You could investigate the VBQ toolbox for SPM (linked). Although intended for more complicated multi-parameter mapping, I think you can use it for just estimating R2* from multiple echoes. Alternatively, it should be relatively easy to implement voxel-wise log-linear least-squares or nonlinear fitting using e.g. MATLAB's lsqnonlin or the method in Eqn.5 in Hagberg's paper (linked).

http://www.fil.ion.ucl.ac.uk/Research/physics_info/QuantMRI_VBM.html

http://doi.org/10.1002/mrm.10283

Thank you very much for your kindness. I try it.

Altough this is not MATLAB,  you could check this ImageJ plugin:

http://www.ironbymr.fr/index.html