CIF: Small: Efficient Model-Based Iterative Reconstruction For High Resolution CT

点击翻译按钮获取中文摘要

Enabling image reconstruction from low-dose X-ray data has been a major challenge in Computed Tomography (CT) imaging for several decades. Model-Based Iterative Reconstruction (MBIR) algorithms enable high-resolution image reconstruction from low-dose data by incorporating models for X-ray physics, the acquisition process and image priors into an optimization process for reconstruction. Despite promising dose-reduction results from these modern image reconstruction algorithms, the classical filtered back projection algorithm and its variants are widely employed in practical settings, especially when rapid imaging is necessary. The main impediment in translating these low-dose imaging technologies to practice has been their computational cost — long reconstruction time compared to classical methods.This project develops a novel algorithmic approach that leverages techniques from approximation theory together with performance optimization tools for tackling the computational challenges in MBIR. The project has a set of inter-connected goals that develop the proposed framework for 3-D optics, specializing to common X-ray and detector geometries. Moreover, signal processing tools are developed within this framework for increasing image resolution, in a computationally efficient way, in a broad set of CT inverse problems. The project also includes an extensive evaluation plan using established benchmarks as well as a repository of data maintained by the US National Library of Medicine specifically for validation of acceleration methods that seek to enable routine use of MBIR methods. These developments could play a game changing role in the applicability of low-dose imaging in practice and, ultimately, adoption in a wider range of applications.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.