensors stored in a nifti file.
Parameters:
-----------
B :
not currently used
data : array_like (..., 6)
diffusion tensors elements stored in lower triangular order
Returns
-------
dti_params
Eigen values and vectors, used by the Tensor class to create an
instance
"""
data, wrap = _makearray(data)
data_flat = data.reshape((-1, data.shape[-1]))
dti_params = np.empty((len(data_flat), 4, 3))
for ii in xrange(len(data_flat)):
tensor = from_lower_triangular(data_flat[ii])
eigvals, eigvecs = decompose_tensor(tensor)
dti_params[ii, 0] = eigvals
dti_params[ii, 1:] = eigvecs
dti_params.shape = data.shape[:-1] + (12, )
dti_params = wrap(dti_params)
return dti_params
def _getD(self):
"""Calculates the 3x3 diffusion tensor for each voxel"""
params, wrap = _makearray(self.model_params)
evals = params[..., :3]
evecs = params[..., 3:]
evals_flat = evals.reshape((-1, 3))
evecs_flat = evecs.reshape((-1, 3, 3))
D_flat = np.empty(evecs_flat.shape)
for ii in xrange(len(D_flat)):
Q = evecs_flat[ii]
L = evals_flat[ii]
D_flat[ii] = np.dot(Q * L, Q.T)
D = _filled(wrap(D_flat))
D.shape = self.shape + (3, 3)
return D
def _getD(self):
"""Calculates the 3x3 diffusion tensor for each voxel"""
params, wrap = _makearray(self.model_params)
evals = params[..., :3]
evecs = params[..., 3:]
evals_flat = evals.reshape((-1, 3))
evecs_flat = evecs.reshape((-1, 3, 3))
D_flat = np.empty(evecs_flat.shape)
for ii in xrange(len(D_flat)):
Q = evecs_flat[ii]
L = evals_flat[ii]
D_flat[ii] = np.dot(Q * L, Q.T)
D = _filled(wrap(D_flat))
D.shape = self.shape + (3, 3)
return D
def fa(self, fill_value=0, nonans=True):
r"""
Fractional anisotropy (FA) calculated from cached eigenvalues.
Parameters
----------
fill_value : float
value of fa where self.mask == True.
nonans : Bool
When True, fa is 0 when all eigenvalues are 0, otherwise fa is nan
Returns
---------
fa : array (V, 1)
Calculated FA. Note: range is 0 <= FA <= 1.
Notes
--------
FA is calculated with the following equation:
.. math::
FA = \sqrt{\frac{1}{2}\frac{(\lambda_1-\lambda_2)^2+(\lambda_1-
\lambda_3)^2+(\lambda_2-lambda_3)^2}{\lambda_1^2+
\lambda_2^2+\lambda_3^2} }
"""
evals, wrap = _makearray(self.model_params[..., :3])
ev1 = evals[..., 0]
ev2 = evals[..., 1]
ev3 = evals[..., 2]
if nonans:
all_zero = (ev1 == 0) & (ev2 == 0) & (ev3 == 0)
else:
all_zero = 0.0
fa = np.sqrt(
0.5
* ((ev1 - ev2) ** 2 + (ev2 - ev3) ** 2 + (ev3 - ev1) ** 2)
/ (ev1 * ev1 + ev2 * ev2 + ev3 * ev3 + all_zero)
)
fa = wrap(np.asarray(fa))
return _filled(fa, fill_value)
def fa(self, fill_value=0, nonans=True):
r"""
Fractional anisotropy (FA) calculated from cached eigenvalues.
Parameters
----------
fill_value : float
value of fa where self.mask == True.
nonans : Bool
When True, fa is 0 when all eigenvalues are 0, otherwise fa is nan
Returns
---------
fa : array (V, 1)
Calculated FA. Note: range is 0 <= FA <= 1.
Notes
--------
FA is calculated with the following equation:
.. math::
FA = \sqrt{\frac{1}{2}\frac{(\lambda_1-\lambda_2)^2+(\lambda_1-
\lambda_3)^2+(\lambda_2-lambda_3)^2}{\lambda_1^2+
\lambda_2^2+\lambda_3^2} }
"""
evals, wrap = _makearray(self.model_params[..., :3])
ev1 = evals[..., 0]
ev2 = evals[..., 1]
ev3 = evals[..., 2]
if nonans:
all_zero = (ev1 == 0) & (ev2 == 0) & (ev3 == 0)
else:
all_zero = 0.
fa = np.sqrt(0.5 * ((ev1 - ev2)**2 + (ev2 - ev3)**2 + (ev3 - ev1)**2) /
(ev1 * ev1 + ev2 * ev2 + ev3 * ev3 + all_zero))
fa = wrap(np.asarray(fa))
return _filled(fa, fill_value)
def ols_fit_tensor(design_matrix, data, min_signal=1):
r"""
Computes ordinary least squares (OLS) fit to calculate self-diffusion
tensor using a linear regression model [1]_.
