System and method for silent multi-gradient echo dynamic magnetic resonance imaging
By acquiring functional MR data sets in the MRI system and combining gold angle division and multi-scale low-rank matrix reconstruction technology, the problems of acoustic noise and anatomical image registration error in functional MRI are solved, and high-resolution functional image acquisition is achieved and anatomical image generation, suitable for patients with loud noise sensitive.
Patent Information
- Application Number
- CN202111095534.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Priority Date
- 2020-09-30
- Filing Date
- 2021-09-17
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2041-09-17
AI Technical Summary
The existing functional MRI technology causes high sound noise due to the rapid switching of magnetic field gradients when acquiring brain functional images, and requires the acquisition of high-resolution anatomical images separately to identify the location of functional activity, increasing scanning time and introducing errors.
By collecting the functional MR data set of the subject's 3D k-space volume in the MRI system, combining the free induction attenuation (FID) data set and the gradient echo data set, the method of gold angle and golden segmentation is used to divide the k-space volume. Reduce acoustic noise and generate high-resolution functional and anatomical images through multi-scale low-rank matrix reconstruction techniques.
The silent acquisition function MR dataset is realized without the need for separate anatomical image acquisition, which reduces scanning time, avoids the need for anatomical image registration, improves image resolution and signal-to-noise ratio, and reduces artifacts. for patients who are sensitive to loud noise.
Smart Images

Figure CN114325527B_ABST
Abstract
Description
Background Art
[0001] The field of the present disclosure relates generally to systems and methods for dynamic magnetic resonance (MR) imaging, and more particularly to systems and methods for silent multi-gradient echo dynamic MR imaging.
[0002] Dynamic MR imaging has been used to study the temporal response of subjects. For example, functional MRI (fMRI) has been used to diagnose neural function in the brain. Typical fMRI data acquisition is loud due to the rapid switching of magnetic field gradients, which hinders the application of fMRI to patients who are sensitive to loud acoustic noise. In addition, in order to increase the sampling rate in the time direction, functional images have low image resolution. Therefore, a separate anatomical scan with higher image resolution is required to identify the location of functional activity in the brain. This additional requirement increases the scan time and introduces errors due to the registration of functional images with anatomical images. Summary of the invention
[0003] In one aspect, a magnetic resonance (MR) imaging method for dynamically imaging a subject is provided. The method includes acquiring a functional MR dataset of a three-dimensional (3D) k-space volume of the subject including multiple frames of the k-space dataset while applying a functional stimulus to the subject. Acquiring the functional MR dataset includes acquiring a frame of the k-space dataset of the 3D k-space volume by setting the orientation to an initial angle for each of the multiple repetitions, and acquiring a free induction decay (FID) dataset. Acquiring the FID dataset includes applying a gradient sequence to the subject, each gradient in the gradient sequence corresponding to a k-space spoke in a k-space spoke sequence, wherein the k-space spoke sequence defines a k-space segment having an orientation in the 3D k-space volume. Acquiring the FID dataset also includes applying one or more radio frequency (RF) excitation pulses to the subject while applying the gradient sequence, and collecting the FID dataset corresponding to the RF emission from the subject immediately after each RF excitation pulse. Acquiring frames of k-space datasets also includes acquiring a gradient echo dataset corresponding to the FID dataset by reapplying a gradient sequence to the subject while acquiring echoes emitted by RF to obtain a gradient echo dataset, and updating the orientation to a golden angle along which the k-space segments divide the 3D k-space volume according to the golden section. Acquiring frames of k-space datasets also includes repeatedly acquiring the FID dataset, acquiring the gradient echo dataset, and updating the orientation until a predetermined condition is met. Acquiring a functional MR dataset has a sound level similar to the sound level of background noise of an MRI system before imaging. The method also includes: generating an anatomical image based on one of the frames of multiple k-space datasets, generating an anatomical MR image and a functional image based on the functional MR dataset; and generating a frame of a functional MR image based on the functional MR dataset.
[0004] On the other hand, an MR imaging method for dynamically imaging a subject is provided. The method includes acquiring a k-space data set of a 3D k-space volume of the subject, including setting the orientation to an initial angle, and acquiring a free induction decay (FID) data set. Acquiring the FID data set includes applying a gradient sequence to the subject, each gradient in the gradient sequence corresponding to a k-space spoke in a k-space spoke sequence, wherein the k-space spoke sequence defines a k-space segment having an orientation in the 3D k-space volume. Acquiring the FID data set also includes applying one or more radio frequency (RF) excitation pulses to the subject while applying the gradient sequence, and collecting a FID data set corresponding to the RF emission from the subject immediately after each RF excitation pulse. Acquiring the k-space data set also includes: acquiring a gradient echo data set corresponding to the FID data set by reapplying the gradient sequence to the subject while acquiring echoes of the RF emission to obtain a gradient echo data set; updating the orientation to a golden angle along which the k-space segment divides the 3D k-space volume according to the golden section; and repeatedly acquiring the FID data set; acquiring the gradient echo data set; and updating the orientation until a predetermined condition is met. BRIEF DESCRIPTION OF THE DRAWINGS
[0005] The patent or application file contains at least one drawing executed in color. Copies of the patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.
[0006] Figure 1 is a schematic diagram of an exemplary magnetic resonance imaging (MRI) system.
[0007] Figure 2A is a schematic diagram of an exemplary method of dynamic MR imaging.
[0008] Figure 2B yes Figure 2A Flowchart of the method shown.
[0009] Figure 3A is a flow chart of an exemplary method of acquiring a k-space dataset.
[0010] Figure 3B yes Figure 3A Exemplary pulse sequences used in the methods shown.
[0011] Figure 3C When applied Figure 3B Cumulative k-space trajectory for the pulse sequence shown in .
[0012] Figure 3D It is used Figure 3A k-space volume acquired using the method shown.
[0013] Figure 3Eis a diagram showing an exemplary definition of the orientation of a k-space segment.
[0014] Figure 3F Exemplary orientations with golden angles in 3D k-space are shown.
[0015] Figure 4A An exemplary pulse sequence is shown.
[0016] Figure 4B Shows that when applied Figure 4A Echoes formed when the pulse sequence is shown in .
[0017] Figure 4C Another exemplary pulse sequence is shown.
[0018] Figure 4D Shows that when applied Figure 4C Echoes formed when the pulse sequence is shown in .
[0019] Figure 4E Another exemplary pulse sequence is shown.
[0020] Figure 4F Shows Figure 4E The pulse sequence shown is similar to Figure 4A Comparison of k-space trajectories for the pulse sequences shown.
[0021] Figure 5A are anatomical images based on data sets acquired using known methods.
[0022] Figure 5B is based on Figure 5A Plots of functional images of the same dataset as the anatomical images shown.
[0023] Figure 5C It is based on Figure 2A Anatomical images of the functional MR dataset acquired using the method shown.
