Abstract: For image quality scoring of an image from a medical scanner, a generative model of an expected good quality image may be created using deep machine-learning. The deviation of an input image from the generative model is used as an input feature vector for a discriminative model. The discriminative model may also operate on another input feature vector derived from the input image. Based on these input feature vectors, the discriminative model outputs an image quality score.

Abstract: A method for performing Computed Tomography (CT) reconstruction includes acquiring a sparse measurement matrix using a CT scanner and applying a reconstruction process over a number of iterations to reconstruct image data from the sparse measurement matrix. The reconstruction process performed during each respective iteration includes generating a random view subset and determining a portion of the sparse measurement matrix corresponding to the random view subset. The reconstruction process further includes performing a stochastic gradient descent on the portion of the sparse measurement matrix to yield an image, applying a proximal total variation regularization to the image, and adjusting a step size associated with the Acquire CT sparse measurement stochastic gradient descent.

Abstract: A method for magnetic resonance (MR) imaging is provided. A first sampling mask is provided for sampling along a first set of parallel lines extending in a first direction in k-space. A second sampling mask is provided for sampling along a second set of parallel lines extending in a second direction in k-space. The second direction is orthogonal to the first direction. A first set of MR k-space data is sampled using an MR scanner, by scanning a subject in the first direction using the first sampling mask. A second set of MR k-space data is sampled using the MR scanner, by scanning the subject in the second direction using the second sampling mask. An MR image is reconstructed from a combined set of MR k-space data including the first set of MR k-space data and the second set of MR k-space data.

Abstract: A method for performing a magnetic resonance image reconstruction with spatially varying coil compression includes using a non-Cartesian acquisition scheme to acquire a multi-coil k-space dataset fully sampled along a fully sampled direction and decoupling the multi-coil k-space dataset along the fully sampled direction to yield a plurality of uncompressed coil data matrices. The plurality of uncompressed coil data matrices are compressed to yield a plurality of virtual coil data matrices which are aligned along the fully sampled direction to yield a plurality of aligned virtual coil data matrices. The aligned virtual coil data matrices are coupled along the fully sampled direction to yield a compressed multi-coil k-space dataset. Intensity values in the plurality of aligned virtual coil data matrices are normalized based on the plurality of uncompressed coil data matrices and an image is reconstructed using the compressed multi-coil k-space dataset.

Abstract: A computer-implemented method for sparse recovery of fiber orientations using a multidimensional Prony method for use in tractography applications includes performing magnetic resonance imaging to acquire a plurality of sparse signal measurements using a q-space sampling scheme which enforces a lattice structure with a predetermined number of collinear samples. Next, for each voxel included in the plurality of sparse signal measurements, a computer system is used to perform a parameter estimation process. This process includes translating a portion of the sparse signal measurements corresponding to the voxel into a plurality of Sparse Approximate Prony Method (SAPM) input parameters, and applying a SAPM process to the SAPM input parameters to recover a number of fiber populations, a plurality of orientation vectors, and a plurality of amplitude scalars.

Abstract: A computer-implemented method for providing image quality optimization individualized for a user includes a computer receiving raw image data acquired from an image scanner and identifying one or more raw image quality features based on the raw image data. The computer automatically determines one or more target image quality features by applying one or more user preferences to the one or more raw image quality features. The computer also automatically determines one or more processing parameters based on the one or more target image quality features. The computer may then process the raw image data using the one or more processing parameters to yield an image.

Abstract: A computer-implemented method of performing image reconstruction with sequential cycle-spinning includes a computer system acquiring an input signal comprising k-space data using a magnetic resonance imaging (MRI) device and initializing an estimate of a sparse signal associated with the input signal. The computer system selects one or more orthogonal wavelet transforms corresponding to a wavelet family and performs an iterative reconstruction process to update the estimate of the sparse signal over a plurality of iterations. During each iteration, one or more orthogonal wavelet transforms are applied to the estimate of the sparse signal to yield one or more orthogonal domain signals, the estimate of the sparse signal is updated by applying a non-convex shrinkage function to the one or more orthogonal domain signals, and a shift to the orthogonal wavelet transforms. Following the iterative reconstruction process, the computer system generates an image based on the estimate of the sparse signal.

Abstract: For image quality scoring of an image from a medical scanner, a generative model of an expected good quality image may be created using deep machine-learning. The deviation of an input image from the generative model is used as an input feature vector for a discriminative model. The discriminative model may also operate on another input feature vector derived from the input image. Based on these input feature vectors, the discriminative model outputs an image quality score.

Abstract: For correction of an image from an imaging system, a deep-learnt generative model is used as a regularlizer in an inverse solution with a physics model of the degradation behavior of the imaging system. The prior model is based on the generative model, allowing for correction of an image without application specific balancing. The generative model is trained from good images, so difficulty gathering problem-specific training data may be avoided or reduced.

