Registration of intraoperative optical image sequence

Abstract

In neurosurgery, cortical warping is one of the significant sources of noises during optical imaging after the skull and the dura have been removed. The optical image sequence must be registered for further data analysis. The registration algorithms in widely used medical image tools, e.g. Automated Image Registration (AIR), Statistical Parametric Mapping (SPM), usually express the cortical warping as polynomials or terms of cosine basis. However, these nonlinear models do not faithfully fit the elastic warping of the cortexes, and thus can not achieve a satisfactory result. Based on the elastic model, i.e., the approximating thin-plate splines (aTPS), we propose herein an improved aTPS (iaTPS) algorithm to deal with the elasticity of the cortical warping. In the cost function of the original aTPS algorithm, landmarks with different localization uncertainties should be given different weights, however, due to the absence of a convincing method to specify these weights, a same weight value was manually set for all landmarks in practice. In our iaTPS algorithm, landmarks are categorized into several classes (usually 3~5) by their localization uncertainties, and the weights for each class are decided by an optimization process. The comparison experiment on the intraoperative optical data of human brain has shown that the new algorithm can offer better registration accuracy than the aTPS algorithm.