In a magnetic resonance imaging procedure, multiple slices are initially spatially selectively excited in a first time interval by respective RF pulses followed by at least one RF refocusing pulse that causes one echo signal from each slice, with a time interval of two consecutive echo signals equal to the first time interval. A second RF refocusing pulse is emitted at a second time interval from the last echo signal that causes, one further echo signal per slice, with the time interval of two consecutive echo signals equal to the first time interval. At least one further RF refocusing pulse is emitted in a third time interval following the preceding RF refocusing pulse producing multiple temporally separated echo signals per refocusing pulse. The third time interval is selected so that the number of echo signals per RF refocusing pulse is twice the number of excited slices.

A magnetic resonance imaging (MRI) system, method and/or computer readable medium is configured to effect MR imaging with reduced artifact by generating one or more image reconstruction maps from one or more prescans, acquiring a main scan dataset from a main MRI scan of an object, warping one or more image reconstruction maps to have geometric distortions substantially corresponding to geometric distortions in the main scan dataset, and forming a diagnostic MR image of the object using the main scan dataset and the warped one or more image reconstruction maps.

Systems and methods for combined ghost artifact correction and parallel imaging reconstruction of simultaneous multislice (SMS) magnetic resonance imaging (MRI) data are provided. Dual-polarity training data are used to generate ghost-free slice data, which are used as target data in a reconstruction kernel training process. The training data are used as source data in the reconstruction kernel training. As a result, reconstruction kernels are computed, which can be used to reconstruct images from SMS data in which slice-specific ghosting artifacts are removed.

Disclosed herein is a medical system (100, 300) where execution of machine executable instructions (120) causes a computational system (104) to: receive (200) a time series of a R2-star map (122) for a brain volume (500); receive (202) a stimulus signal (124) descriptive of an occurrence of a sensory stimulus; receive (204) a selection of one or more seed voxels (126) identified in the time series of the R2-star map; calculate (206) a denoised time series of the R2-star map (128); calculate (208) a correlation map (130) between the seed voxels and the denoised time series of the R2-star map; determine (210) an activated region (132) of the brain volume using voxels identified in the correlation map; provide (212) a hemodynamic response (134) function for each voxel and each occurrence of the sensory stimulus; and provide (214) a subject specific hemodynamic response function (136) by averaging the hemodynamic response functions.

An echo planar imaging method capable of reducing image distortion is provided. Two radio frequency (RF) pulses and are used to respectively flip over a longitudinal magnetization into a transverse plane at different moments, so as to obtain observable transverse magnetization; the transverse magnetization obtained by flipping over the longitudinal magnetization by means of and are respectively rephased in two different echo trains by means of gradient pulses. Navigator echo signals are respectively collected before the two gradient echo trains, and the amplitude difference between two gradient echo train signals is corrected on the basis of the amplitude difference between the two navigator echo signals. Correcting the signals collected by the two gradient echo trains, so as to obtain two pieces of image data, and the two pieces of image data are averaged to obtain a final image.

A computer-implemented method of reducing noise in magnetic resonance (MR) images is provided. The method includes executing a neural network model of analyzing MR images, wherein the neural network model is trained with a pair of pristine images and corrupted images. The pristine images are the corrupted images with noise reduced, and target output images of the neural network model are the pristine images. The method also includes receiving first MR signals and second MR signals, reconstructing first and second MR images based on the first MR signals and the second MR signals, and analyzing the first MR image and the second MR image using the neural network model. The method further includes deriving a denoised MR image based on the analysis, wherein the denoised MR image is a combined image based on the first MR image and the second MR image and outputting the denoised MR image.

In one embodiment, a phase correction method comprising: acquiring first k-space data acquired in a first readout direction and second k-space data acquired in a second readout direction that is opposite to the first readout direction; weighting first real space data obtained from the first k-space data to generate first adjusted data with a predetermined weighting in which weight coefficients vary depending on a pixel position in a readout direction and become lower in a region where a variance of a phase difference is larger than a predetermined variance; weighting second real space data obtained from the second k-space data to generate second adjusted data with the predetermined weighting; calculating a correction amount for correcting a phase difference; and correcting a phase difference between data that are different from each other in polarity of a gradient pulse in the readout direction during acquisition, by using the correction amount.

For distortion correction in magnetic resonance (MR) echo planar imaging (EPI), a combination of supervised and unsupervised training provides deep learning with generalized applicability across different use cases as well as sufficient training data (e.g., less ground truth data is needed due to the inclusion of unsupervised learning). The neural network for displacement estimation may include deformable convolution and/or layers for a non-diffeomorphic displacement field. The network architecture features and/or combined supervised and unsupervised learning may be used together or individually.

A method can include obtaining a scaling factor for a location proximate a metallic object by optimizing a function of an acquired dataset and a simulated dataset. The simulated dataset can include a first signal from a first pulse having a first excitation flip angle and a first refocusing flip angle. The simulated dataset can include a second signal from a second pulse having a second excitation flip angle and a second refocusing flip angle.

A method for avoiding artifacts in measurement data captured using a magnetic resonance system which has a gradient unit. The method includes loading data which characterizes the gradient unit of the magnetic resonance system; loading a measurement protocol to be used for capturing the measurement data, wherein the measurement protocol includes gradients to be switched and RF excitation pulses and RF refocusing pulses to be irradiated, wherein, after irradiation of an RF excitation pulse, a train of at least two RF refocusing pulses is irradiated and measurement data is captured after each RF refocusing pulse; determining compensation gradients which, after the capture of the measurement data, are to be switched after a final RF refocusing pulse of the train of RF refocusing pulses associated with the RF excitation pulse and before a following RF excitation pulse as a function of the loaded measurement protocol and of the data which characterizes the gradient unit; and carrying out the measurement protocol using the determined compensation gradients.