Optimized GPU Framework for Ultrasound B-Mode Imaging

Ultrasound B-mode imaging is the basic image mode which can offer anatomic information of organs for clinical diagnosis. Because of the massive computation involved in baseband processing from focused radio-frequency (RF) signals followed by envelop detection, compression and scan conversion required for high quality B-mode imaging, existing medical systems always rely on complicated hardware in real time implementation. In this paper, we have proposed the parallel processing of RF signals to B-mode image based on commercial graphics processing unit (GPU) under NVIDIA CUDA platform. Experiments show that our GPU implementation can achieve a frame rate about 48 fps from RF signals of 206 × 2172 samples to an image with a size of 256 × 512. This is about 52 times faster than that based on CPU platform with the same image quality.