Magnetic resonance fingerprinting (MRF) has been proposed as a technique to quantify tissue parameters, such as T1 and T2 relaxation times, which are biomarkers for various pathologies. One assumption of MRF is that the signal in each voxel is generated by exactly one set of tissue parameters. Due to MRI resolution at the range of millimeter in combination with the cellular structure of biological tissue, each voxel consists of multiple tissue compartments. An over-simplified single compartment model results in apparent relaxation times that are influenced by the relaxation times and the fractional proton densities of all contributing compartments. This can lead to a misinterpretation of signal changes. For example, in diseases that causes demyelination in white matter (Multiple Sclerosis, Dementia), a reduction of the myelin water fraction would result in a misleading change of the apparent relaxation time of the voxel. We propose a multi-compartment MRF method that allows to identify multiple tissue contributions within a voxel, including the fractional proton density (PD) of different compartments. Our machine learning based approach automatically identifies the number of compartments within each voxel that can be identified with the available SNR in that voxel. We will correct for partial-volume effects at the borders of two types of tissues, as well as analyze tissue microstructure. For the second case our learned model will also include chemical exchange between compartments. After an initial validation phase using numerical simulations, we will first perform MRF scans of dedicated 3D printed phantoms with multiple compartments. Our quality criterion is successful estimation of all simulated tissue compartments for all voxels with a relative error of less than 5% to the ground truth. We will then perform in-vivo MRF measurements of healthy volunteers (n=5). We will generate synthetic FLAIR and MP-RAGE contrasts from parameter maps estimated with conventional and the proposed multi-compartment MRF technique. We will compare them with currently used clinical contrasts acquired using established pulse sequences and validate the performance of our approach by measuring the cortical thickness. Further, we will validate the performance for microstructure composition in white matter. Our hypothesis is that it will be possible separate the compartments for myelin, intra- and extra-cellular water and compare the results to ex-vivo data found in literature. In summary, the methods developed in this R21 proposal will provide a novel technique to accurately and reproducibly identify biomarkers beyond the resolution of a voxel. It will allow to identify changes in tissue composition and fractional proton density at the microstructure level.
The overarching goal of this proposal is to establish a MR imaging technique for quantifying biomarkers at the sub-voxel level. We will develop a machine learning based MR fingerprinting image reconstruction that enables the identification and characterization of multiple compartments within a voxel in terms of fractional proton density and relaxation times. We will generate accurate synthetic fluid suppressed images such as the FLAIR and MP- RAGE with accurate partial volume behavior and analyze the myelin water fraction and the corresponding relaxation times in white matter.
