Debackgrounding

Having performed phase unwrapping you are now ready for the second stage of QSM: debackgrounding. In QSM, debackgrounding is necessary because the total magnetic field shift (TFS) is dominated by strong, unwanted background fields caused by the scanner's main magnet imperfections and tissue-air susceptibility interfaces. These background fields are often orders of magnitude larger than the weak local fields generated by target tissues (such as iron or myelin), making it impossible to accurately map local tissue susceptibility without first stripping them away.

Jim uses the Resharp method for debackgrounding. The settings for this stage are:

• Set the kernel radius, in mm. In the RESHARP method, the kernel radius setting controls the size of the spherical mean value (SMV) convolution kernel used to estimate the background magnetic field. This is trade-off between tissue masking area and frequency filtering accuracy:

  • A larger radius better filters out low-frequency background fields, but results in a larger eroded margin at the surface. This means you may lose valuable data near the boundaries.

  • A smaller radius preserves more tissue near the boundaries by minimising mask erosion, but can lead to poorer estimation of the background field and residual artifacts.

• Set the regularisation parameter that controls the balance between smoothing out residual background artifacts and preserving the true local magnetic field variations. Because the spherical mean value (SMV) deconvolution step is mathematically ill-posed, it contains "zero-crossings" in the frequency domain that wildly amplify noise. The RESHARP method introduces Tikhonov regularisation to stabilise this process.

  The regularisation parameter value sets the balance between immunity to noise and data fidelity:

  • Too large (over-regularisation): the algorithm over-smooths the local phase image. This dampens the peak values of your local tissue fields, leading to an underestimation of true susceptibility values in high-contrast structures.

  • Too small (under-regularisation): the algorithm fails to suppress the ill-posed frequencies. This allows noise and streaking artifacts to leak back into the local field map, which will corrupt your final QSM inversion.

You may need to experiment with the above settings for best results with your data.

Now click the to button start the debackgrounding.

The output from this debackgrounding stage is a new image with name taken from the input image (or first input image, if you multiple input images), but with a suffix _LFS appended to the name. This is an image of the Local Field Shift with the gross background fields removed.

One slice of the local field shift (LFS) image produced after de-backgrounding

You can now go on to the next stage, which is dipole inversion.