JPH0636596B2 - Radiation image processing method - Google Patents

Description

The present invention relates to a radiation image processing method mainly used for medical diagnosis.

[Prior Art] In recent years, a radiation image is often converted into an electric signal and processed and reproduced.

The term “radiation image” as used herein refers to an image formed by irradiating a subject with high-energy electromagnetic waves such as radiation, γ rays, and neutron rays, and detecting the attenuation of the electromagnetic waves by the subject two-dimensionally. For example, there are a radiograph by radiation, an image obtained by two-dimensionally scanning the amount of light emitted from a stimulable phosphor and converting it into an electric signal, or an image obtained by a so-called indirect photography television.

All of the processing performed in these processes corrects image deterioration due to radiation detection elements, radiation tubes, imaging methods, etc., or reproduces images that match human visual characteristics, and provides images with better diagnostic properties. It is what

If these processings are roughly classified into two, gradation conversion processing and
There are two types of spatial frequency processing.

For example, Japanese Unexamined Patent Application Publication No. Sho.
Japanese Patent Laid-Open No. 5-88740 and Japanese Patent Laid-Open No. 54-121043 disclose density information by reading image information, converting it into an electric signal, and performing signal processing for changing the level of the electric signal. Also, an image with good contrast can be obtained. Further, for example, the density of a portion important for diagnosis in the reproduced image is 0.5 to
There is a method of setting the range to about 1.5 and increasing the contrast in the density region.

[Problems to be Solved by the Invention] With such a gradation conversion process, signal processing is simple, but an image suitable for the human visual characteristics and the noise characteristics of the accumulative phosphor is obtained. Was difficult.

On the other hand, the one that performs spatial frequency processing corrects the deterioration of the frequency response due to the system until the image is obtained, for example, the radiation tube in the case of film, the movement of the subject, the fluorescent material for sensitization, the film, etc. In addition, according to the human visual characteristics,
An image that is visually easy to see can be obtained by emphasizing in the spatial frequency domain.

By the way, conventionally, in order to perform emphasis in the frequency domain, as shown in Japanese Patent Laid-Open No. 56-138735, the method of the formula (1) is used.

D = Dorg + β (Dorg−Dus) (1) where β is an enhancement coefficient, Dorg is an original image signal, and Dus
Is an unsharp mask signal of the original image signal Dorg.

In this method, in order to emphasize a certain spatial frequency or more, an unsharp mask signal Dus corresponding to the low spatial frequency component of the original image is obtained at each scanning point, and a high spatial frequency component is obtained by subtracting from the original image signal Dorg, The original picture signal Do
It is realized by adding to rg.

When the original image filter is formed by this method, as shown in FIG.
A filter as shown in (b) can be realized. That is, it changes greatly in the ranges a and b which are the spatial frequency regions,
The filter does not change much in other cases, and the ratio of changes in the ranges a and b is not sharp mask signal D.
It is only the characteristic of us.

Therefore, in order to realize an arbitrary filter for correction of the image acquisition system by this method and correct the average transfer function of the image acquisition system, a plurality of non-sharp mask signals Dus are expressed by the formula (2). It needs to be calculated, but the calculation time is extremely long and not practical.

D = Dorg + β1 (Dorg-Dusl) + β2 (Dorg-Dus2) + ... (2) The present invention has been made in view of the above circumstances, and it is possible to correct well even a small portion of an image by high-speed processing. It is possible to obtain a good image excellent in diagnosability without graininess, especially by using the convolution operation processing, the shape of the spatial frequency filter can be freely designed for each pixel,
It is an object of the present invention to provide a radiation image processing method capable of improving image characteristics and minimizing image noise while improving image quality.

[Means for Solving the Problems] In order to achieve the above-mentioned object, the present invention involves scanning a radiation image, reading out radiation image information, converting the information into an electric signal, and reproducing it as a visible image. For each pixel of the scan, a convolution function of a two-dimensional predetermined range is multiplied, and a convolution operation for adding the respective values is performed on a predetermined image, and the spatial frequency of the maximum response of the convolution function is calculated. Alternatively, the spatial frequency and intensity of the maximum response of the convolution function are changed according to the intensity of the original image signal to correct the image.

[Embodiments] Hereinafter, embodiments of the present invention using X-rays as radiation will be described in detail with reference to the accompanying drawings.

When the radiation image is scanned, the radiation image information is read out, converted into an electric signal, and then reproduced as a visible image,
For each pixel in each scan, a convolution function of a two-dimensional predetermined range is multiplied, and a convolution operation for adding the respective values is performed for a predetermined image.

