JPH07222724A - Image correcting method - Google Patents

Abstract

PURPOSE:To eliminate image shading due to the sensitivity nonuniformity of an RF probe without increasing a noise. CONSTITUTION:Sensitivity distribution is found by applying a low-pass filter to an MR image or MR signal in advance, and the sensitivity correction coefficient of each pixel is found based on the sensitivity distribution. The sensitivity correction coefficient is provided with a value proportional to the inverse of sensitivity in a high sensitive area, and a value to be the increasing function of the sensitivity in the extremely low sensitive area via the maximum value in a low sensitive area. A correction image can be obtained by multiplying an original image by the correction coefficient. Moreover, the low-pass filter with coefficient different at every pixel is applied to the correction image at need.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to a nucleus for detecting nuclear magnetic resonance (hereinafter referred to as "NMR") signals from hydrogen, phosphorus, etc. in an object to visualize the density distribution and relaxation time distribution of the nucleus. The present invention relates to a method for correcting an image (hereinafter referred to as an MR image) captured by a magnetic resonance imaging (MRI) device.

[0002]

2. Description of the Related Art An MRI apparatus obtains a tomographic image of the human body by measuring the proton density in the human body and the relaxation state of the spins of the protons by using the NMR phenomenon. Magnetic resonance (M
In order to measure the (R) signal, a radio frequency (RF) probe which is a kind of radio frequency antenna is provided.

As such an RF probe, for example, a saddle type coil, a multiple element resonator (ME
R), slotted tube resonator (STR), solenoid coil, etc. are used. Surface coils are also used for spinal and local applications. The surface coil has higher sensitivity than the above-mentioned MER and STR, but has a large spatial nonuniformity of sensitivity. Although the quadrature reception method and the multiple coil (also called phased array) method have been conventionally used to increase the sensitivity of the surface coil, the nonuniformity of sensitivity cannot be avoided because it is an essential property of the surface coil.

This nonuniformity of sensitivity causes so-called shading, in which the MR image corresponding to the low sensitivity portion of the coil becomes dark. FIG. 3A schematically shows a sagittal image when a cervical spine is imaged by a surface coil as a typical example. Here, the RF probe 20 is arranged on the back side of the cervical spine. This image has shading due to the sensitivity distribution of the RF probe 20. That is, the image profile on the cross section A-A sharply decreases from the right side of the image to the left side as shown in FIG.

As a method of correcting the shading due to the non-uniformity of the RF probe, the sensitivity distribution of the RF probe is obtained by previously photographing the human equivalent phantom, and the image data I photographed based on this sensitivity distribution data is obtained. There have been proposed several methods for correcting shading and for correcting shading by performing image processing on a captured MR image.

An example will be described. First, the taken image of the subject is I (x, y). The sensitivity distribution s (x, y) of the coil, which is separately calculated, is stored in the computer.
As the correction coefficient F, F (x, y) = 1 / s (x, y) is calculated, and the sensitivity nonuniformity is corrected by the calculation of I (x, y) × F (x, y). In principle, the sensitivity can be corrected by this method. But in reality, s is far from the coil
Since (x, y) is very small (dark), F (x, y)
The value of becomes extremely large. Further, outside the subject (background), the value of I (x, y) is small and the S / N is low. Therefore, at a location far from the coil and outside the subject (most backgrounds satisfy this condition), I (x, y) × F (x, y)
In many cases, the noise of I (x, y) is amplified by the calculation of, and the background portion of the corrected image has a large value, that is, becomes bright. As a result, even if one tries to see the detailed structure of the subject in the corrected image, the background is often too bright and difficult to see.