Parameters
----------
design_matrix : array (g, 7)
Design matrix holding the covariants used to solve for the regression
coefficients. Use design_matrix to build a valid design matrix from
bvalues and a gradient table.
data : array ([X, Y, Z, ...], g)
Data or response variables holding the data. Note that the last
dimension should contain the data. It makes no copies of data.
min_signal : default = 1
All values below min_signal are repalced with min_signal. This is done
in order to avaid taking log(0) durring the tensor fitting.
Returns
-------
eigvals : array (..., 3)
Eigenvalues from eigen decomposition of the tensor.
eigvecs : array (..., 3, 3)
Associated eigenvectors from eigen decomposition of the tensor.
Eigenvectors are columnar (e.g. eigvecs[:,j] is associated with
eigvals[j])
See Also
--------
WLS_fit_tensor, decompose_tensor, design_matrix
Notes
-----
This function is offered mainly as a quick comparison to WLS.
.. math::
y = \mathrm{data} \\
X = \mathrm{design matrix} \\
\hat{\beta}_\mathrm{OLS} = (X^T X)^{-1} X^T y
References
----------
.. [1] Chung, SW., Lu, Y., Henry, R.G., 2006. Comparison of bootstrap
approaches for estimation of uncertainties of DTI parameters.
NeuroImage 33, 531-541.
"""
data, wrap = _makearray(data)
data_flat = data.reshape((-1, data.shape[-1]))
evals = np.empty((len(data_flat), 3))
evecs = np.empty((len(data_flat), 3, 3))
dti_params = np.empty((len(data_flat), 4, 3))
# obtain OLS fitting matrix
# U,S,V = np.linalg.svd(design_matrix, False)
# math: beta_ols = inv(X.T*X)*X.T*y
# math: ols_fit = X*beta_ols*inv(y)
# ols_fit = np.dot(U, U.T)
inv_design = np.linalg.pinv(design_matrix)
for param, sig in zip(dti_params, data_flat):
param[0], param[1:] = _ols_iter(inv_design, sig, min_signal)
dti_params.shape = data.shape[:-1] + (12,)
dti_params = wrap(dti_params)
return dti_params
def wls_fit_tensor(design_matrix, data, min_signal=1):
r"""
Computes weighted least squares (WLS) fit to calculate self-diffusion
tensor using a linear regression model [1]_.
Parameters
----------
design_matrix : array (g, 7)
Design matrix holding the covariants used to solve for the regression
coefficients.
data : array ([X, Y, Z, ...], g)
Data or response variables holding the data. Note that the last
dimension should contain the data. It makes no copies of data.
min_signal : default = 1
All values below min_signal are repalced with min_signal. This is done
in order to avaid taking log(0) durring the tensor fitting.
Returns
-------
eigvals : array (..., 3)
Eigenvalues from eigen decomposition of the tensor.
eigvecs : array (..., 3, 3)
Associated eigenvectors from eigen decomposition of the tensor.
Eigenvectors are columnar (e.g. eigvecs[:,j] is associated with
eigvals[j])
See Also
--------
decompose_tensor
Notes
-----
In Chung, et al. 2006, the regression of the WLS fit needed an unbiased
preliminary estimate of the weights and therefore the ordinary least
squares (OLS) estimates were used. A "two pass" method was implemented:
1. calculate OLS estimates of the data
2. apply the OLS estimates as weights to the WLS fit of the data
This ensured heteroscadasticity could be properly modeled for various
types of bootstrap resampling (namely residual bootstrap).
.. math::
y = \mathrm{data} \\
X = \mathrm{design matrix} \\
\hat{\beta}_\mathrm{WLS} = \mathrm{desired regression coefficients (e.g. tensor)}\\
\\
\hat{\beta}_\mathrm{WLS} = (X^T W X)^{-1} X^T W y \\
\\
W = \mathrm{diag}((X \hat{\beta}_\mathrm{OLS})^2),
\mathrm{where} \hat{\beta}_\mathrm{OLS} = (X^T X)^{-1} X^T y
References
----------
.. _[1] Chung, SW., Lu, Y., Henry, R.G., 2006. Comparison of bootstrap
approaches for estimation of uncertainties of DTI parameters.
NeuroImage 33, 531-541.
"""
if min_signal <= 0:
raise ValueError("min_signal must be > 0")
data, wrap = _makearray(data)
data_flat = data.reshape((-1, data.shape[-1]))
dti_params = np.empty((len(data_flat), 4, 3))
# obtain OLS fitting matrix
# U,S,V = np.linalg.svd(design_matrix, False)
# math: beta_ols = inv(X.T*X)*X.T*y
# math: ols_fit = X*beta_ols*inv(y)
# ols_fit = np.dot(U, U.T)
ols_fit = _ols_fit_matrix(design_matrix)
for param, sig in zip(dti_params, data_flat):
param[0], param[1:] = _wls_iter(ols_fit, design_matrix, sig, min_signal=min_signal)
dti_params.shape = data.shape[:-1] + (12,)
dti_params = wrap(dti_params)
return dti_params
def ols_fit_tensor(design_matrix, data, min_signal=1):
r"""
Computes ordinary least squares (OLS) fit to calculate self-diffusion
tensor using a linear regression model [1]_.
