Temporal Super-Resolution for Fast T1 Mapping.

Xunkang Zhao,Jun Lyu,Fanwen Wang,Chengyan Wang,Jing Qin

DOI: https://doi.org/10.1007/978-3-031-52448-6_42

2024-01-01

Abstract:Cardiac T1 mapping can provide important biomarkers for many cardiovascular diseases. However, the acquisition time of the T1 mapping sequence is relatively long, which not only affects image quality but also causes discomfort to patients. Existing strategies to accelerate T1 mapping imaging mostly involve reducing the sampling at each time point and using reconstruction algorithms to predict complete images from undersampled k-space data. These in-plane acceleration methods often neglect the correlation between frames in the time series and may introduce more artifacts with increasing acceleration factors, without effectively reducing the overall scan time. We propose a novel inter-plane acceleration method for T1 mapping, which is completely different from the in-plane acceleration methods. This method can truly shorten the scan time. Specifically, instead of using reconstruction algorithms at each sampling time point, we reduce the number of sampling time points and employ a time super-resolution algorithm to generate missing time point images. For this purpose, we design a cross-frame adaptive enhancement network to achieve temporal super-resolution, which effectively utilizes input frame features to generate missing frame features and employs cross-frame deformable attention mechanism to enrich the expression of missing frame features, enhancing the reconstruction of missing frames and avoiding artifacts. Experimental results on the CMRxRecon T1 mapping dataset demonstrate that the proposed network predicts missing frames with clearer details and fewer artifacts compared to the comparative algorithms. This method opens up a new pathway for accelerating T1 mapping imaging and holds promise for clinical applications.