Accelerating the Whole-Brain Multi-Parametric Imaging Through Joint Deep Learning Reconstruction and Physical Model Integration

The three-dimensional (3D) multi-parametric MR imaging technique MULTIPLEX quantifies T1/T2*/PD/QSM maps within a single scan. However, the whole brain MULTIPLEX imaging suffers from a long scan time due to its dual-TR, dual-flip angle, and multi-echo signal acquisition strategy, limiting the achievable spatial resolution. Although the traditional compressed sensing (CS) method can handle under-sampled data, it suffers from prolonged reconstruction time and cannot guarantee quantitative accuracy. This work develops a multi -contrast joint deep learning model integrated with the MR physical model, named Joint DeepMTP, significantly enhancing the accuracy of reconstruction and quantification. Our proposed method achieves comparable reconstruction and quantification performance to the reference at 9- fold CAIPI acceleration for 3D high-resolution MULTIPLEX imaging with a reconstruction time of 3 minutes.