Abstract

The measurement situation we consider involves multiple source and detector locations around the perimeter of a highly scattering domain. A frequency domain measurement involves modulating each light source with a CW signal and performing a coherent detection at this modulation frequency. The multiple scatter and absorption within the domain results in a spatially-dependent magnitude and phase of the light intensity (the measurement data). This measured data then must be inverted to form an image based on the spatially varying absorption and scattering parameters in a diffusion equation model [1]. With high scatter and relatively low loss, the diffusion equation provides an accurate forward model. Forming an image involves inverting the diffusion equation to obtain the spatially-dependent scattering and absorption coefficient. This nonlinear inverse problem requires an iterative solution to minimize an appropriate cost function [1]. We have presented a physically-based Bayesian formulation (where instrument shot noise is incorporated) and an iterative coordinate descent (ICD) optimization for the optical diffusion imaging problem [2]. In particular, we have proposed an ICD/Born method which provides high quality reconstructions and is computationally efficient when compared to the conventional iterative Born approximation methods. However, the computational complexity of ICD/Born is still prohibitive for large three dimensional problems Perhaps more importantly, local optimization methods such as ICD/Born can become trapped in local minima.

© 2000 IEEE