High-resolution quantitative T2 mapping of the human brain at 7 T using a multi-echo spin-echo sequence and dictionary-based modeling

Abstract

Quantitative (Formula presented.) mapping offers a unique contrast for detailed brain imaging. At ultra-high field strengths (7 T), the higher signal-to-noise ratio (SNR) enables higher spatial resolution and the delineation of smaller structures. The translation of multi-echo spin-echo-based acquisitions to higher field strength, however, is complicated by inhomogeneities in the radio frequency (RF) transmit field resulting in stronger stimulated echoes and multi-echo refocusing pathways. The (Formula presented.) decay will thus depend on the specific sequence details and other experimental properties. The signal can be modeled by Bloch equation simulations to create a dictionary of possible signal patterns to fit the experimental data and estimate (Formula presented.). Particularly at smaller voxel sizes and shorter (Formula presented.) times, noise will affect the dictionary matching of the data by the introduction of a bias in the acquired signal magnitude dependent on the SNR. This study aims to develop a robust, accurate, and fast (Formula presented.) mapping approach at 7 T, addressing RF inhomogeneity and noise bias. We employed a 2D multi-echo spin-echo sequence combined with a Bloch equation simulation-aided dictionary matching technique. The method incorporated a pre-measured (Formula presented.) map for regularization of the dictionary fit and applied a patch-based PCA denoising algorithm with magnitude bias correction to mitigate noise-induced errors. The method was tested in simulations, phantom validations, and in five human participants. In vivo, (Formula presented.) isotropic high-resolution (Formula presented.) maps showed detailed contrast within cortical and subcortical areas. Notably, regions with high iron content, such as the substantia nigra or nucleus ruber, were distinctly visible. The proposed method provided consistent (Formula presented.) values across different brain regions that aligned well with the literature where available. Simulations and experiments demonstrated the importance of the noise correction to achieve high-quality maps. The proposed method can significantly contribute to studies on brain microstructure and pathology, since it produces reliable (Formula presented.) maps at high resolution.