In order to realize an arbitrary shape filter for the correction of the image acquisition system by this method and correct the average transfer function of the image acquisition system, the non-sharp mask signal Du is expressed by the above equation (2).
It is well known to use equation (3) to achieve this without finding s.

That is, as shown in FIG. 2, for example, with respect to the I-th pixel 2 in the horizontal direction and the J-th pixel 2 in the vertical direction of the original image 1, the horizontal direction centered on the pixel 2 is −k / 2 to k / 2. Direction-l / 2-
Within the range of 1/2, all convolution functions for all the images centered on I and J are all multiplied, and the result is the result of the so-called two-dimensional convolution operation to obtain a new image 2 '. ing. This method is performed for all the pixels of the original image I.

With this method, as shown in FIG. 2, the number of calculations becomes very large.

As an example, if the vertical size of the image is M, the horizontal size of the image is N, and the vertical size of the convolution function h is K, the horizontal size of the convolution function h is L. It is necessary to perform multiplications and additions times, and to access the image information, which is (K × L) × (M × N) times for the entire image.

For example, in the case of an X-ray image, K = 10-60, L = 10
60 and M = 1000 to 2000, the maximum case is 60 × 60 × 2000 × 2000 = 144 × 10.
It becomes 0 × 10 ⁶ times.

The frequency characteristic of the system for obtaining the radiation image normally has a shape as shown in FIG. 3. To correct this frequency characteristic, the fourth characteristic is used.
If the spatial filter has a frequency characteristic that changes smoothly as shown in the figure, the characteristic becomes optimum as shown in FIG. 5 when the average transfer function of the system is corrected.

That is, the important frequency components 0 to 1.0 lp /
The response of the system is about 1.0 in the mm range. On the other hand, the simple filter as shown in FIG. 1 cannot correct the correction in this way.

By the way, although this filter corrects fine details in the image well, it has been found that graininess is noticeable. It was found that the cause of this is that most of the image information in the radiation image is at a spatial frequency of 0.51 lp / mm or less, and at a spatial frequency higher than that, there is a large proportion of noise although there is image information. .

The present invention can finely correct even a small portion of an image by high-speed processing, and can obtain a good image excellent in diagnosability without graininess. In particular, by using a convolution operation process, The shape of the spatial frequency filter can be freely designed for each pixel, the characteristics of the image can be improved, and the image quality can be improved while minimizing the image noise.

Therefore, when the radiation image is scanned, the radiation image information is read out, converted into an electric signal, and then reproduced as a visible image, a convolution function within a predetermined two-dimensional range is set for each pixel in each scan. The convolution operation of multiplying each and adding the respective values is performed on a predetermined image, and the spatial frequency of the maximum response of the convolution function or the spatial frequency and intensity of the maximum response of the convolution function are used as the intensity of the original image signal. To correct the image.

Here, how to specifically determine the filters hx and hy will be described.

First, it is necessary to know the frequency response of the image acquisition system.
Since the line imaging system is represented by any of the curves a to d in FIG. 4, it should be selected appropriately.
Or, the characteristics of the image processing system Even if 2a is specified by approximating with, almost the same result can be obtained.

f: lp / mm 2a: frequency at which the response becomes 1 / e e: low natural logarithm Then, the reciprocal of the frequency response value is taken as the filter response.

Here, the filter shown in FIG.
In the part where the response of the filter is more than (1.5 to 6.0), the value of Q is set like B or the spatial frequency is set to 3 to 5 lp / mm like the value of C, and gradually increases. It may be 0.

There are various combinations of hx and hy, but here, h
The description will be made assuming that x = hy, that is, the frequency components of the image are almost equal in X and Y.

Next, the square root of the frequency response obtained in FIG. 6 is taken, and this is taken as the frequency response of hx and hy, and is shown in FIG.

Further, this is subjected to inverse Fourier transform to obtain the counts in the real regions of hx and hy.

The above calculation may be performed each time the curve a, b, c, d in FIG. 3 is selected, but it is preferably calculated in advance and stored in a ROM or a floppy disk. It should be noted that it is possible to save time for each calculation by simply reading.

Further, it is known that the value of Q is preferably about 2.0 to 4.0, though there are individual differences depending on the preference of the diagnostician.

FIGS. 8 (a) and 8 (b) are graphs showing the judgments of 20 doctors by changing Q variously.

