General Electric Company

Ge Precision Healthcare LLC

Sangtae Ahn

Christopher Judson Hardy

Ravindra Mohan Manjeshwar

Steven Gerard Ross

Rafael Shmuel Brada

Itzik Malkiel

Ron Wein

Evren Asma

Michael Rotman

Bruno Kristiaan Bernard De Man

Dattesh Dayanand Shanbhag

Florian Wiesinger

Sandeep Suryanarayana Kaushik

Isabelle Heukensfeldt Jansen

Floribertus P M Heukensfeldt Jansen

Nitin Jain

Lin Fu

Xue Rui

Sathish Ramani

Hua Qian

David Erik Chevalier

Dirk Bequé

Anne Menini

Uri Wollner

Graeme C Mckinnon

Lishui Cheng

Kris Thielemans

A computer-implemented method of processing complex magnetic resonance (MR) images is provided. The method includes receiving a pair of corrupted complex data and pristine complex images. The method also includes training a neural network model using the pair by inputting the corrupted complex data to the neural network model, setting the pristine complex images as target outputs, and processing the corrupted complex data using the neural network model to derive output complex images of the corrupted complex data. Training a neural network model also includes comparing the output complex images with the target outputs by computing a phase-sensitive structural similarity index measure (PS-SSIM) between each of the output complex images and its corresponding target complex image, wherein the PS-SSIM is real-valued and varies with phases of the output complex image and phases of the target complex image, and adjusting the neural network model based on the comparison.

Systems and methods of processing magnetic resonance images using phase-sensitive structural similarity index measure

K-space data obtained from a magnetic resonance imaging scan where motion was detected is split into two parts in accordance with the timing of the motion to produce first and second sets of k-space data corresponding to different poses. Sub-images are reconstructed from the k first and second sets of k-space data, which are used as inputs to a deep neural network which transforms them into a motion-corrected image.

System and method for correcting for patient motion during MR scanning

A magnetic resonance (MR) imaging method of detecting and scoring motion artifacts in MR images of an object is provided. The method includes computing a k-space difference map based at least in part on first MR signals of the object acquired with a first coil and second MR signals of the object acquired with a second coil. The method also includes generating a difference plot based on the k-space difference map, the difference plot including a curve. The method further includes calculating a motion score based on the curve in the difference plot, wherein the motion score indicates the level of motion artifacts in the image caused by motion of the object during acquisition of the first MR signals and the second MR signals, and the motion score includes an area under the curve. Moreover, the method includes outputting the motion score.

Systems and methods of automatically detecting and scoring motion artifacts in magnetic resonance imaging

The subject matter discussed herein relates to a fast magnetic resonance imaging (MRI) method to suppress fine-line artifact in Fast-Spin-Echo (FSE) images reconstructed with a deep-learning network. The network is trained using fully sampled NEX=2 (Number of Excitations equals to 2) data. In each case, the two excitations are combined to generate fully sampled ground-truth images with no fine-line artifact, which are used for comparison with the network generated image in the loss function. However, only one of the excitations is retrospectively undersampled and inputted into the network during training. In this way, the network learns to remove both undersampling and fine-line artifacts. At inferencing, only NEX=1 undersampled data are acquired and reconstructed.

Reconstruction of MR image data

A magnetic resonance imaging (MRI) system includes control and analysis circuitry having programming to acquire magnetic resonance (MR) data using coil elements of the MRI system, analyze the MR data, and reconstruct the MR data into MR sub-images. The system also includes a trained neural network associated with the control and analysis circuitry to transform the MR sub-images into a prediction relating to a presence and extent of motion corruption in the MR sub-images. The programming of the control and analysis circuitry includes instructions to control operations of the MRI system based at least in part on the prediction of the trained neural network.

MRI system and method using neural network for detection of patient motion

The present disclosure relates to performing an unrolled iterative reconstruction of image data. Such approaches may include use of neural networks used for one or both of data fidelity and/or image update steps, resulting in improved image quality and accelerated reconstruction for various imaging modalities including CT, PET and MR.

Iterative image reconstruction

A system and method for detecting, timing, and adapting to patient motion during an MR scan includes using the inconsistencies between calculated images from different coil-array elements to detect the presence of patient motion and, together with the k-space scan-order information, determine the timing of the motion during the scan. Once the timing is known, various actions may be taken, including restarting the scan, reacquiring those portions of k-space acquired before the movement, or correcting for the motion using the existing data and reconstructing a motion-corrected image from the data.