As a means to reduce this, Roemer et al.
In the background portion, a method of making the correction coefficient F constant is proposed, but in this case as well, the background noise could not be sufficiently reduced. In addition, the sensitivity distribution s (x, y) of the coil is obtained by applying two-dimensional low-frequency passage to the obtained image data itself.
Although there is also a method of quasi-obtaining the above, the filter reported so far is not stable in the examination by the present inventors, and there is no filter that can be applied to all imaging (Journal of the Japanese Society for Magnetic Resonance Medicine, Vol. 13 (S ), P. 323, "Evaluation of self-correction method for non-uniform luminance image", Ando et al. (1993), American Journal of Rent Geology, 148, 418-420 (198).
7 years), Axel and others, "Intensity Collection"
In-Surface Coil MR Imaging "(Americ
an Journal of Rentogenology, Axel et al., Intensit
y correction in surface-coil MR imaging), Magnetic Resonance in Medicine, 16: 192 (1
990), Roomer et al., "The Phased Array" (M
agnetic Resonancein Medicine, PB Roemer et al,
The NMR phased array)).

[0008]

As described above, in the conventional correction method, noise may be emphasized in the low sensitivity region, and it may be difficult to make a diagnosis during image diagnosis. In addition, the correction may not be sufficiently performed depending on the imaging condition. Further, generally, filtering is performed in order to reduce image noise and improve image diagnostic ability. If low-pass filtering is performed here, noise is reduced and diagnostic performance is improved in a portion having a low S / N on the image, but this effect is small in a portion having a high S / N, and conversely, the space is reduced by the filtering. The resolution is reduced and the diagnostic ability is reduced. In particular, in the image shown in FIG. 3 in which the sensitivity of the RF probe drastically changes, this tendency is strong, and appropriate filtering has not been performed.

In order to improve such a situation, conventionally, an adaptive filter has been studied. For example, as an adaptive filter, the variance of the signal of the target portion is sequentially examined and local filtering is performed (for example, , IEEE
E Transaction on Medical Imaging, Volume 12, Issue 2, 322-327, Itagaki (Improvements of
nuclear magnetic resonance image quality using ite
rations of adaptive nonlinear filtering, Itagak
i). However, such a conventional process has a problem that the calculation time is long because the repeated dispersion is calculated for each part, which is not practical.

It is an object of the present invention to provide a practical image correction method which can solve the above conventional problems and obtain an image with high diagnostic ability. That is, it is an object of the present invention to provide an image correction method that solves the problem of noise enhancement in a low sensitivity region in an MR image of a magnetic resonance imaging apparatus and that can easily and appropriately correct the sensitivity of a high frequency probe. Another object of the present invention is to provide an image correction method capable of effectively filtering an MR image without deteriorating the spatial resolution.

[0011]

An image correction method according to the present invention that achieves the above object includes a first image obtained by image reconstruction based on an MR signal detected by an RF probe in a magnetic resonance imaging apparatus. An image correction method for correcting shading by using a second image having the shading information, wherein a sensitivity correction coefficient for each pixel of the first image is used as a sensitivity correction coefficient of the corresponding pixel of the second image. It is defined by the sensitivity, is substantially inversely proportional to the sensitivity in the high sensitivity region, has the maximum value in the low sensitivity region, and becomes an increasing function of the sensitivity in the lower sensitivity region than the sensitivity giving the maximum value. Using the function, the corrected image is obtained as the product of the first MR image and the sensitivity correction coefficient.

According to a preferred aspect of the present invention, the second image is created by applying a low-pass filter to the first image or the MR signal, and has sensitivity distribution information which is shading information. As the low-pass filter, one having a coefficient that is a variable corresponding to at least the field of view at the time of imaging or the type of RF probe is used. Such a low-pass filter is, for example, a Butterworth type filter of second order.

Yet another aspect of the image correction method of the present invention is
A second image is created by applying a first low-pass filter to the first image or MR signal, and a sensitivity correction coefficient defined by the sensitivity of the picture element of the second image is applied to the first MR image. A corrected image is created by multiplying the corrected image by a second low-pass filter, and the coefficient of the second low-pass filter is defined by the second image. It is different for each. Also in this aspect, preferably, the sensitivity correction coefficient for correcting the first image is
A function that is approximately inversely proportional to the sensitivity in the high sensitivity region of the second image, has a maximum value in the low sensitivity region, and becomes an increasing function of the sensitivity in the sensitivity region lower than the sensitivity giving the maximum value. To use.

