University of Minnesota

Beth Israel Deaconess Medical Center, Inc.

Samsung Electronics Co., Ltd.

Mehmet Akcakaya

Steen Moeller

Reza Nezafat

Sebastian Weingartner

Warren J Manning

Jaime Shaw

Tamer Basha

Gulaka Praveen

Sebastien Roujol

Sudhir Ramanna

Sebastian Weingaertner

Seng-Wei Chieh

Chi Zhang

Burhaneddin Yaman

Seyed Amir Hossein Hosseini

Self-supervised training of machine learning ('ML') algorithms for reconstruction in inverse problems are described. These techniques do not require fully sampled training data. As an example, a physics-based ML reconstruction can be trained without requiring fully-sampled training data. In this way, such ML-based reconstruction algorithms can be trained on existing databases of undersampled images or in a scan-specific manner.

Systems and methods for training machine learning algorithms for inverse problems without fully sampled reference data

Images are reconstructed from undersampled k-space data using a residual machine learning algorithm (e.g., a ResNet architecture) to estimate missing k-space lines from acquired k-space data with improved noise resilience. Using a residual machine learning algorithm provides for combining the advantages of both linear and nonlinear k-space reconstructions. The linear residual connection can implement a convolution that estimates most of the energy in k-space, and the multi-layer machine learning algorithm can be implemented with nonlinear activation functions to estimate imperfections, such as noise amplification due to coil geometry, that arise from the linear component.

Methods for scan-specific artifact reduction in accelerated magnetic resonance imaging using residual machine learning algorithms

Methods for reconstructing images from undersampled k-space data using a machine learning approach to learn non-linear mapping functions from acquired k-space lines to generate unacquired target points across multiple coils are described.

Methods for scan-specific k-space interpolation reconstruction in magnetic resonance imaging using machine learning

Systems and methods for late gadolinium enhancement ('LGE') tissue viability imaging in a dynamic (e.g., temporally-resolved) manner using magnetic resonance imaging ('MRI') are provided. Dynamic LGE images can be generated throughout the entire cardiac cycle at high temporal resolution in a single breath-hold. Dynamic, semi-quantitative longitudinal relaxation maps are acquired and retrospective synthetization of dynamic LGE images is implemented using those semi-quantitative longitudinal relaxation maps.

System and method for producing temporally resolved images depicting late-gadolinium enhancement with magnetic resonance imaging

Magnetic resonance imaging ('MRI') data are corrected from corruptions due to physiological changes using a self-navigated phase correction technique. Unlike motion correction techniques, the effects of physiological changes (e.g., breathing and respiration) are corrected by making the MRI data self-consistent relative to an absolute uncorrupted phase reference. This phase correction information can be extracted from the acquisition itself, thereby eliminating the need for a separate navigator scan, and establishing an accelerated acquisition. This absolute reference can be computed in a data segmented space, and the subsequent data can be corrected relative to this absolute reference with low-resolution phases.

Efficient multi-shot EPI with self-navigated segmentation

A fully sampled calibration data set, which may be Cartesian k-space data, is used to obtain targeted and optimal interpolation kernels for non-regularly sampled data. The calibration data are self-calibration data obtained from a time-averaged image, or re-sampled data. ACS data are resampled for calibration of region-specific kernels. Subsequently, an explicit noise-based regularized solution can be utilized to estimate region-specific kernels for reconstruction.

Scalable self-calibrated interpolation of undersampled magnetic resonance imaging data

Described here are systems and methods for volumetric excitation in magnetic resonance imaging ('MRI') using frequency modulated radio frequency ('RF') pulses. In general, quadratic phase modulation along the slice encoding direction is implemented for additional spatiotemporal encoding, which better distributes signal content in the slice direction and enables higher acceleration rates that are robust to slice-undersampling.

Method for magnetic resonance imaging using slice quadratic phase for spatiotemporal encoding

Methods for fast magnetic resonance imaging ('MRI') using a combination of outer volume suppression ('OVS') and accelerated imaging, which may include simultaneous multislice (&#x201c;SMS&#x201d;) imaging, data acquisitions amenable to compressed sensing reconstructions, or combinations thereof. The methods described here do not introduce fold-over artifacts that are otherwise common to reduced field-of-view (&#x201c;FOV&#x201d;) techniques.

