TRAINING OF NEURAL NETWORKS TO GENERATE SYNTHETIC THREE-DIMENSIONAL MAGNETIC RESONANCE IMAGES

    公开(公告)号:EP4431967A1

    公开(公告)日:2024-09-18

    申请号:EP23161410.8

    申请日:2023-03-13

    IPC分类号: G01R33/56 G01R33/565 G06N3/02

    摘要: Disclosed herein is a method of training an image processing neural network (100).
    The method comprises repeatedly: receiving (200) a training set of magnetic resonance images (600) each descriptive of a field of view (409) of a subject (418), wherein the training set of magnetic resonance images are acquired according to a synthetic magnetic resonance imaging protocol; receiving (202) at least one two-dimensional training magnetic resonance image (602) acquired from the field of view of the subject; positioning (204) the at least one two-dimensional training magnetic resonance image within a data structure (300) equivalent to the output layer of the image processing neural network; aligning (206) the training set of magnetic resonance images to match the positioning of the at least one two-dimensional training magnetic resonance image; and training (208) the image processing neural network using the aligned training set of magnetic resonance images and the data structure.

    DUAL-ECHO DIXON-TYPE WATER/FAT SEPARATION MR IMAGING

    公开(公告)号:EP3413070A1

    公开(公告)日:2018-12-12

    申请号:EP17175127.4

    申请日:2017-06-09

    发明人: EGGERS, Holger

    摘要: The invention relates to a method of Dixon-type MR imaging. It is an object of the invention to provide a method that enables efficient and reliable water/fat separation using bipolar readout magnetic field gradients and avoids flow-induced leaking and swapping artifacts. According to the invention, an object (10) is subjected to an imaging sequence, which comprises at least one excitation RF pulse and switched magnetic field gradients, wherein two echo signals, a first echo signal and a second echo signal, are generated at different echo times (TE1, TE2). The echo signals are acquired from the object (10) using bipolar readout magnetic field gradients. A first single echo image is reconstructed from the first echo signals and a second single echo image is reconstructed from the second echo signals. A zero echo time image is computed by extrapolating the phase of the first single echo image at each voxel position to a zero echo time using the phase difference between the first and the second single echo image at the respective voxel position. Flow-induced phase errors are identified and estimated in the zero echo time image, and the phase of the first single echo image is corrected according to the estimated flow-induced phase errors. Finally, a water image and/or a fat image are reconstructed from the echo signals, wherein signal contributions from water and fat to the echo signals are separated using the phase-corrected first single echo image and the second single echo image. Moreover, the invention relates to a MR device (1) and to a computer program to be run on a MR device (1).

    DTI WITH CORRECTION OF MOTION-INDUCED DIFFUSION GRADIENT INCONSISTENCY

    公开(公告)号:EP3394628A1

    公开(公告)日:2018-10-31

    申请号:EP16809833.3

    申请日:2016-12-15

    IPC分类号: G01R33/563 G01R33/565

    CPC分类号: G01R33/56341 G01R33/56509

    摘要: The invention relates to a method of MR imaging of an object (10) placed in an examination volume of a MR device (1). It is an object of the invention to enable diffusion tensor imaging (DTI) with minimization of artifacts caused by motion-induced changes of the diffusion encoding. The method of the invention comprises the steps of: - subjecting the object (10) to at least one shot of a multi-echo imaging sequence, a train of echo signals being generated by the shot, wherein the multi-echo imaging sequence comprises diffusion gradients; - acquiring the echo signals for different diffusion gradients; -detecting motion of the object (10) during acquisition of the echo signals; - determining a plurality of motional states of the object (10) based on the detected motion, wherein the object (10) is stationary in each detected motional state; - determining directions of the diffusion gradients relative to a reference frame of the object (10) for detected motional states of the object (10), respectively; and - deriving diffusion coefficients from the acquired echo signals for the determined relative directions of the diffusion gradients. Moreover, the invention relates to a MR device for carrying out this method as well as to a computer program to be run on a MR device.