[0024] Figure 5D Based on the reconstruction Figure 5C Graph of functional images of the same functional data set acquired for the images shown.
[0025] Figure 5E It is used Figure 2A Another anatomical image acquired by the method shown and having a Figure 5C Images shown are higher image resolution.
[0026] Fig. 5F Based on the reconstruction Figure 5E Graph of functional images of the same functional data set acquired for the images shown.
[0027] Fig. 6A yes Figure 5A Enlargement of the image shown.
[0028] Figure 6B yes Figure 5B Enlargement of the graph shown.
[0029] Figure 6C yes Figure 5C Enlargement of the image shown.
[0030] Fig.6D yes Figure 5D Enlargement of the graph shown.
[0031] Fig. 6E yes Figure 5E Enlargement of the image shown.
[0032] Fig. 6F yes Fig. 5F Enlargement of the graph shown.
[0033] Figure 7 is a block diagram of an exemplary computing device. DETAILED DESCRIPTION
[0034] The present disclosure includes systems and methods for performing silent dynamic magnetic resonance (MR) imaging of a subject. In dynamic imaging, images of the subject are acquired to study dynamic changes in the subject over time. The subject used herein is an object placed in the bore of an MR imaging (MRI) system and imaged by the MRI system. The subject can be a human, an animal, or a phantom. In addition to being silent, the systems and methods described herein allow for the acquisition of multiple gradient echo data sets, thereby allowing for adjustment of image contrast. In addition, anatomical images and functional MR images are generated from the same functional MR data set without the need to acquire anatomical images separately. Methodological aspects will be apparent in part and will be discussed explicitly in part in the following description.
[0035] In magnetic resonance imaging (MRI), a subject is placed in a magnet. When the subject is in a magnetic field generated by a magnet, the magnetic moments of nuclei such as protons attempt to align with the magnetic field, but precess around the magnetic field in a random order at the Larmor frequency of the nuclei. The magnetic field of the magnet is called B0 and extends in the longitudinal or z direction. During the acquisition of MRI images, a magnetic field in the xy plane and close to the Larmor frequency (called the excitation field B1) is generated by a radio frequency (RF) coil and can be used to rotate or "tilt" the net magnetic moment Mz of the nucleus from the z direction toward the transverse or xy plane. After the excitation signal B1 terminates, the nucleus emits a signal, which is called an MR signal. In order to use the MR signals to generate an image of the subject, magnetic field gradient pulses (Gx, Gy, and Gz) are used. Gradient pulses are used to scan through k-space, the space or distance of spatial frequency in reverse. There is a Fourier relationship between the acquired MR signals and the image of the subject, so the image of the subject can be derived by reconstructing the MR signals.
[0036] Figure 1 A schematic diagram of an exemplary MRI system 10 is shown. In an exemplary embodiment, the MRI system 10 includes a workstation 12 having a display 14 and a keyboard 16. The workstation 12 includes a processor 18, such as a commercially available programmable machine running a commercially available operating system. The workstation 12 provides an operator interface that allows a scan plan to be entered into the MRI system 10. The workstation 12 is coupled to a pulse sequence server 20, a data acquisition server 22, a data processing server 24, and a data storage server 26. The workstation 12 and each of the servers 20, 22, 24, and 26 communicate with each other.
[0037] In an exemplary embodiment, the pulse sequence server 20 responds to instructions downloaded from the workstation 12 to operate the gradient system 28 and the radio frequency ("RF") system 30. The instructions are used to generate gradient waveforms and RF waveforms in the MR pulse sequence. The RF coil 38 and the gradient coil assembly 32 are used to perform the prescribed MR pulse sequence. The RF coil 38 is shown as a whole body RF coil. The RF coil 38 can also be a local coil that can be placed near the anatomical structure to be imaged, or a coil array including multiple coils.
[0038] In the exemplary embodiment, a gradient waveform for performing a delimiting scan is generated and applied to a gradient system 28, which excites the gradient coils in a gradient coil assembly 32 to generate a magnetic field gradient G for position encoding of the MR signals. x , G y and G z The gradient coil assembly 32 forms part of a magnet assembly 34, which also includes a polarizing magnet 36 and an RF coil 38. The magnet assembly 34 forms a bore 35 in which a subject 37, such as a patient, is received and scanned.
[0039] In an exemplary embodiment, the RF system 30 includes an RF transmitter for generating RF pulses used in an MR pulse sequence. The RF transmitter responds to the scanning scheme and direction from the pulse sequence server 20 to generate RF pulses with a desired frequency, phase and pulse amplitude waveform. The generated RF pulses can be applied to the RF coil 38 by the RF system 30. The response MR signal detected by the RF coil 38 is received by the RF system 30 and amplified, demodulated, filtered and digitized under the instruction of the command generated by the pulse sequence server 20. The RF coil 38 is described as a transmitter and receiver coil, so that the RF coil 38 transmits RF pulses and detects MR signals. In one embodiment, the MRI system 10 may include a transmitter RF coil that transmits RF pulses and a separate receiver coil that detects MR signals. The transmission channel of the RF system 30 can be connected to the RF transmission coil, and the receiver channel can be connected to a separate RF receiver coil. Typically, the transmission channel is connected to the whole body RF coil 38, and each receiver section is connected to a separate local RF coil.
[0040] In an exemplary embodiment, the RF system 30 also includes one or more RF receiver channels. Each RF receiver channel includes an RF amplifier that amplifies the MR signals received by the RF coil 38 to which the channel is connected; and a detector that detects and digitizes the I quadrature component and the Q quadrature component of the received MR signal. The magnitude of the received MR signal can then be determined as the square root of the sum of the squares of the I component and the Q component, as shown in the following equation (1):
[0041]
[0042] And the phase of the received MR signal can also be determined as shown in the following equation (2):
[0043]
[0044] In an exemplary embodiment, the digitized MR signal samples generated by the RF system 30 are received by the data acquisition server 22. The data acquisition server 22 can operate in response to instructions downloaded from the workstation 12 to receive real-time MR data and provide a buffer memory so that no data is lost due to data overflow. In some scans, the data acquisition server 22 only passes the acquired MR data to the data processing server 24. However, in scans where information derived from the acquired MR data is needed to control further execution of the scan, the data acquisition server 22 is programmed to generate the required information and transmit it to the pulse sequence server 20. For example, during a pre-scan, MR data is acquired and used to calibrate the pulse sequence executed by the pulse sequence server 20. In addition, navigator signals can be acquired during a scan and used to adjust operating parameters of the RF system 30 or the gradient system 28, or to control the order of views for sampling k-space.
[0045] In an exemplary embodiment, the data processing server 24 receives MR data from the data acquisition server 22 and processes the MR data according to instructions downloaded from the workstation 12. Such processing may include, for example, Fourier transforming the raw k-space MR data to produce a two-dimensional or three-dimensional image, applying filters to the reconstructed image, performing back-projection image reconstruction on the acquired MR data, generating functional MR images, and calculating motion or flow images.
