JPH04317263A - Irradiation field recognition pre-processing method utilizing neural network - Google Patents

Abstract

PURPOSE:To prevent the recognition accuracy of an irradiation field or the filtering accuracy of an irradiation field recognition pre-processing as well as learning speed from dropping due to a neural network even when an image in which the area outside the irradiation field is formed only in a part of the circumference of the irradiation field is read. CONSTITUTION:An image space processing is conducted for image data SQ corresponding to an image 40 consisting of 256 picture elements by 256 picture elements read from a recording sheet, and it is converted into image data S'Q corresponding to the image consisting of 300 by 300 picture elements which is obtained by adding a frame area 41 having a density below that of the area outside the irradiation field around the image 40. Then the obtained image data S'Q are inputted to the neural network, an irradiation field recognition processing is conducted for it, and outputted as code data binarized so that an area outside the irradiation field 42 can be expressed by '1' while an area 43 other than the area 42 can be expressed by '0'.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention reads image data from a recording sheet on which a radiation image of a subject is recorded through an irradiation field aperture, and uses a neural network to recognize the radiation irradiation field on the sheet based on the read image data. Alternatively, it relates to a preprocessing method for irradiation field recognition that emphasizes edges.

0002

2. Description of the Related Art It is practiced in various fields to read a recorded radiation image to obtain image data, perform appropriate image processing on the image data, and then reproduce and record the image. For example, an X-ray image is recorded using an X-ray film with a low gamma value designed to be compatible with later image processing, and the X-ray image is read from the film on which it is recorded and converted into an electrical signal. By performing image processing on this electrical signal (image data) and then reproducing it as a visible image in a photocopy, etc., contrast, sharpness,
Systems that can obtain reproduced images with good image quality performance such as graininess have been developed (for example,
(See Publication No. 93).

[0003] Also, the applicant of the present application has reported that radiation (X-rays, α
When irradiated with rays, β-rays, γ-rays, electron beams, ultraviolet rays, etc., a part of this radiation energy is accumulated, and then when excited light such as visible light is irradiated, the accumulation exhibits stimulated luminescence depending on the accumulated energy. Using stimulable phosphor (photostimulable phosphor),
A radiation image of an object such as a human body is recorded on a sheet of stimulable phosphor, and this stimulable phosphor sheet is scanned with excitation light such as a laser beam to generate stimulated luminescence. A radiation image recording and reproducing system has already been proposed in which image data is obtained by photoelectrically reading the exhaustion light, and based on this image data, a radiation image of the subject is output as a visible image on a recording material such as a photographic light-sensitive material, a CRT, etc. (Japanese Unexamined Patent Publication Nos. 55-12429, 56-11395, 5
No. 5-163472, No. 56-104645, No. 5
5-116340 etc.).

This system has the practical advantage of being able to record images over a much wider range of radiation exposure compared to conventional radiographic systems using silver halide photography. In other words, in stimulable phosphors, it is recognized that the amount of emitted light that is stimulated and emitted by excitation after accumulation is proportional to the amount of radiation exposure over an extremely wide range, and therefore the amount of radiation exposure varies depending on various imaging conditions. Even if the value varies considerably, the amount of stimulated luminescence light emitted from the stimulable phosphor sheet is read by the photoelectric conversion means by setting the reading gain to an appropriate value and converted into an electrical signal. Recording materials such as photographic materials, C
By outputting a radiation image as a visible image on a display device such as an RT, it is possible to obtain a radiation image that is not affected by fluctuations in radiation exposure amount.

[0005] In the above system, the stimulable phosphor sheet is preliminarily exposed to a low-level light beam before reading it under optimal reading conditions depending on the dose of radiation applied to the stimulable phosphor sheet to obtain image data. Pre-reading is performed to read the outline of the radiation image recorded on this sheet by scanning, the pre-read image data obtained by this pre-reading is analyzed, and then a light beam of a higher level than the light beam used in the pre-reading is applied to the sheet. There is also a system configured to perform main reading to obtain image data by irradiating and scanning a light beam and reading the radiation image under optimal reading conditions (Japanese Patent Application Laid-open Nos. 58-67240 and 58-67241). , 58-
No. 67242, etc.).

[0006] Here, the reading condition is a general term for various conditions that affect the relationship between the amount of stimulated luminescence light and the output of the reading device during reading, such as the reading gain that determines the input/output relationship, It means the scale factor or the power of excitation light in reading.

