New Dynamic model was implemented:
 "Variable Flip Angle T1 recovery  S = M0*(1-exp(-TR/T1))*sina(A)/(1-exp(-TR/T1)/cos(A)) "

It appears as the model #24 on the FireVoxel Dynamic Experiment Framework

Model has 2 hyperparameters defining the Local optimization as in Deoni2003 and FireVoxel global optimization.  In case of the Global optimization we also can set the # of iterations, but it can be kept default.

There are 3 output parameters: T1 (ms), M0 and fitting residual.

To test the model a Brain ROI (mask) was detected on the max signal timepoint using the standard EdgeWave algorithm and parameters.

We ran algorithm with both linear and global optimization and found out that global optimization finds a 20% smaller residual on average, while reducing the Max residual  by x2.5 times.

Residual statistics for Linear optimization:  Mean=12.1, Max=215.8

Residual statistics for the Global Optimazation:   Mean=10.4  Max=97.5

 


The following T1 map was obtained:

 ... with the corresponding Fitting Residual map
 

Typical  fitting for a single voxel looks like this:

For this particular fitting: 
Linear Optimization:   T1 = 1238  ms
Global  Optimization:  T1 = 12.64 ms
i.e. about   ~3% difference