The General Hospital Corporation

Berkin Bilgic

Kawin Setsompop

Lawrence Leroy Wald

Stephen Farman Cauley

Qiyuan Tian

Susie Yi Huang

Jonathan R Polimeni

Daniel Polak

Thomas Witzel

Huihui Ye

Zijing Zhang

Sohyun Han

JaeJin Cho

Mary Kate Manhard

Congyu Liao

Magnetic resonance imaging ('MRI') using a spin- and gradient-echo (“SAGE”) pulse sequence with blip up-down acquisition (“BUDA”) encoding enables distortion-free, high-resolution diffusion-weighted imaging and/or quantitative parameter mapping. Phase-encoding polarities are alternated across shots during a multi-shot acquisition. In each shot, multi-contrast data are acquired at echo times associated with a gradient echo, a mixed gradient-and-spin echo, and a spin echo. High in-plane resolution and distortion-free quantitative parameter maps can be generated, such as T2 maps. T2* maps, paramagnetic susceptibility maps, and diamagnetic susceptibility maps. Diffusion-weighted data can be acquired using diffusion encoding gradients and BUDA encoding, where multi-contrast data are acquired in the b=0 acquisition. Diffusion parameter maps can be generated from the b=0 and diffusion-weighted data.

Distortion-free diffusion and quantitative magnetic resonance imaging with blip up-down acquisition of spin- and gradient-echoes

Higher quality diffusion metrics and/or diffusion-weighted images are generated from lower quality input diffusion-weighted images using a suitably trained neural network (or other machine learning algorithm). High-fidelity scalar and orientational diffusion metrics can be extracted using a theoretical minimum of a single non-diffusion-weighted image and six diffusion-weighted images, achieved with data-driven supervised deep learning. As an example, a deep convolutional neural network ('CNN') is used to map the input non-diffusion-weighted image and diffusion-weighted images sampled along six optimized diffusion-encoding directions to the residuals between the input and output high-quality non-diffusion-weighted image and diffusion-weighted images, which enables residual learning to boost the performance of CNN and full tensor fitting to generate any scalar and orientational diffusion metrics.

Fast diffusion tensor MRI using deep learning

Super-resolution images are generated from standard-resolution images acquired with a magnetic resonance imaging ('MRI') system. More particularly, super-resolution (e.g., sub-millimeter isotropic resolution) images are generated from standard-resolution images (e.g., images with 1 mm or coarser isotropic resolution) using a deep learning algorithm, from which accurate cortical surface reconstructions can be generated.

Super-resolution anatomical magnetic resonance imaging using deep learning for cerebral cortex segmentation

Systems and methods are provided for improving MRI data acquisition efficiency while providing more detailed information with high resolution and isotropic resolution without gaps. Improved data acquisition efficiency may be achieved by implementing a machine learning algorithm with a hardware processor and a memory to estimate imperfections in fast imaging sequences, such as a multi-shot echo planar imaging (MS-EPI) sequence. These imperfections, such as patient motion, physiological noise, and phase variations, may be difficult to model or otherwise estimate using standard physics-based reconstructions.

Multi-shot echo planar imaging through machine learning

Methods for reducing scan time in magnetic resonance imaging ('MRI'), particularly when imaging three-dimensional image volumes, using a simultaneous time-interleaved multislice ('STIMS') acquisition are described. The unused time in each repetition time (&#x201c;TR&#x201d;) period is exploited to provide an additional reduction in encoding time for a three-dimensional acquisition (e.g., a 3D whole brain coverage). Groups of spatially interleaved slices are excited in a single TR, with the excitation and acquisition of the groups of slices being interleaved in time.

Method for simultaneous time-interleaved multislice magnetic resonance imaging

Systems and methods for simultaneously acquiring high-resolution images of a subject from multiple different slice locations using magnetic resonance imaging ('MRI') are described. The present invention overcomes the aforementioned drawbacks by providing method for producing a plurality of images of a subject with a magnetic resonance imaging ('MRI') system. A radio frequency (RF) excitation field is applied by the MRI system to a portion of a subject that includes a plurality of slice locations. First data are simultaneously acquired from each of the plurality of slice locations by the MRI system.

Systems and methods for slice dithered enhanced resolution simultaneous multislice magnetic resonance imaging

Systems and methods for simultaneous multislice ('SMS'} magnetic resonance imaging ('MRI'}, in which a random blip gradient encoding scheme is utilized to impart a different phase to each of a plurality of different slice locations. Because of the random blip gradient encoding, the amount of the imparted phase is randomized for each phase encoding step in a Cartesian k-space trajectory. This data acquisition strategy leads to incoherent aliasing artifacts across the simultaneously excited slices. Images of the individual slices can be reconstructed using a compressed sensing framework.

Simultaneous multislice MRI with random gradient encoding

Systems and methods for reconstructing images using a hierarchically semiseparable ('HSS') solver to compactly represent the inverse encoding matrix used in the reconstruction are provided. The reconstruction method includes solving for the actual inverse of the encoding matrix using a direct (i.e., non-iterative) HSS solver. This approach is contrary to conventional reconstruction methods that repetitively evaluate forward models (e.g., compressed sensing or parallel imaging forward models).

Systems and methods for fast magnetic resonance image reconstruction using a heirarchically semiseparable solver

Described here are systems and methods for quantitative susceptibility mapping ('QSM') using magnetic resonance imaging ('MRI'). Susceptibility maps are reconstructed from phase images using an automatic regularization technique based in part on variable splitting. Two different regularization parameters are used, one, &#x3bb;, that controls the smoothness of the final susceptibility map and one, &#x3bc;, that controls the convergence speed of the reconstruction. For instance, the regularization parameters can be determined using an L-curve heuristic to find the parameters that yield the maximum curvature on the L-curve. The &#x3bc; parameter can be determined based on an l-regularization and the &#x3bb; parameter can be determined based on the iterative l-regularization used to reconstruct the susceptibility map.

Growing community of inventors

Boston, MA, United States of America

Berkin Bilgic