[0007] The high level/low level of the light beam refers to the intensity of the light beam irradiated per unit area of the sheet, or the intensity of the stimulated luminescence light emitted from the sheet, respectively. If it depends on the wavelength of the light beam (has a wavelength sensitivity distribution), the weighted intensity after weighting the intensity of the light beam irradiated per unit area of the sheet by the wavelength sensitivity. Yes, there are ways to change the level of the light beam, such as using light beams of different wavelengths, changing the intensity of the light beam itself emitted from a laser light source, etc., and inserting or removing an ND filter, etc. on the optical path of the light beam. Various known methods can be used, such as a method of changing the intensity of the light beam, a method of changing the scanning density by changing the beam diameter of the light beam, and a method of changing the scanning speed.

[0008] In addition, regardless of whether the system performs this pre-reading or the system does not perform pre-reading, the obtained image data (including the pre-read image data) is analyzed to determine the optimal image when performing image processing on the image data. Some systems are designed to determine processing conditions. The method for determining the optimal image processing conditions based on this image data is not limited to systems using stimulable phosphor sheets, but for example, image data is obtained from a radiation image recorded on a recording sheet such as a conventional X-ray film. It is also applied to the system.

[0009] The above image data (including pre-read image data)
Various methods have been proposed to find the optimal reading conditions and image processing conditions by analyzing the image data, and one known method is to create a histogram of the image data (
For example, Japanese Patent Application Laid-open No. 156055/1983).

[0010] By obtaining a histogram of image data, it is possible to know, for example, the maximum value and minimum value of the image data, the value of the image data at the point where the frequency is maximum, etc., and from these values it is possible to determine the stimulable phosphor. It is possible to grasp the characteristics of a radiation image recorded on a recording sheet such as a sheet or an X-ray film. Therefore, by determining optimal reading conditions and image processing conditions based on this histogram, it becomes possible to reproduce and output a radiation image that is highly suitable for observation.

On the other hand, when photographing and recording radiation images on a recording sheet, it is necessary to avoid irradiating radiation to parts of the subject that are not necessary for observation, or to prevent radiation from being irradiated to parts unnecessary for observation. Scattered radiation enters the necessary areas, reducing image quality. Therefore, an irradiation field diaphragm is used to limit the radiation irradiation area so that the radiation is applied only to the necessary areas of the subject and part of the recording sheet. I often take pictures.

However, when analyzing image data and determining reading conditions and image processing conditions as described above,
If the image data used for analysis is image data obtained from a recording sheet taken using an irradiation field aperture, even if the image data is analyzed while ignoring the existence of this irradiation field, the recorded radiation image will still be lost. There are cases in which radiation images that are suitable for observation cannot be reproduced and recorded because the information is not correctly understood and incorrect reading conditions and image processing conditions are determined.

In order to solve this problem, before determining the reading conditions and image processing conditions, it is necessary to recognize the irradiation field and determine the reading conditions and image processing conditions based on the image data within the irradiation field.

[0014] There are various methods for recognizing the radiation field, but among them, the following is a versatile method that can accurately recognize the radiation field even if the radiation field has an irregular shape. For example, a contour point that is considered to be on the contour of the irradiation field based on image data corresponding to each pixel along a plurality of radial line segments connecting a predetermined point included in the irradiation field and the edge of the sheet. The applicant has already proposed a method that can be called circular irradiation field detection, in which the area surrounded by lines along these contour points is recognized as the irradiation field (Japanese Patent Application Laid-Open No. 1983-1993). No. 259538)
.

[0015] Various similar improved irradiation field recognition methods have also been proposed. (Unexamined Japanese Patent Publication No. 2-42436
This is image data that corresponds to each pixel on each of a plurality of radial line segments connecting a predetermined point included in the irradiation field on the recording sheet and the edge of the recording sheet. Based on the above, for each of these line segments, contour candidate points that are considered to be intersection points between the contour of the irradiation field on one or more recording sheets and the above line segment are determined along with the candidate ranking, and the contour candidate points on each line segment are determined. exists both within a predetermined peripheral range close to the edge of the recording sheet and within a central range inside this peripheral range, and when a contour candidate point of the first rank exists within the peripheral range, it is within the central range. The candidate ranking of the contour candidate point with the highest ranking among the contour candidate points in This allows the irradiation field to be correctly recognized even if a foreign object other than the object is photographed together with the object.

