Sequential model-based free-water suppression of T1, T2, and proton density maps for synthetic FLAIR and DIR MRI.
Abstract
Synthetic MRI (SynMRI) enables flexible generation of contrast-weighted images from quantitative T1, T2, and proton density (PD) maps. However, synthetic inversion-recovery (IR) images are often degraded by cerebrospinal fluid (CSF) partial-volume effects (PVE). We developed a sequential model-based framework for suppressing free-water signals in quantitative maps and improving synthetic IR imaging and tissue volume fraction mapping. Using four spin-echo acquisitions, including a heavily T2-weighted image, the free-water component was estimated from a physical signal model and sequentially removed before conventional quantitative map estimation. The resulting water-suppressed (Wsup) quantitative maps enabled generation of synthetic FLAIR using a conventional spin-echo signal model and synthetic WAIR and GAIR using either double-inversion recovery (DIR) or a simplified single-inversion recovery (SIR) model without requiring joint nonlinear optimization, specialized pulse sequences, or dedicated hardware. Digital phantom simulations showed marked reductions in relative error for synthetic IR images in mixed-tissue voxels, from 370% to 20% for WAIR and from 300% to 12% for GAIR, with further reductions to 5.4% and 3.0%, respectively, using the simplified SIR strategy. Synthetic FLAIR also improved gray matter-white matter contrast, while tissue volume fraction maps closely matched the ground truth. A proof-of-concept study in a healthy volunteer demonstrated reduced CSF-related partial-volume artifacts, improved synthetic IR image contrast in CSF partial-volume regions, and quantitative parameters in CSF-PVE regions that approached those of the corresponding pure tissues. The proposed sequential model-based framework demonstrates the feasibility of a practical and computationally efficient approach for free-water suppression, synthetic IR imaging, and tissue volume fraction mapping from only four acquired images.