MRI system and method for detection and correction of patient motion

A method for sparse image reconstruction includes acquiring coil data from a magnetic resonance imaging device. The coil data includes undersampled k-space data corresponding to a subject. The method further includes processing the coil data using an image reconstruction technique to generate an initial undersampled image. The method also includes generating a reconstructed image based on the coil data, the initial undersampled image, and a plurality of iterative blocks of a flared network. A first iterative block of the flared network receives the initial undersampled image. Each of the plurality of iterative blocks includes a data consistency unit and a regularization unit and the iterative blocks are connected both by direct connections from one iterative block to the following iterative block and by a plurality of dense skip connections to non-adjacent iterative blocks. The flared network is based on a neural network trained using previously acquired coil data.

System and method for sparse image reconstruction

An imaging system is provided that includes at least one detector configured to acquire imaging information, a processing unit, and a display unit. The processing unit is operably coupled to the at least one detector, and is configured to reconstruct an image using the imaging information. The image is organized into voxels having non-uniform dimensions. The processing unit is configured to perform a penalized likelihood (PL) image reconstruction using the imaging information. The PL image reconstruction includes a penalty function. Performing the penalty function includes interpolating a voxel size in at least one dimension from an original size to an interpolated size before determining a penalty function, determining the penalty function using the interpolated size to provide an initial penalty, interpolating the initial penalty to the original size to provide a modified penalty, and applying the modified penalty in the PL image reconstruction.

Systems and methods for imaging with anisotropic voxels

The present approach relates to the use of machine learning and deep learning systems suitable for solving large-scale, space-variant tomographic reconstruction and/or correction problems. In certain embodiments, a tomographic transform of measured data obtained from a tomography scanner is used as an input to a neural network. In accordance with certain aspects of the present approach, the tomographic transform operation(s) is performed separate from or outside the neural network such that the result of the tomographic transform operation is instead provided as an input to the neural network. In addition, in certain embodiments, one or more layers of the neural network may be provided as wavelet filter banks.

Tomographic reconstruction based on deep learning

A method includes acquiring scan data for an object to be imaged using an imaging scanner. The method also includes reconstructing a display image using the scan data. Further, the method includes determining one or more aspects of a quantitation imaging algorithm for generating a quantitation image, wherein the one or more aspects of the quantitation imaging algorithm are selected to optimize a quantitation figure of merit for lesion quantitation. The method also includes reconstructing a quantitation image using the scan data and the quantitation imaging algorithm; displaying, on a display device, the display image; determining a region of interest in the display image; determining, for the region of interest, a lesion quantitation value using a corresponding region of interest of the quantitation image; and displaying, on the display device, the lesion quantitation value.

Systems and methods for emission tomography quantitation

Aspects of the invention relate to generating an emission activity image as well as an emission attenuation map using an iterative updation based on both the raw emission projection data and the raw radiography projection data, and an optimization function. The outputs include an optimized emission activity image, and at least one of an optimized emission attenuation map or an optimized radiography image. In some aspects an attenuated corrected emission activity image is obtained using the optimized emission activity image, and the optimized emission attenuation map.

Systems and methods for reconstruction of emission activity image

A computer-implemented method for penalized-likelihood reconstruction of a Positron Emission Tomography (PET) image includes generating a regularization function in which a smoothing parameter is modulated by one or more data-independent spatially variable modulation factors to compensate for sensitivity variations in a PET voxel dataset, and reconstructing the PET image from the PET emission dataset using the regularization function.

Systems and methods for PET penalized-likelihood image reconstruction with spatially varying smoothing parameters

Imaging system and method are presented. Emission scan (ES) and anatomical scan (AS) data corresponding to a target volume in a subject are received. One or more at least partial AS images are reconstructed using AS data. An image-space certainty (IC) map representing a confidence level (CL) for attenuation coefficients of selected voxels in AS images and a preliminary attenuation (PA) map based on AS images are generated. One or more of selected attenuation factors (AF) in projection-space are initialized based on PA map. A projection-space certainty (PC) map representing CL for the selected AF is generated based on IC map. An emission image of the target volume is initialized. The selected AF and emission image are iteratively updated based on the ES data, PC map, initial AF, and/or initial emission image. A desired emission image and/or AF values are determined based on the iteratively updated AF and/or emission image.

System and method for joint estimation of attenuation and activity information

A computer-implemented method for partial volume correction in Positron Emission Tomography (PET) image reconstruction includes receiving emission data related to an activity distribution, reconstructing the activity distribution from the emission data by maximizing a penalized-likelihood objective function to produce a reconstructed PET image, quantifying an activity concentration in a region of interest of the reconstructed PET image to produce an uncorrected quantitation, and correcting the uncorrected quantitation based on a pre-calculated contrast recovery coefficient value to account for a partial volume error in the uncorrected quantitation.

Growing community of inventors

Guilderland, NY, United States of America

Sangtae Ahn