基于GPU的快速三维医学图像刚性配准技术

Fast 3D rigid medical image registration based on GPU

自动三维配准将多个图像数据映射到同一坐标系中,在医学影像分析中有广泛的应用。但现有主流三维刚性配准算法(如FLIRT)速度较慢,2563大小数据的刚性配准需要300 s左右,不能满足快速临床应用的需求。为此提出了一种基于CUDA(compute unified device architecture)架构的快速三维配准技术,利用GPU(gra-phic processing unit)并行计算实现配准中的坐标变换、线性插值和相似性测度计算。临床三维医学图像上的实验表明,该技术在保持配准精度的前提下将速度提高了一个数量级,非常适合海量快速的临床应用。

Automatic 3D image rigid registration technique, which maps multiple images into a common coordinate space, is an important clinical image analysis tool. The state of art rigid registration techniques like FLIRT take 300s to complete a rigid registration for a 2563 data set. This makes it hard to apply this technique in time critical clinical environment. This paper proposed a parallel accelerating registration algorithm based on the new CUDA. By using the GPU to compute the coordinate transformation, interpolation and similarity metric computation, great improvement of speed has been achieved without compromi sing the registration accuracy. The proposed technique is valuable for time critical clinical application.