KR101575798B1 - Apparatus and method for magnetic resonance image processing - Google Patents

Abstract

A method of processing a magnetic resonance image according to an embodiment of the present invention includes the steps of: (a) receiving data of a plurality of frames through a receiving coil; (b) estimating a space-time kernel vector representing a correlation between temporally successive frames of the received data; (c) applying a low-coefficient approximation to the space-time kernel matrix made up of the space-time kernel vector to presume a space-time kernel vector; and (d) reconstructing an image based on the re-estimated space-time kernel vector.

Description

TECHNICAL FIELD [0001] The present invention relates to a magnetic resonance imaging apparatus,

The present invention relates to an apparatus and a method for processing a magnetic resonance image (MRI).

Generally, an apparatus for processing magnetic resonance imaging (MRI) is a device for acquiring a tomographic image of a specific region of a patient by using a resonance phenomenon in response to the supply of electromagnetic energy. The X- It is widely used because it does not have exposure and can acquire a tomographic image relatively easily.

To briefly describe a method of generating a magnetic resonance image, a RF signal of a high frequency is applied to a subject for imaging a magnetic resonance image a plurality of times to excite a spin of a nucleus in the subject. Through the application of a pulse train to such a magnetic resonance apparatus, various signals such as a free induction attenuation signal (FID) and a spin echo are generated in a magnetic resonance imaging apparatus, and these signals are selectively acquired to generate a magnetic resonance image.

However, such a magnetic resonance image has a limitation in that the imaging time is longer than that of other imaging devices. However, a compressed sensing technique has been introduced as a method of reducing the imaging time. Compression-sensing techniques have shown that if the original signal has sparsity, the original signal can be reconstructed even if it is sampled at a frequency lower than the Nyquist sampling frequency. In this case, in order to reconstruct the original signal by the compression sensing technique, it is necessary to transform the original signal into the region where the sparseness of the original signal is maximized. In this conversion region, the original signal is repeatedly The original signal is reconstructed through an estimation method.

Meanwhile, a parallel magnetic resonance imaging (pMRI) method uses coil sensitivity information to estimate a non-measured signal during acquisition of image data. The unmeasured signal can be estimated using the preceding information such as the amount of spatial variation of the coil sensitivity in the image space or the spatial interaction value between adjacent video signals on the k-space. Therefore, in order to avoid defects and noise in image reconstruction, it is important to collect accurate precursor information for calibration.

In a conventional parallel magnetic resonance imaging, the calibration signal is generally obtained before or after the acquisition of the actual image data, which is referred to as a separate calibration. However, separate calibrations are subject to subordinate movements, potentially resulting in inconsistencies between the calibration signal and the image data. In order to solve this problem, it is known that it is desirable to perform a self-calibration in which a calibration signal and image data are obtained simultaneously in a single image measurement process.

Meanwhile, in general, a parallel magnetic resonance imaging (SRI) method is known in which a sense (sens) encoding method based on an image and a generalized auto-calibrating Partial Parallel Acquisition (GRAPPA) method are known.

An image-based imaging method, such as sense, computes coil sensitivity information by separating individual coil images through self-calibration in the image space, where the individual coil images are inversely transformed into inverse Fourier transforms Fourier transform. Since the coil sensitivity inverse matrix is performed pixel by pixel in the image reconstruction using the coil sensitivity information, self-calibration based on the image requires highly accurate coil sensitivity information. As a result, a large number of calibration signals are required at the center of the k-space, and the time taken to construct the image is inevitable. In addition, if the field of view is smaller than the subject being imaged, the self-calibration based on the image may also indicate residual aliasing artifacts in the image reconstruction.

On the other hand, the k-space based image method such as Grafa calculates the spatial correlation (convolution kernels) between the calibration signal and the adjacent measured source signal by self-calibration. Unlike the sense, high coil sensitivity information There is no limitation in reconstructing the viewing angle without necessity. Nevertheless, when the data of the video signal is damaged by noise or the spatial interaction value changes, a residual aliasing defect and amplified noise are generated in the reconstructed image.