[0016] Furthermore, an irradiation field recognition method suitable for detecting a rectangular irradiation field has already been proposed by the present applicant (Japanese Patent Laid-Open No. 39039/1983).

[0017] When the irradiation field is rectangular, the X and Y axes are selected along two adjacent sides of the rectangle, and the image information read from the entire recording area where the irradiation field aperture photography has been performed is used to Obtain image data at each position on the recording area, differentiate this image data in the X-axis direction, and calculate the differential value as Y.
Project it in the axial direction to determine the position on the X-axis of the contour extending in the Y-axis direction among the rectangular irradiation field contours, similarly differentiate the above image data in the Y-axis direction, and calculate the differential value in the X-axis direction. This method determines the position on the Y-axis of the rectangular irradiation field contour extending in the X-axis direction, and recognizes the area surrounded by these contours extending in the X and Y-axis directions as the irradiation field. .

Furthermore, the present applicant has already proposed an irradiation field recognition method that can properly recognize the irradiation field even when the outline of the rectangular irradiation field is blurred and has a certain width (Japanese Patent Application Laid-Open No. No. 64-11465).

This is a method for recognizing the irradiation field when radiation image information is recorded by applying a rectangular irradiation field aperture on the recording medium. Set the X-axis and Y-axis along two adjacent sides of the outline, and
Differential processing is performed in the axis (and Y-axis) direction, and the position where the absolute value of the differential value obtained by the processing is greater than or equal to a predetermined threshold value is
Extract the contour candidate points that are responsible for the irradiation field contour in the Y-axis (and X-axis) direction of the book, and then
From the projection results, calculate the spread in the X-axis (and Y-axis) direction of the contour candidate points responsible for each of the two Y-axis (and X-axis) direction irradiation field contours, and Y-axis (and X-axis) corresponding to
A feature is that the position on the X-axis (and Y-axis) of the directional irradiation field contour is determined, and the area surrounded by the X-axis and Y-axis direction irradiation field contours in which each position is determined is recognized as the irradiation field. This is an irradiation field recognition method.

[0020] The present applicant has also proposed many other irradiation field recognition methods. (Unexamined Japanese Patent Publication No. 62-15536, 62-15536,
2-15537, 62-15539, etc.) By using such an irradiation field recognition method, the irradiation field is first determined, and then the image data corresponding to the determined irradiation field is analyzed to determine appropriate reading conditions and images. Processing conditions are required.

On the other hand, in any of the many radiation field recognition methods described above, if the outline of the radiation field is blurred, causing so-called edge blur, it may be difficult to accurately recognize the radiation field.

Furthermore, even if the irradiation field is narrowed down using a mask, the scattered radiation rays wrap around the mask side and irradiate the outside of the irradiation area of the recording sheet, causing so-called density fog in the outside irradiation area of the recording sheet. Even when such a phenomenon occurs, it may be difficult to accurately recognize the irradiation field using the irradiation field recognition method described above.

In such a case, it is desirable to perform image filtering on the radiation image signal so as to increase the density difference between the inside and outside of the edge boundary of the radiation image field as a preprocessing for radiation field recognition of the radiation image.

The present applicant has already proposed a method of applying such image filtering to radiographic image signals (Japanese Patent Application No. 76534/1999).

By the way, in recent years, the concept of neural networks has emerged and is being applied to various fields.

[0026] This neural network can control each part inside the neural network by inputting information (teacher signal) indicating whether the output signal output when a certain input signal is given is a correct signal or an incorrect signal. It is equipped with an error back propagation learning function that corrects the weights of connections between units (weights of synaptic connections), and by repeatedly 'learning', when a new signal is input. This can increase the probability of outputting the correct answer. (for example,"
D. E. Rumelhart, G. E. Hinton
and R. J. Williams: Learning
representations by back-
propagating errors, Nature
, 323-9, 533-356, 1986a", "Hideki Aso: Backpropagation Computrol N
o. 24 53-60'', ``Neural Computer by Kazuyuki Aihara, Tokyo Denki University Press'').

[0027] Using this neural network,
It is possible to perform the above-mentioned irradiation field recognition or image filtering using the image data of the radiation image as input, and after making corrections based on the processing results, the above-mentioned reading conditions and/or image processing conditions are set. be able to.

That is, in the various types of recognition and filtering described above, the image data of the radiation image is input to the neural network, the feature values representing the processing results are output, and this neural network is repeatedly 'learned' in advance. It is possible to gradually obtain correct feature quantities.