[0046] In an exemplary embodiment, the image reconstructed by the data processing server 24 is transmitted back to the workstation 12 and stored there. In some embodiments, the real-time image is stored in a database memory cache ( Figure 1 In the example of FIG. 1 , the real-time images can be output from the database memory cache to the operator display 14 or a display 46 located near the magnet assembly 34 for use by the attending physician. Batch mode images or selected real-time images can be stored in a disk storage device 48 or a host database on the cloud. When such images have been reconstructed and transferred to the storage device, the data processing server 24 notifies the data storage server 26. The operator can use the workstation 12 to archive images, generate films, or send images to other facilities via the network.
[0047] Functional MRI (fMRI) has been used to study and diagnose brain function based on the observation that when neural activity increases in a specific area of the brain, the MR signal increases by a small amount in that specific area, because of an effect known as the blood oxygen level dependent (BOLD) effect. In fMRI image acquisition, the subject alternates between periods of performing a task and periods of a control state, for example, 30-second blocks of looking at a visual stimulus alternate with 30-second blocks of eyes closed. The acquired fMRI data is analyzed to identify brain areas where the MR signal has a matching pattern of change, and these identified areas are the areas in the brain that are activated by the stimulus, in this example, the visual cortex at the back of the subject's head. The activated areas are highlighted on the subject's MR image.
[0048] Because fMRI studies the temporal changes of MR images, MR images are acquired quickly to achieve the desired temporal resolution. At the same time, the image should have a certain resolution level that enables the identification of activated areas in the brain. Although the repetition of image acquisition increases the signal-to-noise ratio (SNR) of the image, the scanning time for acquiring MR images is limited due to the comfort of the subject. Therefore, fMRI often faces the challenge of obtaining sufficient image resolution and temporal resolution within a reasonable scanning time. The typical pulse sequence used for fMRI research is the echo planar imaging (EPI) sequence. In EPI, after selecting and exciting a slice, the phase encoding gradient is quickly switched to scan the k-space through the slice along a Cartesian grid. After excitation, the magnetic moment in the xy plane tends to phase shift due to field inhomogeneity, and therefore the MR signal attenuates. Because the EPI sequence acquires the entire k-space plane after one excitation, the MR signal acquired by the EPI sequence has a relatively low SNR. In order to compromise, EPI images are acquired at a relatively low image resolution such as 64×64 (the number of voxels in the x-direction and the y-direction). Therefore, usually before acquiring functional images, anatomical images with higher image resolution, such as 256 × 256, are acquired separately using different pulse sequences. Brain regions with specific neural activity identified by the functional images acquired using the EPI sequence are then overlaid on the anatomical images.
[0049] In addition, because the EPI sequence acquires the entire k-space plane after one RF excitation, the EPI image is subject to more chemical shifts and geometric distortions than the anatomical image. In addition, due to the rapid switching of the gradient, the EPI sequence produces loud, sharp noises in the bore of the magnet that receives the subject. The noise is disturbing to the subject, especially for patients who are sensitive to loud noises, such as those with tinnitus and post-traumatic stress disorder (PTSD). The loud noise also limits the functional study of auditory nerve function and resting state networks in the brain.
[0050] Functional MRI is used in one example herein.The methods and systems disclosed herein can be used for any dynamic imaging, such as in vivo contrast uptake imaging.
[0051] The systems and methods disclosed herein provide a silent method for acquiring a functional MR data set without the need for a separate anatomical image acquisition. The silent method as used herein means acquiring a functional data set with a sound level that is much lower than the sound level of acquisition using an EPI sequence, and can be similar to the sound level of the background noise of an MRI system prior to imaging. Functional images and anatomical images are generated based on the same functional MR data set. The method is advantageous because it reduces scan time and avoids the need to register anatomical images to functional images and registration errors. Figure 2A and Figure 2B An exemplary method 200 of dynamic MR imaging is shown. Figure 2A is a schematic diagram of method 200 . Figure 2B is a flow chart of method 200 .
[0052] In an exemplary embodiment, method 200 includes acquiring 202 a functional MR dataset of a 3D k-space volume of a subject. While acquiring 202 the functional MR dataset, frames of the k-space dataset of the 3D k-space volume are repeatedly acquired 204 multiple times while a functional stimulus is applied to the subject, for example, a visual pattern of 30 seconds with eyes open and 30 seconds with eyes closed is shown. The functional MR dataset includes multiple frames of the k-space dataset. Anatomical images and functional images are generated 206 based on the functional MR dataset. Generating 206 the anatomical images and functional images includes generating 208 anatomical images based on the functional MR dataset. The functional MR dataset has a large data volume. For example, if the matrix size is 392×318×165, where the number of voxels in the x-direction, y-direction, and z-direction are 392, 318, and 165, respectively, and if the number of frames is 500, the associated data volume is approximately 125 gigabytes (GB). Due to the large amount of data, the MR image can be reconstructed 218 by expanding the functional MR dataset into a sum of multi-scale low-rank (MSLR) matrices to increase the computational speed and the volume of available memory storage space. MSLR matrices are low-rank matrices of various scales of block size. For example, the functional MR dataset is a space-time matrix 212 and can be represented as a sum of block-wise low-rank matrices 214-a, 214-b, 214-c, where 214-a, 214-b, 214-c have reduced block size scales ( Figure 2A ). To further reduce computational load, improve memory locality, and minimize communication between threads and processors, random optimization 216 may be used.
[0053] In an exemplary embodiment, an anatomical image is generated 208 based on the same functional MR dataset. A frame of the k-space dataset may be used to generate the anatomical image. The anatomical image may be reconstructed 210 by expanding the frame of the k-space dataset as a sum of MSLR matrices.
[0054] Generating 206 the anatomical image and the functional MR image includes generating 218 the functional image based on the functional MR dataset. In one embodiment, a frame of the functional image may be reconstructed using a subset of the functional MR dataset acquired at a time point adjacent to the frame to reduce the amount of data. In another embodiment, because temporal changes on the functional image are of interest, the functional image may be reconstructed by reconstructing a difference image, which is an image representing the difference between one frame and a previous frame. In some embodiments, the functional image is reconstructed by extending the functional MR dataset having a smaller MSLR matrix set than the MSLR matrix set of the anatomical image, wherein the smaller MSLR matrix set reflects temporal changes.
[0055] In some embodiments, the temporal resolution of the functional MR image is greater than the sampling rate determined by the time of imaging the 3D volume. The functional image of the 3D volume is generated by repeatedly imaging the 3D volume, where the interval between repetitions serves as the imaging time of the 3D volume. The temporal resolution of the functional images of the known methods is limited by the temporal sampling rate of the imaging time of the 3D volume. In contrast, the functional images in the systems and methods of the present invention can be generated for each repetition using a subset of the MR signal of the entire 3D volume, and therefore, the temporal resolution of the functional images can be extended to a sampling rate greater than the imaging time of the 3D volume. For example, in EPI fMRI or other known fMRI, the length of the time segment of each frame is fixed and defined at the acquisition point, for example, for a 3D volume of 2.1 s, a data set acquired with a repetition time (TR) can only be reconstructed at a TR of 2.1 s. With the systems and methods described herein, functional images can be flexibly reconstructed using different time segments after acquisition, for example, for Figure 5C The 3D volume shown in FIG. 1 was acquired at a TR of 580 ms. Using the same dataset, functional images can be reconstructed with different TRs, such as 200 ms or 300 ms.
