ROI-based data modelling

nordicICE can perform different types of data modelling (curve fitting) to the active ROI curve. When the curve fit is completed, a results window will appear which displays the resulting fitting parameters. For all curve fitting procedures, general information about the image type and fixed parameters must be set prior to curve fitting. Five different curve fits can be applied:

Linear: 

Performs a linear fit (Ax + b) to the data within the specified range.

Exponential fit: 

Fits the curve to a single exponential of the form: y=A*exp(- B*x) where A and B (=1/T2) are the model parameters . X can be either the echo time, TE (for T2/T2* fitting) or b-value (for ADC analysis). The type of output depends on the value of x set under Axis options.

T1-relaxation: 

Estimates T1 relaxation times from one of the following sequence types:

· Inversion recovery – X-values must be set to ‘Inversion time’ (TI) under Axis options

· Saturation recovery (SR). X-values must be set to ‘delay time’ under Axis options

· Multi-flip angle spoiled gradient echo (SPGR) sequence. X-values must represent flip angles

Gamma variate: 

Fits the curve to a gamma variate curve of the form: F(x)=f(x).(x-T₀)^a .exp[(x-T₀)*b], where T₀, a and b are fitting variables and f(x)=0 for x<T₀ and f(x)=1 for x>=T₀. For more details see: Bjørnerud A et al. Magn Reson Med (2002) 47;298-304.

Kinetic analysis: 

Fits the curve to a two-compartment kinetic model which describes the flux/reflux of a contrast agent from the blood pool to the extravascular space. For more details, see Tofts et al. JMRI (1999) 10(3):52-61.

All the above curve fitting functions, except for linear fit, are solved using a non-linear variance weighted least squares iterative algorithm. The variance is obtained from the measured ROI variance for each dynamic image in the series. As for all iterative methods, the ability to obtain a good curve fit solution is dependent on the iteration start values (initial 'guess' values) for the model parameters. nordicICE attempts to automatically find good start values based on the detected curve characteristics, but this may sometimes fail due to noisy data or uncharacteristic curve shapes. This may be a problem for the more complex functions like gamma variate and kinetic analysis. If the curve fitting fails (e.g. the fitted curve is clearly not describing the data to be modelled) using default settings, try changing the iteration start values and uncheck the option for automatic detection of iteration start values.

Related topics:

Data modelling Results