Module inversion 

Dipole inversion methods for QSM

This module provides various methods to solve the inverse problem of estimating magnetic susceptibility from local field measurements.

Methods include:

• TKD: Truncated k-space division (fast, simple)
• TSVD: Truncated SVD (zeros small values)
• Tikhonov: L2 regularization (closed-form)
• TV: Total variation regularization via ADMM (iterative)
• NLTV: Nonlinear-data-term TV via ADMM (QSM.jl nltv)
• RTS: Rapid two-step method
• MEDI: Morphology-enabled dipole inversion (plain, with SMV, or MEDI-mSMV)
• TGV: Total Generalized Variation (single-step from a field in ppm or one wrapped echo; α in ppm)
• COSMOS: multi-orientation least squares (closed form, well-posed)
• STI: susceptibility tensor imaging (multi-orientation, 6-component tensor)
• STAR-QSM: streaking artifact reduction (STI Suite 3.0’s QSM_star)

Re-exports

pub use iqsm_echo_weighting::IqsmEchoWeighting;

pub use cosmos::cosmos;

pub use cosmos::cosmos_weighted;

pub use cosmos::uncovered_cone;

pub use cosmos::CosmosParams;

pub use cosmos::CONE_THRESHOLD;

pub use cosmos::MAX_UNCOVERED_CONE;

pub use sti::sti;

pub use sti::sti_forward;

pub use sti::tensor_maps;

pub use sti::StiParams;

pub use sti::SusceptibilityTensor;

pub use sti::TensorMaps;

pub use sti::N_TENSOR;

pub use tkd::tkd;

pub use tkd::tsvd;

pub use tkd::TkdParams;

pub use tikhonov::tikhonov;

pub use tikhonov::TikhonovParams;

pub use tikhonov::Regularization;

pub use tv::tv_admm;

pub use tv::TvParams;

pub use nltv::nltv;

pub use nltv::NltvParams;

pub use rts::rts;

pub use rts::RtsParams;

pub use medi::extract_csf;

pub use medi::medi;

pub use medi::EchoMagnitudes;

pub use medi::MediParams;

pub use msmv::msmv;

pub use msmv::MsmvOutput;

pub use msmv::MsmvParams;

pub use tfi::tfi;

pub use tfi::TfiKernel;

pub use tfi::TfiParams;

pub use tfi::TfiPrecision;

pub use tgv::tgv_qsm;

pub use tgv::TgvInput;

pub use tgv::TgvParams;

pub use tgv::get_default_alpha;

pub use tgv::get_default_iterations;

pub use tgv::rad_per_ppm;

pub use ilsqr::ilsqr;

pub use ilsqr::IlsqrParams;

pub use ilsqr::IlsqrPrecision;

pub use ilsqr::STI_DEFAULT_NITER;

pub use ilsqr::STI_DEFAULT_PAD_MM;

pub use ilsqr::STI_GUI_NITER;

pub use ilsqr::STI_GUI_PAD_MM;

pub use ilsqr::STI_PAD_MM;

pub use star::star_qsm;

pub use star::StarParams;

pub use star::STAR_GUI_PAD;

pub use star::QSM_STAR_DEFAULT_PAD;

pub use star::STI_TEMPLATES_STAR_PAD;

pub use lsqr_qsm::lsqr_qsm;

pub use lsqr_qsm::LsqrQsmParams;

pub use lsqr_qsm::RESIDUAL_WEIGHTING_PER_TESLA;

pub use heidi::cone_fill_iterations;

pub use heidi::heidi;

pub use heidi::heidi_cone_fill;

pub use heidi::homogeneity_masks;

pub use heidi::HeidiParams;

pub use heidi::HomogeneityMasks;

pub use admm::FansiKernel;

pub use ndi::ndi;

pub use ndi::NdiParams;

pub use fansi::fansi;

pub use fansi::FansiParams;

pub use l1qsm::l1qsm;

pub use l1qsm::L1QsmParams;

pub use whqsm::whqsm;

pub use whqsm::WhQsmParams;

pub use whqsm::WhQsmWeight;

pub use hdqsm::hdqsm;

pub use hdqsm::hdqsm_magnitude;

pub use hdqsm::HdQsmParams;

pub use amp_pe::amp_pe;

pub use amp_pe::AmpPeParams;

pub use tile_budget::max_wasm_patch_edge;

pub use tile_budget::tile_bytes;

pub use tile_budget::tile_patch_size;

pub use tile_budget::tiles_within_wasm_budget;

pub use tile_budget::wasm_tiles_in_flight;

pub use tile_budget::TileConfig;

pub use tile_budget::WASM_MAX_PATCH_ACTIVATION_BYTES;

pub use tile_budget::WASM_TILE_ACTIVATION_BUDGET;

