Abstract
The significant progress of medical imaging devices in recent years has enabled the fast acquisition of high precision volumetric data. Due to this, the need of extension of traditional 2D image processing algorithm to volumetric data has become higher. Because the amount of computational cost increases as the size of 3D data, the application of GPGPU (general-purpose computing on GPUs) based on GPU (graphics processing unit) has been widely researched and several preceding studies were reported. The authors have focused on the improvement of performance and automation for ultra-sound image registration. Instead of using original intensity value, we are currently researching the merit of local features based on 3D local statistics as the similarity measure. As the course of research, we implemented the parallel data processing of local features on GPU and it showed the performance improvement by 19 to 230 times.
