LINEAR: Studying Implicit Neural Illustration With Express Bodily Priors for Accelerated Quantitative T1rho Mapping
Authors: Yuanyuan Liu, Jinwen Xie, Zhuo-Xu Cui, Qingyong Zhu, Jing Cheng, Dong Liang, Yanjie Zhu
Summary: Quantitative T1rho parameter mapping has proven promise in medical and analysis research. Nevertheless, it suffers from lengthy scan instances. Deep learning-based strategies have been efficiently utilized in accelerated quantitative MR parameter mapping. Nevertheless, most strategies require fully-sampled coaching dataset, which is impractical within the clinic. On this examine, a novel subject-specific unsupervised methodology primarily based on the implicit neural illustration is proposed to reconstruct photos from extremely undersampled k-space information and estimate parameter maps from reconstructions, which solely takes spatiotemporal coordinates because the enter. Particularly, the proposed methodology discovered a implicit neural illustration of the MR photos pushed by two express priors of photos (or k-space information), together with the low-rankness of Hankel matrix, and the self-consistency of k-space information. The ablation experiments present that the proposed methodology can characterize the bodily priors of MR photos effectively. Furthermore,experimental outcomes of retrospective and potential information present that the proposed methodology outperforms the state-of-the-art strategies when it comes to supressing artifacts and attaining the bottom error
