Mei-Lan Chu1, Hsiao-Wen Chung1, Yi-Ru Lin2, Tzu-Cheng Chao3, 4
1Graduate Institute of Biomedical Electronics and Bioinformatics, National Taiwan University, Taipei, Taiwan; 2Department of Electronic Engineering, National Taiwan University of Science and Technology, Taipei, Taiwan; 3Institute of Medical Informatics, National Cheng-Kung University, Tainan, Taiwan; 4Dept. of Computer Science and Information Engineering, National Cheng-Kung University, Tainan, Taiwan
The fidelity of training data has been one major limitation for k-t reconstruction method at present. The traditional acquisition of the training data is performed in a separate scan, which may not exactly follow the same procedure of the real acquisition stage. Another acquisition scheme is variable density k-t sampling pattern that is the lines of central k-space are fully sampled while the peripheral lines are under-sampled. The training data can be extracted from central k lines, and the temporal resolution is maintained. However, the number of acquired central k lines for training data implies a tradeoff between the net accelerating ratio and the quality of training data. In this work, we propose to solve the reconstructing problem by exploiting data-driven method to acquire the prior knowledge of imaged object. The method uses the idea that each x-f profile can be transformed into a linear combination of features, which are the principal components of the distribution of x-f profiles. The principal components can be extracted from several existing homogenous data, instead of the data acquired directly from the imaged object. This is the main difference between traditional k-t method and the proposed method. We demonstrate its feasibility with numerical simulations of cine cardiac imaging. The results show that the proposed method can achieve comparable temporal resolution and leads to reduced artifacts even from substantially down-sampled k-space data.