A five-dimensional blood flow magnetic resonance imaging method without external gating

Abstract

The present invention belongs to the field of magnetic resonance imaging technology, and is specifically a five-dimensional blood flow magnetic resonance imaging method without external gating. The method of the present invention includes: data acquisition; continuous alternating sampling of the center and periphery of the K space to obtain training data and imaging data; image reconstruction; image restoration using the established model according to the training data and imaging data, generating three-dimensional blood flow images corresponding to each time frame during the scan, and calculating the physiological movement information of breathing and heartbeat, so as to realize real-time presentation of five dimensions; image post-processing; according to the extracted breathing and heartbeat signals, the blood flow image obtained by image reconstruction is further divided to form a five-dimensional matrix of heartbeat and breathing, which is used for comprehensive analysis of the influence of breathing on blood flow. The present invention can also flexibly set the required number of heartbeat phases and respiratory phases, and then generate a corresponding five-dimensional data matrix; and can capture the blood flow changes related to physiological pathology between heartbeat cycles.

Description

Technical Field

[0001] The invention belongs to the technical field of magnetic resonance imaging, and in particular relates to a five-dimensional blood flow magnetic resonance imaging method. Background Art

[0002] The traditional four-dimensional blood flow magnetic resonance imaging method usually uses gated K-space segmented scanning based on synchronously acquired ECG signals under the condition of free breathing of the subject, and only collects end-expiratory data with respiratory gating to suppress respiratory artifacts. In actual clinical practice, most patients with cardiovascular diseases cannot guarantee a stable heart rate and breathing, so it is very easy to have more data rejections due to the current acquisition data not meeting the preset conditions, which in turn leads to longer acquisition time. In order to solve the problem of low data acquisition efficiency and further extended acquisition time, researchers have successively proposed real-time blood flow magnetic resonance imaging technology and five-dimensional blood flow magnetic resonance imaging technology based on self-gating. Among them, real-time blood flow magnetic resonance imaging technology has currently realized continuous data acquisition and image reconstruction of a two-dimensional layer without applying ECG and respiratory gating, so it can detect blood flow changes in different heartbeat cycles, but its limitation is that it is still unable to image three-dimensional blocks. The five-dimensional blood flow magnetic resonance imaging technology based on self-gating has currently realized multi-dimensional blood flow magnetic resonance imaging with distinguishable respiratory status. Its blood flow imaging dimensions include three-dimensional space plus the time phase dimension within a heart cycle and the phase dimension within a respiratory cycle. However, this method cannot capture the blood flow changes related to physiological and pathological changes between heart cycles. Summary of the invention

[0003] The purpose of the present invention is to provide a five-dimensional blood flow magnetic resonance imaging method without external gating, that is, to achieve five-dimensional real-time blood flow imaging including three spatial dimensions and heartbeat and respiratory cycle change dimensions without using electrocardiogram and respiratory gating, thereby providing clinicians with more comprehensive physiological and pathological information about patients with cardiovascular diseases.

[0004] The five-dimensional blood flow magnetic resonance imaging method without external gating proposed in the present invention has a process framework as follows: Figure 1 As shown, the specific steps are:

[0005] (i) Data acquisition: Continuously and alternately sampling the center and periphery of the K space to obtain two sets of data: training data and imaging data;

[0006] (ii) Image reconstruction: Utilizing the low-rank characteristics of dynamic blood flow images and the sparse characteristics of time-frequency space to model, the model is used to perform image restoration based on the acquired training data and imaging data, generating three-dimensional blood flow images corresponding to each time frame during the scan, and also calculating the physiological movement information of breathing and heartbeat based on the acquired training data, thereby achieving real-time presentation of five dimensions;

[0007] (iii) Image post-processing: Based on the extracted respiratory and heartbeat signals, the blood flow image obtained by the previous image reconstruction step is further divided to form a five-dimensional matrix of xyz-heartbeat-respiration, which is used for comprehensive analysis of the impact of respiration on blood flow.