If such a neural network is used to perform irradiation field recognition and image enhancement filtering of radiation images, the above recognition and filtering can be performed simply by inputting image data corresponding to each pixel of the radiation image to the neural network. Since it is possible to obtain feature quantities representing the processing results, various types of highly accurate processing can be performed with a simple configuration.

[0030]

[Problems to be Solved by the Invention] By the way, in the above-mentioned neural network, an image in which outside irradiation areas are formed on four sides of the irradiation field area, that is, an image in which all ends of the irradiation field area are surrounded by outside irradiation areas. It is set to perform the above-mentioned irradiation field recognition or image filtering using an algorithm that assumes an image.

[0031] However, there are cases in which an image is obtained in which the out-of-irradiation area is formed only in a part of the periphery of the irradiation field area, and when such an image is input to the above neural network, the neural network Since an image with characteristics different from those of the input image currently being used will be input, there is a risk that the recognition accuracy or filtering accuracy of the neural network will decrease.

[0032] Furthermore, there is a possibility that the learning speed of the neural network will decrease.

The present invention has been developed in view of the above circumstances, and even when an image in which an outside irradiation area is formed in a part of the periphery of the irradiation field is read, the irradiation field can be recognized by a neural network. It is an object of the present invention to provide a preprocessing method for irradiation field recognition using a neural network that can prevent a decrease in accuracy or filtering accuracy in irradiation field recognition preprocessing and a decrease in learning speed.

[0034]

[Means for Solving the Problems] The first of the preprocessing methods for irradiation field recognition using a neural network of the present invention
In this method, frame image data is added to the image data of the above image so that frame areas having a density equal to or lower than that of the outside irradiation area are formed on four sides of the image, and then these image data are subjected to neural processing. The feature is that the irradiation field is recognized by inputting it into the network.

That is, in a preprocessing method for irradiation field recognition in which the irradiation field of the radiation image is recognized by a neural network based on image data read from a recording sheet on which a radiation image of the subject is recorded through an irradiation field aperture, ◆ Frame areas having a density equal to or lower than that of the outside irradiation area are added to four sides of the image formed by the image data read from the recording sheet,
The present invention is characterized in that frame image data is added to the image data, and then these image data are sent to the neural network.

Furthermore, the second method of the preprocessing method for irradiation field recognition using the neural network of the present invention is as follows:
Frame image data is added to the image data of the above image so that frame areas having a density equal to or lower than that of the outside irradiation area are formed on four sides of the image, and then these image data are input into a neural network. This method is characterized by emphasizing the edges of the irradiation field to make it easier to recognize the irradiation field.

That is, based on the image data read from the recording sheet on which the radiation image of the subject is recorded through the irradiation field aperture, image filtering is performed to emphasize the irradiation field of the radiation image using a neural network, and then the irradiation field is recognized. In the preprocessing method for irradiation field recognition, ◆ the above-mentioned method is applied so that frame areas having a density equal to or lower than that of the outside irradiation area are added to four sides of the image formed by the image data read from the recording sheet. This method is characterized in that frame image data is added to image data, and then these image data are sent to the neural network.

Here, the density equivalent to the outside irradiation area means a density that is approximately the same as the minimum density in the outside irradiation area.

[0039]

[Operations and Effects] According to the preprocessing method for irradiation field recognition using a neural network of the present invention, by adding predetermined frame image data to image data, four
In this case, image data corresponding to the outside irradiation area or a frame area with a density lower than the density equivalent to the outside irradiation area is created, and then these image data are input to the neural network. .

Therefore, even if the image data read from the recording sheet corresponds to an image having an outside irradiation area only in a part of the periphery of the irradiation field area, this image data is input into the neural network. The image data corresponds to an image in which an outside irradiation area or a frame area is formed around the irradiation field area of the image.

[0041] The neural network performs irradiation field recognition or preprocessing for this irradiation field recognition based on such image data, that is, image data in which a region that can be distinguished from the irradiation field region in terms of density exists in the peripheral part of the image. Since it is possible to perform irradiation field edge enhancement as a irradiation field, learning can be performed using a normal algorithm in the same way as with image data corresponding to an image in which the irradiation field area is surrounded by an outside irradiation area.

[0042] This makes it possible to prevent a decrease in the recognition accuracy or filtering accuracy of the irradiation field by the neural network with respect to an image in which an outside irradiation area exists only in a part of the periphery of the irradiation field area, and also to prevent a decrease in the learning speed.

