Steady-state ihMT sequences using multiband pulses

This software can be used for simulating the steady-state of bSSFP and SPGR MRI sequences when employing multi-band pulses in tissues with inhomogeneous MT effects and builds on prior work using non-selective multi-band RF pulses for controlling saturation of MT effects.

A paper describing the theory and experiments implemented in this repo is currently under review and links will be posted as soon as it is published. A related abstract was presented at ISMRM 2019 (#427)

This code is distributed under the GNU General Public License v3 (see the included LICENSE). Please cite our work appropriately if you use it and link any extensions back to this repo

This code has been tested using Matlab R2018b. Please get in touch to report any bugs.

Shaihan Malik, King's College London, June 2019. @shaihanmalik

Biophysical model

The code implements the model described by the following diagram:

Tissue consists of a main compartment of free water, and two semisolid compartments s1 and s2 that are both in contact with the free water pool but not each other. Compartment s1 consists only of Zeeman ordered magnetization, while s2 contains both Zeeman and Dipolar ordered terms. The relative fractions of these are f and (1-f) respectively.

Tissue properties are contained within a struct tissuepars which is constructed as follows:

tissuepars.
			free.
				R1 <----- Free pool longitudinal relaxation rate R1 (s^-1)
				R2 <----- Free pool transverse relaxation rate R2 (s^-1)
			semi.
				M0 <----- Semisolid fraction of M0 (i.e. M0s in paper)
				R1 <----- Semisolid Zeeman longitudinal relaxation rate (R1Z, s^-1)
				R1D<----- Semisolid Dipolar longitudinal relaxation rate (R1D, s^-1)
				T2 <----- Semisolid Transverse relaxation TIME (s)
				f  <----- fraction of semisolid pool in compartment s2
			k	   <----- exchange rate between free and semisolid pools (directionless; s^-1)
			lineshape <-- Flag that decides lineshape model, either 'SL' (Super Lorentzian) or 'Gaussian'

† The symbol 'f' will be replaced by 'δ' in a future revision

Tissues can be initialised by calling function init_tissue:

• init_tissue.m

  Takes a string argument, current options:

  • 'ic' (internal capsule model from Mchinda et al 2017)
  • 'pl161' prolipid 161, using data from paper
  • 'simple' simplified rounded parameter values, similar to human brain

The tissuepars struct contains a flag for the lineshape. The lineshapes are implemented by two separate functions SuperLorentzian_lineshape or gauss_lineshape.

• SuperLorentzian_lineshape.m

    [G,w_loc] = SuperLorentzian_lineshape(T2s,fsample,varargin)
    
    INPUTS:     T2s = T2 of semisolid compartment, seconds
                fsample   = frequencies at which function is to be evaluated (can
                be vector)
                optional: 'interpzero' - interpolate between ±1.5kHz as per
                Gloor et al 2008, to remove high peak at f=0.
    
    OUTPUTS:    G   = lineshape value, s
                w_loc = local field term (rad/s)
  

The 'interpzero' flag is used to interpolate across the zero frequency term which has a singularity. This approach was proposed by Bieri and Scheffler and also used by Gloor et al 2008

• gauss_lineshape.m

     [g,w_loc] = gauss_lineshape(T2s,f)
     
     INPUTS:       T2s = T2 of semisolid compartment, seconds
                   f   = frequency at which function is to be evaluated (can
                   be vector)
     OUTPUTS:      g   = lineshape value, s
                   w_loc = local field term (rad/s)
  

RF pulses

The work that this code relates to uses multiband RF pulses for simultaneous saturation and excitation. These pulses may be designed to have equal RF power that is distributed differently (as needed) over one, two, or three different frequency bands with offset frequencies Δ.

The illustration shows three different pulses all with the same total power. These can be generated using gen_MB_pulse.m:

• gen_MB_pulse.m

  This function will generate multiband pulses which produce a required on-resonance flip angle (θ) for a specified pulse duration τ and TR that has a given specific total RMS B1. As in the diagram above, these could either have 2 or 3 bands (single or dual off-resonance frequency). The 2-band pulses can have either a positive or negative offset; the names of the pulses are therefore 2+B, 2-B, 3B.

  [pulseMB, b1sqrd, pulse_per_band,TBP] = gen_MB_pulse(theta,tau,TR,b1rms_total,delta,nband,varargin)
  
  arguments:
  -  theta:         on-resonance flip angle (rad)
  -  tau:           pulse duration (s)
  -  TR:            repetition time (s)
  -  b1rms_total:   overall RMS B1 of the whole sequence (uT)
  -  delta:         offset frequency for off-resonant bands (Hz) 
  -  nband:         string argument for number of bands. Options are '2+', '2-', or '3'.
  
  optional arguments:
  -  alpha:         scalar value determining width of Gaussian envelope for pulse (default is 3)
  -  dt:            RF sampling duration. Default 6.4us
  

Simulation code

The main simulation functions are contained within the src folder. Details of the implementations are given in an upcoming paper, link to be added when available. Most functions assume that there are 3 frequency bands that need to be considered though some may have zero contribution. This could be readily extended if necessary.

