This effort puts forth powerful models capturing the characteristics of big dynamic magnetic resonance imaging (MRI) data, and then offering architectures and algorithms, while revealing fundamental insights into various analytical and implementation trade-offs involved. The proposed framework will extract salient global trends to enable imputation for missing MRI data entries due to imaging speed limitations, and obtain parsimonious representations to process and draw inferences from big pools of MRI data. Leveraging advances in low-rank and sparsity-aware signal processing, learning and optimization, online, parallel, and decentralized algorithms based on matrix and tensor models, will enable streaming analytics of sequential measurements using parallel processors, and tracking dynamically evolving datasets.
This project will directly impact high-resolution 3D dynamic MRI technology to improve medical diagnosis and treatment. The developed algorithms and tools will enable technology transfer to benefit a wide population and improve healthcare. Insights gained from this project's large-scale analytics context will also benefit big data mining, neuroscience, smart grid, and health informatics. Broader impact will be further effected by the integration of the proposed research with an educational plan designed to train the new cadre of next-generation of medical data science professionals, as well as promote cross-fertilization of academic research with health industry needs.
