Struct HeidiParams 

pub struct HeidiParams {

    pub cone_threshold: f64,
    pub gradient_threshold: f64,
    pub apply_laplacian_correction: bool,
    pub laplacian_threshold: f64,
    pub denoise: Option<AnisotropicDiffusionParams>,
    pub gradient_mask_floor: f64,
    pub continuation_steps: usize,
    pub inner_iterations: usize,
    pub mu_min: f64,
    pub tol: f64,
    pub b0: f64,
    pub echo_times: Vec<f64>,
    pub te: EchoTime,
}

Parameters for heidi.

Fields

cone_threshold: f64

|D(k)| above which a coefficient counts as well-conditioned and is taken from chi_init unchanged.

gradient_threshold: f64

Field-gradient threshold in ppm/mm. Below it, a direction is considered homogeneous and its TV weight is switched on.

apply_laplacian_correction: bool

Open the homogeneity masks where the field’s Laplacian is large.

laplacian_threshold: f64

Field-Laplacian threshold in ppm per voxel² (the stencil is in voxel units, matching upstream).

denoise: Option<AnisotropicDiffusionParams>

Anisotropic-diffusion smoothing applied to the field (in upstream’s phase units) before the gradients are taken. None skips it. Its reference_itk reproduces the ITK executable upstream runs (see the module docs).

gradient_mask_floor: f64

TV weight kept at voxels that read as an edge in every direction.

The default, 0, is what the shipped code does: computesusceptibility_heidi sets these voxels to gradientMaskReplaceValue (0.1) and then passes the masks through fourierbasedresize(...) > 0.5, which rounds the 0.1 away. (Its changelog, v10.6, says the masks “must contain values of 0.1”; set 0.1 to follow that instead.)

continuation_steps: usize

NESTA continuation steps on μ.

This is the dominant quality knob, and more is not better. μ always lands on mu_min, so the count controls how many levels of TV minimisation are actually performed — past the first couple of levels μ is already far below the field’s gradient scale and the term is effectively pure TV, so extra levels just keep flattening. On the CI phantom, at otherwise-default settings: 4 steps → r=0.8984, 6 → r=0.9017, 8 (upstream’s DEFAULT_TV_MAXINTITER, kept here) → r=0.8757. Worth sweeping on your own data before trusting the default.

inner_iterations: usize

Accelerated-gradient iterations per continuation step.

mu_min: f64

Final (smallest) smoothing parameter μ.

tol: f64

Relative objective change that ends an inner loop early.

The driver inversion_heidi.m sets PostProcCone.tol = eps, which never triggers: every continuation step runs all inner_iterations. (computesusceptibility_heidi’s own default, used when the driver sets none, is DEFAULT_TV_TOLVAR = 1e-3.)

b0: f64

Field strength in tesla (scan parameter, for the conversion to upstream’s units)

echo_times: Vec<f64>

Echo times in seconds, in acquisition order (scan parameter, for the conversion to upstream’s units).

te: EchoTime

Which echo time the field stands for the phase of, in the conversion to upstream’s units. Default EchoTime::Last: qsm_heidi’s phase2susceptibilityfactor.m uses the last echo time unless an echo is named (heidi/phase2susceptibilityfactor.m lines 70-71, qsm_heidi 4592631), and inversion_heidi.m names none (line 58).

Trait Implementations

impl Clone for HeidiParams

fn clone(&self) -> HeidiParams

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for HeidiParams

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

Formats the value using the given formatter.

impl Default for HeidiParams

fn default() -> Self

inversion_heidi.m’s settings.

impl Serialize for HeidiParams

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

where __S: Serializer,

Serialize this value into the given Serde serializer.