用于压缩感知磁共振成像的分割字典学习算法

Dictionary Learning with Segmentation for Compressed-Sensing Magnetic Resonance Imaging

字典学习算法可以根据数据本身的特点构建稀疏域中的基,从而使数据的表示更加稀疏.该文在传统的字典学习算法基础上提出了分割字典学习算法,由于部分磁共振图像组织结构简单、可以进行图像分割,因此可根据此特点来优化字典中基函数的构建,使磁共振图像的表达更为稀疏,从而获得更高的重建图像质量.该文利用模拟数据和真实数据进行了重建实验,结果表明与传统的字典学习算法相比,分割字典学习算法能进一步改善重建图像质量.

Dictionary learning （DL） builds a set of basis functions from the input data, such that the data can be represented more sparsely. Based on the fact that certain magnetic resonance （MR） images can be easily segmented, we propose an algorithm named dictionary learning with segmentation （DLS）. The algorithm achieves better image reconstruction quality by optimizing construction of the dictionary and to making representation of the MR images sparser though incorporating image segmentation into dictionary learning. The experimental results on simulated datasets and in vivo images demonstrated that the proposed algorithm can yield better reconstruction relative to the traditional dictionary learning algorithm.