[0043]

Embodiments Hereinafter, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a perspective view showing an example of a radiation image reading and reproducing apparatus using the first method of the irradiation field recognition preprocessing method of the present invention. This radiation image reading and reproducing device uses the above-mentioned stimulable phosphor sheet.

The stimulable phosphor sheet 1 on which the radiation image has been recorded is set at a predetermined position in the reading means 100. The stimulable phosphor sheet 1 set in this predetermined position is
The sheet is transported (sub-scanning) in the direction of arrow Y by a sheet transporting means 15 such as an endless belt driven by a driving means (not shown). On the other hand, a light beam 17 emitted from a laser light source 16 is reflected and deflected by a rotating polygon mirror 18 that is driven by a motor 24 and rotates at high speed in the direction of the arrow.
After passing through a converging lens 19 such as a lens, the optical path is changed by a mirror 20 and enters the sheet 1 in the sub-scanning direction (
Main scanning is performed in the direction of the arrow X, which is substantially perpendicular to the direction of the arrow Y). From the part of the sheet 1 irradiated with the light beam 17, stimulated luminescence light 2 is emitted in an amount corresponding to the radiographic image information stored and recorded.
1 is emitted, and this stimulated luminescence light 21 is guided by a light guide 22 and passes through a photomultiplier (photomultiplier tube) 2.
photoelectrically detected by 3. The light guide 22 is made by molding a light-guiding material such as an acrylic plate, and the linear entrance end surface 22a is connected to the stimulable phosphor sheet 1.
A photomultiplier 23 is arranged on an annular output end surface 22b extending along the upper main scanning line.
The light-receiving surfaces of the two are combined. The stimulated luminescent light 21 that has entered the light guide 22 from the incident end surface 22a is transmitted through the light guide 22.
The photomultiplier 23 converts the amount of stimulated luminescence light 21 representing a radiation image into an electrical signal.

The analog output signal S output from the photomultiplier 23 is logarithmically amplified by the log amplifier 26 and digitized by the A/D converter 27, so that image data SQ corresponding to each pixel of the radiation image is obtained. can get.

The obtained image data SQ is input to the image space enlarging means 28 and undergoes image space enlarging processing to form a frame area around the image. This image space expansion process is a process in which a frame area of a predetermined density is created around an input image using software using a computer system, and
The image data S'Q whose image space has been expanded by the image space expansion means 28 is then input to the neural network 29.

The neural network 29 recognizes a highly reliable irradiation field based on the input image data S'Q.

When the irradiation field is recognized in this way, the image data S'Q corresponding to this irradiation field is sent to the image processing means 30, where appropriate image processing is performed.

The image data S'Q subjected to image processing is sent to the reproducing means 31, and the reproducing means 31 reproduces and records a radiation image based on this image data S'Q.

Now, the image space enlarging process performed on the image data S'Q by the image space enlarging means 28 will be further explained with reference to FIG.

The image data SQ outputted from the A/D converter 27 has, for example, 256 pixels×as shown in FIG.
The image space enlarging means 28 has data corresponding to an image 40 of 256 pixels.
is input.

In the image space enlarging means 28, frame areas 41 are added to the four sides of this image 40 so that the frames in all directions have approximately the same width, and the enlarged image is made up of 300 pixels x 300 pixels. The image data SQ is converted to image data S'Q so that the size of the image data SQ is the same as that of the image data SQ.

The density of the pixels in this frame area is set to the minimum density of the input image 40 (density equivalent to the outside irradiation area) or the density corresponding to the light emission amount of 0 of the stimulable phosphor sheet 1 described above.

Thereafter, the image data S'Q is input to the neural network 29 and subjected to irradiation field recognition processing.

In this neural network 29, the same algorithm as for the image data of the image whose irradiation field area is surrounded by the outside irradiation area is used for the image data S'Q that has been subjected to image space expansion processing. learning for irradiation field recognition, and outputs the feature amounts that are the recognition results to the image processing means 30.

The image whose irradiation field has been recognized by the neural network 29 is outputted with the irradiation field area 42 marked with "1" and the area 43 other than the irradiation field area marked with "0".

Note that, as shown in FIG. 3, of the image signals corresponding to each pixel, only the image signal corresponding to each diagonally shaded pixel may be extracted and input to the neural network 29. This allows the neural network 29
The number of input points can be reduced. At this time, instead of uniformly thinning out the entire image, considering that the main part of the image often exists near the center of the image, it may be thinned out so that the center of the image is dense and the edges are coarse. good.

