UCLA

Current medical ultrasound datasets used in training simulators lack adequate coverage due to imaging hardware limitations. We present software solutions both for generating and for interacting with large-scale ultrasound datasets. The generation process combines a physical simulation with a ray tracing rendering technique to create synthetic ultrasound volumes for use as ground-truth test data. These datasets allow us both to train users and to evaluate an automatic registration solution used to align multiple real ultrasound volumes. We merge the aligned results with a multiresolution functional-based convex optimization technique to achieve seamless blends between adjacent volumes. A content-aware embedding algorithm places the merged data into a clean background template. We enable end-users to interact with the final results through a real-time mannequin-based translational tracking system.