JPS6262378B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to an image on an X-ray photographic film (hereinafter referred to as "X").
The present invention relates to an X-ray image processing method and apparatus that improve diagnostic performance by performing unsharp mask processing when copying an X-ray image. It goes without saying that it is desirable to obtain as much information as possible with a single X-ray photograph, since X-rays can be harmful to the human body if the exposure dose is high. However, in general, X-ray photographic film must have sufficient sensitivity and wide exposure range for imaging, as well as high contrast, high sharpness, and fine graininess necessary for observation and interpretation. On the other hand, these conditions often conflict with each other, and it is difficult to create an X-ray photographic film that satisfies all of them. It's reality. Therefore, the present inventors read out the image on this X-ray photographic film, converted it into an electrical signal, processed it and reproduced it as a copy photograph, thereby improving the contrast, sharpness, and graininess of the X-ray photographic film. In Japanese Patent Application No. 53-28533, a method was proposed that would improve the diagnostic performance of images, obtain as much diagnostic information as possible, and at the same time give X-ray photographic film better suitability for imaging. This method is
Although this technology has the potential to dramatically improve the diagnostic performance of X-ray images compared to the past, the image processing method used to reproduce X-ray images is not necessarily clear in detail, and is left to future research. Ta. Therefore, the inventor has repeatedly researched this point and
An X-ray image processing method was proposed in Japanese Patent Application No. 163,575/1983 in which a non-sharp mask process was applied when copying a radiation image to improve diagnostic performance. This method emphasizes the knowledge that frequencies important for diagnosis are in the extremely low frequency (hereinafter referred to as "ultra-low frequency") region, although there are some differences depending on each part of the human body, and emphasizes high frequency components. It is known that trying to improve sharpness only emphasizes noise components in the case of X-ray image processing, which actually reduces diagnostic performance, and that noise accounts for a high proportion in high frequency regions. If you reduce the emphasis in this high frequency range, the noise will become less noticeable.
Based on the knowledge that this makes it easier to see, the idea is to emphasize very low frequency components and at the same time relatively reduce high frequency components, which account for a large proportion of noise, so that images that are easier to see visually can be obtained. Basically, by scanning the original X-ray photograph,
After reading out the X-ray image information recorded in this and converting it into an electrical signal, when reproducing it as a copy photograph, etc., the unsharp mask density Dus corresponding to the ultra-low frequency is determined at each scanning point, and the original photograph is concentration
Dorg, the emphasis coefficient is β, and the density reproduced in a photocopy, etc. is D′, then D′=Dorg+β(Dorg−Dus) is performed for the unsharp mask processing, and the frequency This is an X-ray image processing method characterized by emphasizing components. Here, the ultra-low spatial frequency means a spatial frequency of approximately 0.5 cycles/mm or less. Here the unsharp mask density corresponding to very low frequencies
Dus refers to the density at each scanning point of an unsharp image (hereinafter referred to as an "unsharp mask") that blurs the original image so that it only contains frequency components lower than ultra-low frequency components. A sharp mask is used whose modulation transfer function is 0.5 or more at a spatial frequency of 0.01 cycles/mm and 0.5 or less at a spatial frequency of 0.5 cycles/mm. As for how to create,
(1) When reading image information, the density of the measurement point is averaged with the surrounding density by changing the spot diameter of the reading light beam (for this purpose, the spot diameter of the light beam is (2) Store the density of each measurement point and change it according to the size of the non-sharp mask. , a method of reading out data along with peripheral data and calculating the average value (simple average or average value using various weighted averages) of Dus (in this method, two methods are used:
In some cases, it is created after A/D conversion as a digital signal, and in some cases, the analog signal is de-sharpened with a low-pass filter only in the main scanning direction before A/D conversion, and digital signal processing is performed in the sub-scanning direction. included. ) can be used. The inventors further compared and studied the non-sharp mask creation methods described in (1) and (2) above, and found that method (2) is the most preferable in order to provide flexibility in image processing. However, in this case, ideally, the following calculation is normally required to obtain the unsharp mask density Dus at each scanning point. Here, i and j are the coordinates of a circular area centered on each scanning point (the number of pixels falling within that area is N in the diametrical direction), and aij is a weighting coefficient that is equal in all directions. It is preferable to have directional and smooth changes.

[Formula]. However, if such an operation were to be performed simply, it would be necessary to perform approximately π/4N ² multiplications and π/4N ² additions for each scan point, and if N is large, the operation becomes extremely difficult. The disadvantage is that it is time consuming and impractical. In fact, when reading out a normal X-ray image by scanning the original photograph, it is necessary to ensure that the frequency components of the image are not lost; Although there are some differences, it is usually necessary to scan at a sampling rate of about 5 to 20 pixels/mm (200 to 50μ in terms of pixel size), whereas the non-sharp mask in the present invention supports ultra-low frequencies. Therefore, in order to create this mask, it is necessary to perform calculations using an extremely large number of pixels. For example, for a mask with Gaussian weighting coefficients, if the pixel size is 100μ x 100μ, then if _c = 0.1 cycles/mm, N will be approximately 50, and _c = 0.02
In the case of cycles/mm, N is approximately 250, so the calculation time becomes enormous. (Here, _c is the modulation transfer function of the unsharp mask.