[0008] The specific process of data collection in step (1) is as follows:

[0009] First, based on the traditional four-dimensional blood flow magnetic resonance imaging sequence, the sampling trajectory is modified to obtain two sets of (k, t)-space data in a continuous and interlaced manner in time; one set is training data (also known as self-navigation data) and the other is imaging data. The three-dimensional sampling strategy based on the Cartesian coordinate system is adopted here, in which the training data mainly samples the central K space, and the imaging data samples the peripheral K space area, and a random distribution sampling mode is adopted in the time dimension. In the specific scanning process, considering the need to obtain physiological movement information such as breathing and heartbeat, it is required to set the head and foot direction as the frequency encoding direction of the K space data when placing the acquisition frame.

[0010] Here, in addition to adopting a three-dimensional sampling strategy based on a Cartesian coordinate system, data collection can also adopt a sampling strategy based on a non-Cartesian coordinate system.

[0011] The specific process of image reconstruction in step (ii) is as follows:

[0012] (1) Solving the time domain basis matrix V: First, the four sets of velocity encoding training data collected are arranged together and a one-dimensional Fourier transform is performed along the frequency encoding direction to the image domain to form a dynamic series matrix with low spatial resolution and high temporal resolution. Where M is the number of data points in the frequency encoding direction, N c is the number of coil channels, N v is the dimension collected in the speed encoding direction, T represents the number of sampled time frames, and then the matrix is ​​subjected to singular value decomposition to obtain the time domain basis matrix L represents the rank of the matrix P;

[0013] (2) Extracting breathing and heartbeat motion signals: Based on the low spatial resolution and high temporal resolution dynamic series matrix generated in the previous step First, average the signal modulus along the speed encoding direction to obtain Then, principal component analysis and cluster analysis based on multi-coil channels are used to obtain the changing signals of breathing and heartbeat during the scanning period;

[0014] (3) Calculate the sensitivity distribution map There are two ways to do this. One is to directly use the ratio of the low-resolution multi-channel image obtained by the pre-scan to the body coil image. The other is to average the acquired K-space imaging data along the time axis to obtain the full sampling data of each channel K-space, and then transform it to the image domain through Fourier transform, and then divide each channel image by the square root of all channel images to obtain the image. Where N x ,N y ,N z Indicates the number of pixels in the scanned 3D region of interest in the three directions of coordinate axes x, y, and z;

[0015] (4) Solving the spatial domain basis matrix U: Based on the time domain basis matrix V and sensitivity distribution map S obtained above, the following problem-solving model is established:

[0016]

[0017] Calculate the spatial domain basis matrix where d i is the data actually collected by the ith channel in the Cartesian coordinate system, Ω is the downsampling operator, and F s is the Fourier transform matrix from K space to spatial domain, Φ is the sparse transform operator, which is set here as the operator for Fourier transform to frequency domain along the time axis; λ is the weight of the sparse constraint term; its value can be determined based on experience or by imaging the same volunteer using the traditional four-dimensional blood flow movie imaging method and the proposed method, adjusting the value of λ so that the results of the two imaging methods within one heartbeat cycle are approximated; U on the right side of equation (1) is a variable, and the optimal U obtained by solving the optimization algorithm is a determined quantity, which is recorded here as

[0018] (5) Reconstructing real-time blood flow images: Based on the obtained spatial domain basis functions With time domain basis functions Through matrix multiplication operation A set of dynamic real-time blood flow images can be calculated; in addition, considering that the respiratory movement and heartbeat signals extracted previously have the same time resolution as the dynamic blood flow images reconstructed here, five-dimensional synchronized real-time blood flow information such as xyz-heartbeat-respiration can be formed, which can be used for subsequent real-time hemodynamic evaluation.

