UC Berkeley

In magnetic resonance imaging (MRI), the human body’s magnetic susceptibility causes a slight resonance frequency shift of the nuclear spins. This frequency shift is reflected in the phase of MRI signals. By solving a magnetic dipole model using the frequency shift map, we can recover the tissue’s local magnetic susceptibilities using a method called quantitative susceptibility mapping (QSM). However, QSM is inaccurate if non-homogeneous magnetic susceptibility sources exist within one imaging voxel. To address this issue, we developed a compartmentalized tissue signal model named DECOMPOSE-QSM.

The DECOMPOSE-QSM considers three susceptibility components within one imaging voxel: paramagnetic component, diamagnetic component, and neutral component which is serving as the susceptibility reference. The multi-echo gradient echo signal of a voxel with a susceptibility mixture can be expressed as the summation of the complex single exponential signal coming from each component. By solving for the parameters that characterize this summation of three complex exponentials, paramagnetic and diamagnetic component susceptibility maps can be resolved. An optimization-based solver is proposed to solve this highly non-linear model. The solver alternates between solving three sub-problems: 1) a constraint linear least square problem, 2) a log-modified constraint linear least square problem, and 3) a log-modified nonlinear constraint least square problem. Additionally, in an attempt to accelerate the calculation, a deep learning-based solver is developed, which uses a multi-layer perceptron framework to explore the signal model's behavior in higher dimensions. The DECOMPOSE-QSM is validated by imaging susceptibility mixture gel phantoms made in-house and verifying the Curie's Law relation for the paramagnetic components with a series of temperature-varying imaging experiments.

Using the DECOMPOSE-QSM technique, the anisotropic paramagnetic susceptibility in the brain was discovered with a multi-orientation dataset on a postmortem chimpanzee brain. The paramagnetic susceptibility anisotropy is believed to be reflecting a microstructural order of oligodendrocytes wrapping around axons. The anisotropy due to the arrangements of oligodendrocytes has been observed in previous studies using polarized light imaging. Our finding is the first observation of such anisotropy using an MRI-based technique.