Optical tomography (OT) recovers the cross-sectional distribution of opticalparameters inside a highly scattering medium from information contained inmeasurements that are performed on the boundaries of the medium. Theimage reconstruction problem in OT can be considered as a large-scaleoptimization problem, in which an appropriately defined objective functionneeds to be minimized. In the simplest case, the objective function is theleast-square error norm between the measured and the predicted data. Inbiomedical applications that apply near-infrared light as the probing toolthe predictions are obtained from a model of light propagation in tissue.Gradient techniques are commonly used as optimization methods, whichemploy the gradient of the objective function with respect to the opticalparameters to find the minimum. Conjugate gradient (CG) techniques thatuse information about the first derivative of the objective function haveshown some good results in the past. However, this approach is frequentlycharacterized by low convergence rates. To alleviate this problem we haveimplemented and studied so-called quasi-Newton (QN) methods, which useapproximations to the second derivative. The performance of the QN and CGmethods are compared by utilizing both synthetic and experimental data.
Read full abstract