[0019] The specific process of image post-processing in step (iii) is as follows:

[0020] Based on the above results, the blood flow image obtained by reconstructing the image is divided into multiple respiratory states according to the extracted respiratory motion signal, and after each cardiac cycle is identified, each cardiac cycle is further divided to form multiple time phases. That is, a five-dimensional blood flow image matrix corresponding to the comprehensive cardiac cycle in which both respiration and heartbeat can be distinguished is established by data binning, so that the overall evaluation of the impact of respiratory motion on blood flow can be achieved at the same time.

[0021] Compared with the prior art, the present invention has the following characteristics and advantages:

[0022] (1) Compared with the existing traditional four-dimensional blood flow magnetic resonance imaging methods, the present invention does not require electrocardiogram and respiratory gating, and the acquisition time is not affected by the changes in the subject's heart rate and respiration. The acquired data can be efficiently used in subsequent image reconstruction, and the acquisition time is short and stable. The blood flow imaging results of the present invention are of higher quality because the present invention can obtain higher-quality respiratory motion information based on the high-frequency self-navigation data obtained during the acquisition process, and has a more detailed division of different respiratory states during the reconstruction process, thereby obtaining a dynamic blood flow image with fewer artifacts;

[0023] (2) Compared with the existing real-time blood flow magnetic resonance imaging methods, the present invention further expands the spatial dimension, not only realizing real-time blood flow imaging of three-dimensional blocks, but also providing additional synchronization information of heartbeat and breathing dimensions;

[0024] (3) Compared with the existing five-dimensional blood flow magnetic resonance imaging method, the present invention can not only provide the same imaging information as the existing five-dimensional blood flow imaging method, but also flexibly set the required cardiac phase number and respiratory phase number, thereby generating a corresponding five-dimensional data matrix with a smaller amount of calculation; in addition, it can also simultaneously capture the blood flow changes related to physiological and pathological changes between cardiac cycles. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flowchart of the five-dimensional blood flow magnetic resonance imaging method without external gating of the present invention.

[0026] Figure 2 It is a schematic diagram of the respiratory movement and heart beat signals extracted in the present invention.

[0027] Figure 3 It is a schematic diagram of data binning according to the extracted breathing and heartbeat signals in the present invention.

[0028] Figure 4The results of reconstruction of an atrial fibrillation patient using the method proposed in the present invention include the flow waveforms corresponding to the ascending aorta, aortic arch and descending aorta, as well as synchronous physiological motion signals such as breathing and heartbeat.

[0029] Figure 5 The results of different respiratory phases of an atrial fibrillation patient reconstructed using the method proposed by the present invention include the modulus image (a) and the systolic flow velocity distribution diagram (b) under four respiratory states from the end of exhalation to the end of inspiration. DETAILED DESCRIPTION

[0030] Taking the thoracic aorta blood flow magnetic resonance imaging based on four-point velocity encoding as an example, the present invention is further explained. It is divided into the following three parts:

[0031] Part I: Data Collection

[0032] First, based on the traditional four-dimensional blood flow magnetic resonance imaging sequence, the sampling trajectory is modified to achieve a continuous interlaced approach in time to obtain two sets of (k, t)-space data, one set is training data (also known as self-navigation data), and the other set is imaging data. The three-dimensional sampling strategy based on the Cartesian coordinate system is adopted here, in which the training data is mainly sampled from the central K space, while the imaging data is sampled from the peripheral K space area, and a random distribution sampling pattern is adopted in the time dimension. In the specific scanning process, considering that we need to obtain physiological movement information such as breathing and heartbeat, it is required to set the head and foot direction as the frequency encoding direction of the K space data when placing the acquisition frame.

