JP2758715B2 - Background noise removal device for subject cross-sectional images - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial Application Field) The present invention relates to a cross-section of an object for removing noise in a blank portion of a background from a plurality of cross-section images of an object having the same contour imaged by, for example, an MRI apparatus. The present invention relates to a background noise removal device for an image.

(Prior Art) When imaging a cross-sectional image of a human body or the like using a conventional MRI apparatus or the like, a large number of images are taken while changing parameters indicating the echo time, pulse repetition time, and the like as imaging conditions for the same part of the human body. Is generally taken. As a matter of course, in the image obtained by imaging the same site, the outline of the real image portion matches.

Further, in the cross-sectional image taken, air is imaged in a background portion occupying the periphery of the real image portion. Therefore, theoretically, the intensity of the image signal becomes zero, and the image becomes a dull image.

However, actually, noise is generated in the background portion, and an image in which the difference between the signal intensity of the generated noise and the signal intensity of the real image portion is hardly reduced depending on the setting of the imaging conditions in particular.

Further, depending on the image processing performed after the imaging, the background noise is amplified and the signal intensity becomes higher than that of the actual image, and the image may be extremely invisible.

As described above, when noise is generated in a background image portion that is not originally required, and when the noise is conspicuous, or is amplified and emphasized by processing, not only the actual image portion is hardly seen but also diagnosis becomes difficult. Therefore, it is necessary to remove noise generated in the background portion.

Therefore, conventionally, noise has been removed by the processing shown in FIG.

In this method, when the captured image F shown in FIG. 7A is input, the distribution of the signal intensity of each image is obtained for the image F as shown in the histogram of FIG. The distribution on the right side of the histogram is the distribution of the signal intensity for the real image portion Fo showing the human body cross section, and the distribution on the left side is the distribution of the signal intensity for the background image portion Fb including noise. Since the two distributions are clearly separated from each other, the signal intensity at the valley where the intermediate frequency becomes zero is set as the threshold Th3. Next, the signal intensity of each pixel of the image F is compared with the threshold value Th3, and the pixels having the threshold value Th3 or less are uniformly converted to the signal intensity of zero or the threshold value Th3 or less. When these processes are performed, an image H shown in FIG. C is obtained, and the surrounding background image is uniformly converted to a plain dark color.

(Problems to be Solved by the Invention) However, this method is good when the signal intensity distribution of the real image portion Fo and the signal intensity distribution of the background image portion Fb including noise are almost completely separated. As shown in b
If the distribution of the real image portion Fo and the distribution of the background image portion Fb overlap, noise may not be removed properly.

As described above, when both distributions overlap, the signal intensity closer to the maximum value on the background image portion Fb side than the middle valley of both distributions is set to the threshold so as not to delete the image signal for the real image portion Fo. Th4. When noise removal processing is performed using this threshold value Th4, an image shown in FIG. 3C is obtained, and the low-level portion of the real image portion Fo remains without being removed. However, the background image part
Noise having a large signal strength in Fb is not removed and remains as a speckled noise image N.

When the noise image N is to be completely removed, the processing shown in FIG. 9 is performed. That is, the ninth
As shown in FIG. B, the signal intensity at the middle valley between the signal intensity distributions of the real image portion Fo and the background image portion Fb is set to a threshold value Th5.
And When noise removal processing is performed using this threshold value Th5, an image shown in FIG. 3C is obtained. Although noise in the background image portion Fb can be completely removed, a portion L with a low signal intensity in the real image portion Fo is removed. Would. A human body part having a low signal intensity in the real image part Fo includes a cavity such as the inside of the intestine and the lungs, and erasing the image of this part may not be preferable for diagnosis.

As described above, the conventional noise removal method is effective only for a limited number of images, and has a disadvantage that it cannot be applied to general images.

The present invention has been made in order to solve the above-described problems, and an object of the present invention is to include noise even in a cross-sectional image of an object in which the signal intensity of a real image and a background image including noise overlap. An object of the present invention is to provide a background noise removing device for a subject cross-sectional image which can effectively remove only a background image.

