Next Generation Fast Methods for Medical and Nanoscale Technology

PI: Jacob White, Department of Electrical Engineering & Computer Science, MIT
PI: Athanasios Polimeridis, Center for Computational and Data-Intensive Science and Engineering, Skoltech

Voxel-based 3-D structure generation is so fast and reliably automatic, it has become a standard in a broad range of applications such as: Medical-Imaging-based anatomical reconstruction, 3-D printer model creation, and virtual nanofabrication. Its speed has enabled engineers and clinicians with hand-held tablets to visualize designs, diagnose diseases, and even plan surgery. But they cannot easily see their structures in action, nor assess and optimize performance, even though such a capability would be disruptively enabling. The problem is that state-of-the-art 3-D simulators, often finite- or volume-element-based, are not automatic enough, reliably accurate enough, or fast enough. And they are so mismatched to the billion-cube geometries produced by voxel-based structure generation, simulations can take days. Instead, far better performance can be achieved by exploiting the voxelization, as in our recently developed FFT-accelerated volume integral equation method (FFT-VIEM). For example, the FFT-VIEM approach, implemented in the open-source package MARIE, only needs minutes to accurately compute MR fields in a billion-voxel humanhead model. We propose generalizing FFT-VIEM to medical and nanoscale applications such as electromagnetic, thermal, and ultrasonic field analysis in human tissues, and performance analysis in emerging photonic and terahertz-circuit nanotechnology. And since material properties are accessibly exposed in FFT-VIEM, we also propose using fast simulation for material property extraction (the inverse problem). We plan to expand our public-domain MARIE software in to a catalog of open-source scripting-language-based tools for a variety of applications, to both facilitate commercialization by others while also enabling the creation of a case-study-based interdisciplinary course on numerical solution of PDEs for use at both Skoltech and MIT. In addition, we plan to accelerate the maturation of the nascent Skoltech student research community by involving Skoltech PhD students in numerical research and broader educational activities (such as the handson approach to teaching feedback control using hardware textbooks and MITx/EdX), both at Skoltech and as visiting students at MIT. The effort at MIT will be facilitated primarily by the MIT PI and a full-time post-doctoral researcher, supported by an extensive travel budget for students and faculty. Finally, as a SkolTech industrial partner, Datadvance will integrate the open-source FFTVIEM with their state-of-the-art optimization and cloud-based distributed solution methods, and demonstrate improved availability and efficiency of MR imager design optimization and inverse problem solving.

Back to the list >>