Parameters
----------
design_matrix : array (g, 7)
Design matrix holding the covariants used to solve for the regression
coefficients. Use design_matrix to build a valid design matrix from
bvalues and a gradient table.
data : array ([X, Y, Z, ...], g)
Data or response variables holding the data. Note that the last
dimension should contain the data. It makes no copies of data.
min_signal : default = 1
All values below min_signal are repalced with min_signal. This is done
in order to avaid taking log(0) durring the tensor fitting.
Returns
-------
eigvals : array (..., 3)
Eigenvalues from eigen decomposition of the tensor.
eigvecs : array (..., 3, 3)
Associated eigenvectors from eigen decomposition of the tensor.
Eigenvectors are columnar (e.g. eigvecs[:,j] is associated with
eigvals[j])
See Also
--------
WLS_fit_tensor, decompose_tensor, design_matrix
Notes
-----
This function is offered mainly as a quick comparison to WLS.
.. math::
y = \mathrm{data} \\
X = \mathrm{design matrix} \\
\hat{\beta}_\mathrm{OLS} = (X^T X)^{-1} X^T y
References
----------
.. [1] Chung, SW., Lu, Y., Henry, R.G., 2006. Comparison of bootstrap
approaches for estimation of uncertainties of DTI parameters.
NeuroImage 33, 531-541.
"""
data, wrap = _makearray(data)
data_flat = data.reshape((-1, data.shape[-1]))
evals = np.empty((len(data_flat), 3))
evecs = np.empty((len(data_flat), 3, 3))
dti_params = np.empty((len(data_flat), 4, 3))
#obtain OLS fitting matrix
#U,S,V = np.linalg.svd(design_matrix, False)
#math: beta_ols = inv(X.T*X)*X.T*y
#math: ols_fit = X*beta_ols*inv(y)
#ols_fit = np.dot(U, U.T)
inv_design = np.linalg.pinv(design_matrix)
for param, sig in zip(dti_params, data_flat):
param[0], param[1:] = _ols_iter(inv_design, sig, min_signal)
dti_params.shape = data.shape[:-1] + (12, )
dti_params = wrap(dti_params)
return dti_params
def wls_fit_tensor(design_matrix, data, min_signal=1):
r"""
Computes weighted least squares (WLS) fit to calculate self-diffusion
tensor using a linear regression model [1]_.
Parameters
----------
design_matrix : array (g, 7)
Design matrix holding the covariants used to solve for the regression
coefficients.
data : array ([X, Y, Z, ...], g)
Data or response variables holding the data. Note that the last
dimension should contain the data. It makes no copies of data.
min_signal : default = 1
All values below min_signal are repalced with min_signal. This is done
in order to avaid taking log(0) durring the tensor fitting.
Returns
-------
eigvals : array (..., 3)
Eigenvalues from eigen decomposition of the tensor.
eigvecs : array (..., 3, 3)
Associated eigenvectors from eigen decomposition of the tensor.
Eigenvectors are columnar (e.g. eigvecs[:,j] is associated with
eigvals[j])
See Also
--------
decompose_tensor
Notes
-----
In Chung, et al. 2006, the regression of the WLS fit needed an unbiased
preliminary estimate of the weights and therefore the ordinary least
squares (OLS) estimates were used. A "two pass" method was implemented:
1. calculate OLS estimates of the data
2. apply the OLS estimates as weights to the WLS fit of the data
This ensured heteroscadasticity could be properly modeled for various
types of bootstrap resampling (namely residual bootstrap).
.. math::
y = \mathrm{data} \\
X = \mathrm{design matrix} \\
\hat{\beta}_\mathrm{WLS} = \mathrm{desired regression coefficients (e.g. tensor)}\\
\\
\hat{\beta}_\mathrm{WLS} = (X^T W X)^{-1} X^T W y \\
\\
W = \mathrm{diag}((X \hat{\beta}_\mathrm{OLS})^2),
\mathrm{where} \hat{\beta}_\mathrm{OLS} = (X^T X)^{-1} X^T y
References
----------
.. _[1] Chung, SW., Lu, Y., Henry, R.G., 2006. Comparison of bootstrap
approaches for estimation of uncertainties of DTI parameters.
NeuroImage 33, 531-541.
"""
if min_signal <= 0:
raise ValueError('min_signal must be > 0')
data, wrap = _makearray(data)
data_flat = data.reshape((-1, data.shape[-1]))
dti_params = np.empty((len(data_flat), 4, 3))
#obtain OLS fitting matrix
#U,S,V = np.linalg.svd(design_matrix, False)
#math: beta_ols = inv(X.T*X)*X.T*y
#math: ols_fit = X*beta_ols*inv(y)
#ols_fit = np.dot(U, U.T)
ols_fit = _ols_fit_matrix(design_matrix)
for param, sig in zip(dti_params, data_flat):
param[0], param[1:] = _wls_iter(ols_fit,
design_matrix,
sig,
min_signal=min_signal)
dti_params.shape = data.shape[:-1] + (12, )
dti_params = wrap(dti_params)
return dti_params