pub use tiled::check_wasm_patch_fits;

pub use tiled::tile_concurrency;

pub use tiled::tiled_field_inversion;

pub use tiled::tiled_scatter;

pub use tiled::tiled_volume_algorithm;

pub use tiled::Tile;

pub use xqsm::xqsm;

pub use xqsm::xqsm_tiled;

pub use qsmnet::qsmnet;

pub use qsmnet::qsmnet_tiled;

pub use qsmnet::QsmnetNorm;

pub use qsmnetme::QsmnetmeNorm;

pub use qsmnetme::QsmnetmeParams;

pub use qsmnetme::qsmnetme;

pub use autoqsm::autoqsm;

pub use iqsm::iqsm;

pub use iqsm::iqsm_multi_echo;

pub use iqsm_plus::iqsm_plus;

pub use iqsm_plus::iqsm_plus_multi_echo;

pub use iqfm::iqfm;

pub use iqfm::iqfm_multi_echo;

pub use qsmgan::qsmgan;

pub use lpcnn::lpcnn;

pub use lpcnn::lpcnn_tiled;

pub use lpcnn::LPCNN_ALPHA;

pub use lpcnn::LPCNN_GT_MEAN;

pub use lpcnn::LPCNN_GT_STD;

pub use ir2qsm::ir2qsm;

pub use ir2qsm::ir2qsm_tiled;

pub use modl_qsm::modl_qsm;

pub use modl_qsm::modl_qsm_tiled;

pub use modl_qsm::MODL_ALPHA;

pub use modl_qsm::MODL_MEAN;

pub use modl_qsm::MODL_STD;

pub use nextqsm::nextqsm;

pub use nextqsm::nextqsm_padded;

pub use nextqsm::nextqsm_tiled;

pub use nextqsm::NEXTQSM_LAMBDAS;

Modules

admm

Shared ADMM iteration infrastructure for TV-based inversion methods.

amp_pe

AMP-PE: Approximate Message Passing with built-in Parameter Estimation for QSM.

autoqsm

AutoQSM single-step deep-learning reconstruction (onnx feature).

cosmos

COSMOS: calculation of susceptibility through multiple orientation sampling

fansi

FANSI nonlinear TV / TGV dipole inversion.

hdqsm

HD-QSM: Hybrid data-fidelity two-stage linear dipole inversion.

heidi

HEIDI — Homogeneity Enabled Incremental Dipole Inversion

ilsqr

iLSQR: LSQR dipole inversion with streaking-artefact removal

iqfm

iQFM single-step deep-learning tissue-field mapping (onnx feature).

iqsm

iQSM single-step deep-learning reconstruction (onnx feature).

iqsm_echo_weighting

Echo weighting for the multi-echo iQSM, iQFM and iQSM+ combinations.

iqsm_plus

iQSM+ single-step deep-learning reconstruction (onnx feature).

ir2qsm

IR2QSM dipole inversion (onnx feature).

l1qsm

L1-QSM: nonlinear L1 data-fidelity dipole inversion with TV regularization

lpcnn

LPCNN dipole inversion (onnx feature).

lsqr_qsm

Minimally regularised LSQR dipole inversion (Schweser 2010)

medi

MEDI (Morphology Enabled Dipole Inversion), a port of MEDI_L1.m from the MEDI toolbox (release 2024.11.26)

modl_qsm

MoDL-QSM dipole inversion (onnx feature).

msmv

mSMV (maximum spherical mean value) filtering, the field step of MEDI-mSMV

ndi

Nonlinear Dipole Inversion (NDI) for QSM

nextqsm

NeXtQSM single-step deep-learning reconstruction (onnx feature).

nltv

Nonlinear total variation (NLTV) dipole inversion

qsmgan

QSMGAN dipole inversion (onnx feature).

qsmnet

QSMnet deep-learning dipole inversion (onnx feature).

qsmnetme

QSMnetME deep-learning dipole inversion (onnx feature).

rts

Rapid Two-Step (RTS) dipole inversion

star

STAR-QSM: streaking artifact reduction for QSM

sti

STI: susceptibility tensor imaging

tfi

TFI (preconditioned Total Field Inversion)

tgv

TGV-QSM: Total Generalized Variation for Quantitative Susceptibility Mapping

tikhonov

Tikhonov regularization for QSM

tile_budget

How much heap a tiled network’s tiles take, and how many a browser keeps in flight.

tiled

Overlap-tiled inference for fully-convolutional deep-learning inversions (onnx).

tkd

Truncated k-space division (TKD) / TSVD for QSM

tv

Total Variation (TV) regularized dipole inversion using ADMM

whqsm

Weak-Harmonic QSM (WH-QSM) dipole inversion

xqsm

xQSM deep-learning dipole inversion (onnx feature).