The circles are easier to judge than the original (+1) The triangles are the same as the original (0) × are harder to judge than the original (-1).

The frequency response of the filter thus obtained is shown in FIG.

It has been found that this filter has noticeable graininess that is well corrected even in small areas of the image.

It has been found that the cause of this is that in the X-ray image, most of the image information is at a spatial frequency of 0.5 lp / mm or less, and at a spatial frequency higher than that, although there is image information, the proportion of noise is large. .

The present invention is characterized in that the spatial frequency response of the convolution function is changed according to the intensity of the original image signal to correct the image.

That is, as shown in FIG. 9, 0.5-1.0 lp / mm
With the above spatial frequencies, it is possible to obtain good results by weakening the degree of enhancement and enhancing in the spatial frequency domain.

In the above description, Q is set to obtain the frequency fp in FIG. 6, and the above is set to B or C from the frequency fp reaching that Q, but conversely, the frequency fp is 0.5 to 1.0 lp /
It has been found that it is preferable to set the frequency to mm, set the frequency response at that time to Q, and gradually lower it like B or C at frequencies higher than that.

However, depending on the system that obtains the image, the maximum value of Q of the spatial frequency response may not be corrected unless it is tripled or more.

In this case, if a film of this type is always used, an unnatural image may occur in the image. For example, if the above-described processing is performed while keeping the spatial frequency characteristic Q constant regardless of the signal size (concentration) of the X-ray image of the stomach using barium contrast agent, a large area of low signal containing a large amount of contrast agent. Parts are emphasized more than necessary, and a suspicious image such as a white edge or a black band is generated, which deteriorates the image quality. In addition, graininess at a low image signal level conspicuously deteriorates the image quality.

In order to prevent the generation of these suspicious images, FIG. 10 (a),
The processes shown in (b) and (c) are effective.

That is, in FIG. 10 (a), in order to prevent the occurrence of a false image,
In the example in which the spatial frequency characteristic Q is small according to the size of the image signal S in the weak portion S1 or less of the original image signal S, is large in the range of a signal that originally has many signals, and is small in the strong portion S2 or more. is there.

The curve A shows a case where the curve is smoothly changed, and the curve B is a curve which is folded at S1 and S2.
Almost no difference was seen between the two. In the case of this example, Q is reduced in the weak portion of the original image signal to reduce the graininess of the image. Also, the portion S2 where the original image signal is strong
As a result, Q is reduced to prevent black edges and white edges.

Specifically, the low-signal part and the high-signal part occupy a large part of the entire image, and this region is not important for diagnosis and the middle-signal part is particularly important for diagnosis, such as gallbladder and liver angiography. This is the case, and if noise or the like is emphasized, the diagnosis is hindered. Therefore, it is an example of FIG. 10A that it is desirable to emphasize only the middle signal portion other than these.

Further, in the case of front chest imaging, if the characteristic of the convolution function h is fixed, noise in the low signal region of the spine and the heart portion increases, which makes the image quality very noticeable and deteriorates the diagnostic quality.

In this case, noise can be suppressed by decreasing Q of the frequency response of the convolution function h in the low signal region of the spine and heart and increasing it in the high luminance part of the lung.

This is the case in FIG. 10 (b).

The one shown in FIG. 10 (c) is suitable for the case of diagnosing a low-signal region such as blood vessel formation or lymphatic vessel formation, and that region does not occupy a very large part in the entire image.

As an example of these, the case where the frequency response Q of FIG. 10A is used in front chest imaging will be described.

The histogram of the image taken from the front of the chest has a shape as shown in FIG. U is the spine part, V is the heart part, and W is the lung field part. In general radiography, I am mainly interested in the lung field part W, and the spine part U and the heart part V part are not so much. It is not important, but it is said that it is good if the shape can be decoded to some extent. Therefore, the lower value S1 and the upper value S2 of the original image signal as shown in FIG. 10A can be determined from this histogram. The lower value S1 or less and the upper position S2 or more do not need to be corrected so much. That is, the lower value S1 or less has poor S / N, and the upper value S2 or more has good S / N, but if emphasized, the above-mentioned white edge or black edge may occur. This lower value S1 is defined as min + (ma
x-min) α1 and α1 are set to about (0.1 to 0.5), and the upper value S2 is min + (max-min) α2 and α
2 may be about (0.8 to 1.5). Further, the point at which the Louis product of the histogram is 30 to 50% may be S1, and the signal intensity at which the Louis product of the histogram is 80 to 90% of the entire image is 0.8 to 1.5 times S2.

