Skip to content

Parameters.md

Latest commit

 

History

History
163 lines (126 loc) · 6.46 KB

Parameters.md

File metadata and controls

163 lines (126 loc) · 6.46 KB

Fitting Parameters

By writing all relevant fitting parameters into a json or toml (for more details see) file, correct fitting of the image data and storage of your fitting parameters is ensured. Due to strong dependencies on initial fitting parameters in some of the implemented approaches, it is strongly recommended to use a specified file with an adapted parameter set for each model (and image region). The file is structured in three main parts, General, Model and Boundaries. For a full entry by entry explanation based on the final file see

General Parameters

name description value
Class corresponding parameter class "IVIMParams", "IVIMSegmentedParams", "NNLSParams", "NNLSCVParams"
description Description of the fit str
fit_type fitting approach used "single", "multi", "gpu"
max_iter maximum iterations int
n_pools number of pools (CPUs) (multi only) int
fit_tolerance tolerance for convergence check (gpu only) float
b-values x_axis data list of ints

Example toml code:

[General]
Class = "IVIMParams"
description = "IVIM fitting parameters for DWI data"
fit_type = "multi"
max_iter = 250
n_pools = 4
b_values = [0, 50, 100, 200, 400]

Model Specific parameters

Every model has model string attribute defining the fitting model used. The available vary between the different parameter classes. The model specific parameters are defined in the Model section of the parameter file. For a more detailed description, take a look here.

IVIM

The available models are:

  • "monoexp" - mono-exponential fitting
  • "biexp" - bi-exponential fitting
  • "triexp" - tri-exponential fitting

For the basic IVIM fitting, the model-specific parameters are defined in the Model section of the parameter file. These are as follows:

name description value
fit_reduced whether reduced fitting should be used bool
fit_S0 wether S0 should be calculated instead directly bool
fit_t1 whether T1 should be calculated in the first step bool
repetitioon_time repetition time for IVIM fitting int, float
fit_t1_steam whether T1 decay during mixing time for STEAM should included bool
mixing_time mixing time for STEAM fitting int, float

Example toml code:

[Model]
model = "biexp"
fit_reduced = false
fit_S0 = false
fit_t1 = true
repetition_time = 2000
fit_t1_steam = true
mixing_time = 20

Segmeted IVIM

The segmented IVIM fitting uses the same models as the normal IVIM fitting but adds some additional parameters To the General section. The model specific parameters are:

name description value
fixed_component name of component to fix (e.g. "D_1") str
fixed_t1 whether T1 should be calculated in the first step bool
reduced_b_values can befalse or a list of b_values for initial fitting bool, list

Example toml code:

[General]
# ... other general parameters ...
fixed_component = "D_1"
fixed_t1 = true
reduced_b_values = [0, 50, 100, 200, 400]

NNLS

The NNLS Model adds some additional parameters to the Model section. The model specific parameters are:

name description value
reg_order regularization order (0-3 or "CV") int, str
mu regularization factor (0 - 3 only) float
tol tolerance for convergence check (CV only) float

Example toml code:

[Model]
reg_order = 2
mu = 0.02
tol = 1e-6

Boundaries

The boundaries value ist a dictionary holding sub dictionaries for each component. For IVIM there are generally three options. The first is the classic model with non-relative fractions and without S0 fitting. The second is the S0 approach with relative fractions and S0 and the third is the reduced bi-exponential fitting, which assumes the data to be normalized.

IVIM

For the IVIM fitting each parameter has its own boundaries. The boundaries are defined in the Boundaries section of the parameter file. Additionally if the fit_t1 parameter is set to true, the T1 boundaries are defined as well. Example toml code for reduced bi-exponential:

[Boundaries]
# Boundaries for reduced bi-exponential fitting
[Boundaries.D.1]
# Array with [x0, lb, ub]
# [0.02, 0.003, 0.3]
0 = 0.02
1 = 0.003
2 = 0.3

[Boundaries.D.2]
# [0.001, 0.0007, 0.05]
0 = 0.001
1 = 0.0007
2 = 0.05

[Boundaries.f.1]
# [85, 10, 500]
0 = 85
1 = 10
2 = 500

[Boundaries.f.2]
# [20, 1, 100]
0 = 20
1 = 1
2 = 100

NNLS

Since the NNLS method takes a different approach in calculating the underlying diffusion parameters a different set of boundaries and additional parameters is needed.

Boundaries:

name description type
d_range diffusion value range for NNLS fitting list(float, float)
n_bins number of exponential terms used for fitting int