With the popularity of real-time 3D/4D data capturing capabilities, there is an emerging need to compute the deformable models over the networks. Traditional model-driven deformable models have their bottlenecks since the spatial information for large-scale datasets cannot be efficiently solved and adaptively minimized over different network conditions. This project centers on the concept of spectrum-driven deformable models. Analogous to Fourier analysis being applied to image processing, the spectral deformable models employ manifold harmonics to efficiently and effectively perform segmentation, registration, physics-based simulation, compression, and streaming of 3D deformable surfaces and volumes.
In this project, manifold harmonics are used to transform arbitrary scanned datasets into a reduced diffusion subspace, in which real-time segmentation, registration, and physics-based simulation can be performed. Besides the stretching deformations, the rotational and tensor fields are encoded with manifold harmonics, which provides a "spectral multiresolution" structure to compress and stream deformable models over different network conditions.
The PI works with the medical imaging experts at UT Southwestern Medical Center, to build a tele-diagnosis system for evaluating each component in the spectral deformable models. The test-bed on tele-medicine has impacts on the next-generation diagnosis and treatment services, as well as on clinical education. The theoretical and technical breakthrough can benefit our society at large, from tele-immersion, remote sensing, to speech training, through the PI?s further outreach activities. The research and education are integrated by taking research advances into existing and future courses; developing the visualization software for education; and attracting more undergraduate students into research.
