Abstract: A computerized system and method of modeling myocardial tissue perfusion can include acquiring a plurality of original frames of magnetic resonance imaging (MRI) data representing images of a heart of a subject and developing a manually segmented set of ground truth frames from the original frames. Applying training augmentation techniques to a training set of the originals frame of MRI data can prepare the data for training at least one convolutional neural network (CNN). The CNN can segment the training set of frames according to the ground truth frames. Applying the respective input test frames to a trained CNN can allow for segmenting an endocardium layer and an epicardium layer within the respective images of the input test frames. The segmented images can be used in calculating myocardial blood flow into the myocardium from segmented images of the input test frames.

Abstract: Systems and methods for denoising a magnetic resonance (MR) image utilize an unsupervised deep convolutional neural network (U-DCNN). Magnetic resonance (MR) image data of an area of interest of a subject can be acquired, which can include noisy input images that comprise noise data and noise free image data. For each of the noisy input images, iterations can be run of a converging sequence in an unsupervised deep convolutional neural network. In each iteration, parameter settings are updated; the parameter settings are used in calculating a series of image feature sets with the U-DCNN. The image feature sets predict an output image. The converging sequence of the U-DCNN is terminated before the feature sets predict a respective output image that replicates all of the noise data from the noisy input image. Based on a selected feature set, a denoised MR image of the area of interest of the subject can be output.

Abstract: Methods, computing devices, and magnetic resonance imaging systems that improve image quality in turbo spiral echo (TSE) imaging are disclosed. With this technology, a TSE pulse sequence is generated that includes a series of radio frequency (RF) refocusing pulses to produce a corresponding series of nuclear magnetic resonance (NMR) spin echo signals. A gradient waveform including a plurality of segments is generated. The plurality of segments collectively comprise a spiral ring retraced in-out trajectory. During an interval adjacent to each of the series of RF refocusing pulses, a first gradient pulse is generated according to the gradient waveform. The first gradient pulses encode the NMR spin echo signals. An image is then constructed from digitized samples of the NMR spin echo signals obtained based at least in part on the encoding.

Abstract: Methods, computing devices, and MRI systems that reduce artifacts produced by Maxwell gradient terms in TSE imaging using non-rectilinear trajectories are disclosed. With this technology, a RF excitation pulse is generated to produce transverse magnetization that generates a NMR signal and a series of RF refocusing pulses to produce a corresponding series of NMR spin-echo signals. An original encoding gradient waveform comprising a non-rectilinear trajectory is modified by adjusting a portion of the original encoding gradient waveform or introducing a zero zeroth-moment waveform segment at end(s) of the original encoding gradient waveform. During an interval adjacent to each of the series of RF refocusing pulses a first gradient pulse is generated. At least one of the first gradient pulses is generated according to the modified gradient waveform. An image is constructed from generated digitized samples of the NMR spin-echo signals obtained.

Abstract: A method of cardiac strain analysis uses displacement encoded magnetic resonance image (MRI) data of a heart of the subject and includes generating a phase image for each frame of the displacement encoded MRI data. Phase images include potentially phase-wrapped measured phase values corresponding to pixels of the frame. A convolutional neural network CNN computes a wrapping label map for the phase image, and the wrapping label map includes a respective number of phase wrap cycles present at each pixel in the phase image. Computing an unwrapped phase image includes adding a respective phase correction to each of the potentially-wrapped measured phase values of the phase image, and the phase correction is based on the number of phase wrap cycles present at each pixel. Computing myocardial strain follows by using the unwrapped phase image for strain analysis of the subject.

Abstract: Training a neural network to correct motion-induced artifacts in magnetic resonance images includes acquiring motion-free magnetic resonance image (MRI) data of a target object and applying a spatial transformation matrix to the motion-free MRI data. Multiple frames of MRI data are produced having respective motion states. A Non-uniform Fast Fourier Transform (NUFFT) can be applied to generate respective k-space data sets corresponding to each of the multiple frames of MRI; the respective k-space data sets can be combined to produce a motion-corrupted k-space data set and an adjoint NUFFT can be applied to the motion-corrupted k-space data set. Updated frames of motion-corrupted MRI data can be formed. Using the updated frames of motion corrupted MRI data, a neural network can be trained that generates output frames of motion free MRI data; and the neural network can be saved.

