Fast registration of contrast-enhanced magnetic resonance images of the breast

Abstract

Contrast-enhanced magnetic resonance (MR) imaging is an effectivetool for the detection and analysis of female breast cancer. Theimaging protocol consists of 3D volumes acquired at different timesbefore and after the administration of a contrast agent.Intensity-time profiles are constructed for every voxel to aid inthe diagnosis process. However, because of the voluntary orinvoluntary movements of the patients, the images have to beregistered before a diagnosis can be reliably given.Nonrigid deformation based on B-splines optimizing the normalizedmutual information (NMI) criterion has proved to be successful inthis registration task involving elastic deformations. In the firstpart of this work, we have proposed a fast approximation algorithmto estimate the gradient of NMI using a set of auxiliary imagesconstructed from the image conditional probability distributions.Our method could speed up the registration process by an order ofmagnitude with similar registration quality. In the second part, weaimed to further speed up the registration process by offloading thebulk of the computational load to the GPU hardware for efficientprocessing. We exploited the single instruction multiple data (SIMD)processing capabilities and the dedicated interpolation hardware toobtain a further speed up of 15-30 times compared to CPUimplementation.