Parameter decomposition for iteratively regularized
Gauss-Newton solutions in optical tomography

Rosemary Renaut
Dept. of Mathematics and Statistics
Arizona State University, Tempe AZ 85287-1804
renaut@asu.edu
Taufiquar Khan, Alexandra Smirnova

We extend evaluation of the iteratively regularized Gauss Newton method for the solution of the parameter estimation problem in Optical Tomography. The general problem of optical tomography requires the estimation of the underlying model parameters $ {\mathbf q}$, for example the coefficient of diffusion $ D$ and the coefficient of absorption $ \mu_a$, (i.e. $ {\mathbf q}=(D,\mu_a)^T$) that belong to a parameter set $ Q$. The conditioning of the problem with respect to each parameter set is different. We investigate the use of an alternating parameter decomposition approach for solution of the nonlinear inverse problem with regularization. Contrary to statements on the general nonlinear least squares problem in standard references eg Bjorck 1996 , we find that decomposition with respect to the parameter set allows solution of the regularized problem with the use of appropriately chosen weighting schemes.