Settings

The Fat/Water Separation uses both the magnitude and the phase information from the input images to fit a model that represents the different proportions of fat and water in each pixel. The model also includes the spatial variation in magnetic field strength (inhomogeneity), which is assumed to have a smooth variation across the image.

The processing is in 3 stages:

  1. Estimation of the residue function (the squared error between a model fit and the data) for every image voxel, for a range of B0 inhomogeneity offset frequencies (the VARPRO method) and the optimal T2* value at each frequency offset.
  2. Iterative search for the optimal smooth B0 inhomogeneity 'surface' that fits the data within smoothness constraints.
  3. Having found the optimal estimate of the B0 inhomogeneity, then the VARPRO method is used again with a narrow range of B0 inhomogeneity related frequency offsets, and a more closely-spaced set of T2* values. With default settings, the frequency resolution is approximately 1Hz, and the T2* resolution is about 1ms. Reducing the ΔB search range and T2* search range will result in better resolution in frequency and T2*.

The tool has the following settings:

  • Main magnetic field strength. The program needs to know the field strength in order to correctly model the chemical shift dispersion. This may be recorded in the image header. If so, you can select Check-box to try to get the B0 field strength from the image header. You will see an error message if the field strength can't be found in the header. If it is not in the image header, or can't be found by the program, you need to deselect this checkbox and enter the field strength manually: Setting the B0 field strength manually.

  • Echo times. If the echo times are recorded in the image header, then they will be read if you select Check-box to try to get the echo times from the image header. You will see an error message if the echo times can't be found in the header. If not present, or can't be found, then enter the echo times, in milliseconds, in the text box as a comma-separated list, such as: Entering the echo times manually.

  • ΔB search range. The amount (in parts per million) by which the B0 field strength is allowed to vary in the search for the B0 inhomogeneity. Use smaller values (e.g. 1.5 PPM) if you know that the field is pretty uniform over the subject.

  • T2* search range minimum/maximum. The fitting involves a search for the 'best' T2* value that fits the data. If you know the range of actual T2* values for your data, you can set this here. If you do not want T2* to be estimated, but instead to be a fixed value then set both the minimum and maximum T2* to this fixed value, and T2* will not be estimated. Fixing the T2* value is useful when your data is not of sufficient quality (small number of echoes, or not enough echo time range) to estimate T2*. In this case, just set the min and max to a plausible T2* value (e.g. 200 ms).

  • Fat frequency spectrum. The tool has a built-in default fat frequency spectrum, with 6 peaks, as shown in the table below:

    The default fat spectrum.
    Frequency (PPM) Relative Intensity
    0.8996 0.0870
    1.2997 0.6930
    2.1007 0.1280
    2.7598 0.0040
    4.3227 0.0390
    5.3001 0.0480

    This spectrum is used if you select Check-box to use the default fat spectrum. If you want to use a different fat spectrum, you can specify it in a text file with two columns of data. In the first column are the chemical shifts (referenced to TMS) and in the second column are the relative intensities of the fat peaks. There should be some whitespace or a comma between the two columns. Select the text file after unchecking the Check-box to use the default fat spectrum checkbox.

    Selecting a text file containing a custom fat spectrum

  • B0 field smoothness. This affects the constraints on the algorithm that is used to estimate the B0 inhomogeneity. If you select 'Lowest' this allows larger spatial variations in the B0 field, with lower variations for 'Medium', and lower still if you select 'Highest' smoothness. You should only need to alter this if the program fails to follow large changes in B0 (in which case select 'Lowest') or if the data quality is poor, but you know that the field has fairly smooth variations (select 'Highest').

  • Max. nodes in B0 inhomogeneity search. The B0 inhomogeneity is estimated at a set of "nodes" that cover the whole field-of-view of the image. As the number of nodes increases, it can take a very long time to converge on the best estimate of B0. In order to make the algorithm run in an acceptable time, the number of nodes is reduced by down sampling the input image. This option sets the maximum number of nodes.

    If you increase it, the program will take longer to run, but may give better results when there are large B0 inhomogeneities. Note, however, that the final search for the B0 inhomogeneity (done in stage 3 of the algorithm) always runs at full spatial resolution, using the output of the of the initialisation from the search over the nodes.

  • Process slice-by-slice rather than in 3-D. Selecting this will process each slice independent of the others. For images acquired with 3-D pulse sequence, you would not normally do this, because to do so would discard the information about 3-D smoothness of the B0 inhomogeneity. However, this option can be useful when your block of slices was not acquired in 3-D, but in multi-slice 2-D, possibly with gaps between the slices, or when slices are not parallel. Then, the assumption about B0 smoothness does not hold. If you are constrained by computer memory, processing slice-by-slice also uses less memory.

  • Perform hybrid complex/magnitude fitting. When selected, this applies, as a final step, the hybrid fitting described in this reference. This may be useful where phase errors occur, possibly as a result of eddy-currents.

  • Reverse fat/water assignments. Occasionally, if the scanner has set the transmitter reference frequency to fat instead of water, the fat and water are assigned in reverse. Selecting this option will reverse them back again. Mis-assignment can also occur if the phases evolve in your images in the opposite direction to that expected (positive phase change for positive relative frequencies).

  • Zero the image background. Selecting this option will apply a simple intensity threshold which will remove most of the background noise in the output images.