(Means for Solving the Problems) In order to achieve the above-described embodiment, the first invention extracts one image from a series of subject cross-sectional image groups having substantially the same contour and forms the image. Threshold value determining means for determining a gray level value of a pixel for dividing a real image portion and a background image portion around the real image portion from a gray level distribution of pixels and using the obtained threshold value as a threshold value; A binarized image converting means for converting the extracted image into a binarized image; and a run length of pixels which are at H level for each pixel column from at least four directions, up, down, left, and right of the converted binarized image. A position where the obtained run length exceeds a predetermined value is regarded as a boundary position between the real image portion and the background image portion, and a portion in front of the boundary position is defined as an L-level pixel column, and subsequent portions are defined as an H-level pixel column. How to detect run length of contour pattern The contour pattern creating means created for each direction is superimposed on the contour pattern created for each direction, and the logical product of the H level is obtained for each pixel, and a mask pattern including all the H level pixels is created. Mask pattern creating means for performing mask processing on each of a series of subject cross-sectional images using the created mask pattern, and background image removing means for removing a background image portion including noise around the real image. It is characterized by the following.

According to a second aspect of the present invention, the threshold value determining means according to the first aspect of the present invention comprises: a means for extracting one image from a series of subject cross-sectional image groups having substantially the same contour; and extracting a corner portion from the extracted image. Means for obtaining the gray level distribution of the pixels constituting the image of the extracted corner portion, and obtaining the gray level value of the pixel for separating the real image portion and the background image portion from the obtained gray level distribution. And means for setting a threshold value.

(Operation) In the first invention, one image is taken out of a series of cross-sectional image groups of the subject having substantially the same contour, and the actual image portion and the actual image portion are extracted from the gray level distribution of pixels constituting the image. The gray level value of the pixel that separates the background image portion around the image portion is obtained and becomes a threshold value. Based on the threshold value, the extracted image is converted into a binarized image.

Next, the run lengths of the pixels at the H level are obtained for each pixel column from at least four directions, up, down, left, and right, of the converted binary image. When the obtained run length exceeds a predetermined value, the boundary position between the real image portion and the background image portion is detected, and the trouble of the boundary position is set to the L-level pixel column, and the subsequent portions are set to the H-level pixel column. An outline pattern is created for each run length detection direction.

Further, the contour pattern created for each run-length detection direction is superimposed, the logical product of the H level is obtained for each pixel, and a mask pattern including all the H level pixels is created.

Next, mask processing using the created mask pattern is performed on each of a series of subject cross-sectional images,
The background image around the real image is removed, leaving only the real image within the outline.

In the second aspect, when one image is taken out from a group of a series of cross-sectional images of the subject having substantially the same contour, a corner portion of the image is extracted, and a gray level of a pixel constituting the corner portion image is obtained. The distribution is determined.

Next, from the obtained gray level distribution of the pixels, the gray level value of the pixel that separates the real image portion and the background image portion is obtained and becomes a threshold value. Converted to a binarized image.

(Example) Hereinafter, an example of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the first and second inventions.

In the figure, images F ₁ , F ₂ ,..., F _n are images captured by an MRI apparatus or the like and have the same contour but different imaging conditions. The image processing unit 1 receives the input images F ₁ , F ₂ ,.
・ In addition to various image processing for F _n , images G ₁ , G ₂ ,
Send to the mask processing unit 5 as _Gn .

On the other hand, the images F ₁ , F ₂ ,... _Fn are immediately sent to the mask processing unit 5 if image processing is unnecessary.

At the same time, the image F _1, F _2, the image F ₁ beginning of the · · · F _n is input to the threshold calculating section 2.

The threshold value calculation unit 2 detects the distribution state of the signal intensity, which is the gray level of each pixel constituting the image F _1, and obtains the image F ₁
A threshold value for distinguishing a middle real image portion from a background image portion containing noise is obtained and sent to the binarization portion 3.

The binarizing unit 3 converts the image F ₁ into 2 using the obtained threshold value.
The image is converted into a value image and sent to the mask pattern creation unit 4.

