JPH0538334A - Method of recognizing field of view of collimation of radiation - Google Patents

Abstract

PURPOSE:To precisely recognize a field of view of collimation when a radiation image which has been picked up with the use of a collimation field stop is read. CONSTITUTION:When a profile of a look-ahead image signal Sp is inputted to a computer system 222 in a collimation field recognizing circuit 220, a neural network in the computer system 222 outputs edge candidate points. A computing section 223 creates a collimation field pattern from the edge candidate points. Further the neural network detects a threshold value for detecting the edge candidate points, and accordingly, the edge candidate points may also be obtained in accordance with the threshold value.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention irradiates a stimulable phosphor sheet, on which radiation image information is stored and recorded, with excitation light to photoelectrically emit stimulated emission light emitted from the stimulable phosphor sheet. The present invention relates to a method of recognizing a radiation irradiation field on the sheet and a device for carrying out the method when the radiation image information is read by the above-mentioned method.

[0002]

2. Description of the Related Art Radiation (X rays, α rays,
When β-rays, γ-rays, electron beams, ultraviolet rays, etc.) are irradiated, a part of this radiation energy is accumulated in the phosphor, and when this phosphor is irradiated with excitation light such as visible light, It is known that phosphors exhibit stimulated emission, and phosphors exhibiting such properties are called stimulable phosphors (stimulable phosphors).

Using this stimulable phosphor, radiation image information of a human body or the like is once recorded on a sheet of the stimulable phosphor, and the stimulable phosphor sheet is scanned with excitation light such as laser light to stimulate the irradiation. The emitted light is generated, and the obtained stimulated emission light is photoelectrically read to obtain an image signal. Based on this image signal, a radiation image is made visible on a recording material such as a photographic light-sensitive material or a display device such as a CRT. A radiation image information recording / reproducing system for outputting has already been proposed by the applicant. (JP Sho
55-12492, 56-11395, etc. ) In this system, in order to eliminate the influence of fluctuations in imaging conditions or to obtain a radiation image with excellent observation and interpretation suitability, the recording state of the radiation image information accumulated and recorded on the stimulable phosphor sheet, or the chest, A recording pattern (hereinafter referred to as “stored record information” when collectively referred to as “collective record information”) that is determined by the region of the subject such as the abdomen, the imaging method such as simple imaging, contrast imaging, etc. is visible for observation and interpretation. Grasping is performed prior to image output, the reading gain is adjusted to an appropriate value based on this accumulated recording information, and the recording scale factor is determined so that the resolution is optimized according to the contrast of the recording pattern. It is desirable to do.

As a method of grasping the accumulated record information of the radiation image prior to the output of the visible image in this way, Japanese Patent Laid-Open No. 58-6
The method disclosed in 7240 is known. This method
Prior to the main reading, a stimulable phosphor is used in advance by using excitation light having a lower level than the excitation light to be irradiated during a reading operation for obtaining a visible image for observation and interpretation (hereinafter referred to as "main reading"). A read operation for grasping the accumulated record information of the radiation image accumulated and recorded on the sheet (hereinafter referred to as "pre-reading") is performed to grasp the outline of the accumulated record of the radiation image, and the pre-reading is performed when the main reading is performed. The reading gain is appropriately adjusted based on the information, the recording scale factor is determined, or the signal processing condition is determined.

Various methods have been considered for grasping the accumulated record information of the stimulable phosphor sheet from the pre-reading image signal obtained by the pre-reading as described above. One of such methods is a histogram of the pre-reading image signal. It is known how to create. That is, since the accumulated record information can be grasped from the signal maximum value, the minimum value, the signal value which becomes the frequency maximum point, etc. of the histogram, the reading conditions such as the reading gain, the recording scale factor, etc. can be obtained based on the histogram. By determining the image processing conditions, it is possible to reproduce a radiation image with excellent diagnostic suitability.

On the other hand, at the time of recording (imaging) radiation image information, in order not to irradiate a portion not necessary for diagnosis with radiation, or when a portion not necessary for diagnosis is irradiated with radiation, the portion necessary for diagnosis starts from that portion. Scattered light enters,
To prevent this, the contrast resolution will decrease.
In many cases, the radiation field is narrowed down in the entire recording area of the stimulable phosphor sheet to perform imaging.