In this connection, United States Patent No. 2006-0050981 (Technique for parallel MRI imaging (kt grappa)) relates to calibration in a low frequency band of kt space, but does not consider the noise correlation between the coils / RTI >

An object of the present invention is to provide a dynamic parallel magnetic resonance imaging apparatus and method capable of maintaining a high resolution while suppressing noise that may appear when the number of acquired data decreases.

According to a first aspect of the present invention, there is provided a magnetic resonance imaging method comprising the steps of: (a) receiving data of a plurality of frames through a reception coil; (b) estimating a space-time kernel vector representing a correlation between temporally successive frames of the received data; (c) applying a low-coefficient approximation to the space-time kernel matrix made up of the space-time kernel vector to presume a space-time kernel vector; and (d) reconstructing an image based on the re-estimated space-time kernel vector.

According to a second aspect of the present invention, there is provided a magnetic resonance imaging apparatus comprising: a magnetic resonance apparatus including a main magnet, a gradient coil, and an RF coil, the magnetic resonance apparatus being configured to surround the subject; And a signal processing unit for generating a magnetic resonance image based on the magnetic resonance signal received from the signal transmitting and receiving unit, wherein the signal processing unit comprises: Estimates a space-time kernel vector indicating a correlation between temporally successive frames in the data of the frame, reconstructs the image based on the re-estimated space-time kernel vector by applying a low-coefficient approximation to a space-time kernel matrix composed of space-time kernel vectors .

According to the above-mentioned problem solving means of the present invention, correlation between temporally consecutive frame data is estimated, and when low-coefficient approximation is applied to the frame data, Can be maintained. Particularly, in the process of estimating the correlation between frame data, noise can be modeled considering the noise between the receiving coils, and the correlation can be estimated not only with respect to temporally continuous frame data but also with all data in the data block.

1 is a diagram illustrating a magnetic resonance imaging apparatus according to an embodiment of the present invention.
2 is a flowchart illustrating a method of processing a magnetic resonance image according to an embodiment of the present invention.
3 is a view for explaining a method of processing a magnetic resonance image according to an embodiment of the present invention.
FIG. 4 is a view showing a specific algorithm of a method for processing a magnetic resonance image according to an embodiment of the present invention.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings, which will be readily apparent to those skilled in the art. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In order to clearly illustrate the present invention, parts not related to the description are omitted, and similar parts are denoted by like reference characters throughout the specification.

Throughout the specification, when a part is referred to as being "connected" to another part, it includes not only "directly connected" but also "electrically connected" with another part in between . Also, when an element is referred to as "comprising ", it means that it can include other elements as well, without departing from the other elements unless specifically stated otherwise.

1 is a diagram illustrating a magnetic resonance imaging apparatus according to an embodiment of the present invention.

The magnetic resonance imaging apparatus 10 includes a magnetic resonance imaging apparatus 100, a signal transmission and reception unit 200, a signal processing unit 300, an image output unit 400, a control unit 500, and a user interface 600.

The magnetic resonance apparatus 100 includes a shield of a cylindrical structure surrounding a subject to be photographed, a main magnet provided inside the shield, a gradient coil, an RF coil, and the like. The main magnet, the gradient coil, the RF coil, etc., generate a magnetic field for inducing magnetic resonance signals from the nuclei in the human body. The gradient coil generates an inclined magnetic field that varies in a plurality of directions, for example, a constant gradient for each of the x direction, the y direction, and the z direction, in proportion to the distance from the reference position in the sperm field generated by the main magnet. do. Here, the reference position may be the origin of the three-dimensional coordinate system when the space in which the static magnetic field generated by the main magnet exists is represented by a three-dimensional coordinate system. By the gradient magnetic field generated by the gradient coil, each of the magnetic resonance signals received through the RF coil has position information in the three-dimensional space. On the other hand, the gradient coil may be composed of an X gradient coil that generates a gradient magnetic field varying in the x direction, a Y gradient coil that generates a gradient magnetic field that varies in the y direction, and a Z gradient coil that generates a gradient magnetic field that varies in the z direction.