Mask pattern creating section 4, the actual image of the contour outside the image F ₁ to the pixels of the L level, to create a mask pattern by converting the internal contour pixels of H level using binarized image input, Send to the mask processing unit 5.

Image F _1, F ₂ mask processing unit 5 by using the mask pattern input, · · F _n or images G _1, G _2, adding masking to · · G _n, the pixels of the background image portion including noise uniformly converted into L level, the background noise removed image H _1, H _2, and outputs it as · · · H _n.

By these series of processing, the image of the same contour captured
F _1, F _2, the · · · F _n is inputted, the image H _1, H ₂ was removed diagnostically unsightly noise, · · · H _n is obtained.

This image _{H 1, H 2, ··· H} n is the contour outside becomes low signal strength uniform background section, the contour original image is kept intact.

Note that the image input to the threshold value calculation unit 2 is
Is not limited to F _1, it is possible to input any image.

FIG. 2 shows an embodiment of the second invention, and is an explanatory diagram of the processing of the threshold value calculation unit 2 in FIG.

In general, screen obtained by imaging the human body cross-sectional image by the MRI apparatus and the like, as shown in image F ₁ in the drawing, the center actual image portion Fo is positioned representing the target body cross-section, the periphery A blank background image portion Fb is surrounding. Normally, the generated noise is uniformly distributed in the background image portion Fb which is an image of the air.

Therefore, a part of the background image portion Fb is extracted from the corner portions Fc at the four corners of the screen that are not normally occupied by the real image portion Fo, and the distribution of the signal intensity of each pixel included therein is obtained. Next, the maximum value of the signal intensity of the background image portion Fb including noise is detected (block 7), and the detected value is used as a threshold value used in the binarization conversion processing of the image (block 8).

These processes compared to the case of calculating the threshold based on the distribution of the signal intensities of all images contained in the image F _1, faster processing time is greatly shortened device is worn.

FIG. 3 and FIG. 4 are explanatory diagrams when binarizing an image performed by the binarizing unit 3 in FIG.

In FIG. 3, the histogram of the signal intensity of pixels constituting the input image F ₁ as shown in the figure a is a diagram b, the distribution of the actual image portion Fo and the background image portion Fb are clearly separated. In this case, the real image portion Fo and the background image portion in the histogram
The signal intensity of the valley located at the middle of the Fb distribution and having a frequency of zero is defined as a threshold Th1.

Using this threshold Th1, the image F ₁ binarized is the pattern P shown in FIG. C. That is, the signal intensity for each pixel constituting the image F ₁ as compared to the threshold value Th1 and the pixel if higher than the threshold Th1 level H
The level is converted to a level, and if the level is low, the pixel is converted to an L level. The pattern P thus obtained is
Portion Po composed of H-level pixels corresponding to real image portion Fo
And a portion Pb composed of L-level pixels corresponding to the background image portion Fb.

In this example, since the signal intensities of the real image portion Fo and the background image portion Fb are completely separated into different levels, the threshold
A clear pattern P is obtained only by a simple binarization process using Th1. In this case, the pattern P can be sent to the mask processing unit 5 as a mask pattern as it is.

However, in general, the real image portion Fo and the background image portion
The signal intensity distributions of Fb often overlap each other.

FIG. 4 shows a processing example in such a case. Histogram of signal intensities of the pixels constituting the input image F ₁ as shown in the figure a is a diagram b, the distribution of the actual image portion Fo and the background image portion Fb overlap. Therefore, the actual image part in the histogram and
The signal intensity of the valley located in the middle of the distribution of Fo and the background image portion Fb is set as a threshold Th2.

Using this threshold Th2, the image F ₁ binarized is the pattern P shown in FIG. C. Pattern P is a real image part
The outline of Fo is composed of a portion Po composed of H-level pixels, a portion Pi composed of L-level pixels, and a portion Pb composed of L-level pixels corresponding to the background image portion Fb. The portion Pi of the contour, which is composed of L-level pixels, is obtained by converting the pixels of the real image portion Fo whose signal intensity is lower than the threshold value Th2 to the L level without being converted to the H level. As described above, since a portion Pi composed of L-level pixels is generated inside a portion Po composed of H-level pixels corresponding to the real image portion Fo, when the pattern P is used as a mask pattern, a part of the real image portion Fo is Usually, it cannot be used as it is. The image F ₁
Is obtained by imaging a lung or the like of a human body, and if an air portion is clearly imaged and included in the real image portion Fo, the obtained pattern P is used as a mask pattern as it is. There is a case where there is no problem even if used.