When grasping the accumulated record information of the stimulable phosphor sheet as described above, the radiation irradiation field B is applied to the image recording area of the stimulable phosphor sheet 103 as shown in FIG.
Is narrowed down, and the look-ahead is performed on a region considerably larger than the irradiation field B (for example, the entire region),
There is a problem that the accumulated record information actually recorded in the irradiation field B is erroneously grasped. That is, in the above case, the histogram is formed including the pre-reading image signal of the area outside the radiation field B, and therefore this histogram does not correctly reflect the actual accumulated record information in the field B. Will be.

The present applicant has already proposed some methods for recognizing the radiation irradiation field B as described above (for example, Japanese Patent Laid-Open No. 63-259538), and based on an image signal, the irradiation field inside and outside the irradiation field can be detected. If the edge candidate points that are the boundaries are obtained, the irradiation field is recognized by connecting them with lines, and the histogram is created only for the recognition area, the above problem can be solved, and the irradiation field is only rectangular. Of course, even in the case of an irregular polygon or a shape surrounded by a curve such as a circle or an ellipse, the irradiation field can be accurately recognized.

In the above-mentioned Japanese Patent Laid-Open No. 63-259538, as an operation for obtaining an edge candidate point based on the above image signal, image data is obtained along a radial direction from a predetermined point in the irradiation field toward the sheet end. , A method using a method such as differential processing or pattern matching is described.

Since the value of the image data corresponds to the magnitude of the energy of the radiation incident on the stimulable phosphor sheet, the image data outside the irradiation field generally has a low quantum level, and the image data inside the irradiation field generally has a high quantum level. It becomes a level. Therefore, the difference between the image data of the portion where the contour of the irradiation field is located (edge portion) is generally a larger quantum level than the difference between the image data of the other portions. Therefore, in the method based on the differential processing, the irradiation field edge candidate points are obtained based on this difference and the irradiation field is recognized.

[0011]

However, when a difference value of the quantum level equal to or higher than that of the edge portion of the irradiation field appears in the irradiation field depending on the type of image, in the method by the differentiation processing, the local processing called differentiation is performed. Because of this method, edge candidate points may be detected in the irradiation field, and the irradiation field may not be recognized accurately. Moreover, if there are many cases where the irradiation field cannot be accurately recognized, it may be necessary to make corrections each time.

Therefore, in the irradiation field recognition, the edge candidate points are not detected from a partial value of the image data, but
It is desirable to detect edge candidate points globally from all values of the image data, and even if the irradiation field is mistakenly recognized, learning does not cause that error, and knowledge is accumulated according to the irradiation field shape etc. It is desirable that the system changes itself.

In view of the above circumstances, it is an object of the present invention to provide a method for accurately recognizing an irradiation field and an apparatus capable of implementing the method.

[0014]

One of the methods for recognizing a radiation field of the present invention is to emit stimulated emission light from a stimulable phosphor sheet on which radiation image information is recorded by narrowing down the field as described above. When reading and obtaining an image signal, ◆ Digital image data at each position on the stimulable phosphor sheet is obtained from this image signal, and a plurality of radial patterns from a predetermined point included in the radiation irradiation field toward the sheet end are obtained. Based on the image data along the direction, the neural network is used to find the edge candidate points that are considered to be the edges of the radiation field on the stimulable phosphor sheet, and the area enclosed by the lines along these edge candidate points is determined. It is characterized in that it is recognized as a radiation irradiation field.

In order to obtain digital image data at each position on the stimulable phosphor sheet, it is necessary to first define the position on the sheet. This position may be defined on a pixel-by-pixel basis, or a plurality of pixels having a fixed relationship, for example, a plurality of 3 to 5 pixels arranged in a fixed direction may be collectively set as one position. The digital image data at each position in the former case means a digitalized image signal of a pixel corresponding to that position, and the digital image data at each position in the latter case means a plurality of pixels included in that position. What is determined based on an image signal, for example, digital image data obtained by averaging image signals of a plurality of pixels. Defining the position like this latter means in other words, pre-processing the image signal for each pixel obtained by the reading process by linear or non-linear filtering, for example, 3 to 3 for the image signal for each pixel. This means that one-dimensional smoothing is performed every 5 lines.