Abstract: A system and method including receiving magnetic resonance (MR) imaging data from a first MR scanner device, the MR imaging data including data for a plurality of MR scans of different structural or anatomical regions; generating, based on the MR imaging data, normalized reference data including statistical information for each MR scan; learning a transformation, based on the normalized reference data, to correlate a set of input MR imaging data to the normalized reference data; and storing a record of the transformed imaging data.

Abstract: Machine training and application of machine-trained classifier are used for image-based tumor phenotyping in a medical system. To create a training database with known phenotype information, synthetic medical images are created. A computational tumor model creates various examples of tumors in tissue. Using the computational tumor model allows one to create examples not available from actual patients, increasing the number and variance of examples used for machine-learning to predict tumor phenotype. A model of an imaging system generates synthetic images from the examples. The machine-trained classifier is applied to images from actual patients to predict tumor phenotype for that patient based on the knowledge learned from the synthetic images.

Abstract: A computer-implemented method for denoising image data includes a computer system receiving an input image comprising noisy image data and denoising the input image using a deep multi-scale network comprising a plurality of multi-scale networks sequentially connected. Each respective multi-scale network performs a denoising process which includes dividing the input image into a plurality of image patches and denoising those image patches over multiple levels of decomposition using a threshold-based denoising process. The threshold-based denoising process denoises each respective image patch using a threshold which is scaled according to an estimation of noise present in the respective image patch. The noising process further comprises the assembly of a denoised image by averaging over the image patches.

Abstract: A method comprises acquiring navigation data during navigation scans at each of a plurality of points in time. A plurality of magnetic resonance imaging (MM) k-space data corresponding to an imaged object are acquired at the plurality of points in time using Cartesian sampling, the k-space data including at least two spatial dimensions, a time. The respective motion state for each of the k-space data are computed based on the navigation data. At least one image is reconstructed from the plurality MM k-space data using k-space data corresponding to at least two motion states and the same point in time to reconstruct the at least one image.

Abstract: A learning-based magnetic resonance fingerprinting (MRF) reconstruction method for reconstructing an MR image of a tissue space in an MR scan subject for a particular MR sequence is disclosed. The method involves using a machine-learning algorithm that has been trained to generate a set of tissue parameters from acquired MR signal evolution without using a dictionary or dictionary matching.

Abstract: Disclosed herein is a method obtaining a magnetic resonance image of an object, comprising obtaining a first time evolution signal from a magnetic resonance signal from the object; performing a search of a compressed dictionary of magnetic resonance fingerprints to select a magnetic resonance fingerprint representative of the first time evolution signal, wherein the selected magnetic resonance fingerprint is an exact or approximate nearest neighbor match to the first time evolution signal; obtaining a magnetic resonance parameter associated with the selected fingerprint; generating the magnetic resonance image of the object from the obtained magnetic resonance parameter; and performing a second search of the compressed dictionary using the magnetic resonance image.

Abstract: A method of generating Magnetic Resonance (MR) parameter maps includes creating one or more parameter maps, each respective parameter map comprising initial parameter values associated with one of a plurality of MR parameters. A dynamical update process is performed over a plurality of time points. The dynamical update process performed at each respective time point includes applying a randomized pulse sequence to subject using an MR scanner to acquire a k-space dataset. This randomized pulse sequence is configured to excite a distinct range of values associated with the plurality of MR parameters. The dynamical update process further includes applying a reconstruction process to the k-space dataset to generate an image and using a tracking process to update the one or more parameter maps based on the randomized pulse sequence and the image.

Abstract: A computer-implemented method for providing image quality optimization individualized for a user includes a computer receiving raw image data acquired from an image scanner and identifying one or more raw image quality features based on the raw image data. The computer automatically determines one or more target image quality features by applying one or more user preferences to the one or more raw image quality features. The computer also automatically determines one or more processing parameters based on the one or more target image quality features. The computer may then process the raw image data using the one or more processing parameters to yield an image.

Abstract: A method for denoising Magnetic Resonance Imaging (MRI) data includes receiving a noisy image acquired using an MRI imaging device and determining a noise model comprising a non-diagonal covariance matrix based on the noisy image and calibration characteristics of the MRI imaging device. The noisy image is designated as the current best image. Then, an iterative denoising process is performed to remove noise from the noisy image.

Abstract: A method for denoising magnetic resonance images includes estimating a normalization field corresponding to a magnetic resonance imaging device and acquiring a non-normalized image from the magnetic resonance imaging device. A noise level estimation process is performed with the non-normalized image to yield a noise level. The normalization field is applied to the noise level to yield a potentially inhomogeneous noise-level map and to the non-normalized image to yield a normalized image. An adaptive polynomial filtering process is performed using the normalized image and the potentially inhomogeneous noise-level map to yield a denoised image.

Abstract: A method for denoising Magnetic Resonance Imaging (MRI) data includes receiving a noisy image acquired using an MRI imaging device and determining a noise model comprising a non-diagonal covariance matrix based on the noisy image and calibration characteristics of the MRI imaging device. The noisy image is designated as the current best image. Then, an iterative denoising process is performed to remove noise from the noisy image.