FIG. 5 is an explanatory diagram of the processing performed in the mask pattern creating section 4 of FIG.

The mask pattern creation unit 4 scans the pixels from the four directions of up, down, left, and right with respect to the pattern P input from the binarization unit 3 to obtain the run length of the pixels at the H level. If the obtained run length exceeds a preset value, the position is regarded as the position where the pixel changes from the L level to the H level, that is, the boundary between the background image portion Fb and the real image portion Fo. This boundary is also the outline of the real image portion Fo. When the contour of each scanning row is obtained by setting the L level pixel before the detected boundary position and the H level pixel thereafter, the contour pattern Pl corresponding to the contour in one direction with respect to the real image portion Fo is obtained. , Pr, Pu, Pd are obtained.

Next, when the obtained contour patterns Pl, Pr, Pu, and Pd are overlapped to obtain a logical product of the pixels at the H level (block 11), a pattern Pa is obtained. This pattern Pa is
The portion Pi of the pixel at the L level included in the pattern P is removed to obtain a shape corresponding to the contour of the real image portion Fo.

The value set in advance as a criterion for determining whether the detected run-length value of the pixel at the H level is the boundary of the image is larger than the maximum length of the noise generated in the background image portion Fb. This is a large value.

In addition, when the value set in this way exceeds the detected run length, it is determined that the image is at the boundary of the image. The positions displayed as the patterns Pl, Pr, Pu, Pd are positions shifted to the front by the set value. Therefore, the contour represented in the pattern Pa is closer to the inside than the actual contour of the real image portion Fo. Therefore, in order to correct the actual contour, the logical sum of the pattern Pa and the original pattern P is obtained for the pixel at the H level (block 12), and is set as a mask pattern Pm. That is,
This mask pattern Pm has the same contour as the pattern P, and is obtained by removing the portion Pi of the pixel at the L level which is inside.

In this embodiment, the positional deviation between the contour of the pixel portion Po at the H level in the pattern P and the contour patterns Pl, Pr, Pu, Pd is finally corrected by the logical sum. , Pr, Pu, Pd, at the time of conversion, if the contour pattern Pl, Pr, Pu, Pd is formed at the position returned in the scanning direction by the set run length value, it is possible to obtain an accurate contour pattern from the beginning Can be. In this case, since it is not necessary to correct the pattern Pa synthesized from the logical product of the contour patterns, the pattern Pa can be used as it is as a mask pattern.

Further, in the figure, the contour pattern is created by calculating the run lengths of the pixels at the H level from the four directions of up, down, left, and right every 90 degrees, but if the contour of the target image has a complicated shape, In this case, a contour pattern is created from another direction to obtain an accurate shape.

FIG. 6 is an explanatory diagram of the mask processing performed in the mask processing section 5 of FIG.

The mask processing unit 5, processing the mask pattern Pm mask pattern creating section 4 creates image G ₁ is inputted. Of course the contour of the shape and processing the actual image portion of the image G ₁ of the mask pattern Pm match.

Here, superimposed image G ₁ and the mask pattern Pm, calculated and H-level of the pixels constituting the mask pattern Pm, the logical product of the signal intensity of each pixel of the image G ₁ (block 1
3) obtaining a background noise cancellation images H _1. That is, the portion of the pixel at the H level of the mask pattern Pm, the intensity of the pixel signal in the image G ₁ is represented by it. Also,
The L level pixels of the mask pattern Pm are uniformly represented as L level pixels. As a result, noise is generated in the background portion in G ₁ which is input is all removed to the L level of the pixel, uniformly a background portion in which the dark.