The neural network inputs each information in 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. Error reverse propagation learning (back propagation) that corrects the weight of connection between units (weight of synapse connection)
It is equipped with a function, and by repeating'learning ', it is possible to increase the probability of outputting a correct answer when a new signal is input (for example, "DE
Rumelhart, GEHinton and RJWilliams: Learning rep
resentations by back-propagating errors, Nature, 323
-9,533-356, 1986a "," Hideki Aso: Back Propagation Computrol No.24 53-60 "," Kazuyuki Aihara Neural Computer Tokyo Denki University Press ").

Image signals can be input to this neural network and edge candidate points can be input. Even if an incorrect output is made, the correct output is input as a teacher signal for learning, and the accuracy of irradiation field recognition is further improved. Can be improved.

After obtaining a large number of edge candidate points by the neural network and then obtaining a straight line or a curve along those points by a known method, the straight line or the curve becomes an irradiation field contour. Therefore, if the area surrounded by this line is regarded as the radiation field, the field is correctly recognized.

In the above-described method, the stimulable phosphor sheet on which the radiation image information is accumulated and recorded by irradiating the radiation with the irradiation field narrowed is irradiated with the exciting light, and the exciting light is emitted from the sheet. Photodetection means for photoelectrically reading the stimulated emission light to obtain an image signal carrying the radiation image information, and signal conversion for obtaining digital image data at each position on the stimulable phosphor sheet from the image signal. Means and ◆ is an edge portion of the radiation field on the stimulable phosphor sheet, the image data being input along a plurality of radial directions from a predetermined point included in the radiation field toward the sheet end. Neural network that outputs edge candidate points that are considered to be: ◆ Recognizing means that recognizes a region surrounded by lines along these edge candidate points as a radiation field Thus it may be performed.

Another method for recognizing a radiation field according to the present invention is to irradiate a stimulating light on a stimulable phosphor sheet on which a radiation field is focused and radiation image information is stored and recorded. When obtaining the image signal that carries the radiation image information by photoelectrically reading the stimulated emission light emitted from the sheet by light irradiation by the photodetector, each of the stimulable phosphor sheets on the stimulable phosphor sheet is obtained from the image signal. Obtaining digital image data at a position, the edge portion of the radiation field on the stimulable phosphor sheet based on image data along a plurality of radial directions from a predetermined point included in the radiation field toward the sheet end The threshold value for edge candidate point detection that is considered to be is obtained by a neural network, and the edge candidate point is obtained based on the threshold value. The region surrounded by line along the et edge candidate points is characterized in that to recognize the irradiation field.

As a method for obtaining the edge candidate points based on the threshold value, the image data is differentiated, and all points having a differential value exceeding the threshold value are set as the edge candidate points. Alternatively, a method of using a point closest to the predetermined point among the points exceeding the threshold value as an edge candidate point, or the like can be considered.

Further, in the apparatus for carrying out another method according to the present invention described above, the stimulating light is applied to the stimulable phosphor sheet on which the radiation field is focused and the radiation image information is stored and recorded. Photodetection means for photoelectrically reading the stimulated emission light emitted from the sheet by this excitation light irradiation to obtain an image signal carrying the radiation image information, and from the image signal, each on the stimulable phosphor sheet A signal conversion unit for obtaining digital image data at a position, and the image data along a plurality of radial directions from a predetermined point included in the radiation irradiation field toward the sheet end portion as an input, and on the stimulable phosphor sheet A neural network that outputs a threshold value for detecting an edge candidate point that is considered to be an edge portion of the radiation field, and an essence based on the threshold value. An edge candidate point detecting means for determining a candidate point is a region surrounded by a line along these edge candidate points characterized in that comprising a recognizing means for recognizing the irradiation field.

[0023]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below based on the embodiments shown in the drawings.

FIG. 1 shows a radiation image information recording / reproducing system for recognizing a radiation irradiation field by the first method of the present invention. This radiographic image information recording / reproducing system basically includes a radiographic image capturing unit 20, a pre-reading reading unit 30, a main-reading reading unit 40, and an image reproducing unit 50. In the radiation image capturing unit 20, radiation 102 is emitted from a radiation source 100 such as an X-ray tube toward a subject (subject) 101. At the position where the radiation 102 transmitted through the subject 101 is irradiated, as described above, the stimulable phosphor sheet 10 that accumulates the radiation energy is stored.
3 are arranged, and the subject 101 is placed on the storage phosphor sheet 103.
The transmission radiation image information of is stored and recorded. A diaphragm 104 that narrows down the irradiation field of the radiation 102 is arranged between the radiation source 100 and the subject 101.

