MitkGenericFitting User Guide

Table of Contents

• Overview
• Supported Models
  • Linear Model
  • Generic Model (2 parameters)
• - The current version supports 2 model parameter (a, b)
• Usage
  • Required Arguments
  • Optional Arguments
• Output
• Fitting Modes
  • Pixel-based Fitting (Default)
  • ROI-based Fitting
• Examples
  • 1. Linear Model Fit (Pixel-based)
  • 2. Exponential Decay (Generic, Pixel-based)
  • 3. ROI-based Linear Fit

Overview

MitkGenericFitting is a command-line tool for fitting model functions to time-resolved imaging data (3D+t). It supports voxel-wise or region-based fitting using a mask and produces parameter images as output.

Supported image types for input include .nrrd, .nii, and DICOM files.

Supported Models

Linear Model

• Model: y = a + b*t
• Specify with: --function Linear

Generic Model (2 parameters)

• User-defined model formula using two free parameters (a, b)
• Example: y = a * exp(-b*x)
• The parser is able to recognize:
  • sums, differences, products and divisions (a + b, 4 - 3, 2 * x, 9 / 3)
  • algebraic signs ( +5, -5)
  • exponentiation ( 2 ^ 4 )
  • parentheses (3 * (4 + 2))
  • following unary functions: abs, exp, sin, cos, tan, sind (sine in degrees), cosd (cosine in degrees), tand (tangent in degrees)
  • variables (x, a, b, ... j)

Remarks

• The variable "x" is reserved. It is the signal position / timepoint.

- The current version supports 2 model parameter (a, b)

Usage

MitkGenericFitting -i <inputImage> -o <outputTemplate> -f <functionType> [options]

Required Arguments

Argument	Short	Type	Description
--input	-i	file	Input dynamic image (3D+t).
--output	-o	file	Output template file path for resulting parameter images. File extensions determine output format.
--function	-f	string	Model type: "Linear" or "Generic-2" for generic 2 parameter model. Default is Linear.

Optional Arguments

Argument	Short	Type	Description
--formular	-y	string	Model formula string (required for generic models).
--mask	-m	file	Binary mask image defining ROI. Required for ROI-based fitting.
--roibased	-r	flag	Use ROI-based fitting (fit mean signal in mask).
--verbose	-v	flag	Enable verbose output.
--help	-h	flag	Show help text.

Output

For each fitted model parameter, a separate image will be created. The output filenames will be based on the --output template plus a suffix per parameter (e.g., _Param_0.nrrd, _Param_1.nrrd).
A fitting summary and evaluation metric image (e.g., Chi^2) will also be generated.

Fitting Modes

Pixel-based Fitting (Default)

• The model is fitted to each voxel individually.
• Mask is optional (if provided, fitting is constrained to masked region).

ROI-based Fitting

• The average time-intensity curve within the mask is computed.
• The model is fitted to the ROI curve.
• Output will be constant-valued images reflecting the ROI fit result.

Examples

1. Linear Model Fit (Pixel-based)

MitkGenericFitting -i dyn.nrrd -o result/linFit.nrrd -f Linear

Description:
Performs voxel-wise fitting of a linear model on dyn.nrrd.

Naming Strategy:

• Output parameter images: linFit_y-intercept.nrrd (intercept), linFit_slope.nrrd (slope)

Generated Values:

• Each voxel gets individual values for intercept and slope.

---

2. Exponential Decay (Generic, Pixel-based)

MitkGenericFitting -i dyn.nrrd -o result/expFit.nrrd -f Generic-2 -y "a*exp(-b*x)"

Description:
Fits an exponential decay model to each voxel.

Naming Strategy:

• Output files: expFit_a.nrrd (a), expFit_b.nrrd (b)

Generated Values:

• Each voxel gets a fitted p0 and p1 value for the exponential model.

---

3. ROI-based Linear Fit

MitkGenericFitting -i dyn.nrrd -o result/roiLinFit.nrrd -f Linear -m mask.nrrd -r

Description:
Fits a linear model to the mean curve inside the mask.

Naming Strategy:

• Output parameter images: roiLinFit_y-intercept.nrrd (intercept), roiLinFit_slope.nrrd (slope)

Generated Values:

• Each output image has constant values representing ROI fit result.

---