When these processes are completed, the next images G ₂ , G _{3 ,.}
Is input, and a mask processing unit using the same mask pattern Pm is performed.

Incidentally, the mask processing unit 5 in addition to the image _{G 1, G 2 ··· G n} , the image F _1, F ₂ ··· F _n is input mask processing is performed.

In the embodiment, as described above, any one of the images F ₁ , F _2, ... F _n representing the human body cross section to be processed is extracted,
A threshold value for extracting a corner portion of the image and binarizing the image is obtained in a short time, and the image extracted based on the obtained threshold value is binarized, and the H level of the binarized image is obtained. By generating a mask pattern by obtaining run lengths of pixels from four directions and adding mask processing to the target image, noise in the background image portion can be easily and reliably removed.

(Effects of the Invention) As described above, according to the first invention, in many subject cross-sectional images having the same contour imaged by an MRI apparatus or the like, for example, the background image including the signal intensity distribution of the real image portion and the noise Even in the case where the signal intensity distributions of the portions partially or overlap, only the background image including noise can be effectively removed.

Further, according to the second aspect, the threshold value used for binarizing the extracted image is obtained only from the signal intensity distribution of the pixel in the corner portion of the image, so that the processing time is reduced. .

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment of the first and second inventions, FIG. 2 is an explanatory diagram showing a process of a threshold value calculator according to the embodiment of the second invention, and FIGS. FIG. 5 is an explanatory view showing the processing of the binarizing section, FIG. 5 is an explanatory view showing the processing of the mask pattern creating section, FIG. 6 is an explanatory view showing the processing of the mask processing section, and FIGS. FIG. 9 is an explanatory diagram illustrating a processing example. 1 image processing unit, 2 threshold value calculation unit, 3 binarization unit, 4 mask pattern creation unit, 5 mask processing unit, F ₁ , F ₂ ,. _n ...... input image, Fb ...... background image portion, Fc ...... corners, Fo ...... real image _{portion, G 1, G 2, ···} G n
…… Processed image, H ₁ , H ₂ ,... H _n …… Background noise removed image,
P: pattern, Pa: pattern, Pi: pixel portion at L level, Pl, Pr, Pu, Pd ... contour pattern, Pm ...
Mask pattern, Po ... H-level pixel portion, Th
1, Th2 …… Threshold

──────────────────────────────────────────────────続 き Continuing from the front page (72) Inventor Shiro Nakatoge 1-1, Tanabe Nitta, Kawasaki-ku, Kawasaki City, Kanagawa Prefecture Inside Fuji Electric Co., Ltd. (58) Field surveyed (Int. Cl. ⁶ , DB name) A61B 5/055 G06T 3/00-5/50

Claims (2)

(57) [Claims] An image is extracted from a series of cross-sectional images of a subject having substantially the same contour, and a real image portion and a background image portion surrounding the real image portion are obtained from a gray level distribution of pixels constituting the image. Threshold value deciding means for determining a gray level value of a pixel for classifying the image and setting the threshold value, and converting the extracted image into a binary image using the determined threshold value Means for determining a run length of a pixel at an H level for each pixel column from at least four directions of up, down, left, and right of the converted binary image, and determining that the obtained run length exceeds a predetermined value as an actual image portion. Contour pattern creating means for creating, for each run-length detection direction, an outline pattern in which the front of the boundary position is regarded as an L-level pixel row and the succeeding H-level pixel row is regarded as a boundary position; Created for each And a mask pattern generating means for generating a logical product of the H level for each pixel and generating a mask pattern composed of pixels having all the H levels, and a series of mask patterns using the generated mask pattern. A background image removing unit that performs a mask process on each of the subject cross-sectional images and removes a background image portion including noise around the actual image. 2. The apparatus for removing background noise from a cross-sectional image of an object according to claim 1, wherein: means for extracting one image from a group of a series of image of the object cross-section having substantially the same contour; Means for extracting a portion, a means for obtaining a gray level distribution of pixels constituting an image of the extracted corner portion, and a gray level value of a pixel for separating a real image portion and a background image portion from the obtained gray level distribution And means for determining a threshold value, and a threshold value determining means by: