Configuration file settings

The EPTlib application is configured by a .toml file.

Title and annotations
Method
Mesh
Input
1. Wildcard characters
Output
Postprocessing
Parameters

In .toml files, the symbol # marks the rest of the line as a comment, except when inside a string.

Title and annotations

title = "Example"
description = "Just an example of configuration file"

The title and the description will appear in the log of the EPTlib application.

Method

method = 0

The method is selected according to the following table.

Code	Method
0	Helmholtz EPT
1	Convection-reaction EPT
2	Gradient EPT
100	Helmholtz EPT with automatically selected kernel

Starting from the code 100, the implemented methods are still experimental.

Mesh

[mesh]
    size = [128,128,8]
    step = [1e-3,1e-3,4e-3] # [m]

size is the number of voxels of the input data in each direction.
step is the size of a voxel in meters.

Take care that Python h5py package stores the data in row-major order. If you are using Python h5py to create or inspect the input HDF5 datasets, the size to be reported in the configuration file is the shape of the dataset ordered from the last to the first dimension. It can be obtained like reported in the tutorial, where
    nn = tx_sens.shape
    size = nn[::-1]

Input

[input]
    frequency = 64e6 # [Hz]
    tx-channels = 2
    rx-channels = 4
    tx-sensitivity = "phantom.h5:/tx_sens>"
    trx-phase = "phantom.h5:/trx_phase><" # [rad]
    wrapped-phase = false
    reference-image = "phantom.h5:/ref_img"

frequency is the Larmor frequency of the input data in hertz.
tx-channels is the number of transmit channels in the input data.
rx-channels is the number of receive channels in the input data.
tx-sensitivity is the address of the transmit (Tx) sensitivity (magnitude). It must be a dataset in an .h5 file.
trx-phase is the address in an .h5 file of the transceive (TRx) phase in radians. It must be a dataset in an .h5 file. Alternatively, it is possible to provide the address of the transmit (Tx) phase in radians with the option tx-phase. It is not possible to define both trx-phase and tx-phase in the same configuration file.
wrapped-phase is equal to true if the input transceive phase maps are wrapped. Currently, only the phase-based methods take advantage of it (default: false).
reference-image is the address in an .h5 file of a reference image that can be used by the EPT methods and the postprocessing. It must be a dataset in an .h5 file. If it is an empty string, all the methods will be applied with no reference image (default: empty string).

For some EPT methods, the tx-sensitivity or the trx-phase (or tx-phase) could be optional.

In case of multiple transmit channels, the wildcard character > is used in the tx-sensitivity address in order to distinguish the data from each channel. It will be replaced by an increasing integer number from zero up to the number of transmit channels minus one.

Similarly, multiple receive channels are distinguished in the trx-phase (or tx-phase) address by means of the wildcard character <, which will be replaced by an increasing integer number from zero up to the number of receive channels minus one.

For example, with the previous input settings the transmit sensitivity data would be read from
"phantom.h5:/tx_sens0"
"phantom.h5:/tx_sens1"
whereas the transceive phase data would come from
"phantom.h5:/trx_phase00" "phantom.h5:/trx_phase10"
"phantom.h5:/trx_phase01" "phantom.h5:/trx_phase11"
"phantom.h5:/trx_phase02" "phantom.h5:/trx_phase12"
"phantom.h5:/trx_phase03" "phantom.h5:/trx_phase13"

Wildcard characters

It is possible to change the default wildcard characters, the starting index and the increasing step with the following optional instructions in the .toml configuration file.

[input.wildcard]
    tx-character = '>'
    rx-character = '<'
    start-from = 0
    step = 1

Output

[output]
    electric-conductivity = "example.h5:/sigma" # [S/m]
    relative-permittivity = "example.h5:/epsr"

electric-conductivity is the address where the electric conductivity, expressed in siemens per meter, will be written. It must be a dataset in an .h5 file.
relative-permittivity is the address where the relative permittivity will be written. It must be a dataset in an .h5 file.

If an output address is not provided in the configuration file, the corresponding properties evaluation will not be stored.

Postprocessing

The following set of instruction is optional. It can be used to apply a filter in postprocessing to the EPT results.

[postprocessing]
    output-electric-conductivity = "example.h5:/sigma-postpro"
    output-relative-permittivity = "example.h5:/epsr-postpro"
    perform-only-postprocessing = false
    input-electric-conductivity = ""
    input-relative-permittivity = ""
    input-reference-image = ""
    input-electric-conductivity-variance-map = ""
    input-relative-permittivity-variance-map = ""
    kernel-size = [5, 5, 5]
    kernel-shape = 1
    weight-param = 0.10

output-electric-conductivity is the address where the postprocessed electric conductivity, expressed in siemens per meter, will be written. It must be a dataset in an .h5 file. If it is an empty string, the postprocessing will not be applied to the electric conductivity map (default: empty string).
output-relative-permittivity is the address where the postprocessed relative permittivity will be written. It must be a dataset in an .h5 file. If it is an empty string, the postprocessing will not be applied to the relative permittivity map (default: empty string).
perform-only-postprocessing is equal to false if the postprocessing must be applied to the result of the EPT method. It is equal to true if only the postprocessing must be performed in a provided set of input maps (default: false).
input-electric-conductivity is the address of the electric conductivity map to be filtered. It must be a dataset in an .h5 file. It is used only if perform-only-postprocessing is true, otherwise the electric conductivity computed by the EPT method is used as input (default: empty string).
input-relative-permittivity is the address of the relative permittivity map to be filtered. It must be a dataset in an .h5 file. It is used only if perform-only-postprocessing is true, otherwise the relative permittivity computed by the EPT method is used as input (default: empty string).
input-reference-image is the address of the reference image used by the filter. It must be a dataset in an .h5 file. It is used only if perform-only-postprocessing is true, otherwise the reference image provided to the EPT method is used as input, when available (default: empty string).
input-electric-conductivity-variance-map is the address of the electric conductivity variance map used by the filter. It must be a dataset in an .h5 file. It is used only if perform-only-postprocessing is true, otherwise the variance map estimated by the EPT method is used as input, when available (default: empty string).
input-relative-permittivity-variance-map is the address of the relative permittivity variance map used by the filter. It must be a dataset in an .h5 file. It is used only if perform-only-postprocessing is true, otherwise the variance map estimated by the EPT method is used as input, when available (default: empty string).
kernel-size is the number of voxels along each semi-axis of the filter kernel (default: [1,1,1]).
kernel-shape selects the kernel shape according to the following table (default: 2).
weight-param is a parameter of the weight function used for the reference image (default: 0.10).

Code	Shape	Example with `size = [2,2,1]`
0	Cross
1	Ellipsoid
2	Cuboid

The postprocessing filter is selected according to the following decision tree:

If neither the reference image nor the variance map are available, a median filter is applied.
If the reference image is available, but the variance map is not, a median filter whose kernel is anatomically adapted is applied.
If the reference image is not available, but the variance map is, a moving weighted average is applied, using the reciprocal of the variance map values as weight.
If both the reference image and the variance map are available, a moving weighted average whose kernel is anatomically adapted is applied, using the reciprocal of the variance map values as weight.

The filter kernel is anatomically adapted with respect to the reference image according to a hard threshold rule: only the voxels whose relative contrast with respect to the central voxel is lower than weight-param are kept in the kernel.

Parameters

In addition to the previous settings, that are common to all EPT methods, each method could require some additional parameters. The settings for each specific EPT method are described in their pages, where examples of configuration files are also provided (see Methods).