It means the spatial frequency value of 0.5. ) Also, averaging a circular area means changing the addition range for each scanning line, but having to make such a judgment during calculation requires a significant number of calculation mechanisms. , it is uneconomical. An object of the present invention is to provide an X-ray image processing method and apparatus that can improve diagnostic performance economically and at high speed. In order to achieve this objective, the present inventor has conducted intensive research and found that in the above image processing method,
To obtain the unsharp mask density, we simply calculate the original image density Dorg at each scanning point within a rectangular area surrounded by two sides parallel to the main scanning direction and two sides parallel to the sub-scanning direction. It has been found that a method of determining the unsharp mask density Dus for ultra-low spatial frequencies at each scanning point by averaging is in accordance with the above objective. In other words, such a method of creating a non-sharp mask has a uniform weight in a rectangular area, and therefore its transfer characteristic causes oscillations, compared to a mask whose weight decays smoothly with, for example, a Gaussian distribution weight. And,
Although it has drawbacks such as the degree of unsharpness varying depending on the direction, there is no substantial difference in terms of improvement in diagnostic performance from the case of the ideal mask calculation described above. It has been found that since it is a simple averaging of , it is possible to significantly reduce the calculation time and the cost of the device, as will be described later. The present invention scans an original X-ray photograph, reads out the X-ray image information recorded therein, converts it into an electrical signal, and then reproduces it as a copy photograph. Find the density Dus of the unsharp mask corresponding to the frequency, and when the density of the original photo is Dorg, the emphasis coefficient is β, and the density reproduced in the copy photo, etc. is D′, D′ = Dorg + β (Dorg − Dus). In an X-ray image processing method that emphasizes frequency components higher than ultra-low spatial frequencies by performing calculations, the unsharp mask density Dus is divided into two sides parallel to the main scanning direction of the scanning and one parallel to the sub-scanning direction. This is an X-ray image processing method characterized by calculating the original image density Dorg at each scanning point within a rectangular non-sharp mask surrounded by two sides by simple averaging. The apparatus of the present invention also includes a photodetector that scans an original photograph, reads out an X-ray image recorded therein, and converts it into an electrical signal, and an output of this photodetector that is transmitted in the main scanning direction of the scanning. A circuit that calculates the unsharp mask density Dus corresponding to ultra-low spatial frequency by simply averaging over the range of a rectangular unsharp mask surrounded by two parallel sides and two sides parallel to the sub-scanning direction. , this unsharp mask density Dus, the original image density Dorg which is the output of the photodetector,
This X-ray image processing apparatus is equipped with an arithmetic unit that performs an arithmetic operation expressed by the arithmetic expression D'=Dorg+β(Dorg-Dus), where the density of the reproduced image is D' from the enhancement coefficient β. In the present invention, the unsharp mask density Dus corresponding to ultra-low frequencies refers to an unsharp image (hereinafter referred to as "unsharp mask") that blurs the original image so that it only contains frequency components lower than the ultra-low frequency components.
refers to the signal corresponding to the density of each scanning point. As this unsharp mask, the modulation transfer function is
0.5 or more when the spatial frequency is 0.01 cycles/mm, and when the spatial frequency is 0.5 cycles/mm
0.5 or less, or the integral value of the modulation transfer function with 0.001 as the lower end in the spatial frequency range of 0.01 to 0.5 cycles/mm is 0.001 to 10
90% of the integral value of the modulation transfer function in cycles/mm
Something like this is used. In addition, patent application (3) dated November 22, 1978 (applicant:
Fuji Photo Film Co., Ltd.), the modulation transfer function is 0.5 or more at a spatial frequency of 0.02 cycles/mm and 0.5 or less at a spatial frequency of 0.15 cycles/mm. With sharp masks, the improvement in diagnostic performance is significantly favorable. Here, the above-mentioned unsharp mask has a modulation transfer function _c of 0.01 to 0.5 cycles, preferably 0.02 to
It can be said to be in the range of 0.15 cycles/mm. In the present invention, when simple averaging is performed using a rectangular unsharp mask, in other words, when the unsharp mask is rectangular and the weight of the original photographic density (Dorg) of the pixels included in the mask is constant, The regulation that _c is 0.01 to 0.5 cycles/mm (preferably 0.02 to 0.15 cycles/mm) theoretically limits the length of one side of a rectangular non-sharp mask to 60 mm to
It is synonymous with 1.2 mm (preferably 30 mm to 4 mm). Note that even when the non-sharp mask has a rectangular shape, the length of each side only needs to be within the above range; for example, a rectangular mask with a large aspect ratio is effective for image processing of linear tomography. In the present invention, the enhancement coefficient β includes cases where it is a constant and cases where it is a function of the original photographic density (Dorg) or the unsharp mask density (Dus), but especially in the latter case, that is, when the enhancement coefficient β is It is preferable to change the density according to the photographic density (Dorg) or the unsharp mask density (Dus), since the diagnostic performance can be further improved. Furthermore, depending on how the enhancement coefficient β and the unsharp mask density (Dus) are selected, the maximum value B of the modulation transfer function of the system that provides the enhanced copy photograph according to the present invention and the modulation transfer function near zero frequency can be determined. Although the ratio of A values (B/A) changes, when B/A<1.5, there is almost no difference in diagnostic performance compared to conventional X-ray photography. Furthermore, when performing the processing of the present invention with the emphasis coefficient β as a constant, if B/A exceeds 6, unnatural image parts may appear due to excessive emphasis, or the image may appear white or black. This is not desirable as it often causes a partial appearance and interferes with the diagnosis. On the other hand, when the enhancement coefficient β is changed according to the original photographic density Dorg or the unsharp mask density Dus, the preferred range of B/A (in this case, B/A also changes according to Dorg or Dus) , B/A is its maximum value) is expanded, and even if B/A exceeds 6, as long as it is 10 or less, the above-mentioned false image was not noticeable. Furthermore, when the value of B/A was set in the range of 2 to 5.5 when β was fixed, and in the range of 2 to 8 when β was variable, the diagnostic performance was significantly improved. In addition, the emphasis coefficient β is set so that B/A is within the above range, but in addition to β of B/A, it changes slightly depending on the shape of the unsharp mask, that is, Dus, but B/A =1.5 to 10 is synonymous with setting β to 0.4 to 8 when a simple averaging mask is used. In the present invention, in addition to the above operations, smoothing processing can also be performed. In general, in the frequency region of very low spatial frequencies or higher, there is a lot of noise and it is often difficult to see, so it is often preferable to perform further smoothing processing to further improve the diagnostic performance. The smoothing process is preferably such that the modulation transfer function is 0.5 or more when the spatial frequency is 0.5 cycles/mm, and 0.5 or less when the spatial frequency is 5 cycles/mm. What kind of smoothing process is preferable?
