磁共振扫描速度长期受到梯度系统性能和人体生理条件的限制，存在数据采集时间过长的问题。课题组前期研究的G-SPEED技术，基于数据在变换域的稀疏特性，突破了香农-奈奎斯特采样定理的限制，通过灵活地减少数据采集总量来缩短扫描时间；基于解析求解的过程使得G-SPEED的重建快速且易于并行实现；结合GPU图形处理器的并行加速，能获得实时的重建速度。本项目旨在通过基于数据稀疏表示的实时G-SPEED磁共振成像技术研究，减少磁共振成像的数据采集和重建时间，实现对病人实时监护和检查，以减少检查时间，提高重危病人的存活率,提高医院昂贵MRI设备的利用率。本项目主要研究1）基于多分辨率鬼影模型的G-SPEED数据稀疏表示方式；2）图像弱小鬼影的来源与抑制方法；3）鬼影敏感的图像客观量化准则和优化采集方案；4）基于GPU的并行加速实现；5）G-SPEED理论技术与其他快速成像方法的组合优化与应用。

The scanning time of magnetic resonance imaging (MRI) is relatively long. This is due to the limitations of gradient system performance and physiological considerations. The original G-SPEED technique proposed here can largely shorten the scanning time of MRI by reducing the amount of acquired data. It overcomes the Shannon-Nyquist sampling limitation based on sparse representation of data in some transform domain. The reconstruction of G-SPEED is very fast with analytical solution and could be easily implemented with parallel computation. Real time reconstruction can be achieved by using graphic processing unit (GPU). This proposal is aimed to shorten the time of data acquisition and reconstruction, achieving real time monitoring and examination of patients, improving survival rate of emergency patients and the utilization of expensive MRI scanner. The study will mainly focus on 1) Sparse data representation of G-SPEED based on multi-resolution ghost model; 2) Mechanism of minor ghost and its suppression methods; 3) Objective quantification of ghosts for image evaluation and optimization of sampling scheme; 4) Parallel implementation based on GPU; 5) Optimization and combination with other fast imaging methods for practical applications.