• ssSSFP_ihMT.m - Direct simulation of steady state for balanced SSFP sequence under instantaneous RF pulse approximation (i.e. relaxation and exchange are neglected during RF pulses, pulses are applied instantaneously).

  function [Mss,Mz] = ssSSFP_ihMT(flipangle,b1sqrd,Delta_Hz,TR,tau,dphi, tissuepars,varargin)
  
  Steady-state ihMT bSSFP sequence. For non-selective multiband sequences
  
  INPUTS:         
     flipangle  = flip angle on resonance (rad)
     b1sqrd     = mean square B1+ per frequency band of
                  multiband pulse (over the duration of the pulse). 
                  1x3 vector (so far we assume 3 bands maximum but 
                  this could be changed). Units uT^2
     Delta_Hz   = 1x3 vector of frequencies of each band. Typically 
                  [-delta 0 delta]. Units Hz
     TR         = repetition time, sec
     tau        = pulse duration, sec
     dphi       = Off-resonance phase gained per TR, unit=radians
     tissuepars = structure containing all tissue parameters. See
                  init_tissue()
  OUTPUTS:
     Mss        = Steady-state Mxy (after excitation pulse)
     Mz         = Longitudinal magnetization, including semisolid
                  terms
  

This code will directly compute the steady-state for bSSFP. The argument b1sqrd is the mean square B1 in each frequency band over the pulse duration (not TR). This is output by the gen_MB_pulse function. The tissue properties are passed to this function using the struct that is initialised by init_tissue. The absorption line is computed within this function, using the flag tissuepars.lineshape to decide which function to use. If the Super-Lorentzian is used then the interpzero option is also used here.

• ssSPGR_ihMT.m - Identical functionality to ssSSFP_ihMT.m except for SPGR sequence, the only difference is that the argument dphi is not needed. Perfect spoiling of transverse components is assumed.

• ssSSFP_ihMT_integrate.m - Integration method for direct computation of steady-state for bSSFP without assuming instantaneous pulses. Uses eigenvector based method for steady-state computation (see paper when available).

      function [Mss,Mz] = ssSSFP_ihMT_integrate(b1pulse,dt,Delta_Hz,TR,dphi, tissuepars)
  
     Steady-state ihMT bSSFP sequence with eigenvector based time integration method
  
  INPUTS:         
      b1pulse    = RF pulse, Mx3 array (M=#timepoints, 3=frequency
                   bands). Units are uT
      dt         = dwell time, sec
      Delta_Hz   = 1x3 vector of frequencies of each band. Typically 
                   [-delta 0 delta]. Units Hz
      TR         = repetition time, sec
      dphi       = off-resonance phase per TR (rad)
      tissuepars = structure containing all tissue parameters. See
                  init_tissue()
  
  OUTPUTS:
      Mss        = Steady-state Mxy (after excitation pulse)
      Mz         = Longitudinal magnetization, including semisolid
                          terms
  

Note that unlike ssSSFP_ihMT.m this function takes the sampled RF waveform, decomposed into separate frequency bands in the time domain. The flip angle and b1rms therefore do not need to be defined.

• ssSPGR_ihMT_integrate.m - Identical functionality to ssSSFP_ihMT_integrate.m except for SPGR sequence, the only difference is that the argument dphi is not needed. Perfect spoiling of transverse components is assumed.

• Bieri_Scheffler_finite_correction.m - implements finite RF pulse duration correction to R2 of the free water pool for bSSFP as outlined in this paper. We assume pulses have a Gaussian envelope as generated by gen_MB_pulse - the user supplies the time-bandwidth-product (TBP) needed for this correction. Takes a tissuepars struct and returns a copy with modified tissuepars.free.R2.

Experiments

The scripts in the folder experiments may be used to replicate certain experiments. Make sure the lib and src subfolders are on the path before executing these.

Compare ihMT effect for different sequences in white matter

Uses white_matter_ihMT_comparisons.m. This script performs the following tests:

• Simulate expected signal curves for 1B, 2B and 3B pulses in WM for case where f=0 and f=1, for SPGR and bSSFP
• Display ihMTR and MTR for the same sequences
• Explore ΔihMT and ihMTR for bSSFP with variable Δ, b1rms, θ
• Investigate limits for pulse design using gen_MB_pulse within hardware and safety constraints

Investigate instantaneous pulse approximation

Uses instantaneous_pulse_approx_script.m. This script compares simulations using the instantaneous approximation (i.e. ssSSFP_ihMT and ssSPGR_ihMT) with temporal integration including the shaped RF waveforms (i.e. ssSSFP_ihMT_integrate and ssSPGR_ihMT_integrate respectively). Comparisons are made for different dipolar relaxation times (R1D) and pulse durations, and it is shown that the Bieri Scheffler correction is enough to correct the instantaneous approximation results in most cases. The script also compares the eigenvector based method for finding the steady-state used in ssSSFP_ihMT_integrate and ssSPGR_ihMT_integrate with time-integration over multiple TR periods to reach the steady-state.

Our proposed time integration method is unlike other approaches which simply run a Bloch (or Bloch-McConnell) simulation over many TR periods in order to reach the steady-state; instead we use an eigenvalue approach that directly computes the steady-state after integrating over one TR period. In this script we have also run a simple comparison with this method which is referred to as 'MAMT' after this paper by Portnoy and Stanisz. This result was not included in the paper. The comparison shows that the relative benefit of the proposed direct integration method scales linearly with the number of TR periods that are simulated.

For the white matter example shown here the required number of TR periods to reach very low error is approximately 400 and for this length of simulation the relative speed-up from the eigenvalue method is a factor of approximately 300.

Experimental data fitting

Experiments were performed using SPGR and bSSFP with a range of flip angles on a sample consisting of:

• Water doped win MnCl2, not expected to have MT or ihMT effect
• Bovine Serum Albumin, expected to have MT but not ihMT contrast
• Prolipid 161, used as a test substance for ihMT effects by Swanson et al
• Hair conditioner, also shown to have ihMT effect (e.g. by Varma et al)

Phantom data are available to download from the release page, please place the matrix fitdata.mat into the bin folder. This script will performed constrained fitting to this data using the above listed signal functions and perform error analysis using residual bootstrapping.

Other scripts

• MB_figure_script.m generates the figure with the example RF waveforms used above
• single_pool_comparison_noMT.m compares the signal functions with Ernst and Freeman-Hill equations for the case of no semisolid compartment