Whole-Brain T2 Mapping with Accelerated Stack-of-Stars Acquisition Using Unsupervised Model-Based Implicit Neural Representation Networks

Motivation: Reducing the scan time for T2 mapping is of clinical significance. Goal(s): To develop a fast sampling sequence and an undersampling reconstruction algorithm to significantly reduce acquisition time. Approach: A rapid sampling sequence based on a golden-angle stack-of-stars trajectory is proposed, and a novel unsupervised deep learning algorithm specifically designed for this sequence is introduced. This algorithm utilizes multi-resolution hash encoding with implicit neural representation based on a physical model to significantly shorten scan time. Results: Whole-brain T2 mapping is achieved within a scan time of less than 1.4 minutes. Impact: A T2 quantitative sequence that accelerates scanning using radial trajectory is presented. An unsupervised deep-learning algorithm employing multi-resolution hash-encoding implicit neural representation is introduced to reconstruct T2 maps from undersampled data. This approach has potential to significantly shorten acquisition time.