[0059] Thereafter, learning is repeated using the neural network 29, and the neural network 2
9 describes in detail the method for determining correct image points.

FIG. 4 is a diagram showing an example of a neural network equipped with an error backpropagation learning (backpropagation) function. Error backpropagation learning (backpropagation) is, as mentioned above, by comparing the output of the neural network with the correct answer (teacher signal) to sequentially calculate connection weights (synaptic connection weights) from the output side to the input side. The purpose is to correct the .

As the teacher signal in this case, a radiation image signal obtained in advance from an image without edge blurring of the irradiation field and without irradiation field scattering, for example, an image signal in which irradiation field recognition has been accurately performed, is used.

As shown in the figure, the first layer (input layer), second layer (middle layer), and third layer (
The output layer consists of n1 units, n2 units, and 1 unit, respectively. Each signal F1, F2, ..., Fn1 input to the first layer (input layer) is an image signal SQ corresponding to each pixel of the number thinned out as shown in FIG. 3 among each pixel of the X-ray image. The output Y3,1 from the third layer (output layer) is a signal representing an irradiation field (one of circular, rectangular, etc. irradiation fields).

Let the i-th unit of the k-th layer be Uk,i,
Let each input to the unit Uk,i be Xk,i, each output be Yk,i, the weight of the connection from Uk,i to Uk+1,j be Wk,i;k+1,j, and each unit Uk
, j are the same characteristic functions

[0064]

[Math 1]

Assume that it has the following. At this time, the input Xk,j and output Yk,j of each unit Uk,j are

[
0066

[Math 2]

[0067]

[Math 3]

[0068] However, each input F1, F2,..., Fn1 to each unit U1, i (i = 1, 2,..., n1) constituting the input layer is not weighted and is applied to each unit U1, i (i = 1, 2,...,n1). n1 input signals F1, F2
,...,Fn1 is the weight of each connection Wk,i;k+1,j
It is transmitted to the final output Y3,1 while being weighted by , whereby a binary pattern of the irradiation field is determined.

Here, the weight Wk,i;k+ of each of the above-mentioned connections
The method for determining 1,j will be explained. First, the initial value of the weight Wk,i;k+1,j of each connection is given by a random number. At this time, it is preferable to limit the range of the random number so that even if the inputs F1 to Fn1 fluctuate to a maximum, the output Y3,1 becomes a value within a predetermined range or a value close to this.

The pre-read image signal SQ is thinned out from the stimulable phosphor sheet on which a large number of X-ray images with known irradiation fields are recorded, and the above n1 inputs F1, F2,..., Fn1 are obtained.
is required. These n1 inputs F1, F2,...
, Fn1 are input to the neural network shown in FIG. 5, and the output Yk,i of each unit Uk,i is monitored.

Once each output Yk,i is obtained, the square error between the final output Y3,1 and the teacher signal d indicating the correct image point for this image is calculated.

[0072]

[Math 4]

[0073] is obtained. In order to minimize this squared error E, the weight Wk,i;k of each connection is set as follows.
+1,j is corrected.

To minimize the squared error E, this E should be
Since it is a function of k,i;k+1,j

[0075]

[Math 5]

In this way, the weight of each connection Wk,i;k+1
,j are modified. Here, η is a coefficient called a learning coefficient.

[0077] Here,

[0078]

[Math 6]

[0079], and from formula (2)

[0080]

[Math 7]

[0081] Therefore, formula (6) is

[0082]

[Math. 8]

[0083]

Here, from equation (4),

[0085]

[Math. 9]

[0086] If we transform equation (9) using equation (3), we get

[0087]

[Math. 10]

[0088] Here, from equation (1),

[0089]

[Math. 11]

[0090] Therefore,

[0091]

[Math. 12]

[0092]

In equation (8), set k=2, and (10
) and substituting equation (11) into equation (8), we get

[0094]

[Math. 13]

Substituting this equation (13) into equation (5),

[0096]

[Math. 14]

[0097] According to this equation (14), W2,
The weight of each connection of i; 3, 1 (i=1, 2, . . . , n1) is modified.

