Skip to content

MRICoilSensitivities ​

MRICoilSensitivities estimates coil sensitivity maps, for example with ESPIRiT, and compresses multi-channel data to fewer virtual coils. MRICoilSensitivities depends on MRIBase. It has to be loaded with using MRICoilSensitivities, because MRIReco does not depend on it. The guide Estimate Coil Sensitivities with ESPIRiT shows how to use espirit, and Compress Multi-Coil Data how to reduce the number of coils.

Coil Sensitivities ​

MRICoilSensitivities.espirit Method
julia
espirit(acqData::AcquisitionData, ksize::NTuple{D,Int} = (6,6[,6]), ncalib::Int = 24,
           imsize::NTuple{D,Int}=Tuple(acqData.encodingSize[1:D])
           ; eigThresh_1::Number=0.02, eigThresh_2::Number=0.95, nmaps = 1) where D

espirit(calibData::Array{T}, imsize::NTuple{D,Int}, ksize::NTuple{S,Int} = (6,6[,6])
           ; eigThresh_1::Number = 0.02, eigThresh_2::Number = 0.95, nmaps = 1) where D

Obtains coil sensitivities from a calibration area using ESPIRiT. Works with 2D and 3D data.

The code is adapted from the MATLAB code by Uecker et al. (cf. Uecker et al. "ESPIRiT—an eigenvalue approach to autocalibrating parallel MRI: Where SENSE meets GRAPPA"). The matlab code can be found at: [http://people.eecs.berkeley.edu/~mlustig/Software.html]

Required Arguments

  • acqData::AcquisitionData - AcquisitionData

or

  • calibData::Array{T} - Center of k-space in the format kx × ky [× kz] × coils, where the kz dimension is optional. The type T of the input data determines the type of the calculated maps. Reasonable choices are in the range of Nx = Ny [= Nz] = 24.

  • imsize::NTuple{D,Int} - matrix size of the final maps (optional when passing acqData)

Optional Arguments

  • ksize::NTuple{D,Int} - size of the kernel in each dimension; the default is (6,6) for 2D and (6,6,6) for 3D.

  • ncalib::Int - size of the central part corresponding to the calibration data.

Keyword Arguments

  • eigThresh_1::Number=0.02 - threshold for the singular values of the calibration matrix (relative to the largest value); reduce for more accuracy, increase for saving memory and computation time.

  • eigThresh_2::Number=0.95 - threshold to mask the final maps: for each voxel, the map will be set to 0, if, for this voxel, no singular value > eigThresh_2 exists.

  • nmaps = 1 - Number of maps that are calculated. Set to 1 for regular SENSE; set to 2 for soft-SENSE (cf. Uecker et al. "ESPIRiT—an eigenvalue approach to autocalibrating parallel MRI: Where SENSE meets GRAPPA").

  • use_poweriterations = true - flag to determine if power iterations are used; power iterations are only used if nmaps == 1. They provide speed benefits over the full eigen decomposition, but are an approximation.

  • echo::Int = 1 - select the k-space corresponding to the echo

  • rep::Int = 1 - select the k-space corresponding to the rep

source
MRICoilSensitivities.estimateCoilSensitivities Method
julia
s = estimateCoilSensitivities(I::AbstractArray{T,6}, thresh = 1.e-2)

Estimates the coil sensitivity based on a reconstruction where the data from each coil has been reconstructed individually. Returns a 6D array of the size of I, which is zero where the root sum of squares over the coils is not above thresh times its maximum.

source
MRICoilSensitivities.mergeChannels Method
julia
I4 = mergeChannels(I::AbstractArray{T,6})

Merge the channels of a multi-coil reconstruction. Returns a 6D array.

source

Coil Compression ​

MRICoilSensitivities.applyCoilCompression Method
julia
applyCoilCompression(acqData::AcquisitionData, ccMat; dim = 1)

Apply the compression matrix ccMat to all contrasts, slices and repetitions of acqData.

ccMat can be

  • a matrix (channels × virtual coils), used for every slice,

  • a vector with one matrix per slice, as returned by softwareCoilCompression,

  • a 3D array (positions × channels × virtual coils), as returned by geometricCoilCompression. This requires Cartesian data, with dim fully sampled.

Return a new AcquisitionData. The sampling pattern, the trajectories and the other fields are copied from acqData.

source
MRICoilSensitivities.applyCoilCompression Method
julia
applyCoilCompression(kspace::Array{T,6}, ccMat; dim = 1) where T <: Complex

Apply the compression matrix ccMat to the Cartesian kspace (x, y, z, channels, contrasts, repetitions), as returned by kDataCart, for all contrasts and repetitions.

ccMat can be

  • a matrix (channels × virtual coils), as returned by softwareCoilCompression for an array,

  • a vector with one matrix per slice (dimension 3), as returned by softwareCoilCompression for an AcquisitionData,

  • a 3D array (positions × channels × virtual coils), as returned by geometricCoilCompression. The k-space is then inverse Fourier transformed along dim, compressed position by position and transformed back.