Abstract: Blurring and noise artifacts in magnetic resonance (MR) images caused by off-resonant image components may be corrected with convolutional neural networks, particularly feed forward networks with skip connections. Demodulating complex blurred images with off-resonant artifacts at a selected number of frequencies forms a respective real component frame of the MR data and a respective imaginary component frame for each image. A convolutional neural network is used to de-blur the images. The network has a plurality of residual blocks with multiple convolution calculations paired with respective skip connections. The method outputs, from the convolutional neural network, a de-blurred real image frame and a de-blurred imaginary image frame of the MR data for each complex blurred image.

Abstract: In one aspect the disclosed technology relates to embodiments of a method which, includes acquiring magnetic resonance imaging data, for a plurality of images, of the heart of a subject. The method also includes segmenting, using cascaded convolutional neural networks (CNN), respective portions of the images corresponding to respective epicardium layers and endocardium layers for a left ventricle (LV) and a right ventricle (RV) of the heart. The segmenting is used for extracting biomarker data from segmented portions of the images and, in one embodiment, assessing hypertrophic cardiomyopathy from the biomarker data. The method further includes segmenting processes for T1 MRI data and LGE MRI data.

Abstract: MR image data can be improved by using a complex de-noising convolutional neural network such as a non-blind C-DnCNN, a network for MRI denoising that leverages complex-valued data with phase information and noise level information to improve denoising performance in various settings. The proposed method achieved superior performance on both simulated and in vivo testing data compared to other algorithms. The utilization of complex-valued operations allows the network to better exploit the complex-valued MRI data and preserve the phase information. The MR image data is subject to complex de-noising operations directly and simultaneously on both real and imaginary parts of the image data. Complex and real values are also utilized for block normalization and rectified linear units applied to the noisy image data. A residual image is predicted by the C-DnCNN and a clean MR image is available for extraction.

Abstract: Systems and methods for performing ungated magnetic resonance imaging are disclosed herein. A method includes producing magnetic resonance image MRI data by scanning a target in a low magnetic field with a pulse sequence having a spiral trajectory; sampling k-space data from respective scans in the low magnetic field and receiving at least one field map data acquisition and a series of MRI data acquisitions from the respective scans; forming a field map and multiple sensitivity maps in image space from the field map data acquisition; forming target k-space data with the series of MRI data acquisitions; forming initial magnetic resonance images in the image domain by applying a Non-Uniform Fast Fourier Transform to the target k-space data; and forming reconstructed images with a low rank plus sparse (L+S) reconstruction algorithm applied to the initial magnetic resonance images.

Abstract: Aspects of the present disclosure relate to systems and methods for medical imaging that incorporate prior knowledge. Some aspects relate to incorporating prior knowledge using a non-local means filter. Some aspects relate to incorporating prior knowledge for improved perfusion imaging, such as those incorporating arterial spin labeling.

Abstract: Described herein are systems, methods, and computer-readable medium for magnetic resonance (MR) based thermometry. In one aspect, in accordance with one embodiment, a method for magnetic resonance based thermometry includes: acquiring, by a variable flip-angle T1 mapping sequence, MR data in an area of interest of a subject that is heated by the application of focused ultrasound (FUS) to the brain of the subject, where the MR data includes T1 values over time, and where the acquisition of the MR data includes applying an accelerated three-dimensional ultra-short spiral acquisition sequence with a nonselective excitation pulse; and determining, based at least in part on a mathematical relationship established by T1 mapping thermometry, a temperature change in the area of interest over time, and where the temperature change is caused at least in part by a change in the applied FUS.

Abstract: A method and an apparatus for automatic muscle segmentation are provided. The method includes receiving a high-resolution three-dimensional (3D) image obtained by a magnetic resonance imaging (MM) system and splitting the high-resolution 3D image into high-resolution 3D sub-images. Furthermore, the method includes acquiring low-resolution 3D sub-images of the high-resolution 3D sub-images and determining a boundary box within each corresponding low-resolution 3D sub-image. Moreover, the method includes determining a crop box for each boundary box, cropping each high-resolution 3D sub-image based on its corresponding crop box, and segmenting at least one muscle from the cropped box high-resolution 3D sub-image.

Abstract: Methods, computing devices, and magnetic resonance imaging systems that improve image quality in turbo spiral echo (TSE) imaging are disclosed. With this technology, a TSE pulse sequence is generated that includes a series of radio frequency (RF) refocusing pulses to produce a corresponding series of nuclear magnetic resonance (NMR) spin echo signals. A gradient waveform including a plurality of segments is generated. The plurality of segments collectively comprise a spiral ring retraced in-out trajectory. During an interval adjacent to each of the series of RF refocusing pulses, a first gradient pulse is generated according to the gradient waveform. The first gradient pulses encode the NMR spin echo signals. An image is then constructed from digitized samples of the NMR spin echo signals obtained based at least in part on the encoding.