The RF coil outputs an electromagnetic wave signal having a radio frequency corresponding to the kind of the atomic nucleus in order to transition the atomic nucleus from a low energy state to a high energy state. Also, the RF coil receives an electromagnetic wave signal radiated from the nuclei inside the subject. The electromagnetic wave signal thus received is called a free induction decay (FID) signal or an echo signal. The length of the section from the application time point of the electromagnetic wave signal to the subject, that is, the generation time point of the electromagnetic wave signal to the reception time point of the electromagnetic wave signal from the subject is called an echo time (TE) The length of the section where the application is repeated is called a repetition time (TR).

The signal transmitting and receiving unit 200 generates an AC signal whose frequency varies at a constant slope with respect to each of the x direction, the y direction, and the z direction according to the control signal input from the controller 500, and outputs the generated AC signal to the gradient coil. Further, the control unit 500 generates an AC signal having a pulse train according to the control signal, and outputs the AC signal to the RF coil. Also, the signal transmitting and receiving unit 200 receives the magnetic resonance signal received through the RF coil.

The received magnetic resonance signal is transmitted to the signal processing unit 300, and the signal processing unit 300 generates a magnetic resonance image using the signal. The signal processing unit 300 generates a magnetic resonance image based on the information obtained through the spin echo signal and the gradient echo signal. In the present invention, a method of using a space-time kernel vector that indicates a correlation of temporally successive frames using a dynamic parallel MR image processing in the signal processing is proposed, and a specific method thereof will be described later .

The image output unit 400 outputs a magnetic resonance image generated through the signal processing unit 300 through a display or the like.

The control unit 500 controls operations of the MRI apparatus 100, the signal transmission / reception unit 200, the signal processing unit 300, and the video output unit 400 according to a command input from the user through the user interface 600 do. For example, the signal transmitting and receiving unit 200 controls the gradient coil and the RF coil to output an AC signal, or the magnetic resonance signal received through the RF coil is transmitted to the signal processing unit 300 via the signal transmitting and receiving unit 200 Respectively.

The user interface 600 receives a command from a user and transmits the command to the control unit 500. The user interface 600 may be implemented by a graphic user interface program and a keyboard, a mouse, or the like as an input device, but is not limited thereto.

1 refers to a hardware component such as software or an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit), and performs predetermined roles .

However, 'components' are not meant to be limited to software or hardware, and each component may be configured to reside on an addressable storage medium and configured to play one or more processors.

Thus, by way of example, an element may comprise components such as software components, object-oriented software components, class components and task components, processes, functions, attributes, procedures, Routines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables.

The components and functions provided within those components may be combined into a smaller number of components or further separated into additional components.

Now, the dynamic parallel MR imaging technique according to the present invention will be described.

In order to minimize the patient 's motion, the conventional k - space parallel imaging method uses a variable density sampling form to shorten the data acquisition time and simultaneously restore the absolute value image of each coil to maintain the signal - to - noise ratio. However, the conventional k-space parallel image technique can not reconstruct the correct image if the data is distorted due to noise or the correlation between data varies according to each position of k-space. In order to overcome the disadvantages of the conventional parallel image technique, the present invention model noise when calculating the correlation between data, thereby improving the accuracy of the correlation and dynamically converting the static calibration process, And to restore the image by estimating the correlation between the variable data.

2 is a flowchart illustrating a method of processing a magnetic resonance image according to an embodiment of the present invention.

First, a space-time kernel vector indicating a correlation between temporally successive frames is estimated from data of a plurality of frames received through a reception coil (S210).