[0056] In some embodiments, after generating the functional images, fMRI data processing methods are used to identify brain regions that indicate neural function, such as using the cross-correlation between the time series of voxels and the stimulation pattern to identify voxels of interest. The identified voxels are highlighted in the anatomical image to show neural activity in the brain (see Figure 5B , Figure 5D , Fig. 5F , Figure 6B , Fig.6D and Fig. 6F).
[0057] FIG. 3A to FIG. 3D An exemplary method 204 for acquiring a k-space dataset of a 3D k-space volume 301 is shown. Figure 3D ). Figure 3A is a flow chart of method 204 . Figure 3B is a graph of an exemplary pulse sequence 302 used in method 204 . Figure 3C An accumulated trajectory 304 in k-space is shown when the pulse sequence 302 is applied. Figure 3D is a 3D graph of a 3D k-space volume 301 acquired using a pulse sequence 302 .
[0058] In an exemplary embodiment, method 204 includes acquiring 306 a free induction decay (FID) data set having gradients corresponding to a sequence of k-space spokes or k-space lines having orientations in 3D k-space. A sequence 322 of k-space spokes 324 samples a segment 320 of k-space. Figure 3E An exemplary definition of the orientation of the segment 320 is shown. The orientation used herein is the angle of the segment 320 in the 3D k-space. The orientation of the segment 320 is defined by the polar angle α and the azimuth angle β. The polar angle α is the angle between the kz axis and the normal vector 334 of the segment 320. The azimuth angle β is the angle between the kx axis and the projection 336 of the normal vector 334 on the kx-ky plane. The FID data set includes FID MR signals, which are MR signals acquired immediately after applying the excitation RF pulse. The immediate acquisition used herein means that the time between the end of the RF pulse and the start of the acquisition is negligible. The method 204 also includes acquiring 308 a gradient echo data set corresponding to the sequence 322 of the spoke 324. The method 204 includes updating 310 the orientation to the golden angle. The method 204 also includes repeatedly acquiring 306 the FID data set, acquiring 308 the gradient echo data set, and updating 310 until a predetermined condition is met. The predetermined condition can be that the number of repetitions is equal to the number of preset matrix sizes in the z direction. Thus, a k-space data set of a 3D k-space volume 301 is acquired 204 by iterative application of a pulse sequence 302 that scans a segment 320 of k-space covered by a sequence 322 of spokes 324, the segment 320 being oriented at a golden angle ( FIG. 3B to FIG. 3D ).
[0059] In an exemplary embodiment, Figure 3B The pulse sequence 302 shown corresponds to pulses of a sequence 322 of spokes 324 in k-space to sample a segment 320 of k-space ( FIG. 3B to FIG. 3D). In the pulse sequence 302, a gradient 328 is applied during the acquisition 306 of the FID dataset and the acquisition 308 of the gradient echo dataset. The same sequence 322 of spokes 324 is acquired during the acquisition 306 of the FID dataset and the acquisition 308 of the gradient echo dataset. That is, the same sequence of gradients 328 is applied, and the same segment 320 of k-space is scanned at multiple echo times (TE), one at TE=0 for the FID dataset and one at TE=echo time of the gradient echo. The pulse sequence 302 shows the acquisition of one gradient echo dataset. Multiple gradient echo datasets can be acquired at multiple TEs. TE can be adjusted based on the transverse relaxation time T2* to maximize tissue contrast. Multiple gradient echo datasets also allow for objective separation of BOLD effects from non-BOLD effects, removal of motion effects from the fMRI time course, and measurement of BOLD signals deep in the median prefrontal cortex of the brain.
[0060] When acquiring the FID data set, a gradient G 328 is applied to sample k-space. k is the spatial frequency, i.e., the inverse of the distance. k is related to the gradient as follows:
[0061]
[0062] Where γ is the gyroscope ratio, and r is the direction of the gradient G, such as the x-direction, y-direction, z-direction, or other direction in 3D space. When sampling along the k-space spokes, k may be increased starting from the k-space center 330 along the k-space spokes 324 until the periphery of the k-space is reached using a constant gradient 328 applied along the direction r. When sampling a segment 320 of k-space, a plurality of spokes 324 in the segment 320 are sampled. A spoke 324 in one segment 320 rotates within the segment 320 at a constant angular increment from its neighboring spoke 324, where the angular increment is 360 degrees divided by the number of spokes per loop or segment. Between spokes 324 in the same segment 320, the direction r of the gradient 328 varies slightly, but the magnitude of the gradient 328 may remain constant. Due to this slight variation in the gradient, rather than rapid switching of the gradient, quiet operation of the MRI system 10 is achieved. The acoustic noise generated during the application of the pulse sequence 302 may be similar to the background noise before imaging is performed. This acoustic noise can be further reduced by applying gradients so that k-space is scanned into curved k-space spokes.
[0063] In the depicted example, during acquisition of the FID dataset, the RF pulse 332 may be applied after applying the gradient 328 corresponding to the spoke 324, and the FID dataset sampled by the spoke 324 may be acquired immediately after applying the RF pulse 332. Then, the gradient 328 corresponding to another spoke 324 in the segment 320 is applied, the RF pulse 332 is applied, and the FID dataset for that spoke 324 is acquired.
[0064] After a predetermined number of spokes in the sector have been scanned, a gradient echo dataset is acquired 308. During acquisition 308 of the gradient echo dataset, no RF pulses 332 are applied, but rather the same sequence 322 of spokes 324 is sampled by varying the gradient in the same manner as during acquisition of the FID dataset.
[0065] In an exemplary embodiment, to sample multiple segments 320 of k-space, the pulse sequence 302 is repeated except that a different set of gradients is applied corresponding to another segment 320 of k-space having a different orientation in 3D k-space. The orientation of the segment 320 is at a golden angle to the kx-ky plane. Along the golden angle, the segment 320 divides the k-space volume 301, such as a sphere, according to the golden section.
[0066] In one dimension, the 1D golden mean (or golden section) divides a line segment into a longer segment of length x and a shorter segment of unit length such that:
[0067]
[0068] The larger root of equation (4), 1.618, is called the golden ratio, and the smaller root of 0.618 is called the conjugate golden ratio. Each subsegment of the line segment can be further segmented in the same manner into larger segments and smaller segments having the same ratio as expressed in equation (4), thereby generating a series of partitioning points for the line segment. The partitioning points appear pseudo-random on the line segment. Compared to a random set of points on the line segment that exhibits a cluster, the set of golden points is uniformly distributed in space and time. That is, the partitioning points generated as the golden section uniformly sample the line segment. The golden section can be generated by other mechanisms, such as by using a Fibonacci sequence, or by deriving the eigenvalues of a matrix after consecutive Fibonacci transformations.