[0033] Part II: Image Reconstruction

[0034] (1) Solving the time domain basis matrix V: First, the four sets of velocity encoding training data collected are arranged together and Fourier transformed into the image domain along the frequency encoding direction to form a dynamic series matrix with low spatial resolution and high temporal resolution Where M is the number of data points in the frequency encoding direction, N c is the number of coil channels, N v is the dimension of the velocity encoding direction (here based on the four-point velocity encoding strategy, so N v =4), T represents the number of sampled time frames, and then the matrix is ​​subjected to singular value decomposition to obtain the time domain basis matrix Where L represents the rank of the matrix P;

[0035] (2) Extracting breathing and heartbeat motion signals: Based on the low spatial resolution and high temporal resolution dynamic series matrix generated in the previous step First, average the signal modulus along the speed encoding direction to obtain Then, principal component analysis and cluster analysis based on multi-coil channels are used to obtain the change signals of breathing and heartbeat during the scanning period;

[0036] (3) Calculate the sensitivity distribution map There are two methods. One is to directly use the ratio of the low-resolution multi-channel image obtained by the pre-scan to the body coil image; the other is to average the acquired K-space imaging data along the time axis to obtain the full sampling data of each channel K-space, and then transform it to the image domain through Fourier transform, and then divide each channel image by the square root of all channel images to obtain the image; (where N x ,N y ,N z Indicates the number of pixels in the scanned 3D region of interest in the three directions of the coordinate axes x / y / z)

[0037] (4) Solving the spatial domain basis matrix U: Based on the time domain basis matrix V and sensitivity distribution map S obtained above, the following problem-solving model is established:

[0038]

[0039] Calculate the spatial domain basis matrix where d i is the data actually collected by the ith channel in the Cartesian coordinate system, Ω is the downsampling operator, and F s is the Fourier transform matrix from K space to spatial domain, Φ is the sparse transform operator, which is set here as the operator that performs Fourier transform to frequency domain along the time axis, and λ is the weight of the sparse constraint term;

[0040] (5) Reconstructing real-time blood flow images: Based on the obtained spatial domain basis functions With time domain basis functions Through matrix multiplication operation A set of dynamic real-time blood flow images can be calculated; in addition, considering that the respiratory movement and heartbeat signals extracted previously have the same time resolution as the dynamic blood flow images reconstructed here, five-dimensional synchronized real-time blood flow information such as xyz-heartbeat-respiration can be formed, which can be used for subsequent real-time hemodynamic evaluation.

[0041] Part III: Image post-processing.

[0042] Based on the above results, the reconstructed blood flow image can be divided into multiple respiratory states according to the extracted respiratory motion signal, and each cardiac cycle can be further divided into multiple time phases after each cardiac cycle is identified. That is, a five-dimensional blood flow image matrix corresponding to a comprehensive cardiac cycle in which both respiration and heartbeat can be distinguished can be established by using data binning, so that an overall evaluation of the impact of respiratory motion on blood flow can be achieved at the same time.

[0043] In vivo scanning example:

[0044] The method of the present invention was verified by in vivo experiments. The experiment was conducted on a Philips 3.0T magnetic resonance scanner (Ingenia, Philips Healthcare, Best, The Netherlands), using the original 28-channel chest and abdominal scanning coil. We programmed based on the manufacturer's original four-dimensional blood flow magnetic resonance pulse sequence to achieve data acquisition. We have recruited some subjects to conduct scanning experiments on this imaging method. During the scanning process, no ECG gating or respiratory control was used to assist reconstruction. The specific magnetic resonance scanning parameters are as follows: spatial resolution = 2.4×2.4×2.4mm 3 , TR (repetition time) = 4.3 ms, TE (echo time) = 2.5 ms, FA (flip angle) = 5°, encoding speed = 150 / 200 / 150 cm / s (AP / FH / RL), time resolution 8×TR = 34.4 ms, scanning time 10 minutes. The reconstruction results obtained according to the present invention are given below.