Variations in the coil sensitivity map caused by patient motion are smoothly varying, the background ground image is hardly changed, and there is a very high correlation between frames in different time periods. In order to reflect the correlation in the time direction, a spatiotemporal kernel matrix can be constructed by stacking the space-time kernel vectors g _l of the l-th coil, and this space-time kernel matrix can be expressed by Equation (1) Can be defined.

[Equation 1]

In this case, the space-time kernel vector represents a correlation between two frames temporally adjacent to each other. N _t column vectors of the kernel sequence matrix G _l represent the temporal variation of the reconstruction kernels for a plurality of time frames and N _k row vectors represent the spatiotemporal correlation between each time frame of the adjacent kt signals Represents the kernel weight.

On the other hand, each space-time kernel vector is estimated using a Kalman filter.

Assuming that the spatiotemporal kernel between interframes is independent from each other over a plurality of time frames, space-time kernels in the Nyquist sampled region of K space are modeled as a block diagonal matrix as shown in Equation (2) .

&Quot; (2) "

는 타겟 캘리브레이션 신호를 나타내고,

는 시간 방향 및 공간 방향을 따라 이웃하는 신호들로 이루어진 소스 행렬,

는 시공간 커널 벡터를 나타낸다. 단위 시야(FOV)의 일부분만이 변화하고, 단위 시야의 대부분은 움직이지 않기 때문에, 서로 연속되는 프레임들 간의 시공간 커널 벡터는 미세한 차이를 갖게 될 것이며, 이러한 특성을 이용하여 상태 변화 모델을 적용할 수 있다. 즉, 블록 대각 행렬 구조를 갖는 시공간 커널 벡터를 수학식 3의 상태 모델(state mode)과 수학식 4의 관찰 모델(observation model)을 이용하여 정의할 수 있다.

Represents a target calibration signal,

A source matrix consisting of neighboring signals along the time and space directions,

Represents a space-time kernel vector. Since only a part of the unit field of view (FOV) changes and most of the unit field of view does not move, the space-time kernel vector between successive frames will have a slight difference. . That is, the space-time kernel vector having the block diagonal matrix structure can be defined using the state model of Equation (3) and the observation model of Equation (4).

&Quot; (3) "

&Quot; (4) "

는 프로세스 잡음을 나타내고,

는 측정 잡음을 나타내는데, 각각은 서로 통계적으로 독립적인 관계에 있다.

Lt; / RTI > represents process noise,

Represent measurement noise, each of which is statistically independent of each other.

The model of Equation (3) and Equation (4) can be solved by a Kalman filter estimation method defined by the following Equation (5).

&Quot; (5) "

는 오차 공분산(covariance),

는 프로세스 잡음,

는 측정 잡음,

는 칼만 이득을 각각 나타낸다.

Covariance of error,

Lt; / RTI >

The measurement noise,

Represents the Kalman gain, respectively.

The Kalman filter estimation method is a recursive filter that tracks the state of a linear dynamic system that contains noise. It has a stochastic error in the measured value of an object, and a state at a specific time point of an object is a linear relationship . This Kalman filter is known to be an optimal estimation method that finds the state variable of the system using probabilistic models and measured values of the target system. The Kalman filter is a method of selecting the most probable value as the estimated value according to the probability distribution of the estimated value, and calculates the final estimated value by appropriately correcting the predicted value with the predictive error of the measured value. That is, based on the system model, prediction is made on the state of the next time point and the value of the error covariance, and a new estimated value is calculated by compensating the difference between the measured value and the predicted value. The calculated estimate is the final product, and the process is repeated to update the estimate repeatedly. Since the Kalman filter estimation method is disclosed in the prior art, a more detailed calculation method using the Kalman filter estimation method will be omitted.

In the present invention, the space-time kernel vector is estimated using the Kalman filter estimation method.

Next, the space-time kernel vector is re-estimated by applying a low-coefficient approximation to the space-time kernel matrix composed of space-time kernel vectors (S220).

Since most of the MRI images vary slowly over time, the space-time kernels used in dynamic parallel MR imaging can be approximated with only a few singular values because of the small degree of change. Therefore, by applying a low rank approximation method to the space-time kernel vector accumulated for a predetermined time, additional information can be obtained in the time domain.

The rank of the space-time kernel sequence matrix is much smaller than min (N _k , N _t ), and the noise-canceled space-time kernel sequence matrix G _l can be defined by Equation (6).

&Quot; (6) "

R denotes the coefficient of the kernel sequence matrix G _l , U _r (m) denotes the kernel basis function, and V _r (t) denotes the temporal basis function.

By making the kernel sequence matrix a structured low coefficient matrix, the calibration process of the parallel magnetic resonance imaging method is processed in the form of a structured low coefficient matrix in the Nyquist sampled region of K space over the entire time frame. In order to solve this problem, the minimum coefficients are found through the following equations (7) and (8).

&Quot; (7) "

&Quot; (8) "

는 앞서 칼만 필터를 통해 추정한 시공간 커널 벡터의 집합(

)을 나타내고,

는 주어진 시간 범위내의 전체 K 스페이스를 나타내며,

는 직전 K 스페이스로부터 전체 K 스페이스를 형성하는 재구성 연산자를 나타낸다.At this time,

Is a set of space-time kernel vectors estimated through Kalman filter (

),

Represents the total K space within a given time range,

Represents a reconstruction operator that forms the entire K space from the immediately preceding K space.

Further,?,?,? _L and? _S represent normalization parameters.

와

는 모두 알려지지 않은 상태이고, 수학식 7은 비볼록(non-convex)으로, 이를 해결하기 위해 하나의 변수에 대해 볼록 상태를 갖고 다른 변수는 고정을 시키는 교번 최소화(alternating minimization) 알고리즘을 사용한다. In Equations (7) and (8)

Wow

(7) is non-convex, and an alternating minimization algorithm is used in which the convex state of one variable is fixed and the other variable is fixed in order to solve this problem.

값을 기초로 수학식 9를 통해

를 추정할 수 있다.For example,

≪ RTI ID = 0.0 >

Can be estimated.

&Quot; (9) "

That is, the space-time kernel vector is re-estimated through solving the minimization problem of Equation (9).

Next, the image is reconstructed based on the re-estimated space-time kernel vector (S230).

를 수학식 10의 최소화 문제에 적용하여 이미지 재구성을 수행한다.That is,

Is applied to the minimization problem of Equation (10) to perform image reconstruction.

&Quot; (10) "

Next, when the data of the new frame is received, steps S210 to S230 performed previously are repeatedly performed with additional consideration. At this time, steps S210 to S230 performed previously for data of a new frame through block sliding are repeatedly performed, which will be described with reference to the drawings.

3 is a view for explaining a method of processing a magnetic resonance image according to an embodiment of the present invention.

First, space-time kernel vector estimation and low-coefficient approximation processing are performed on the plurality of temporally continuous frame data 30, 32, 34, and 36. At this time, the frame data to be processed is referred to as a data block, and the number of frame data included in the data block can be changed according to the setting of the user.

Then, when new frame data 38 is received, new frame data 38 is added, except for the data 30 of the first received frame in the data block, Has the same effect. In this manner, the data blocks 32, 34, 36, and 38 of the plurality of frames included in the data block are subjected to the preceding steps S210 to S230 after the data blocks are slid. Therefore, it is possible to estimate not only the correlation between the adjacent frames, but also the correlation between the entire frame data included in the data block.

In the case where the new frame data 40 is received thereafter, steps S210 to S230 in which the data blocks are slid and the data 34, 36, 38, and 40 of the plurality of frames are previously performed is then performed .

FIG. 4 is a view showing a specific algorithm of a method for processing a magnetic resonance image according to an embodiment of the present invention.