[0069] In an exemplary embodiment, the orientation of the segments 320 is a golden angle derived as a golden section for sampling the k-space volume 301. Thus, the plurality of segments 320 pseudo-randomly and uniformly divides the k-space volume. In one example, in the pair of golden angles (α, β) of the orientation of the segments 320, the polar angle α is the corresponding polar angle of the unit vector 340 (the vector having a magnitude of one) on a unit sphere 342 of radius one, where the projection of the unit vector 340 on the kz axis divides the unit length ( Figure 3F ). The azimuth angle β is the corresponding azimuth angle of the unit vector 340, where the projection of the unit vector 340 on the kx-ky plane divides the unit length in the kx-ky plane according to the golden section. Since the span of segment 320 is from -kz to kz, a polar angle α of half the k-space volume 301 is required. In addition, a negative azimuth angle β can be generated as the corresponding azimuth angle of the unit vector 340, where the projection of the unit vector 340 on the kx-ky plane divides the negative unit length in the kx-ky plane according to the golden section. The golden angle (α, β) can be derived in other ways that enable the systems and methods to function as described herein.
[0070] In an exemplary embodiment, segments 320 are generated based on pairs of golden angles. A "mother" segment is generated as a set of spokes 324 on the ky-kx plane. A new segment 320 is generated by rotating the mother segment using a rotation matrix. The rotation matrix first performs a rotation of the segment 320 around the kx axis at an azimuth angle β, and then performs a rotation around the kz axis at a polar angle α. Alternatively, the mother segment is in the kx-kz plane, and a new segment 320 is generated by rotating the mother segment around the kz axis at a polar angle α and then rotating it around the kx axis at an azimuth angle β. Thus, the segment 320 is generated by rotating the mother segment into a segment having an orientation of the golden angle (α, β).
[0071] Images acquired using the pulse sequence 302 may have artifacts caused by a phenomenon known as echo-in-echo-out. Figure 4A A pulse train 302 having six spokes 324 sampled in segment 320 is shown. Figure 4B The echoes or coherence formed when the pulse sequence 302 is applied are shown. In the depicted pulse sequence 302, spokes 1 to 6 are sampled during the acquisition of the FID data set, and spokes 7 to 12 are sampled during the acquisition of the gradient echo data set. For the purpose of illustration, although spokes 1 to 6 and spokes 7 to 12 are the same, different numbers are used. The pulse sequence 302 is a self-refocusing pulse sequence. That is, when a segment 320 of spokes 324 is sampled, the accumulated k-space trajectory 304 forms a loop 401 in which the end point of the last spoke (e.g., spoke 6) returns to the starting point of the first spoke (e.g., spoke 1). Figure 4B). As the gradient changes from spoke 1 to spoke 12, at the time point 402 when the direction of the gradient 328 changes, the echo corresponding to the current spoke begins to form, while the echo of the spoke corresponding to the end point of the loop 401 is about to end. In this way, the acquired data includes MR signals from different spokes (i.e., different positions in k-space), and this mixing causes artifacts in the reconstructed image.
[0072] Figure 4C An exemplary pulse sequence 302 - 2 for reducing echo-in-echo-out artifacts is shown. Figure 4D The echoes formed when the pulse sequence 302-2 is applied are shown. In the pulse sequence 302-2, the gradient 328 is applied and varied in the same manner as in the pulse sequence 302, but the RF pulse 332 is applied once for every two spokes 324 of the pulse sequence 302 instead of every spoke 324. Thus, the echo-in-echo-out artifact is reduced.
[0073] Figure 4E Another pulse sequence 302-3 for reducing echo-in-echo-out artifacts is shown. Figure 4E , only the relative timing between the RF pulse 332 and the spoke 324 is shown. Instead of starting the spoke 324 at the center of k-space, the spoke 324 is a full diameter spoke, where the spoke 324 starts at the periphery of k-space, passes through the center of k-space, and extends to the relative periphery of k-space. When the spoke 324 reaches the center of k-space, the RF pulse 332 is applied. Therefore, the acquired echo corresponds to a specific spoke, rather than being mixed with echoes corresponding to different spokes. Therefore, echo-in-echo-out artifacts are reduced.
[0074] Figure 4FA comparison of the k-space trajectory 410 of pulse sequence 302-3 and the k-space trajectory 412 of pulse sequence 302 is shown. The pulse sequence described herein may be referred to as a loop star pulse sequence. The loop star pulse sequence uses a time-division multiplexed gradient refocusing scheme to generate gradient echoes (GREs) of the FID signal of the initial excitation. The acquisition is divided into segments, where each segment encodes a plane in k-space, similar to the excitation in multiple-shot EPI. For full 3D k-space coverage, multiple segments rotated according to the golden angle scheme are acquired. For pulse sequence 302, in the first loop of the segment, multiple coherences are generated and encoded along different central output radial spokes with effective TE=0. The same gradient waveform is then repeated, but without applying RF pulses, to refocus the magnetization in the loop and generate gradient echoes. Since the magnetization is refocused in consecutive loops, changes in the gradient direction and therefore the acoustic noise remain small, and are even further reduced by distributing the gradient updates along the spokes, which results in a relatively circular loop without sharp edges. To refocus the FID signal, the k-space spokes of the first loop are accumulated toward the closed loop and then repeated identically for one or more subsequent refocusing GRE loops.
[0075] The loop star sequence 302-3 requires only half the excitation and therefore generates only half the coherence as sequence 302. The effective k-space encoding by the full diameter spokes is the same as by the center output spoke, except that the FID signal of the center input portion is missing.
[0076] The loop star pulse sequence 302 suffers from a mixture of inward refocusing and outward phase-shifted signals, referred to as echo-in / echo-out mixing. These signals can be separated by k-space filtering or RF phase cycling, but at the expense of reduced image quality or increased scan time. The loop star pulse sequence 302-3 solves the echo-in / echo-out mixing by separating the RF excitations in time, thereby acquiring full-diameter spokes (without overlapping coherence) during refocusing.
[0077] The number of spokes per loop or per segment (SPL) is a parameter considered in the loop star sequence. Higher SPL produces lower acoustic noise, but increases the echo time given by TE=SPL*TG, where TG is the duration of the flat part of the gradient. The acquisition time (Tacq) of the volume (equivalent to TR in fMRI experiments) is given by Tacq=(TG*SPL*NLoops+2*Tramp)*NSeg, where NLoops is the number of loops, i.e. the number of echoes plus FID, and NSeg is the number of segments determined by the undersampling level in the acquisition. The gradient ramp-up time (Tramp) is typically between 2ms and 5ms to provide silent operation.