Next,

0099

[Math. 15]

Therefore, by substituting equations (2) and (3) into equation (15), we get

[0101]

[Math. 16]

Here, from equation (11),

[0103]

[Math. 17]

Therefore, this equation (17) and (10)
Substituting equation (12) into equation (16),

[0105]

[Math. 18]

In equation (8), set k=1, and (18
) into equation (8), we get

[0107]

[Math. 19]

Substituting this equation (19) into equation (5), setting k=1, we get

[0109]

[Math. 20]

Then, W2,i modified by equation (14)
;3,1(i=1,2,...,n1) is substituted into this equation (20), and W1,i;2,j(i=1,2,...,n1
;j=1,2,...,n2) are corrected.

[0111] Theoretically, by using equations (14) and (20) and setting the learning coefficient η sufficiently small and increasing the number of learnings sufficiently, the weight of each connection Wk,i;k+1,
Although it is possible to converge j to a predetermined value, it is not realistic to make the learning coefficient η too small because it slows down the progress of learning. On the other hand, if the learning coefficient η is set to a large value, the learning may oscillate (the weights of the connections described above may not converge to a predetermined value). Therefore, in practice, an inertia term as shown in the following equation is added to the connection weight correction amount to suppress vibration, and the learning coefficient η is set to a somewhat large value. (For example, D.E. Rumel
hart, G. E. Hinton and R. J. W
illiams: Learning internal
representations by error
propagation In Parallel D
distributed processing, Vol.
ume 1, J. L. McClelland, D. E.
Rumelhart and The PDP Res
search Group, MIT Press, 198
6b”)

[0112]

[Math. 21]

[0113] However, ΔWk,i;k+1,j (t) is the corrected connection weight Wk,i in the t-th learning;
k+1,j to the weight of the connection before modification Wk,i;k+
It represents the amount of correction after subtracting 1,j. Further, α is a coefficient called an inertia term.

For example, α=
0.9 and η=0.25, the weight of each connection Wk,i
;k+1,j is modified (learning) for example 200,000 times, and thereafter the weight Wk,i;k+1,j of each connection is fixed to the final value. At the end of this learning, output Y3
, 1 becomes a signal that approximately correctly represents the irradiation field.

[0115] The above signal representing the irradiation field is as follows:
For example, if the signal only indicates whether it is a circular irradiation field or a rectangular irradiation field, one of the two outputs is set to 1 or 0,
If it is 1, it is a circle, and if it is 0, it is a rectangle, and recognition by a neural network can be made very easily.

[0116] Also, regarding the photographed parts, if some parts predicted in advance are represented by numbers, recognition by a neural network can be easily made possible.

After the learning is completed, an image signal SQ representing an X-ray image with an unknown irradiation field is obtained, and this image signal SQ is input to the neural network shown in FIG. 4, and the output Y3, 1 is a signal representing the irradiation field binary pattern of the X-ray image. This signal is
Since it is obtained after learning as described above, the irradiation field can be represented with high accuracy.

[0118] The neural network described above is not limited to a three-layer structure, and it goes without saying that it may have even more layers. Furthermore, it goes without saying that each layer can be configured with any number of units depending on the number of pixels of the input image signal SQ, the accuracy of the required reading conditions, etc.

The voltage applied to the photomultiplier 21' of the reading means 100', the amplification factor of the amplifier 26', etc. are controlled in accordance with the signal correctly representing the irradiation field obtained by the neural network as described above. The main reading is performed according to controlled conditions.

Next, a radiation image reading and reproducing apparatus using the second method of the irradiation field recognition preprocessing method of the present invention will be described with reference to FIG.

In this apparatus, image space expansion processing is performed on image data SQ in the image space expansion processing means 28 in the same way as the apparatus shown in FIG. 1, and image data S'Q is output.

Thereafter, this image data S'Q is input to the neural network 29' and subjected to irradiation field edge enhancement filtering to facilitate irradiation field recognition.

[0123] That is, this neural network 2
9' is given as input data a group of pixels extracted as a block of 7 pixels x 7 pixels from the image data S'Q, and outputs converted pixel data corresponding to the center pixel of this pixel group as output data. . In other words, when a group of pixels located around a pixel of interest is input to a neural network, based on the density of each pixel included in this pixel group,
One piece of image data corresponding to the pixel of interest is output, the density of which has been converted so that the difference in density becomes large between when the pixel of interest is within the irradiation field and when the pixel of interest is outside the irradiation field.