For example, when interpreting relatively low-frequency shadows such as a chest tomogram, it is preferable to remove as much noise as possible, but conversely, it is preferable to follow fine blood vessel shadows that contain high frequency components such as in an angiogram. If necessary, too strong a smoothing process will make it difficult to see even the shadows you want to see.
Depending on the location of the X-ray, the symptoms, the purpose of the test, etc., the present inventor's research has shown that by performing the smoothing process described above, it is possible to diagnose almost all X-ray images. It was found that this method has the effect of improving performance. Furthermore, this smoothing process is equally effective whether it is applied to D′ after the ultra-low spatial frequency processing of the present invention or to the original photographic density Dorg. It is acknowledged that there is. Further, in the present invention, gradation processing may be performed in addition to frequency emphasis processing using a non-sharp mask. Ultra-low frequency processing has a relatively small effect on diseases where the concentration changes slowly over a large area, such as lung cancer and breast cancer, so it is recommended that 54
It is desirable to use the gradation processing disclosed in No.-23090 etc. in combination. In this case, gradation processing may be performed either before or after ultra-low frequency processing. Embodiments of the present invention will be described in detail below with reference to the drawings. FIG. 1 is a schematic diagram of an X-ray image processing apparatus showing an embodiment of the present invention. In Figure 1, original photograph 1 recorded with an X-ray image by X-ray photography.
is attached to the outer periphery of the transparent drum 2. This transparent drum 2 rotates and simultaneously moves in the axial direction. A reading light source 3 is arranged inside the transparent drum 2. The light emitted from this reading light source 3 is
A light beam is formed by a lens or the like and illuminates the original photograph 1 from behind. The light beam transmitted through the original photograph 1 passes through the aperture 3a and enters the photoelectric converter 4, where it is converted into an electrical signal. This electrical signal is nonlinearly amplified by an amplifier 5, then converted to a digital signal by an A/D converter 6, and stored on a magnetic tape 7. The digital signals of each section stored on the magnetic tape 7 are read out to an arithmetic unit 8, such as a minicomputer, and after determining Dus, the above-mentioned calculation of D'=Dorg+β(Dorg-Dus) is performed. The above Dus is 0.5 or more when the modulation transfer function has a spatial frequency of 0.01 cycle/mm, and 0.5
It must be specified which one is used, which is less than or equal to 0.5 at a spatial frequency of cycles/mm. In addition, when calculating the above formula,
An emphasis factor β must be specified. These values may be specified individually from the outside, or several types may be determined depending on the part of the human body or each case, and these values may be stored in the memory of the computing device. Smoothing processing for reducing high frequency components is performed on D'. This smoothing process makes it possible to reduce noise without damaging information necessary for diagnosis. As described above, the present invention is characterized in that the unsharp mask density Dus is obtained by simple averaging of the signals within the mask using a rectangular unsharp mask. The unsharp mask density Dus can be determined by an extremely simple method. This is an advantage that is common to both digital and analog signal processing, and in practice, this method makes it possible to obtain the unsharp mask density Dus in an extremely short time. This is the first time that unsharp mask processing using the above-mentioned calculation method can be implemented in a practical sense. That is, for example, when calculating by multiplying the original photographic density D _prg(i , _j) of each scanning point by the weighting coefficient a _ij , the unsharp mask density D _us(IJ) is (i, j are pixel numbers indicating the coordinates of each scanning point,
I, J are numbers indicating the coordinates of the unsharp mask density,

[Formula]), the number of calculations is approximately N ² times of multiplication,
It is necessary to repeat the addition N ² times, N (the length of one side of the non-sharp mask expressed in pixels),
That is, when the number of pixels in the unsharp mask increases, it takes a considerable amount of time to obtain the unsharp mask density Dus. For example, the size of the non-sharp mask is 6mm x 6mm.
So, if it contains 3600 pixels (0.1mm x 0.1mm), to calculate the mask at each scanning point,
It is necessary to repeat 3,600 multiplications and 3,600 additions. For example, when calculating with only software using an 8-bit microcontroller, multiplication takes 3 msec and addition takes 5 μs.
Considering that it takes 3 seconds, it takes 3 msec x 3600 + 5 μsec x 3600 ≒ 11 seconds to obtain one non-sharp mask signal, which is completely impractical. On the other hand, according to the present invention, simple addition-average multiplication is no longer necessary, and the calculation time can be greatly reduced. For example, in the above example, each point takes 18 msec. Furthermore, it is possible to simplify various calculations as described below, and depending on the calculation algorithm, the number of calculations can be drastically reduced to just a few times.
It is also possible to obtain the unsharp mask density Dus within 10 μsec, and the practical effects of the present invention are significant.
That is, since D _{us (IJ)} is determined by D _{us (IJ)} = 1/N ² (ΣD _ij ), Dus can be determined by only N ² additions and one division. In more detail, assuming that the size of the unsharp mask is N ₁ in the main scanning direction and N ₂ in the sub-scanning direction, the density of the unsharp mask D _{us (IJ)} is D _{us (IJ)} = 1/N ₁ × N ₂ (ΣD _ij ) (i is −N ₁ −1/2 to +N ₁ −1/2 j is J−N ₂ −1/2 to J+N ₂ −1/2 N ₁ and N ₂ are both positive odd numbers) It can be simply calculated using N ₁ ×N ₂ additions and 1 division. Furthermore, if the calculation procedure is devised using various algorithms described below, the average number of calculations to obtain one non-sharp mask density can be reduced to just four. An example of an algorithm (digital method) that particularly simplifies calculation for determining the unsharp mask density Dus will be described below. As shown in FIG. 2, consider a rectangular non-sharp mask M (indicated by a thick solid line) surrounded by two sides parallel to the main scanning direction and two sides parallel to the sub-scanning direction.
For simplicity, this mask M is a square, and the length of one side is N as the number of pixels. (N is a positive odd number) In Figure 2, D' _IJ performs image processing to obtain the density value of the scanning point (pixel). (final density value), P _IJ is the pixel input at the time of interest when D _IJ is at the tip of the mask in the scanning direction.
The density value of T _IJ is the sum of the signal values of N ² pixels in the mask M, i.e.

[Formula]. Here, the signal value D _IJ of the pixel P _IJ of interest
is first stored at the address corresponding to the density D of the pixel. Each address requires a number of bits (for example, 8 bits) that can express the density value of the pixel. Next, the sum of the densities of N pixels in the main scanning direction C _IJ
(i.e.

Find [Formula]). This is the sum C _I-1 , _J of the density values of N pixels lined up in the column of the pixel of interest P _IJ up to the column before P _IJ , and the sum of the density values of the N pixels before the pixel of interest P _IJ . Concentration value D _IN , _J
and the density _value D _I , _{J of} the pixel P _{I , J} before _the pixel _of interest _.
It can be obtained from D _N , _J. Then, this sum C _IJ is stored at the corresponding address of the sum C of the densities of the columns of pixels in the main scanning direction. Each address requires a sufficient number of bits to perform this operation without overflowing, and this number depends on N. Next, the sum T _IJ of the density values of N ² pixels in the mask M _I , _J is determined. This is the pixel of interest P _I ,
The sum T _I , _J -1 of the density values of the pixels in the mask M _I , _J-1 located one column back in the sub-scanning direction from the mask M _I , _{J that includes J at the tip} , and the mask M _I , _{J -1} 's final column (i.e., column not included in M _I , _J ) sum C
_I , _JN and the sum C I, _J of the density of the column containing the pixel of interest P _I , _J at the tip, the calculation formula T _{I , J} = T _I , _J-1 +
It can be obtained from C _I , _J - C _I , _JN . Then, this value T _I , _J is stored at the corresponding address of the total density T of the pixels in the non-sharp mask. This T _I , _J
is equivalent to N ² times the unsharp mask density Dus, so after obtaining this T _IJ , the calculation formula D′ _I , _J = D _I- 〓〓, _J- 〓〓 + β(D _I- 〓〓, _J- 〓〓- _TI , _J / ^N2 ), the above-mentioned unsharp mask processing can be performed. The memory capacity required for the above calculation will be explained next. FIG. 3a shows a memory for D _IJ , which requires memories for all the number of pixels required in the main scanning direction, and N+1/2 memories in the sub-scanning direction. For example, one memory only needs to have a capacity of 8 bits. FIG. 3b shows memories for C _IJ , and it is sufficient to have the same number of memories as the memories for D _IJ in the main scanning direction, and N+1 memories in the sub-scanning direction. This memory requires two to three times the number of bits as the memory above. FIG. 3c shows a memory for T _IJ , which requires the same number of memories as the above two memories in the main scanning direction, but only needs to have two memories in the sub-scanning direction. FIG. 4 shows an example of a circuit block that performs the above calculation, in which a signal representing the density is sent from the gate 21 to which the pixel input density D _IN is inputted to the memory 22 having the above capacity, and the stored value is stored in the memory 22. The arithmetic circuit 13 performs an arithmetic operation based on the .
These gates 21, memory 22, arithmetic circuit 2
The operation No. 3 is performed by the control circuit 24. Arithmetic circuit 2
3 is output from the gate 21 via the memory 22 as the pixel output density Dorg. According to the above calculation method, the unsharp mask density
The calculation for obtaining Dus is extremely simplified, and the equipment for it is also extremely simplified. this is,
The method of the present invention is based on obtaining the unsharp mask density by simply averaging the density values of pixels within a rectangular mask. That is, according to the simple averaging method of the present invention, an extremely simplified algorithm such as the one described above becomes possible, and calculations can be performed extremely easily. can be realized extremely easily. In addition, in the arithmetic circuit that implements the above algorithm, three types of memory 2 are used as shown in FIG.
8, 29, and 30 can be made into a series of memories with consecutive addresses, but three types of memories 25, 26, and 27 can also be made into three memories by dividing the address bus and data bus as shown in Figure 6. If they can be accessed simultaneously, the calculation time can be further reduced. The control circuit and arithmetic circuit are each made of dedicated hardware such as PLA (programmable
logic array), random logic (random
logic) etc. may also be used. Further, a microcomputer, a minicomputer, etc. may be used for these circuits, or a high-speed microcomputer (eg, bit slice type) may be used for the control circuit, and a dedicated circuit may be used for the arithmetic circuit. This can be determined by selecting an appropriate one depending on the required calculation speed. Next, an example of an algorithm that can reduce the memory capacity even more than the above algorithm will be explained with reference to FIGS. 7, 8, and 9. In this algorithm, the density value D of the pixel of interest, that is, the pixel _P at the tip of the non-sharp mask M _IJ
After storing _IJ in the corresponding address of the memory for D, the sum E of the signals of N pixels of D _IJ in the sub-scanning direction
_IJ , i.e. is calculated and the value is stored at the corresponding address in the memory for E. This is done by the arithmetic expression E _IJ =E _I , _J-1 +D _I , _J -D _I , _JN . Using these stored values, T _IJ corresponding to N ² times the unsharp mask density Dus is determined.
This T _IJ is determined by the arithmetic expression T _IJ =T _I-1 , _J +E _I , _J −E _IN , _J. However, this method cannot calculate when the main scanning returns from the right end to the left end, so at this time, the sum R _J of the first N densities D _IJ in the main scanning direction,

Calculate [Formula] and store it at the corresponding address in the R memory. For example, as shown in FIG. 9, this R _J is N=
5, R ₁ is the sum of S ₁ , ₁ to S ₅ , ₁ , and R ₅
is the sum _of D _1,5 to _{D 5,5} . D ₅ , ₅ to D ₆ ,
Even if the output changes to ₅ , _R5 does not change. Therefore, when the main scan returns from the right end to the left end, T _IJ is calculated using the above R _J according to the arithmetic expression T _IJ =T _I , _J-1 +R _J -R _JN . Using T _IJ obtained in this way, the calculation formula D′ _IJ = D _I- 〓〓, _J- 〓〓 + β (D _I- 〓〓, _J- 〓〓−T _IJ /N ² ) The aforementioned unsharp mask processing can be performed. In this algorithm, the memory for the density value D _IJ of each pixel is as shown in Figure 8a, in the main scanning direction there is a memory for all the necessary number of pixels in the main scanning direction, and in the sub-scanning direction there is N+1 memory. As shown in Figure 8b, c, d, the memories for R, E, and T are N+1 in the main scanning direction, 1 in the sub-scanning direction, and T. It is possible to use a small capacity memory having two memories in the main scanning direction and one memory in the sub scanning direction. For example, 8 bits can be used for each address of the memory for D, but for example 16 bits (depending on the size of N) are required for the memories for R, E, and T. In this algorithm, the memory for S, which requires a small number of bits, is made large, and the other memories, which have a large number of bits, are made smaller, so the overall memory capacity can be significantly reduced. Therefore, the capacity of the memory shown in FIG. 8 can be made much smaller than that shown in FIG. 3, which has a great effect on the simplification of the device. In addition, the positive odd number N in the above two methods is preferably about 10 pixels/mm in order to obtain the image accuracy necessary for diagnosis, and in that case, the size should be in the range of 601 to 11, preferably 301 to 39. good. Both of the above two algorithms utilize a method of digitally processing the density, but the density at each scanning point is integrated in the main scanning direction in an analog manner, and the integrated value is stored in memory. , even if this is numerically integrated, Dus can be similarly obtained by adding the densities of all scanning points within the non-sharp mask. In this case, N analog integration circuits are required because the analog values are integrated and added for each pixel instead of a digital circuit, but the following method can reduce the number of integrators to one. It is possible and advantageous. In other words, the analog output Dorg of each scanning point is 2
one is passed through a delay circuit (the delay time (T) is the scanning time of one pixel (τ) x the number of pixels in the main scanning direction of the non-sharp mask (N), that is, T = τ x N), and the two are connected to a difference density calculation circuit. and integrate the output (Dorg−TDorg) of the difference concentration calculation circuit to obtain ∫ ^t _−∞ (Dorg−TDorg)=∫ ^t _−∞ Dorg
−∫ ^t−N・〓 _−∞ Dorg=∫ ^t _t−N・〓Dorg A method of obtaining the value can also be adopted. This value is the second
This corresponds to C _I , _J in FIGS. 3 and 3, and by digitally adding them in the sub-scanning direction, T _I , _J is obtained, from which the unsharp mask density Dus can be determined. This method is also fast and easy to calculate, and is suitable as an analog method. Note that the unsharp mask density Dus _(IJ) is centered around one scanning point (i, j), and is expressed as N ₁ -1/2<i<N _X -N ₁ -1/2 N ₂ -1/2<j< _{N Y −N 2} −1/2 ( _N Therefore, the unsharp mask density centered around the scanning point at the edge of the image cannot be determined because there is no density outside the edge. As a method of processing this end, the outermost D _IJ
Assuming that the value of is infinitely extended outward, it is natural and advantageous to store the outermost value in memory and use this value as the density outside the edge. Alternatively, the area outside the outermost periphery may be processed as black or white, or may be processed as a constant intermediate value between black and white. In the above method, only one unsharp mask is used to perform unsharp mask processing, but it is also possible to use two unsharp masks of different sizes to provide stages in frequency emphasis. In this case, we will perform the calculation expressed by the equation D'=Dorg+β(Dorg-Dus ₁ ) +α(Dorg-Dus ₂ ), but if we rewrite this equation, we get D'=Dorg+(β+α){Dorg-1/ It can also be expressed as β+α (βDus ₁ +αDus ₂ )}, and the operation corresponds to performing an operation similar to the above-mentioned operation. When Dus ₂ is smaller than the non-sharp mask Dus ₁ and the emphasis coefficient α is positive, the graph of the modulation transfer relationship has an additional peak at the higher component of the emphasized frequencies, and when α is negative In this case, the higher components of the emphasized frequencies are gradually lowered.
The former is particularly suitable for bone regions, angiography, double gastroscopy, etc., and the latter is suitable for chest tomography, cholecystography, hepatography, abdominal plain radiography, head imaging, etc. The arithmetic processing using the non-sharp mask described above will be explained in more detail below with reference to FIG. Figure 10a shows the frequency response when an original photograph is sampled at 10 pixels/mm. This curve is sinc when a rectangular aperture is used as the photodetector aperture.
It is known that when a Gaussian distribution aperture is used for a curve, the curve becomes a Gaussian distribution curve. In Figure 10b, the modulation transfer function is 0.01 cycles/mm.
