Reducing Execution Time in CT Angiography and Dynamic CT Brain Image Registration Through Code Optimisation

Abstract

In this paper we present a preliminary assessment of code optimisation for automatic intensity-based 3D image registration of dynamic CT (used for CT perfusion imaging) to CT angiography (CTA) images. This registration step is intended to be a pre-processing stage for an automated volumetric CT stroke image diagnosis system. The dynamic CT images were acquired with a gantry tilt and which has to be considered in the image registration step. We have previously shown that tilt correction can be performed implicitly and combined with intensity based registration to the CTA image. The processing time though, may hinder its usefulness in a clinical setting. Hence we show how the processing time can be reduced by around 75% simply by changing the implementation of trilinear interpolation in the registration algorithm. Some prospect for further improvement using GPUs is also touched upon.