Return the compressed k-space.

source
MRICoilSensitivities.applyCoilCompressionSensitivityMaps Method
julia
applyCoilCompressionSensitivityMaps(smaps::AbstractArray{T,4}, ccMat; dim = 1) where T <: Complex

Compress the sensitivity maps smaps (x, y, z, channels) with the compression matrix ccMat used for the data, so that they are the sensitivities of the virtual coils:   .

ccMat can be a matrix, a vector with one matrix per slice (dimension 3 of smaps) or a 3D array from geometricCoilCompression, see applyCoilCompression. For geometric coil compression, the matrices are applied position by position along dim; no Fourier transform is needed because the maps are already in image space.

source
MRICoilSensitivities.geometricCoilCompression Method
julia
geometricCoilCompression(acqData::AcquisitionData, numVC::Int = numChannels(acqData); dim = 1, sContr = 1, sRep = 1)

Compress the Cartesian acqData to numVC virtual coils with geometric coil compression. See the method for arrays for the meaning of dim, sContr and sRep.

The compression matrices are computed from the k-space returned by kDataCart and shared by all slices. For 2D data, dim must be 1 or 2. The result keeps the sampling pattern, the trajectories and the other fields of acqData, so undersampled data stay undersampled. Return (acqDataCC, ccMat), where ccMat has size positions × channels × numVC.

source
MRICoilSensitivities.geometricCoilCompression Method
julia
geometricCoilCompression(kspace::Array{T,6}, numVC::Int = size(kspace, 4); dim = 1, sContr = 1, sRep = 1) where T <: Complex

Compress the Cartesian kspace (x, y, z, channels, contrasts, repetitions), as returned by kDataCart, to numVC virtual coils with geometric coil compression.

Dimension dim (1, 2 or 3) must be fully sampled, typically the readout direction. The k-space is inverse Fourier transformed along dim, and one compression matrix is computed for each position along dim from contrast sContr and repetition sRep. The matrices are aligned so that the virtual coils vary smoothly between positions, and applied to all contrasts and repetitions with applyCoilCompression.

Return (kspaceCC, ccMat), where ccMat has size positions × channels × numVC.

Reference

  • Zhang T, Pauly JM, Vasanawala SS, Lustig M. Coil compression for accelerated imaging with Cartesian sampling. Magnetic Resonance in Medicine 2013;69:571–582. doi: 10.1002/mrm.24267
source
MRICoilSensitivities.softwareCoilCompression Method
julia
softwareCoilCompression(acqData::AcquisitionData, numVC::Int = numChannels(acqData); sContr = 1, sRep = 1)

Compress acqData to numVC virtual coils. Works for any trajectory and for undersampled data.

One compression matrix is computed per slice, from contrast sContr and repetition sRep, and applied to all contrasts and repetitions of that slice with applyCoilCompression. Return (acqDataCC, ccMat), where ccMat is a vector with one matrix (channels × numVC) per slice. It can be passed to applyCoilCompressionSensitivityMaps to compress sensitivity maps.

Reference

  • Huang F, Vijayakumar S, Li Y, Hertel S, Duensing GR. A software channel compression technique for faster reconstruction with many channels. Magnetic Resonance Imaging 2008;26:133–141. doi: 10.1016/j.mri.2007.04.010
source
MRICoilSensitivities.softwareCoilCompression Method
julia
softwareCoilCompression(kspace::Array{T,6}, numVC::Int = size(kspace, 4); sContr = 1, sRep = 1) where T <: Complex

Compress the Cartesian kspace (x, y, z, channels, contrasts, repetitions), as returned by kDataCart, to numVC virtual coils.

A single compression matrix is computed from contrast sContr and repetition sRep and applied to all contrasts and repetitions with applyCoilCompression. Return (kspaceCC, ccMat), with ccMat of size channels × numVC.

Reference

  • Huang F, Vijayakumar S, Li Y, Hertel S, Duensing GR. A software channel compression technique for faster reconstruction with many channels. Magnetic Resonance Imaging 2008;26:133–141. doi: 10.1016/j.mri.2007.04.010
source
MRICoilSensitivities.softwareCoilCompression Method
julia
softwareCoilCompression(kdata::Matrix{T}, numVC::Int = size(kdata, 2)) where T <: Complex

Compress the k-space matrix kdata (samples × channels) to numVC virtual coils.

The compression matrix ccMat (channels × numVC) consists of the right singular vectors of kdata that belong to the numVC largest singular values. Return (kdataCC, ccMat) with kdataCC = kdata * ccMat (samples × numVC).

Reference

  • Huang F, Vijayakumar S, Li Y, Hertel S, Duensing GR. A software channel compression technique for faster reconstruction with many channels. Magnetic Resonance Imaging 2008;26:133–141. doi: 10.1016/j.mri.2007.04.010
source