[0078] After acquiring 202 a functional MR data set, an MR image is reconstructed 206 based on the data set. The acquired functional data set has a very large data size and makes the reconstruction process extremely slow. In order to reduce the required computational load and memory, an MSLR reconstruction framework is used in the systems and methods disclosed herein. Stochastic optimization can be used to further reduce the required computational load and memory.
[0079] Specifically, let T be the number of frames, N be the image size, and J be the number of scales of MSLR. And for scale j=1,…,J, let Nj be the block size, Bj be the number of blocks, and Kj be the maximum block matrix rank. Then, for each block, b=1,…,Bj, let Defined as a block space basis, Defined as the block time basis, Defined as a linear operator that embeds the input block matrix into the full image. Volumetric Spatiotemporal Matrix It is expressed as:
[0080]
[0081] For simplicity, the stacking space bases are and time base And this stack is fed into a linear operator embedded in the image is considered to make
[0082]
[0083] The multi-channel MRI acquisition system has C coil channels, and the overall scan is divided into T frames. The image underlying each frame is assumed to be approximately static to obtain sufficiently fine temporal resolution. That is, views sharing short non-overlapping time windows are considered. Let M be the number of measurements per frame, is the overall sensing linear operator that combines the sensitivity map and non-uniform Fourier sampling, and represents the complex white Gaussian acquisition noise. Then, the k-space measurement matrix Given by:
[0084] Y=A(X)+W (7)
[0085] Since the number of measurements is much smaller than the total dynamic image size, the reconstruction problem is severely uncertain. This problem can be solved by using incoherent sampling acquisition, sparse signal representation and nonlinear algorithms for uncertain MRI reconstruction. Following this structure, inherent 3D non-Cartesian trajectories such as 3D meridian trajectories, 1D cones and 3D cones are utilized, where the ordering inherently covers time and k-space throughout the acquisition process. For the model below, MSLR is used to compactly represent the following volume image sequence. Then, the overall forward model is given by:
[0086]
[0087] Using the forward model, one way to impose a low-rank constraint is through convex relaxation by minimizing the nuclear norm or, equivalently, the sum of singular values. Specifically, let ||X|| * represents the nuclear norm of the matrix X. Then, for each scale j, let λ j is the regularization parameter and represents the block nuclear norm, and the convex formula considers the objective function,
[0088]
[0089] Under the identity setting with respect to the basis vectors and certain idealized incoherence conditions, when we choose The MSLR decomposition can be restored when
[0090]
[0091] This relationship is useful because only one scaling parameter needs to be tuned, rather than the J regularization parameter. is selected according to formula (10). However, a disadvantage of the convex formulation is that it uses a lot of memory. This is because the full-size spatiotemporal matrix is stored even when it is of very low rank. Specifically, for image sizes such as 320×320×320 and 500 frames, 125GB is required just to store the image in a complex single precision floating point number. Applying an iterative algorithm will require several times more memory for the workspace, which may be close to a terabyte. Such memory requirements are difficult for a local workstation or even a computing cluster to handle.
[0092] Instead of minimizing the convex problem in Eq. 9, the Burer-Monteiro factorization is considered to directly solve the compressed representation. It considers the following non-convex transformation,
[0093]
[0094] And minimize the following equivalent objective function:
[0095]
[0096] The main benefit of this formulation is that it can use significantly less memory for low-rank matrices. For typical 3D volumes, the variables can even fit the variables in the GPU. However, the main disadvantage is that the problem becomes non-convex. Nevertheless, in practice, applying first-order gradient methods to non-convex objective functions reaches a suitable solution.
[0097] Before turning to the algorithm, the proposed formula can easily incorporate priors on the temporal dynamics, such as smoothness. For example, given the finite difference operator The above objective function can be modified to enforce temporal smoothness as follows:
[0098]
[0099] Even with an explicit LR factorization, performing conventional iterative reconstruction is still too slow, as each iteration requires computing many non-uniform fast Fourier transforms (NUFFTs), once per time frame and per channel. To address this problem, a stochastic optimization technique is applied to speed up the reconstruction, which utilizes a random subset of the measurements to update the variables in each iteration.
[0100] The idea of using a subset of the measurements for iterative updates using stochastic optimization or stochastic gradient descent (SGD) can transform almost computationally infeasible reconstructions into practical ones. For example, for a feature dataset with a matrix size of 320×320×320 and 500 frames, the size of the required memory can be reduced from 125GB to less than 12GB, allowing GPU-accelerated large-scale reconstructions on consumer GPUs.
[0101] Specifically, the objective function is divided into separate frames and coils.
[0102]
[0103] For t = 1, ..., T and c = 1, ..., C, then
[0104] At each iteration, SGD randomly selects a frame t and a coil c and performs the following update:
[0105]
[0106]
[0107] Where α is the step size parameter. In one embodiment, to select the step size α, the step size α is first set to 1, and then the iterative update is run. If the iteration diverges as the gradient norm reaches a numerical limit, the reconstruction is restarted and the step size α is halved.
[0108] With SGD, the number of NUFTS per iteration is dramatically reduced to 1. In addition, SGD can easily support parallel processing by performing micro-batch updates. Given G parallel processors, such as multiple GPUs, each processor g can choose a different frame t and a different coil c to calculate the gradient and And then add synchronization.
[0109] Specifically, in each iteration, SGD randomly selects an index set I = {(t1,c1),(t2,c2),…,(t G ,c G )}. Then, SGD performs:
[0110]
[0111]
[0112] For all t, such that (t,c)∈I, only the sum is synchronized across processors, thus minimizing expensive communication. In one example, the regularization parameter λ is 10 -12 , and the maximum number of iterations is 10.
[0113] The above method and system may be applied to reconstruct anatomical images, wherein the number of frames is set to 1. The frames of the k-space dataset may be selected from the functional MR dataset and expanded as the sum of the MSLR basis.
[0114] In an exemplary embodiment, in order to further reduce computational load and memory requirements, the functional MR image is expanded with a lower rank basis set smaller than the MSLR basis for reconstructing the anatomical image. The frame x of the functional image is represented as:
[0115] x=x 解剖 +∑ i X i (17)
[0116] where x 解剖 is an anatomical image, and X i 's is a set of low-rank bases that are smaller than the low-rank base of the anatomical images. i The number of can be determined empirically. The low-rank basis of the functional image represents the temporal changes of the subject during the fMRI scan. In one example, the regularization parameter λ is 10 -2 .
[0117] 5A to 6F Results of an fMRI study are shown. Figure 5AA is an anatomical image, and panel B shows a functional map of the brain using an fMRI dataset acquired with a known method in which segments of k-space are not pseudo-randomized. Figure 5C and Figure 5E is an anatomical image, and Figure 5D and Fig. 5F is a functional map of the brain using an fMRI dataset acquired using the systems and methods disclosed herein. Figure 5C The image shown is an isotropic 3×3×3mm 3 Compared with the resolution of Figure 5E The images in have a higher isotropy 1mm 3 image resolution. 6A to 6F They are FIG. 5A to FIG. 5F A magnified view of .