Abstract: Training a neural network to correct motion-induced artifacts in magnetic resonance images includes acquiring motion-free magnetic resonance image (MRI) data of a target object and applying a spatial transformation matrix to the motion-free MRI data. Multiple frames of MRI data are produced having respective motion states. A Non-uniform Fast Fourier Transform (NUFFT) can be applied to generate respective k-space data sets corresponding to each of the multiple frames of MRI; the respective k-space data sets can be combined to produce a motion-corrupted k-space data set and an adjoint NUFFT can be applied to the motion-corrupted k-space data set. Updated frames of motion-corrupted MRI data can be formed. Using the updated frames of motion corrupted MRI data, a neural network can be trained that generates output frames of motion free MRI data; and the neural network can be saved.

Abstract: Methods, computing devices, and MRI systems that reduce artifacts produced by Maxwell gradient terms in TSE imaging using non-rectilinear trajectories are disclosed. With this technology, a RF excitation pulse is generated to produce transverse magnetization that generates a NMR signal and a series of RF refocusing pulses to produce a corresponding series of NMR spin-echo signals. An original encoding gradient waveform comprising a non-rectilinear trajectory is modified by adjusting a portion of the original encoding gradient waveform or introducing a zero zeroth-moment waveform segment at end(s) of the original encoding gradient waveform. During an interval adjacent to each of the series of RF refocusing pulses a first gradient pulse is generated. At least one of the first gradient pulses is generated according to the modified gradient waveform. An image is constructed from generated digitized samples of the NMR spin-echo signals obtained.

Abstract: Systems and methods for performing ungated magnetic resonance imaging are disclosed herein. A method includes producing magnetic resonance image MRI data by scanning a target in a low magnetic field with a pulse sequence having a spiral trajectory; sampling k-space data from respective scans in the low magnetic field and receiving at least one field map data acquisition and a series of MRI data acquisitions from the respective scans; forming a field map and multiple sensitivity maps in image space from the field map data acquisition; forming target k-space data with the series of MRI data acquisitions; forming initial magnetic resonance images in the image domain by applying a Non-Uniform Fast Fourier Transform to the target k-space data; and forming reconstructed images with a low rank plus sparse (L+S) reconstruction algorithm applied to the initial magnetic resonance images.

Abstract: Disclosed herein are devices, systems, and methods for use in a magnetic resonance imaging (MRI)-guided procedure in which focused energy is applied to an area of interest of a subject. Disclosed herein are coupling baths comprising an aqueous solution comprising a plurality of paramagnetic particles dispersed in water, wherein, when magnetic resonance images are collected from the area of interest of the subject for MRI guidance: the coupling bath is located proximate to the area of interest, and the composition, the average particle size, the shape, the concentration, the presence or absence of the capping layer, the identity of the plurality of ligands when the capping layer is present, the average thickness of the capping layer when the capping layer is present, or a combination thereof is/are selected such that the coupling bath reduces or prevents imaging artifacts in the magnetic resonance images for the MRI guidance.

Abstract: Systems and methods for denoising a magnetic resonance (MR) image utilize an unsupervised deep convolutional neural network (U-DCNN). Magnetic resonance (MR) image data of an area of interest of a subject can be acquired, which can include noisy input images that comprise noise data and noise free image data. For each of the noisy input images, iterations can be run of a converging sequence in an unsupervised deep convolutional neural network. In each iteration, parameter settings are updated; the parameter settings are used in calculating a series of image feature sets with the U-DCNN. The image feature sets predict an output image. The converging sequence of the U-DCNN is terminated before the feature sets predict a respective output image that replicates all of the noise data from the noisy input image. Based on a selected feature set, a denoised MR image of the area of interest of the subject can be output.

Abstract: A computerized method of reconstructing acquired magnetic resonance image (MRI) data to produce a series of output images includes acquiring a multiband k-space data set from a plurality of multiband slices of spiral MRI data; simultaneously acquiring a single band k-space data set comprising respective single band spiral image slices that are each associated with a respective one of the multiband slices in the multiband k-space data set; using the single band k-space data set, for each individual multiband slice, calculating a respective calibration kernel to apply to the multi-band k-space data set for each individual multiband slice; separating each individual multiband slice from the multiband k space data set by phase demodulating the multi-band k-space data using multiband phase demodulation operators corresponding to the individual multiband slice and convolving phase demodulated multi-band k-space data with a selected convolution operator to form a gridded set of the multi-band k-space data correspond