基于并行成像技术的在线动态磁共振成像的实时重构算法研究

Dynamic magnetic resonance imaging MRI (dMRI) is a key component of many clinical exams such as cardiac, perfusion imaging. To improve the temporal and spatial resolution of real-time reconstruction for online dMRI, we will do some research on the real-time reconstruction for online dMRI. Our work is as follows: for real-time reconstruction based on the SPIRiT model, we will explore regularization terms of multi-coil images to improve the reconstruction quality, such as the dynamic Joint Total Variation (JTV) regularization term, the forest sparsity regularization term, and the low rank regularization term of the global high dimention tensor, and we will design efficient reconstruction algorithms for the responding problems; for real-time reconstruction based on the SENSE model, we will explore the joint reconstruction model based on multi-coil k-space data and regularization terms of multi-coil images to improve the reconstruction quality, and we will design efficient reconstruction algorithms for solving the responding problems. Finally, we will accomplish parallel optimization with GPU to accelerate the reconstruction. The results of our work will benefit the clinical application of cardiac cine MR, MR image guided therapy and surgery.

动态磁共振成像（dMRI）是一些磁共振成像应用（心脏，灌注、功能和食道成像）的关键。针对在线dMRI实时重构的时空分辨率较差的问题，本项目将开展在线dMRI实时重构算法的研究，主要内容包括：针对基于SPIRiT模型的实时重构，探索使用Dynamic JTV正则项、Forest Sparsity正则项、以及全局高阶张量的低秩正则项等多线圈图像的正则项化技术提高重构质量的方法，并设计针对图像域和频域SPIRiT模型的高效重构算法；针对基于SENSE模型的实时重构，探索基于多个线圈k空间数据相关性的联合模型，并探索使用多线圈图像的正则项提高SENSE模型重构质量的方法，并设计相应的高效重构算法。最后，将对设计的算法进行GPU并行优化，最终实现实时。本项目的研究成果将使得心脏电影MR、MR图像引导的治疗和手术等临床应用受益。