[0118] Collected by known methods Figure 5A Compared to image 502, Figure 5C and Figure 5E The artifact 503 in the images 504, 506 shown in FIG. 5 is largely reduced to a negligible level. In the systems and methods disclosed herein, the 3D k-space volume is pseudo-randomly sampled by segments of k-space. Therefore, artifacts and noise are pseudo-randomly dispersed in the acquired MR signals and are therefore greatly reduced. In addition, although image 502 is in the same 3×3×3 mm format as image 504, the 3D k-space volume is pseudo-randomly sampled by segments of k-space. Therefore, artifacts and noise are pseudo-randomly dispersed in the acquired MR signals and are therefore greatly reduced. 3 The image resolution of image 504 is acquired, but the image quality of image 504 is significantly better than that of image 502, showing details such as blood vessel 508. The systems and methods disclosed herein uniformly sample the 3D k-space volume, especially the periphery of the k-space volume, which includes high spatial frequency information, i.e., spatial details. Similarly, Figure 5D and Fig. 5F The functional map shown has a Figure 5B The functional map shown has higher SNR and better image resolution.
[0119] In addition, blood vessels 508 and the blood therein in images 504, 506 are black. Black blood is advantageous for fMRI. The basis of fMRI is the BOLD effect in brain tissue. MR signals are not directly sensitive to neural activity. Instead, MR signal changes are indirect effects related to changes in blood flow following changes in neural activity. When neural activity increases, blood flow increases and the oxygenation level in the blood increases. Due to the presence of hemoglobin in the blood that binds oxygen, oxygen-rich blood has a different magnetic sensitivity than oxygen-poor blood, and therefore has a different level of MR signal. When large blood vessels appear black in MR images in which the signal of large blood vessels is low, contamination of the fMRI map from large blood vessels is significantly reduced.
[0120] The workstation 12 may be any suitable computing device 800 and software implemented therein. The methods 200 and 204 or one or more processes in the methods 200 and 204 may be implemented in any suitable computing device 800 . Figure 7 800. In an exemplary embodiment, the computing device 800 includes a user interface 804 that receives at least one input from a user. The user interface 804 may include a keyboard 806 that enables the user to enter relevant information. The user interface 804 may also include, for example, a pointing device, a mouse, a stylus, a touch-sensitive panel (e.g., a touchpad and a touch screen), a gyroscope, an accelerometer, a position detector, and / or an audio input interface (e.g., including a microphone).
[0121] In addition, in an exemplary embodiment, the computing device 800 includes a display interface 817 that presents information (such as input events and / or verification results) to the user. The display interface 817 may also include a display adapter 808 coupled to at least one display device 810. More specifically, in an exemplary embodiment, the display device 810 may be a visual display device such as a cathode ray tube (CRT), a liquid crystal display (LCD), a light emitting diode (LED) display, and / or an "electronic ink" display. Alternatively, the display interface 817 may include an audio output device (e.g., an audio adapter and / or a speaker) and / or a printer.
[0122] The computing device 800 also includes a processor 814 and a memory device 818. The processor 814 is coupled to the user interface 804, the display interface 817, and the memory device 818 via the system bus 820. In an exemplary embodiment, the processor 814 communicates with the user, such as by prompting the user via the display interface 817 and / or by receiving user input via the user interface 804. The term "processor" generally refers to any programmable system, including system and microcontrollers, reduced instruction set computers (RISC), complex instruction set computers (CISC), application specific integrated circuits (ASICs), programmable logic circuits (PLCs), and any other circuits or processors capable of performing the functions described herein. The above examples are exemplary only and are therefore not intended to limit the definition and / or meaning of the term "processor" in any way.
[0123] In an exemplary embodiment, the memory device 818 includes one or more devices that enable information (such as executable instructions and / or other data) to be stored and retrieved. In addition, the memory device 818 includes one or more computer-readable media, such as but not limited to dynamic random access memory (DRAM), static random access memory (SRAM), solid state disk, and / or hard disk. In an exemplary embodiment, the memory device 818 stores but is not limited to application source code, application object code, configuration data, additional input events, application status, assertion statements, verification results, and / or any other type of data. In an exemplary embodiment, the computing device 800 may also include a communication interface 830 connected to the processor 814 via the system bus 820. In addition, the communication interface 830 is communicatively connected to a data acquisition device.
[0124] In an exemplary embodiment, processor 814 may be programmed by encoding operations using one or more executable instructions and providing the executable instructions in memory device 818. In an exemplary embodiment, processor 814 is programmed to select a plurality of measurements received from a data collection device.
[0125] In operation, a computer executes computer executable instructions embodied in one or more computer executable components stored on one or more computer readable media to implement various aspects of the present invention described and / or shown herein. Unless otherwise specified, the execution or implementation order of the operations in the embodiments of the present invention shown and described herein is not required. That is, unless otherwise specified, these operations can be performed in any order, and embodiments of the present invention may include more or fewer operations than those disclosed herein. For example, it is conceivable that performing or implementing a particular operation before, simultaneously with, or after another operation is within the scope of various aspects of the present invention.
[0126] At least one technical effect of the systems and methods described herein includes (a) silent dynamic imaging; (b) multi-gradient echo acquisition; (c) anatomical and functional images generated from the same functional MR data set; and (d) pseudo-random and uniform sampling of 3D k-space volumes to generate images with reduced artifacts and increased image resolution.
[0127] Exemplary embodiments of systems and methods for silent dynamic imaging are described in detail above. These systems and methods are not limited to the specific embodiments described herein, but the components of the systems and / or the operations of the methods can be used independently and separately from other components and / or operations described herein. In addition, the components and / or operations described can also be defined in other systems, methods and / or devices, or used in combination with other systems, methods and / or devices, and are not limited to being practiced only with the systems described herein.
[0128] Although specific features of various embodiments of the invention may be shown in some drawings and not in others, this is for convenience only. In accordance with the principles of the invention, any feature of a drawing may be referenced and / or claimed in combination with any feature of any other drawing.
[0129] This written description uses examples to disclose the invention, including the best mode, and also to enable those skilled in the art to practice the invention, including making and using any devices or systems and performing any included methods. The patentable scope of the invention is defined by the claims, and may include other examples that occur to those skilled in the art. Such other examples are intended to fall within the scope of the claims if they have structural elements that do not differ from the literal language of the claims, or if they include equivalent structural elements with insignificant differences from the literal language of the claims.