[0014]

As a sensitivity correction coefficient for correcting the first MR image, by using a specific function such that the correction coefficient becomes smaller as the sensitivity decreases in the picture element in the low sensitivity area, the The correction is weaker for the picture element
The brightness is low on the image. Therefore, it is possible to obtain a corrected image that does not hinder diagnosis. In addition, by applying a specific low-pass filter to the first image to be corrected, a sensitivity distribution (shading information) that well reflects the sensitivity distribution of the RF probe can be obtained by image processing without the need for imaging with a phantom or the like. It is possible to obtain the second image that the user has, and since the correction coefficient is defined by this second image, it is possible to effectively correct the shading due to the sensitivity distribution of the RF probe.

Further, according to a study by the inventors, in an image in which shading due to probe sensitivity is remarkable, local S /
Since N is uniquely determined by the probe sensitivity, by defining the coefficient for each picture element of the second low-pass filter to be applied to the corrected image by the sensitivity distribution data of the second image,
Noise can be reduced in the low S / N portion, and high spatial resolution can be achieved in the high S / N portion.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 2 is a sectional view of the main part of a typical MRI apparatus 1 to which the image correction method of the present invention is applied. The static magnetic field magnet 10 normally uses a magnetic field strength of about 0.2T to 2T, and at 0.5T or more, a cylindrical superconducting coil as shown is used as the static magnetic field magnet. The direction of the magnetic flux created by the magnet 10 is perpendicular to the paper surface. Furthermore, the MRI device is NM
A gradient magnetic field coil 11 for applying a magnetic field having a spatial gradient to obtain the spatial information of the R signal is provided, and the gradient magnetic field coil 11 is installed in the bore of the superconducting coil. The RF coil 12 for applying a high frequency magnetic field to the subject 20 is
It is installed inside the gradient coil 11. RF coil 1
A slide type bed 13 is provided inside 2 and a subject (patient) 30 is laid on the slide type bed 13 for inspection. Further, for the purpose of blocking RF noise from the outside between the gradient magnetic field coil 11 and the RF coil 12, removing the high frequency coupling of the RF coil 12 with the gradient magnetic field coil 11 and the like, and preventing its performance deterioration, The RF shield 14 is installed.

The RF coil 20 for reception uses the RF coil for transmission.
It is installed inside the coil 12. FIG. 2 shows an example in which a surface type cervical spine coil is used as such a receiving coil 20. In addition, the MRI system uses these RF coils 1
2, a sequencer for driving and controlling the gradient magnetic field coil 11 and the like in a predetermined sequence, an image processing unit for reconstructing an image based on the MR signal detected by the RF coil 20,
A display unit for storing and displaying the obtained image data is provided.

In the MRI apparatus having such a structure, a high frequency magnetic field is applied to the subject 20 placed in a static magnetic field by the RF coil 12 and a gradient magnetic field is applied by the gradient magnetic field coil 11 to remove the tissue of the subject 20. The constituent nuclei (usually protons) are excited, magnetic resonance (MR) signals emitted from the nuclei in the relaxation process are detected by the RF coil 20 for reception, and a predetermined calculation is performed based on this MR signal. Obtain a tomographic image of the human body. This image has shading due to the sensitivity distribution of the RF probe as described above, and the present invention processes the obtained original image (first MR image) to eliminate sensitivity nonuniformity. to correct.

The image correction method mainly comprises four steps. In the first step, the sensitivity distribution of the RF probe is obtained. In the second step, the correction coefficient is calculated according to the sensitivity distribution obtained in the first step. The third step corrects the original image using the correction coefficient obtained in the second step. The fourth step is done as needed,
The corrected image is filtered according to the sensitivity distribution obtained in the first step. These will be described below in order.