This shows a rectangular unsharp mask using a mask that is 0.5 or more when the spatial frequency is , and 0.5 or less when the spatial frequency is 0.5 cycles/mm. In this example, when an original photo is sampled at 10 pixels/mm, a non-sharp mask is created by taking a simple average of approximately 63 pixels x 63 pixels (this is called "Non-sharp mask size N = 63"). It is. This is equivalent to scanning the original photograph with a large light beam measuring 6.3 mm x 6.3 mm. FIG. 10c is a graph showing the modulation transfer function after calculating (Dorg-Dus). FIG. 10d shows the calculation result D'. Here, β is set to "3". As a result of the above calculation, the maximum value B of the modulation transfer function of the copy photograph is approximately 4.6 times the modulation transfer function A near zero frequency. Figure 10e shows the frequency domain (0.5 to 5 cycles/
This figure shows an example of a smoothing modulation transfer function when performing smoothing processing in mm). Here 5 pixels x
The modulation transfer function of smoothing with 5 pixels is shown. FIG. 10f shows the smoothing process as shown in FIG. 10d.
This shows the modulation transfer function when applied to D′. FIGS. 11A to 11D show an example in which the emphasis coefficient β is continuously changed according to the original photographic density (Dorg) or the unsharp mask density (Dus). Figure 11A shows a flat type with constant β;
Figure B shows the monotonically increasing type (β'≧0), Figures 11C and 11D both include the case where β'<0, Figure 11C shows the low density emphasis type, and Figure 11D shows the medium density emphasis type. These include a stepped change (curve a) and a curved change (curve b). As shown in FIG. 11B, by changing β monotonically, it is possible to prevent false images that tend to occur due to frequency emphasis. As an example, if we perform the frequency processing on an original photograph of the stomach using a barium contrast agent while fixing the enhancement coefficient β, the boundary of a wide uniform low-density area containing a large amount of contrast agent becomes The image is emphasized more than necessary, resulting in a double-contoured false image.
Instead, the occurrence of the double contour can be prevented by making the enhancement coefficient β variable, that is, decreasing β in a low-density area where a large amount of contrast agent has entered, and increasing β in a high-density area such as a gastric subdivision. As another example, in the case of frontal chest photography, if β is fixed, noise will increase in low density areas of the spine and heart, and in extreme cases, details will be washed out. (This is visually very noticeable and has a negative impact on diagnostic performance.) Similarly, if β is made smaller in the low-density areas of the background and the heart, and β is made larger in the high-density areas of the lung field, the above-mentioned noise and white It is possible to prevent an increase in omissions. The low-density enhancement shown in FIG. 11C is suitable when diagnosis of a low-density area is particularly important and the low-density area does not occupy a large portion of the entire image. For example, this applies to angiography and lymphangiography, and in these X-ray images, it is desirable to greatly improve the sharpness of the relevant areas even if the noise increases slightly, so this low-density enhancement is effective. Diagnostic performance is greatly improved. Furthermore, the middle density enhancement shown in Fig. 3D is used when the low density and high density areas occupy a considerable portion of the entire image, and these areas are not important for diagnosis, and the middle density areas are particularly important for diagnosis. Are suitable. For example, cholecystography and hepatography correspond to this case, and if noise or gas areas are emphasized in these X-ray images, it will interfere with diagnosis, so excluding these areas will make the middle-dense area that is the target of diagnosis. It is desirable to emphasize only In any of the above examples, if the emphasis coefficient β is fixed to a small value and frequency processing is performed, various false images will not occur, but gastric subdivisions and other areas that make an important contribution to diagnostic performance will be avoided. The contrast of blood vessels in the lung field and contrast-enhanced vessels does not improve, and diagnostic performance does not improve. By continuously changing the emphasis coefficient β in accordance with the density in this way, an image with improved diagnostic performance can be obtained while preventing the generation of false images. FIG. 12 shows an example of how to increase β. From the histogram of the original photograph, its minimum density D ₀ and maximum density D ₁ are determined, and β is changed almost linearly between these values. D ₀ and D ₁ are determined depending on the type of X-ray image to be processed, and for example, the lowest and highest densities may be the density values when the integral histogram is 0 to 10% and 90 to 100%, respectively. Figures 13 and 14 are low-density emphasis,
This figure shows an example of how to change β in middle density emphasis. In FIG. 13, β decreases between the concentrations A and B from the maximum value βmax to the minimum value βmin. In other words, the emphasis coefficient is increased (βmax) in the low concentration region (from Dmin to A), and decreased (βmin) in the high concentration region (from B to Dmax).
are doing. Concentration A is the minimum concentration (Dmin), the difference between the maximum concentration (Dmax) and the minimum concentration (Dmin) (△
D) plus 0.2 to 0.5 times [Dmin+(0.2 to
0.5) ΔD] is good, and the density B is preferably 0.7 to 1 times larger [Dmin+(0.7 to 1) ΔD]. In Figure 14, β is the first between concentrations A and B.
increases from the minimum value (βmin1) to the maximum value (βmax), and increases from the maximum value (βmax) to the second
decreases to the minimum value (βmin2). That is,
The emphasis coefficients are made small (βmin1, βmin2) in the low density region (from Dmin to A) and the high density region (from D to Dmax), and are made large (βmax) in the medium density region (from B to C). .
Here, the first minimum value (βmin1) and the second minimum value (βmin2) may be equal. In the case of the chevron-shaped dotted line b, β increases between concentrations A and E and decreases between E and D. Concentration A is the minimum concentration (Dmin) plus 0 to 0.2 times the difference (△D) between the maximum concentration (Dmax) and the minimum concentration (Dmin) [Dmin + (0 to 0.2) △D], the concentration B is the average concentration (=Dmin+Dmax/2 or statistical average value) minus 0 to 0.2 times the difference (ΔD) [-(0 to 0.2)ΔD], and concentration E is the average concentration () , the concentration C is the difference between the average concentration and the difference (ΔD) from 0 to
The density D is the maximum density (Dmax) minus 0 to 0.2 times the difference (ΔD) [Dmax-(0 to 0.2)]. )△
D] are respectively desirable. In addition, in the calculations in FIGS. 13 and 14 above,
The maximum density (Dmax) and the minimum density (Dmin) both correspond to the maximum and minimum in the target substantial image, and there are densities larger or smaller than these in areas other than the image. It's also possible. Note that, depending on the case, the maximum and minimum of the entire screen may be simply taken. According to the inventor's experiments, when β is changed depending on the density of the original photograph and when β is changed depending on the density of the non-sharp mask,
The effects were approximately the same. In addition to the frequency emphasis processing using a non-sharp mask as described above, gradation processing can also be used in combination.