[0045] Figure 4 The results of imaging a patient with atrial fibrillation using the method of the present invention are given. In the three-dimensional blood flow map of the aorta obtained by imaging the patient with atrial fibrillation, sections perpendicular to the diameter direction are cut at the ascending aorta, aortic arch and descending aorta, and the waveform of the blood flow in the section changing with time is plotted on the graph. Figure 4 It is shown in the figure, and the waveform of the respiratory and cardiac motion signals synchronized with the time is also given. It can be seen from the figure that the method of the present invention reconstructs the blood flow changes of different continuous cardiac cycles, which is impossible to do with traditional four-dimensional blood flow magnetic resonance imaging. It is worth noting that the method of the present invention also captures the special changes of blood flow in some abnormal cardiac cycles, providing more pathological information for the diagnosis and treatment of patients with arrhythmia in clinical practice.

[0046] Figure 5 The corresponding results of different respiratory phases obtained by imaging an atrial fibrillation patient using the method of the present invention are given, which can be used to analyze the impact of breathing on blood flow. Figure 5The reconstructed value images of the aorta and the maximum intensity projection of the peak flow velocity during systole are displayed under four respiratory states: end-expiration, exhalation, inspiration, and end-inspiration. It can be seen from the figure that the displacement of the aorta and the adjacent liver during breathing can be well captured, and the flow velocity changes between different respiratory phases can also be clearly observed from the maximum intensity projection of the peak flow velocity.

[0047] References

[0048] [1]Joseph AA, Merboldt KD, Voit D, Zhang S, Uecker M, Lotz J, et al. Real-time phase-contrast MRI of cardiovascular blood flow using undersampledradial fast low-angle shot and nonlinear inverse reconstruction. NMRBiomed. 2012; 25:917–924.

[0049] [2] Tan Z, Roeloffs V, Voit D, Joseph AA, Untenberger M, Merboldt KD, et al. Model-based reconstruction for real-time phase-contrast flow MRI: Improved spatiotemporal accuracy. MagnReson Med. 2017;77:1082–1093.

[0050] [3]Kollmeier JM,Tan Z,Joseph AA,Kalentev O,Voit D,Merboldt KD,etal.Real-time multi-directional flow MRI using model-based reconstructions ofundersampled radial FLASH-A feasibility study.NMR Biomed.2019;32(12):e4184.

[0051] [4]Cheng JY,Zhang T,Alley MT,Uecker M,Lustig M,Pauly JM,etal.Comprehensive multi-dimensional MRI for the simultaneous assessment ofcardiopulmonary anatomy and physiology.Sci Rep.2017;7(1):5330.

[0052] [5]Walheim J,Dillinger H,Kozerke S.Multipoint 5D flow cardiovascularmagnetic resonance-accelerated cardiac-and respiratory-motion resolvedmapping of mean and turbulent velocities.J Cardiovasc MagnReson.2019;21(1):42.

[0053] [6]Ma LE,Yerly J,Piccini D,Di Sopra L,Roy CW,Carr JC,et al.5D flowMRI:A fully self-gated,free-running framework for cardiac and respiratorymotion-resolved 3D hemodynamics.RadiolCardiothorac Imaging.2020;2(6):e200219.

[0054] [7]Bastkowski R,Bindermann R,Brockmeier K,Weiss K,Maintz D,GieseD.Respiration dependency of caval blood flow in patients with Fontancirculation:Quantification using 5D flow MRI.RadiolCardiothorac Imaging.2019;1(4):e190005。