First, a method of calculating the sensitivity distribution s (x, y) of the RF probe in the first step will be described. x and y are positions on the image. The sensitivity distribution of the RF probe may be obtained based on an image previously captured using a phantom or the like, but in the present embodiment, basically, a low-pass filter is applied to the original image (first MR image). It is obtained from the second image obtained by the multiplication. The method is
For example, inverse Fourier transform is performed from the image data I (x, y) in the real space to the transformed image J (u, v) in the phase space, and the low-pass filter f (u, v) is transformed into the transformed data in the phase space. It is obtained by multiplying. As described above, the signal in the phase space may be calculated from the original image by the inverse Fourier transform, but since the measurement data (MR signal) of the MR image is obtained as the distribution in the phase space, it is directly calculated from this data. You may calculate.

Here, the shape of the low-pass filter is, for example, the Butterworth type represented by the equation (1).

[0022]

[Equation 1]

(Wherein R (u, v) = √ (u ² + v ² ) and n is the order, which defines the slope of the filter), or Gauss expressed by equation (2). Filter, etc.

[0024]

[Equation 2]

Can be used. In the equations (1) and (2), R ₀ is the cutoff radius of the filter. By appropriately determining the order n and R ₀ which are the parameters of these filters, a sensitivity distribution that well reflects the sensitivity distribution of the RF probe can be calculated. In the case of any of the filters, if the cutoff radius R ₀ is small, it cannot follow the sensitivity distribution of the RF probe that changes rapidly, and the fine structure inside the subject remains, and as a result, the corrected image The contrast resolution is reduced. The optimum R ₀ depends on the field of view (imaging field of view or field of view when performing FFT) L and the characteristics (shape etc.) of the RF coil as described below, and can be uniquely determined by these factors. As an example, the sensitivity of a typical surface coil with a diameter of 100 mm is
Approximately 50 mm (= D) from the coil surface to the inside of the subject
Halves at the depth of. At this time, when a sagittal image is taken with the field of view L being 260 mm, a low-pass filter having a spatial resolution of at least 50 mm is required to obtain the sensitivity distribution.

On the other hand, the spatial frequency corresponding to the m-th picture element from the center is m / L. M that gives the required spatial frequency 1 / D or less is m ≧ L / D, and for that purpose, the cutoff radius R ₀ = L / D is required. That is, R ₀ = 260 /
50. Further, when the array of absolute value data is expanded to the outside by, for example, twice and an appropriate value (for example, zero) is substituted for the well-known image aliasing removal technology or high-definition imaging technology, a substantial visual field L is obtained. Since 'becomes L' = 2L, the above equation becomes R ₀ = 2L / D. On the other hand, when the interpolation is performed so that the number of picture elements is doubled, the field of view does not change, so R ₀ = L
/ D remains. The field of view L is an imaging condition, so it is determined for each imaging region (diagnosis target), and D is a value specific to the coil used. Therefore, if this value is stored at the time of shooting, the optimum R ₀ can be uniquely obtained.

Therefore, the optimum cutoff radius R ₀ of the low-pass filter is obtained by storing these factors at the time of photographing, and the sensitivity is calculated from the image data by using this R ₀ and the equation (1) or (2). The distribution is obtained. Further, the order n in the case of using the Butterworth type filter will be described. If n is small in the Butterworth filter, the high frequency components of the image cannot be removed, and the fine structure of the subject remains. As a result, the contrast in the subject deteriorates in the corrected image. On the other hand, if n is large, the sensitivity distribution of the RF probe that changes abruptly cannot be followed, and as a result, ringing occurs and accurate correction cannot be performed. According to the study by the inventors, a sensitivity distribution that well reflects the actual sensitivity distribution of the RF probe can be calculated when n = 2.

As the low-pass filter, well-known digital filters such as a moving average filter and a median filter may be used in addition to the above Butterworth type filter and Gaussian filter. Next, as a second step, the sensitivity distribution s of the RF probe obtained as described above
The correction coefficient F (s') calculated based on (x, y) will be described. Where s' (x, y) ≡s (x, y) /
s _max (s _max is maximum sensitivity), and (x, y) representing a function of position is omitted for simplification. The same applies to the following.

