Struct TfiParams 

pub struct TfiParams {

    pub lambda: f64,
    pub precond: f64,
    pub merit: bool,
    pub linear: bool,
    pub data_weighting: i32,
    pub percentage: f64,
    pub cg_tol: f64,
    pub cg_max_iter: usize,
    pub max_iter: usize,
    pub tol: f64,
    pub b0: f64,
    pub echo_times: Vec<f64>,
    pub te: EchoTime,
    pub kernel: TfiKernel,
    pub precision: TfiPrecision,
}

TFI parameters (2024 MEDI_L1.m / parse_QSM_input.m defaults; P_B and the linear data term from Liu et al. 2017)

Fields

lambda: f64

Weight of the data-fidelity term ('lambda'): larger fits the data more closely

precond: f64

Preconditioner outside the mask, P_B (1 inside); Liu et al. 2017 use 30 in vivo

merit: bool

MERIT: reweight the noise by the data residual after each iteration ('merit')

linear: bool

Linear data term ||m (D P y − f)||² ('linear', Liu et al. 2017 eq. 5); false gives MEDI_L1’s default exponential term ||m (exp(i D P y) − exp(i f))||²

data_weighting: i32

Data weighting: 0 uniform (1 everywhere), 1 SNR (mask / n_std, mean 1 in the mask)

percentage: f64

Edge fraction for the gradient mask ('percentage')

cg_tol: f64

CG relative residual tolerance

cg_max_iter: usize

Maximum CG iterations per Gauss-Newton step

max_iter: usize

Maximum Gauss-Newton iterations

tol: f64

Stop when ||dy|| / ||y|| falls below this

b0: f64

Field strength in Tesla (scan parameter, for the radian conversion)

echo_times: Vec<f64>

Echo times in seconds, in acquisition order (scan parameter, for the radian conversion).

te: EchoTime

Which echo time the radian conversion uses. Default EchoTime::Spacing: the MEDI toolbox’s MEDI_L1.m works on the field in radians over delta_TE (x/(2*pi*delta_TE*CF)*1e6), which its readers set to TE(2) - TE(1), or TE for one echo (Read_Siemens_DICOM.m lines 176-178).

kernel: TfiKernel

Dipole kernel (default TfiKernel::Dft). MEDI_L1.m calls dipole_kernel.m, whose k-space grid is half a step off on axes with an odd number of voxels; there it is not the dipole field. On QSM-CI’s simulated head (164×205×205) that scales χ by ~1.3 against the DFT-grid kernel and fits the true χ worse (in the mask, demeaned: NRMSE 71.4 % against 62.1 %, r 0.718 against 0.785), so the DFT-grid kernel is the default and TfiKernel::MediOddGrid the parity option. On even grids they are the same.

precision: TfiPrecision

Arithmetic (default TfiPrecision::Double). MEDI_L1.m runs in single precision, by accident of tempn = single(N_std) (see TfiPrecision::Single); the CG solves at cg_tol 0.01 then stop at different points from double precision, which moves χ by about 1 % (KKI travelling head) from the same inputs. Double precision solves the same equations more accurately, so it is the default and single precision the parity option.

Trait Implementations

impl Clone for TfiParams

fn clone(&self) -> TfiParams

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TfiParams

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for TfiParams

fn default() -> Self

Returns the “default value” for a type.

impl Serialize for TfiParams

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.