Claims
1. A magnetic resonance imaging method for performing dynamic imaging of a subject using a magnetic resonance imaging system, the magnetic resonance imaging method comprising: While applying functional stimulation to the subject, acquiring a functional magnetic resonance data set of a three-dimensional k-space volume of the subject including a plurality of frames of the k-space data set by: For each of the plurality of repetitions, a frame of a k-space dataset of the three-dimensional k-space volume is acquired by: Set the orientation to the initial angle; The free induction decay data set was acquired by: applying a gradient sequence to the subject, each gradient in the gradient sequence corresponding to a k-space spoke in a k-space spoke sequence, wherein the k-space spoke sequence defines a k-space segment having the orientation in a three-dimensional k-space volume; applying one or more radio frequency excitation pulses to the subject while applying the gradient sequence; and collecting the free induction decay data set corresponding to radio frequency emissions from the subject immediately after each radio frequency excitation pulse; acquiring a gradient echo dataset corresponding to the free induction attenuation dataset by reapplying the gradient sequence to the subject while acquiring echoes of the radio frequency emission to obtain the gradient echo dataset; updating the orientation to a golden angle, wherein the k-space segment divides the three-dimensional k-space volume according to a golden section along the golden angle; The plurality of repetitions include: repeatedly acquiring a free induction attenuation data set, acquiring a gradient echo data set, and updating the orientation until a predetermined condition is satisfied; and The anatomical MRI image and the functional image are generated based on the functional MRI dataset in the following manner: generating an anatomical image based on one of the plurality of frames of the k-space dataset; and A functional magnetic resonance image is generated based on the functional magnetic resonance dataset. 2 . The method according to claim 1 , wherein acquiring the gradient echo data set further comprises acquiring a plurality of gradient echo data sets at a plurality of echo times.
3. The method according to claim 1, wherein generating a functional magnetic resonance image further comprises reconstructing a frame of the functional magnetic resonance image at a certain time point based on a plurality of free induction decay data sets and a plurality of gradient echo data sets acquired in time intervals before and after the certain time point.
4. The method of claim 1, wherein generating the anatomical image further comprises reconstructing the anatomical image based on one of the multiple frames of the k-space dataset by expanding the one of the multiple frames of the k-space dataset as a sum of multi-scale low-rank matrices. 5 . The method of claim 1 , wherein generating a functional magnetic resonance image further comprises expanding the plurality of frames of the k-space dataset as a sum of multi-scale low-rank matrices.
6. The method according to claim 5, wherein the functional magnetic resonance data set is acquired by a plurality of coils, and generating a functional magnetic resonance image further comprises: Separate the target function into separate frames and separate coils; and For multiple iterations, randomly selecting a coil from the plurality of coils and a frame from the plurality of frames; and The objective function is optimized for the selected coil and the selected frame.
7. The method according to claim 5, wherein generating a functional magnetic resonance image further comprises: A functional magnetic resonance image is reconstructed by expanding a difference k-space dataset as a sum of a multi-scale low-rank matrix indicating temporal variations, wherein the difference k-space dataset comprises a difference between a first frame of the k-space dataset at a first time point and a second frame of the k-space dataset at a second time point adjacent to the first time point.
8. The method according to claim 1, wherein: Applying one or more radio frequency excitation pulses further comprises applying one radio frequency excitation pulse while applying two gradients corresponding to two k-space spokes sequentially to each other.
9. The method according to claim 1, wherein: Applying a gradient sequence further comprises applying the gradient sequence, each gradient in the sequence corresponding to a k-space spoke extending from a first perimeter of k-space, through a center of k-space, and to a second perimeter of k-space opposite the first perimeter; and Applying one or more radio frequency excitation pulses further includes applying radio frequency excitation pulses to the subject as the center of the k-space is scanned by each gradient.
10. A magnetic resonance imaging method for dynamic imaging of a subject, the magnetic resonance imaging method comprising: Acquiring a k-space data set of a three-dimensional k-space volume of the subject, the acquisition comprising: Set the orientation to the initial angle; The free induction decay data set was acquired by: applying a gradient sequence to the subject, each gradient in the gradient sequence corresponding to a k-space spoke in a k-space spoke sequence, wherein the k-space spoke sequence defines a k-space segment having the orientation in a three-dimensional k-space volume; applying one or more radio frequency excitation pulses to the subject while applying the gradient sequence; and collecting the free induction decay data set corresponding to radio frequency emissions from the subject immediately after each radio frequency excitation pulse; acquiring a gradient echo dataset corresponding to the free induction attenuation dataset by reapplying the gradient sequence to the subject while acquiring echoes of the radio frequency emission to obtain the gradient echo dataset; updating the orientation to a golden angle along which the k-space segments divide the three-dimensional k-space volume according to the golden section; and The acquisition of the free induction attenuation data set, the acquisition of the gradient echo data set, and the updating of the orientation are repeated until a predetermined condition is satisfied. The method according to claim 10 , wherein acquiring the gradient echo data set further comprises acquiring a plurality of gradient echo data sets at a plurality of echo times.
12. The method according to claim 10, further comprising: The fMRI datasets were acquired by: The three-dimensional k-space volume is repeatedly acquired for a predetermined number of repetitions while applying functional stimulation to the subject.
13. The method according to claim 12, further comprising: An anatomical image is generated based on the functional magnetic resonance imaging dataset.
14. The method according to claim 12, further comprising: A functional magnetic resonance image is generated based on the functional magnetic resonance dataset.
15. The method of claim 14, wherein generating a functional magnetic resonance image further comprises reconstructing a frame of the functional magnetic resonance image at a certain time point based on a plurality of free induction decay data sets and a plurality of gradient echo data sets acquired in time intervals before and after the time point. 16 . The method of claim 14 , wherein generating a functional magnetic resonance image further comprises reconstructing a functional magnetic resonance image based on the functional magnetic resonance dataset by expanding the functional magnetic resonance dataset as a sum of multi-scale low-rank matrices.
17. The method according to claim 16, wherein the functional magnetic resonance data set is acquired by a plurality of coils and comprises a plurality of frames of a k-space data set, and reconstructing the functional magnetic resonance image further comprises: Separate the target function into separate frames and separate coils; and For multiple iterations, randomly selecting a coil from the plurality of coils and a frame from the plurality of frames; and The objective function is optimized for the selected coil and the selected frame.
18. The method of claim 14, wherein generating a functional magnetic resonance image further comprises: Frames of a functional magnetic resonance image are reconstructed by expanding a difference k-space dataset as a sum of a multi-scale low-rank matrix indicating temporal variations, wherein the difference k-space dataset comprises a difference between a first frame of the k-space dataset at a first time point and a second frame of the k-space dataset at a second time point adjacent to the first time point.
19. The method of claim 10, wherein: Applying one or more radio frequency excitation pulses further comprises applying one radio frequency excitation pulse while applying two gradients corresponding to two k-space spokes sequentially to each other.
20. The method of claim 10, wherein: Applying a gradient sequence further comprises applying the gradient sequence, each gradient in the sequence corresponding to a k-space spoke extending from a first perimeter of k-space, through a center of k-space, and to a second perimeter of k-space opposite the first perimeter; and Applying one or more radio frequency excitation pulses further includes applying radio frequency pulses to the subject as the center of the k-space is scanned by each gradient.