For other parts, it is possible to determine S1 and S2 by the same method from the histogram of that part, but it is difficult to determine what part was photographed according to the part. , Can be realized by leaving it selected.

As described above, when the value of Q was changed according to the intensity of the image signal, the same evaluation as in FIG. 8 (a) was performed, and the result shown in FIG. 12 (a) was obtained.

This example is an evaluation of the example of FIG. 10 (a) with a chest X-ray image. In this case, Q is represented using the maximum frequency response (Qm) of the filter. According to this,
It can be seen that the range of Q is about 3 to 6 as shown in FIGS.

In the above description, the spatial frequency fp at which the filter has the maximum response is fixed, but changing this spatial frequency fp with respect to the image signal can also be used to improve the image quality.

For example, as shown in FIG. 13, the spatial frequency fp is lowered in a portion where the image signal is weak, that is, the low spatial frequency region is corrected for a signal having a poor S / N, and the S / N is good to some extent. In part, it is a method of correcting up to a high spatial frequency.

Further, the frequency fc shown in FIG. 9, that is, the frequency at which the frequency response of the filter becomes about 0.1 is changed according to the image signal, so that there is an effect of suppressing the roughness of the image.

That is, as shown in FIGS. 14A and 14B, the spatial frequency fc is reduced at the original image signal and the signal level with a low noise ratio,
By setting the frequency fc at a high level from a medium to high signal level with a sufficient S / N, noise in the low signal level section can be suppressed.

Further, as shown in FIG. 11, when the spatial frequency fc is raised again in the high signal level portion, the occurrence of suspicion is suppressed for the image which is rapidly changing from the low signal level to the high signal level,
Moreover, it is possible to sufficiently correct the portion that affects the diagnosis.

Further, it is needless to say that the effect is further improved if the Q of the frequency response is changed at the same time as shown in FIG.

[Effects of the Invention] As described above, the present invention enables high-speed image processing suitable for the visual characteristics of the human eye and the noise characteristics of the accumulative phosphor, and particularly at high speeds even for detailed images. It is possible to perform good correction by various processes, and it is possible to obtain a good image with excellent diagnosability without graininess.

In particular, the shape of the spatial frequency filter can be determined by multiplying each pixel of each investigation by a convolution function in a two-dimensional predetermined range and performing a convolution operation for adding the respective values to a predetermined image. , It can be designed freely for each pixel, and the image characteristics can be improved.

Further, by changing the spatial frequency of the maximum response of the convolution function, or the spatial frequency and the intensity of the maximum response of the convolution function according to the intensity of the original image signal, the image noise is improved while the image noise is minimized. Is planned.

[Brief description of drawings]

FIG. 1 is a diagram showing a characteristic example of a conventional spatial frequency filter,
FIG. 2 is an explanatory diagram of convolution calculation, FIG. 3 is a typical frequency characteristic diagram of a system for obtaining a radiation image, FIG. 4 is a diagram showing an example of frequency characteristic of the filter of the present invention, and FIG. FIG. 6 and FIG. 7 are diagrams for explaining an example of the frequency characteristic of the filter of the present invention, FIG. 8 is a diagram showing a doctor's evaluation of the processing, and FIG. 9 is this diagram. FIG. 10 is a diagram showing an example of frequency characteristics of the filter of the invention, FIG. 10 is a diagram showing an example of changes in Q with respect to an image signal, FIG. 11 is a diagram showing a histogram example of a chest X-ray image, and FIG. 12 is a doctor for other processing. Of FIG. 13, FIG. 13 is a diagram showing an example of changes in fp with respect to an image signal, and FIG. 14 is f with respect to an image signal.
FIG. 15 is a diagram showing an example of change in c, and FIG. 15 is a diagram showing an example of filter characteristics for an image signal. 1, 1 '... Image 2, 2' ... Pixel

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁵ Identification code Internal reference number FI Technical display location G06F 15/68 400 A 9191-5L H04N 5/325 7/18 K

Claims (1)

[Claims] 1. When a radiation image is scanned to read radiation image information, converted into an electric signal, and then reproduced as a visible image, each pixel of each scanning is folded in a predetermined two-dimensional range. A convolution operation for multiplying each convolution function and adding the respective values is performed on a predetermined image, and the spatial frequency of the maximum response of the convolution function or the spatial frequency and intensity of the maximum response of the convolution function are calculated. A radiation image processing method, which comprises varying the intensity of an original image signal to correct the image.