It is known that F (s ') = 1 / s' is basically suitable. According to the studies by the inventors, the local S / N of the image is proportional to s'. Also on the image,
In the part of s'<0.1, almost no diagnostic information is included,
On the contrary, it was found that noise was noticeable and hindered image diagnosis. Therefore, it has been found that it is desirable that the correction coefficient is smaller than 1 / s' in this region. In addition, the correction coefficient F
For (s') and its derivative f (s'), if 0 <s'<1 is continuous, an unnatural virtual image on the corrected image can be removed,
It was found to be effective in performing image diagnosis.

Therefore, the correction coefficient F (s ') has a maximum value of 1 / s' in the high-sensitivity region and a maximum value in the low-sensitivity region.
In the extremely low sensitivity region, it is an increasing function of sensitivity. Particularly, the sensitivity that gives the maximum value of the sensitivity correction coefficient is 0.02 of the maximum sensitivity s _max .
To 0.25 (s '= 0.02 to 0.25) and the maximum value F (s') _max of the sensitivity correction coefficient is 4 to 12 times the correction value at the maximum sensitivity, the diagnosis is most easy. . Further, it is more desirable that the sensitivity correction coefficient for the sensitivity of 0 be 0 to 2 times the correction value at the maximum sensitivity.

As a concrete function of the correction coefficient F (s') of such a characteristic,

[0032]

[Equation 3]

[0033]

[Equation 4]

[0034]

[Equation 5]

And the like. The correction coefficient F (s') given by the equation (4) is shown in FIG. 1 as an example. 1 / s' is also shown in FIG. These functions are sequentially s'
It may be calculated according to (x, y), or may be stored in a table format on the computer memory. Even in the case of sequential calculation, since these functions do not include complicated operations such as exponential functions, the calculation time is relatively short.

The third step, the original image I (x, y)
The correction of is given by I (x, y) × F (s ′ (x, y)). In the corrected image I ′ (x, y), the shading caused by the sensitivity distribution of the probe is almost completely corrected, and the portion with a lot of noise and poor diagnostic information is gently suppressed. ) Setting and level (center of concentration) are easy and diagnostic information can be displayed sufficiently. This corrected image I '(x,
In y), noise can be further reduced by filtering.

Next, the filtering of the corrected image according to the sensitivity distribution, which is the fourth step, will be described. As described above, generally, when low pass filtering is applied to a portion having a low S / N on the image, noise in the image is reduced and diagnostic ability is improved. Spatial resolution decreases and diagnostic ability decreases. According to the study by the inventors, it was found that the local S / N is uniquely determined by the probe sensitivity in an image in which the shading of the probe sensitivity is remarkable. Therefore, in this fourth step, the sensitivity distribution data s ′ obtained in the first step
Is used to apply a digital image filter with a different coefficient for each picture element.

This image filter is a low pass filter,
For pixels with a high sensitivity distribution, do not apply a filter, but apply a stronger filter as the sensitivity decreases. As a result, the S / N on the image can be made uniform or the change can be made gentler than before correction. As a result, the contrast resolution of the image becomes uniform, which facilitates image diagnosis.
A known digital filter can be used as the low-pass filter. That is, a median (median) filter in which the average value of the densities in the n × n local area is used as the output density of the picture element at the center of the area, and the average density of the picture element neighboring area (n × n) is set as the output picture element value. It is possible to use a moving average method, a method of selective local averaging in which the average density of a region not including an edge around each picture element is used as the output density of the picture element (Introduction to Computer Image Processing, edited by Tamura, Showa 60
Year, Seiunsha, pp. 103-107). Then, parameters (coefficients) at the time of these filtering, for example, the size of the window (n
Let × n) be a function of the sensitivity data s ′ (x, y) described above. Here, the window size corresponds to a median filter, an n × n region for obtaining a moving average value or a selective local average value. Alternatively, it corresponds to the magnitude of the weight for each picture element when obtaining the moving average value or the selective local average value. If these functions are continuous functions, a step due to a change in the correction parameter cannot be formed in the corrected image, which is desirable. However, it may be discrete as appropriate from the viewpoint of shortening the calculation time. In that case, practically 3 to 10 steps are desirable.