The stimulable phosphor sheet 103 on which the radiation image information of the subject 101 is recorded in this manner is sent to the prereading reading section 30 by the sheet transfer means 110 such as a transfer roller. A laser light source for pre-reading in the pre-reading reading unit 30.
The laser light 202 emitted from 201 is the laser light 202
After passing through the filter 203 that cuts the wavelength region of the stimulated emission light emitted from the stimulable phosphor sheet 103 by the excitation of, the light is deflected linearly by an optical deflector 204 such as a galvanometer mirror, and passes through a plane reflecting mirror 205. And is incident on the stimulable phosphor sheet 103. Laser source 201 here
Is selected so that the wavelength range of the laser light 202 as the excitation light does not overlap with the wavelength range of the stimulated emission light emitted by the stimulable phosphor sheet 103. On the other hand, the phosphor sheet 103 is transferred in the direction of arrow 206 by the sheet transfer means 210 such as a transfer roller and is sub-scanned, and as a result, the phosphor sheet 103 is transferred.
A laser beam 202 is applied to the entire surface of 103. Here, the emission intensity of the laser light source 201, the beam diameter of the laser light 202, the scanning speed of the laser light 202, the stimulable phosphor sheet 103
Is selected such that the energy of the pre-reading excitation light (laser light 202) is smaller than that of the main reading performed by the main reading reading unit 40 described later.

When the laser light 202 is irradiated as described above, the stimulable phosphor sheet 103 emits stimulated emission light of a light amount corresponding to the radiation energy stored and recorded therein, and this emission light is for pre-reading. It is incident on the light guide 207. The stimulated emission light is guided through the light guide 207, emitted from the emission surface, and received by a photodetector 208 such as a photomultiplier. On the light receiving surface of the photodetector 208, a filter is attached which transmits only light in the wavelength range of stimulated emission light and cuts light in the wavelength range of excitation light, and detects only stimulated emission light. Is ready for you. The detected stimulated emission light is converted into an electric signal carrying the stored record information, and is then amplified.
Amplified by 209. The signal output from the amplifier 209 is digitized by the A / D converter 211, and the main reading control circuit 314 of the main reading reading unit 40 serves as the preread image signal Sp.
Entered in. The main reading control circuit 314 determines the reading gain setting value a, the recording scale factor setting value b, and the reproduced image processing condition setting value c based on the accumulated recording information indicated by the preread image signal Sp. In addition, the pre-read image signal S
p is also input to the irradiation field recognition circuit 220 described in detail later.

The stimulable phosphor sheet 103, which has been preread in the above manner, is transferred to the reading unit 40 for main reading. The laser light 302 emitted from the laser light source 301 for main reading in the reading unit 40 for main reading passes through the filter 303 that cuts the wavelength region of the stimulated emission light emitted from the stimulable phosphor sheet 103 by the excitation of the laser light 302. After doing
The beam expander 304 precisely adjusts the beam diameter, and the optical deflector 305 such as a galvanometer mirror is used.
It is linearly deflected by and is incident on the stimulable phosphor sheet 103 via the plane reflecting mirror 306. An fθ lens 307 is arranged between the light deflector 305 and the plane reflecting mirror 306 so that the beam diameter of the laser light 302 scanning the stimulable phosphor sheet 103 becomes uniform. On the other hand, the stimulable phosphor sheet 103 is transferred in the direction of arrow 308 by the sheet transfer means 320 such as a transfer roller and is sub-scanned, and as a result, the entire surface of the stimulable phosphor sheet 103 is irradiated with laser light. When the laser light 302 is irradiated in this way, the stimulable phosphor sheet 103 emits stimulated emission light in an amount corresponding to the radiation energy stored and recorded therein, and this emission light is incident on the main reading light guide 309. To do. The stimulated emission light guided while repeating total reflection in the main reading light guide 309 is emitted from its emission surface and is received by a photodetector 310 such as a photomultiplier. A filter that selectively transmits only the wavelength region of the stimulated emission light is attached to the light receiving surface of the photodetector 310 so that the photodetector 310 detects only the stimulated emission light.

