Realistic rendering of 3D fetal ultrasound via local ambient occlusion

In this paper, we propose a novel method for realistic rendering for 3D ultrasound. Previous methods either failed to deal with the low signal-to-noise ratio (SNR) of 3D ultrasound data or could not achieve real time visualization. We apply a local ambient occlusion illumination to approximate each voxel's intensity coming from incident light for direct volume rendering. As the sampling scheme is very important for local ambient occlusions in terms of both visualization quality and time performance, we propose an entropy-based multiresolution volume management and adaptive sampling scheme to speed up the calculation and support large datasets. Additionally, to obtain a realistic rendering result, we use a 2D color map, which is generated from a real baby image database. Experiments demonstrate our method can produce realistic rendering result, where the depth information and shapes can greatly enhanced when compared to traditional methods. With GPU-based implementation, real time rendering can be achieved without compromising visualization quality.