Claims

1. A five-dimensional blood flow magnetic resonance imaging method without external gating, characterized in that: The specific steps are: (i) Data acquisition: Continuously and alternately sampling the center and periphery of the K space to obtain training data and imaging data; (ii) Image reconstruction: Utilizing the low-rank characteristics of dynamic blood flow images and the sparse characteristics of time-frequency space to model, the model is used to perform image restoration based on the acquired training data and imaging data, and a three-dimensional blood flow image corresponding to each time frame during the scan is generated; in addition, the physiological movement information of breathing and heartbeat is calculated based on the acquired training data, and real-time presentation of five dimensions is achieved; (iii) Image post-processing: Based on the extracted respiration and heartbeat signals, the blood flow image obtained by the previous image reconstruction is further divided to form a five-dimensional matrix of xyz-heartbeat-respiration for comprehensive analysis of the effect of respiration on blood flow; The specific process of data collection in step (1) is as follows: First, the sampling trajectory is modified on the basis of the four-dimensional blood flow magnetic resonance imaging sequence, and two sets of (k, t)-space data are obtained in a continuous and interleaved manner in time; one set is training data, also known as self-navigation data, and the other set is imaging data; a three-dimensional sampling strategy based on the Cartesian coordinate system is adopted here, in which the training data samples the central K space and the imaging data samples the peripheral K space area; and a randomly distributed sampling mode is adopted in the time dimension; during the scanning process, when placing the acquisition frame, the head and foot direction is set as the frequency encoding direction of the K space data to obtain physiological movement information such as breathing and heartbeat; The specific process of image reconstruction in step (ii) is as follows: (1) Solving the time domain basis matrix V: First, the four sets of velocity encoding training data collected are arranged together and a one-dimensional Fourier transform is performed along the frequency encoding direction to the image domain to form a dynamic series matrix with low spatial resolution and high temporal resolution. Where M is the number of data points in the frequency encoding direction, N c is the number of coil channels, N v is the dimension collected in the speed encoding direction, T represents the number of sampled time frames, and then the matrix is ​​subjected to singular value decomposition to obtain the time domain basis matrix Where L represents the rank of the matrix P; (2) Extracting breathing and heartbeat motion signals: Based on the low spatial resolution and high temporal resolution dynamic series matrix generated in the previous step First, average the signal modulus along the speed encoding direction to obtain Then, principal component analysis and cluster analysis based on multi-coil channels are used to obtain the changing signals of breathing and heartbeat during the scanning period; (3) Calculate the sensitivity distribution map There are two ways. One is to directly use the ratio of the low-resolution multi-channel image obtained by the pre-scan and the body coil image to obtain it; the other is to average the acquired K-space imaging data along the time axis to obtain the full sampling data of each channel K-space, and then transform it to the image domain through Fourier transform, and then divide each channel image by the square root of all channel images to obtain the image; where N x ,N y ,N z Indicates the number of pixels in the scanned 3D region of interest in the three directions of coordinate axes x, y, and z; (4) Solving the spatial domain basis matrix U: Based on the time domain basis matrix V and sensitivity distribution map S obtained above, the following problem model is established: Calculate the spatial domain basis matrix Among them, d i is the data actually collected by the ith channel in the Cartesian coordinate system, Ω is the downsampling operator, and F s is the Fourier transform matrix from K space to spatial domain, Φ is the sparse transform operator, which is set here as the operator that performs Fourier transform to frequency domain along the time axis; λ is the weight of the sparse constraint term; (5) Reconstructing real-time blood flow images: Based on the obtained spatial domain basis functions With time domain basis functions Through matrix multiplication operation That is, a set of dynamic real-time blood flow images are calculated; in addition, considering that the respiratory movement and heartbeat signals extracted previously have the same time resolution as the dynamic blood flow images reconstructed here, five-dimensional synchronized real-time blood flow information such as xyz-heartbeat-respiration is formed, which can be used for subsequent real-time hemodynamic evaluation.

2. The five-dimensional blood flow magnetic resonance imaging method according to claim 1, characterized in that: The specific process of image post-processing in step (iii) is as follows: For the blood flow image obtained by reconstructing the image, it is divided into multiple respiratory states according to the extracted respiratory movement signal, and after each cardiac cycle is identified, each cardiac cycle is further divided to form multiple time phases; that is, a five-dimensional blood flow image matrix corresponding to the comprehensive cardiac cycle in which both respiration and heartbeat can be distinguished is established by data binning, so that an overall evaluation of the impact of respiratory movement on blood flow can be achieved at the same time.

Images