Methods and apparatus for scan time reductions in magnetic resonance imaging using outer volume supression

Systems and methods for producing quantitative maps of a longitudinal relaxation parameter, such as a longitudinal relaxation time ('T1'), using magnetic resonance imaging ('MRI') are described. More particularly, a pulse sequence and imaging method for cardiac phase-resolved myocardial T1 mapping are provided.

System and method for dynamic, cardiac phase-resolved quantitative longitudinal relaxation parameter mapping

A system and method for controlling noise in magnetic resonance imaging (MRI) are provided. In one aspect, the method includes reconstructing a series of images of the target using the image data, with each image being defined using signal-to-noise (SNR) units, and selecting an image patch corresponding to the series of images. The method also includes forming a matrix by combining vectors generated using the image patch, and applying a local low rank denoising technique using the matrix and the series of images to generate at least one denoised image.

System and method for controlling noise in magnetic resonance imaging using a local low rank technique

Systems and methods for mapping the transmit sensitivity of one or more radio frequency ('RF') coils for use in magnetic resonance imaging ('MRI') are described. The transmit RF field (&#x201c;B1+&#x201d;) for an RF coil, or an array of RF coils, is mapped using a robust, motion-insensitive technique that implements Bloch-Siegert shifts performed with interleaved positive and negative off-resonance shifts. The motion insensitivity of this technique makes it particularly useful for applications where there is significant motion, such as cardiac imaging, in which previous B1+ mapping techniques are not as accurate or effective.

Motion-robust transmit radio frequency field mapping in magnetic resonance imaging using interleaved bloch-siegert shifting

A system and method for obtaining magnetic resonance images are provided. The system is programmed to control the RF system to apply a saturation pulse at a reference frequency that saturates a selected labile spin species of the subject. The system is programmed to control the RF system to apply an inversion pulse after a variable delay. The system is programmed to control the RF system and the plurality of gradient coils to apply a motion sensitized driven equilibrium (MSDE) preparation pulse. The system is programmed to control the plurality of gradient coils to read imaging data during an acquisition time period. The system is programmed to reconstruct a Tmapping image of the subject with black-blood contrast.

System and method for reducing partial voluming artifacts in quantitative myocardial tissue characterization

An MRI apparatus includes: a data processor configured to acquire a first set of T-weighted imaging data and a second set of T-weighted imaging data; a pulse sequence controller configured to generate a pulse sequence and apply the generated pulse sequence to a gradient coil assembly and RF coil assembly, the generated pulse sequence including: T-preparation modules and associated imaging modules to acquire the first set of T-weighted imaging data, and a saturation pulse sequence and an associated saturation imaging module to acquire the second set of T-weighted imaging data; a curve fitter configured to apply the first and second sets of T-weighted imaging data to a three-parameter model for Tdecay, to determine a Tvalue at a plurality of locations; and an image processor configured to generate a Tmap of the object based on the Tvalue determined at the plurality of locations.

[object Object]

A magnetic resonance imaging (MRI) system and methods are provided for producing images of a subject. In some aspects, a method includes identifying a point in the cardiac cycle, performing an inversion recovery (IR) pulse at a selected time point from the pre-determined point, and sampling a k-space segment at an inversion time from the IR pulse that is substantially coincident with the pre-determined point. The method also includes repeating the IR pulse and k-space sampling for multiple inversion times, and multiple segments of k-space, in an interleaved manner, to generate datasets having T1-weighted contrasts determined by their respective inversion times. The method further includes reconstructing three-dimensional (3D) spatially-aligned images using the datasets, and generating a T1 recovery map by combining the 3D images. In some aspects, a prospective/retrospective scheme may be used to obtain data fully sampled in the center of k-space and randomly undersampled in the outer regions.

System and method for free-breathing volumetric imaging of cardiac tissue

A magnetic resonance (MR) image processing system includes a data collection unit that acquires image data from a target region of an object. A navigator unit acquires a motion signal indicating motion comprising motion of at least a portion of an object. A data processing unit derives k-space data for a k-space data array from the acquired image data, by acquiring k-space data for a first portion of the k-space from the acquired image data in response to the motion signal indicating motion is within a predetermined range and acquiring k-space data for a second portion of the k-space from the acquired image data irrespective of the predetermined range.

Growing community of inventors

Minneapolis, MN, United States of America

Mehmet Akcakaya