When performing gradation processing before ultra-low frequency processing, perform gradation processing using a nonlinear analog circuit and then
Perform D conversion. If it is performed after A/D conversion, digital processing can also be performed using a minicomputer. Further, digital processing is performed after ultra-low frequency processing, or analog processing is performed after D/A conversion. These frequency-emphasized and, if necessary, gradation-processed data are recorded on the magnetic tape 7.
(FIG. 1) The data on the magnetic tape 7 is sequentially read out, converted into an analog signal by the D/A converter 9, and amplified by the amplifier 10.
1 is input. Light generated from this recording light source 11 is irradiated onto a copy film 13 through a lens 12.
This copy film 13 is mounted on a printing drum 14, and since the printing drum 14 rotates and moves, an X-ray image subjected to frequency processing is reproduced and recorded on the copy film 13. When reproducing and recording on the copy film 13, a reduced copy photograph can be obtained by recording at a higher sampling frequency than during input scanning. For example, in the input system, 10
Pixels/mm, if you scan at 20 pixels/mm in the output system, it will be 1/
This is a copy photo reduced to 2. When the photocopy is reduced to 1/2 to 1/3 in this way, the frequency components considered necessary for diagnosis are closer to the frequency range with the highest visibility, so the contrast appears to be visually higher, which is very noticeable. It becomes easier to see. The present invention is not limited to the embodiments described above, and various configuration changes are possible. For example, the original photograph can be read not by a rotating drum but by other optical two-dimensional plane scanning or electronic scanning such as a flying spot scanner.
Further, calculation of the unsharp mask can be performed by digitally processing only the sub-scanning direction by unsharpening the analog signal using a low-pass filter in the main scanning direction before A/D conversion. Furthermore, the above calculation may be performed offline after all data is stored on the magnetic tape, or may be partially stored in a core memory and sequentially processed online. In the embodiment described above, the reproduced image is recorded on a copy film, but photosensitive materials for copying include silver salt photographic film, diazo film,
Electrophotographic materials etc. can also be used. Alternatively, instead of recording on a photosensitive material, it may be displayed on a CRT for observation. Furthermore, this may be recorded optically on the copy material. In the embodiment described above, the original photographic density Dorg is converted into an electrical signal by the photoelectric converter 4 and is further nonlinearly amplified by the amplifier 5, but this is used for band compression and nonlinear correction. This is because a signal that has undergone nonlinear amplification such as logarithmic amplification is more suitable for subsequent signal processing, and in practice, such a signal after nonlinear amplification is often used, but in principle Dorg directly converts the signal into an electrical signal by the photoelectric converter 4.
Needless to say, it is also possible to perform subsequent processing as follows. In addition, the calculation of this unsharp mask should theoretically calculate the average energy, but according to the inventor's experiments, this unsharp mask concentration Dus
When calculating , the result did not change even if the average value was calculated using signal values corresponding to concentrations that were nonlinearly amplified by logarithmic compression. This is practically advantageous in terms of processing. Example A total of 200 cases of the parts shown in Table 1 were compared with original photographs recorded directly on conventional X-ray photographic film and copy photographs created by ultra-low frequency processing according to the present invention. We investigated the improvement of diagnostic performance for the main parts of the body.

[Table] For the reproduced image, the emphasis coefficient β is fixed to 3, and the modulation transfer function in the side direction of the rectangle is calculated using a simple average of image signals in a rectangular area as a non-sharp mask.
It was created by changing the spatial frequency _c , which is 0.5, in six ways. Regarding the existence and degree of improvement in diagnostic performance, the physical evaluation values of ordinary photography (for example,
Because it is virtually impossible to substantiate the
Based on subjective evaluations by 20 radiology interpretation experts. The evaluation criteria were as follows. +2: It is difficult to diagnose with the original photo, but
The lesion area became much easier to see in the photocopy, and diagnostic performance was clearly improved. +1: Compared to the original photo, it became easier to see and the diagnostic performance improved. 0: It is easier to see than the original photograph, but no particular improvement in diagnostic performance is observed. -1: There were areas where diagnostic performance improved, but there were also areas where diagnosis was difficult. -2: There was no area where the diagnostic performance was improved, and there were areas where it was difficult to diagnose. Figure 15 shows the total areas and cases listed in Table 1.
