GPU Accelerated 3D Ultrasound Imaging System

3D ultrasound imaging is a novel technology in medical imaging field. The image display of 3D ultrasound imaging is more intuitional compared with traditional 2D ultrasound imaging. However, the size of 3D ultrasound image data is far higher than 2D ultrasound and the enormous computation amount of volume reconstruction and volume rendering seriously limits the speed of 3D ultrasound imaging system. The development of graphics processing unit (GPU) provides a novel solution to 3D ultrasound's bottleneck. Due to its powerful parallel computation capability and programmability, GPU has been widely used in 3D volume data visualization, image and signal processing and acceleration of computer graphics. This paper applied GPU to 3D ultrasound imaging system and implemented GPU-based volume reconstruction and volume rendering, which is helpful to accelerate system speed, enhance image quality and reduce system cost.The GPU-based volume reconstruction algorithm implemented in this paper maps the coordinate transformation and interpolation respectively into 3D texture coordinate calculation and texture fetching. Serial interpolation of CPU is replaced by parallel interpolation of GPU so that the process of volume reconstruction is greatly accelerated. Furthermore, volume reconstruction is integrated with the resampling operation of volume rendering Reconstruction is only executed for voxels required in volume rendering but not for the entire volume data to prevent from redundant calculation.Ray casting algorithm is a typical volume rendering algorithm with high image quality which is able to display the lower level details of volume data. CPU-based ray casting algorithm is usually very slow because of the large computation amount. The GPU-based ray casting algorithm implemented in this paper accomplished the resampling, color and opacity transfer as well as image composition of the original algorithm in GPU. Real-time rendering of the volume data is achieved by taking the advantage of GPU's high parallel computation power. Furthermore, pixel lighting can be done by calculating the gradient of sampled voxel to increase the sense of reality of displayed image. The GPU-based ray casting algorithm implemented in this paper also introduces the early ray termination technology. For those rays with a accumulated opacity value close to 1, integration will be terminated before the ray exits volume data so that the efficiency of the proposed algorithm is further improved.