According to this embodiment, the shading of an image having strong shading is removed, noise is reduced in a portion having a low S / N, and high spatial resolution is improved in a portion having a high S / N. The diagnostic ability of images is significantly improved. This correction can be automatically processed using data representing several shooting conditions and a target image. Also,
This calculation can be done at high speed.

Although the correction method including the first to fourth steps has been described in the above embodiment, the present invention does not need to perform all of these steps, and the correction method is optional within the scope of the claims. Can be changed to For example, it is possible to perform only the second and third steps by using the image data having the sensitivity distribution acquired in advance. Further, the correction factors other than those described in the second step may be used to perform the first, third and fourth steps.

[0041]

As is apparent from the above description, according to the present invention, by using a specific correction coefficient as a correction coefficient when correcting an image having shading,
It is possible to obtain an image with high diagnostic ability because the shading of the image with remarkable shading is removed. In particular, by defining the correction coefficient based on the shading information (sensitivity distribution) obtained from the original image, it is possible to perform the correction easily and in conformity with the imaging conditions. Further, according to the present invention, when the correction image is filtered, noise having a low S / N ratio is reduced by performing filtering having different coefficients for each picture element corresponding to the sensitivity distribution of each picture element. , The part with high S / N is improved to high spatial resolution,
The diagnostic ability is improved.

[Brief description of drawings]

FIG. 1 is a diagram showing a curve showing an embodiment of a correction coefficient according to the present invention and a conventional correction coefficient curve (1 / s ′).

FIG. 2 is a sectional view showing an entire MRI apparatus to which the present invention is applied.

3A and 3B are respectively a tomographic diagram obtained by an MRI apparatus, and a diagram showing signal intensity in the AA cross section of FIG. 3A.

[Explanation of symbols] 20 ... RF coil

Claims (8)

[Claims] 1. An image correction method for correcting shading of a first image obtained by image reconstruction based on a magnetic resonance signal detected by a high-frequency probe, using a second image having information of the shading. In the first
The sensitivity correction coefficient for each picture element of the image is defined by the sensitivity of the corresponding picture element of the second image, is approximately inversely proportional to the sensitivity in the high sensitivity area, and has the maximum value in the low sensitivity area. And a function such that it becomes an increasing function of sensitivity in a region of lower sensitivity than the sensitivity that gives the maximum value,
An image correction method, wherein a corrected image is obtained by multiplying the first image by the sensitivity correction coefficient. 2. The image correction method according to claim 1, wherein the second image is created by applying a low-pass filter to the first image or the magnetic resonance signal. 3. The image correction method according to claim 2, wherein the low-pass filter has a coefficient whose coefficient corresponds to at least a field of view or a type of the high-frequency probe. 4. The image correction method according to claim 2, wherein the low-pass filter is a Butterworth type filter. 5. The image correction method according to claim 4, wherein the low-pass filter has a second order. 6. An image correction method for correcting shading of a first image obtained by image reconstruction based on a magnetic resonance signal detected by a high frequency probe, using a second image having information on the shading. In the first
The second image by applying a first low-pass filter to the image of
The first image is multiplied by the sensitivity correction coefficient defined by the sensitivity of the picture element of the image to create a corrected image, and a different coefficient is created in the corrected image for each picture element defined by the second image. An image correction method characterized by applying a second low-pass filter that it has. 7. The sensitivity correction coefficient is substantially inversely proportional to the sensitivity in a high sensitivity region of the second image, has a maximum value in a low sensitivity region, and has a sensitivity higher than a sensitivity giving the maximum value. 7. The image correction method according to claim 6, wherein the function is a function that increases the sensitivity in a low region. 8. The image correction method according to claim 6, wherein the coefficient of the first low-pass filter is a variable corresponding to at least the field of view or the type of the high-frequency probe.