[0124] In this way, the converted data output by one processing by the neural network is data for one pixel corresponding to the pixel of interest, so in the next processing, the previous processing in the image data S'Q is For example, a block of 7 pixels x 7 pixels, which is shifted by one pixel with respect to the extracted block, is input to the filter means 28 as input data, and in the end, such processing is repeated for the number of pixels. The image filtering process for the original image data S'Q is completed.

[0125] In this case, the teacher signal is a radiation image signal obtained in advance for an image with no edge blurring of the irradiation field and no irradiation field scattering, for example, an image signal in which irradiation field recognition has been accurately performed. Use images with reduced field irradiation concentrations.

[0126] Therefore, at the end of this learning, the output Y3,1 becomes a signal in which the density differs greatly between the inside and outside of the edge boundary of the irradiation field, which allows easy recognition of the irradiation field.

The image data S'Q, which has been subjected to image filtering processing by the filtering means 28 to increase the difference in density between the inside and outside of the irradiation field edge, is then added to the frame area 41 by the image space expansion processing means 28.
is removed, converted into image data SQ corresponding to the original 256 pixel x 256 pixel image, and input to the irradiation field recognition means 32, where the irradiation field is recognized with high precision.

The image data SQ outputted from the irradiation field recognition means 32 is binary image data.

Note that it is also possible to perform this irradiation field recognition using a neural network.

The image data SQ outputted from the irradiation field recognition means 32 is then subjected to image processing in the image processing means 30 in the same manner as in the apparatus shown in FIG.
Image reproduction is performed at the step.

In the above embodiment, the input image data SQ is 256 pixels x 256 pixels, and is expanded to 300 pixels x 300 pixels by image space expansion processing.
Although pixel image data S'Q is output, the number of pixels of these image data is not limited to this and can be changed as appropriate.

Further, the density of the pixels constituting the frame area 41 is not limited to that of the above-mentioned embodiment, but it is possible to select an appropriate value below the density corresponding to the outside irradiation area.

In the above embodiment, the reading means 100 is used for both pre-reading and main reading, but it is also possible to configure the reading means for pre-reading and the reading means for main reading separately.

The present invention can be used not only for devices using stimulable phosphor sheets but also for devices using conventional X-ray films. That is, the preprocessing method for irradiation field recognition of the present invention is a radiation image reading method that reads a radiation image from a recording sheet on which a radiation image of a subject is recorded to obtain image data, and reproduces and outputs the radiation image based on this image data. It can be widely applied to playback devices in general.

[Brief explanation of the drawing]

[Fig. 1] The first of the preprocessing methods for irradiation field recognition of the present invention.
A perspective view showing an example of a radiation image reading and reproducing apparatus for carrying out an embodiment of the method.

[Fig. 2] A schematic diagram for explaining the processing in the image space enlarging means shown in Fig. 1.

[Figure 3] Diagram schematically representing thinned out pixels of a radiographic image

FIG. 4 is a diagram showing an example of a neural network shown in the device shown in FIG. 1.

[Fig. 5] Second of the preprocessing methods for irradiation field recognition of the present invention
A perspective view showing an example of a radiation image reading and reproducing apparatus for carrying out an embodiment of the method.

[Explanation of symbols]

1 Stimulable phosphor sheet 21 Stimulated luminescent light 23 Photo multiplier 26 Log amplifier 27 A/D converter 28 Image space expansion means 29, 29' Neural network 30
Image processing means 31 Reproduction means 32 Irradiation field recognition means 40 Input image 41 Frame area 42 Irradiation field area 43 Area other than the irradiation field area

Claims (2)

[Claims] 1. A preprocessing method for irradiation field recognition in which an irradiation field of the radiation image is recognized by a neural network based on image data read from a recording sheet on which a radiation image of a subject is recorded through an irradiation field aperture, the method comprising: Frame image data is added to the image data so that frame areas having a density equal to or lower than that of the outside irradiation area are added to four sides of the image formed by the image data read from the recording sheet, and then A preprocessing method for irradiation field recognition using a neural network, characterized in that these image data are sent to the neural network. 2. Based on image data read from a recording sheet on which a radiation image of the subject is recorded through an irradiation field aperture, image filtering is performed to emphasize the irradiation field of the radiation image using a neural network, and then the irradiation field is recognized. In the preprocessing method for irradiation field recognition,
Frame image data is added to the image data so that frame areas having a density equal to or lower than that of the outside irradiation area are added to four sides of the image formed by the image data read from the recording sheet, and A preprocessing method for irradiation field recognition using a neural network, characterized in that the image data is then sent to the neural network.