Improved Gauss-Newton optimisation methods in affine registration of SPECT brain images

In single photon emission computed tomography images, the differences between brains of different subjects require the normalisation of the images with respect to a reference template. The general affine model with 12 parameters is usually chosen as a first normalisation procedure. Usually, the Levenberg-Marquardt or mostly the Gauss-Newton method are used in order to optimise a cost function, which presents an extreme value when the image matches with the template. In this reported work, these optimisation algorithms are compared with two alternative versions of the Gauss-Newton method. Both proposed alternatives include an additional parameter, which allows the adaptive change of the step length along the descent direction. Experimental and simulated results show that the inclusion of this parameter improves the convergence rate considerably.