As shown in the figure, it includes a space-time kernel vector satisfying Equation (7) described above and reconstructing an image based on the space-time kernel vector.

At this time, in estimating the space-time kernel vector, the space-time kernel vector is estimated by considering the noise component of the receiving coil using the Kalman filter estimation method described above. Also, the space-time kernel vector is re-estimated through the low-coefficient approximation, and the image reconstruction is performed based on the estimation.

One embodiment of the present invention may also be embodied in the form of a recording medium including instructions executable by a computer, such as program modules, being executed by a computer. Computer readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. In addition, the computer-readable medium may include both computer storage media and communication media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Communication media typically includes any information delivery media, including computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, or other transport mechanism.

While the methods and systems of the present invention have been described in connection with specific embodiments, some or all of those elements or operations may be implemented using a computer system having a general purpose hardware architecture.

It will be understood by those skilled in the art that the foregoing description of the present invention is for illustrative purposes only and that those of ordinary skill in the art can readily understand that various changes and modifications may be made without departing from the spirit or essential characteristics of the present invention. will be. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive. For example, each component described as a single entity may be distributed and implemented, and components described as being distributed may also be implemented in a combined form.

The scope of the present invention is defined by the appended claims rather than the detailed description and all changes or modifications derived from the meaning and scope of the claims and their equivalents are to be construed as being included within the scope of the present invention do.

100: Magnetic resonance apparatus 200: Signal transmission /
300: signal processing unit 400: video output unit
500: control unit 600: user interface

Claims (6)

A method for processing a magnetic resonance image,
(a) receiving data of a plurality of frames through a receive coil;
(b) estimating a space-time kernel vector indicating a correlation between temporally successive frames of the plurality of frames from the data of the plurality of frames;
(c) applying a low-coefficient approximation to the space-time kernel matrix consisting of the space-time kernel vectors to presume a space-time kernel vector; and
(d) reconstructing an image based on the re-estimated space-time kernel vector.
The method according to claim 1,
The step of estimating the space-time kernel vector
And estimating the space-time kernel vector according to a Kalman filter estimation method, taking into consideration a noise component of the reception coil. The method according to claim 1,
(e) receiving data of a new frame temporally continuous to the plurality of frames;
(f) estimating the space-time kernel vector based on data of a frame other than the data of the first received frame among the plurality of frames and data of the new frame;
(g) applying a low-coefficient approximation to a space-time kernel matrix composed of space-time kernel vectors estimated through the step (f), and re-classifying the space-time kernel vector; And
(h) reconstructing an image based on the spatially resampled kernel vector through the step (g). A magnetic resonance imaging apparatus comprising:
A magnetic resonance apparatus including a main magnet, a gradient coil, and an RF coil,
A signal transmitting and receiving unit for transmitting an electrical signal to the magnetic resonance apparatus and receiving a magnetic resonance signal from the magnetic resonance apparatus;
And a signal processing unit for generating a magnetic resonance image based on the magnetic resonance signal received from the signal transmission / reception unit,
The signal processing unit,
Estimating a space-time kernel vector indicating a correlation between temporally successive frames of the plurality of frames from data of a plurality of frames based on the magnetic resonance signal, applying a low-coefficient approximation to a space-time kernel matrix composed of space-time kernel vectors And reconstructs the image based on the re-estimated space-time kernel vector.
5. The method of claim 4,
The signal processing unit,
And estimating the space-time kernel vector according to a Kalman filter estimation method, taking into consideration a noise component of the reception coil. 5. The method of claim 4,
The signal processing unit,
Space-time kernel vector estimation and image reconstruction are performed on a data block-by-data-block-by-data-block basis including data of a plurality of consecutive frames in time, data of a first received frame is excluded when data of a new frame is received, Wherein the data block is sliced so as to include the data of the space-time kernel vector and the reconstructed image.

Images