This shows the results of averaging the expert evaluation values for 200 cases. The curve shown in FIG. 12 is obtained by further averaging these averaged evaluation results. From FIG. 15, it was confirmed that the range of spatial frequency _c in which the diagnostic performance is particularly improved is in the range of 0.02 to 0.15 cycles/mm. In addition, through this experiment, the value of _c that gives the best evaluation and the evaluation value at that time, in other words, the peak position, are determined based on the evaluator's preference, the area to be imaged, the case, and the purpose of the image (screening or detailed examination, etc.) Although it varies considerably depending on the presence or absence of other clinical test findings, it has been found that the range of _c in which the effects of the treatment according to the present invention are recognized has relatively little variation for all X-ray photographs. In addition, for a total of 20 representative cases shown in Table 1, _c was fixed at 0.05 cycles/mm, while B/
Copy photographs were created using the same method with various changes in A, and similar evaluations were conducted by a total of 20 radiology interpretation experts. FIG. 16 shows the average value of the evaluation for each case. As is clear from Fig. 16, when β is fixed (indicated by the solid line), the diagnostic performance is improved in the range of B/A from 1.5 to 6, and especially in the range of 2 to 5.5. Remarkably, when β was made variable (indicated by the broken line), it was found that the diagnostic performance improved in the range of B/A of 1.5 to 10, and was particularly remarkable in the range of 2 to 8. The improvement of diagnostic performance by combining ultra-low frequency enhancement with other processing (change of emphasis coefficient β, gradation processing, reduction, smoothing processing) was carried out on the various cases mentioned above, and in both cases the diagnosis The results show that the performance is further improved. Since the present invention having the above configuration emphasizes the frequency response from the ultra-low frequency region,
Frequency regions important for diagnosis are greatly emphasized. Therefore, contrast is improved and diagnostic performance is improved. Furthermore, by changing the degree of enhancement depending on the density, shape, etc., it is possible to prevent the generation of false images and to prevent diseases important for diagnosis from becoming difficult to see. Furthermore, since high-frequency components are not emphasized, noise components are reduced, resulting in a smoother image. As a result, an easily visible photographic image can be obtained. All of these image processing techniques are ultimately more effective when they are considered to approach the optimal frequency of the modulation transfer function for human vision, and to achieve this, appropriate image reduction is necessary. is particularly effective.

[Brief explanation of the drawing]

FIG. 1 is a flow sheet showing the process of processing an X-ray image according to the present invention. FIG. 2 is a diagram showing an unsharp mask, pixels, etc. on an image to explain one algorithm for calculating the unsharp mask density.
FIG. 3 is a diagram showing the memory capacity when the above algorithm is used. FIG. 4 is a block diagram showing an example of the configuration of a circuit that performs calculations using the above algorithm. FIGS. 5 and 6 are diagrams showing examples of changes in the memory configuration in the above configuration. FIGS. 7 and 9 are diagrams showing a non-sharp mask, pixels, etc. on an image in order to explain another algorithm for calculating the non-sharp mask density. FIG. 8 is a diagram showing the memory capacity when this algorithm is used. FIG. 10 is a graph showing the steps of frequency emphasis. FIG. 11 is a diagram showing an example of changing the emphasis coefficient β according to the density. FIG. 12 is a graph showing an example of a combination of emphasis coefficient β and original photographic density Dorg. FIG. 13 and FIG. 14 are diagrams showing an example of a specific method for changing the emphasis coefficient β depending on the density. FIG. 15 is a graph showing the results of diagnostic performance evaluation in the example. FIG. 16 shows the ratio B/A between the maximum modulation transfer function B in the emphasized copy photograph and the modulation transfer function A near the zero spatial frequency;
It is a graph showing the relationship with evaluation of diagnostic performance.

Claims (1)

[Claims] 1. After scanning an original X-ray photograph and reading out the X-ray image information recorded therein and converting it into an electrical signal, when reproducing it as a copy photograph etc., Find the density Dus of the unsharp mask corresponding to ultra-low spatial frequencies, and calculate the density of the original photo.
Dorg, the emphasis coefficient is β, and the density reproduced in photocopies, etc. is D′, then the following calculation is performed: D′=Dorg+β(Dorg−Dus) X In the line image processing method, the unsharp mask density Dus is calculated for each scan within a rectangular unsharp mask surrounded by two sides parallel to the main scanning direction of the scan and two sides parallel to the sub-scanning direction. An X-ray image processing method characterized by calculating the density Dorg of an original image at a point by simple averaging. 2 The rectangular non-sharp mask for obtaining the non-sharp mask density Dus has a rectangular side length of 60 mm ~
The X-ray image processing method according to claim 1, characterized in that the size is in the range of 1.2 mm. 3. The X-ray image processing method according to claim 2, wherein the rectangular non-sharp mask has a size such that the length of one side of the rectangle is in the range of 30 mm to 4 mm. 4. The X-ray image processing method according to claim 1, wherein the enhancement coefficient β is a constant. 5. X according to claim 4, characterized in that the maximum modulation transfer function of the copy photograph emphasized by the arithmetic expression is 1.5 to 6 times the modulation transfer function near the zero spatial frequency. Line image processing method. 6. The X-ray image processing method according to claim 1, characterized in that the enhancement coefficient β is changed depending on the density of the original photograph or the density of the non-sharp mask. 7. The method according to claim 6, characterized in that the maximum modulation transfer function of the copy photograph emphasized by the arithmetic expression is 1.5 to 10 times the modulation transfer function near the zero spatial frequency. X-ray image processing method. 8 In addition to emphasizing ultra-low spatial frequency components, when the modulation transfer function is at a spatial frequency of 0.5 cycles/mm,
According to any one of claims 1 to 7, the smoothing process is performed such that the smoothing process is 0.5 or more and 0.5 or less at a spatial frequency of 5 cycles/mm. X-ray image processing method. 9 A photodetector that scans the original photograph, reads out the X-ray image stored therein, and converts it into an electrical signal; and the output of this photodetector is connected to two sides parallel to the main scanning direction of the scanning; A circuit for calculating the unsharp mask density Dus corresponding to an extremely low spatial frequency by simple averaging over the range of a rectangular unsharp mask surrounded by two sides parallel to the sub-scanning direction, and this unsharp mask. From the density Dus, the density Dorg of the original photograph which is the output of the photodetector, and the enhancement coefficient β, when the density of the reproduced image is D′, it is expressed by the formula D′=Dorg+β(Dorg−Dus). An X-ray image processing device that includes a calculation device that performs calculations. 10 The calculation device calculates the original photographic density.
10. The X-ray image processing apparatus according to claim 9, further comprising an emphasis coefficient variable means for increasing or decreasing the emphasis coefficient β in accordance with the magnitude of Dorg or the unsharp mask density Dus.