The output of the photodetector 310 that photoelectrically detects the stimulated emission light showing the radiation image recorded on the stimulable phosphor sheet 103 is based on the reading gain set value a determined by the control circuit 314. Amplifier with read gain set by
Amplifies to an appropriate level electric signal by 1. The amplified electric signal is input to the A / D converter 312, converted into a digital signal with a recording scale factor suitable for the signal fluctuation width based on the recording scale factor setting value b, and input to the signal processing circuit 313. .. The digital signal is
In the signal processing circuit 313, the reproduction image processing condition set value c is set so that a radiographic image excellent in observation and interpretation suitability is obtained.
Signal processing (image processing) based on the above, and output.

The read image signal (main read image signal) So output from the signal processing circuit 313 is input to the optical modulator 401 of the image reproducing section 50. In the image reproducing section 50, the laser light 403 from the recording laser light source 402 is used as the optical modulator 401.
Thus, it is modulated based on the main reading image signal So input from the signal processing circuit 313, deflected by the scanning mirror 404, and scanned on the photosensitive material 405 such as a photographic film. Then, the photosensitive material 405 is transported in a direction (arrow 406 direction) orthogonal to the scanning direction in synchronization with the scanning, and a radiation image based on the main reading image signal So is output on the photosensitive material 405. As a method of reproducing the radiation image,
In addition to such a method, various methods such as the above-described CRT display can be adopted.

Next, even when the radiation field B is narrowed in the stimulable phosphor sheet 103 as shown in FIG. 2, the reading gain setting value a, the recording scale factor setting value b, and the image processing are performed. A mechanism for appropriately determining the condition setting value c will be described with reference to FIG. As shown in FIG. 5, the control circuit 314 includes a signal extraction unit 35.
0, histogram analysis unit 351, reading unit 352, and storage unit
Consists of 353. The prefetch image signal Sp is the signal extraction unit 35.
The pre-reading image signal Sp ′ is input only to the area designated as will be described later. The prefetch image signal Sp ′ output from the signal extraction unit 350 is input to the histogram analysis unit 351. The histogram analysis unit 351 uses the prefetch image signal S
A histogram of p'is created, for example, the maximum value, the minimum value, the maximum frequency value, etc. are obtained, and the information Sr indicating these values is sent to the reading unit 352. The storage unit 353 stores the optimum reading gain setting value a, the recording scale factor setting value b, and the image processing condition setting value c corresponding to the maximum value and the minimum value, and the reading unit 352 stores the information Sr in the information Sr. The corresponding set values a, b, c are read from the storage unit 353 and sent to the amplifier 311, the A / D converter 312 and the signal processing circuit 313, respectively, as described above.

Next, signal extraction in the signal extraction section 350 will be described. The irradiation field recognition circuit 220 stores a neural network (which refers to a coefficient that represents the weight of the connection of the neurons that make up the edge candidate point; hereinafter referred to simply as a neural network) for calculating edge candidate points and a computing unit 223. Consists of.

The pre-read image signal Sp is first extracted in the direction along the direction D ₁ shown in FIG. 3, and thereafter, the directions D ₂ and D are similarly extracted.
₃ ... Extract in the direction along Dn. The plurality of directions D _{1 to} Dn are radial directions from the center O of the stimulable phosphor sheet 103 toward the sheet end, and in this example, the directions D _{1 to} Dn are set at equal angular intervals. ..
Further, in such radial directions D _{1 to} Dn, for example, the size of the stimulable phosphor sheet 103 is a half-cut size (256 × 192 m).
In case of m), about 64 directions are set. Since the level of the image signal in the radiation field B on the stimulable phosphor sheet 103 is obviously higher than the image signal level in the area outside the field B, the pre-read image along a certain direction Di. The value of the signal Sp shows a profile as shown in FIG. When the profile of the image signal Sp along the extracted direction Di is input to the irradiation field recognition circuit 220, the neural network is read from the memory in the computer system 222. Then, as shown in FIG. 4, in order to reduce the number of input points, the profile of the preread image signal Sp is uniformly thinned along the direction Di in which the profile is extracted, and the thinned profile of the preread image signal Sp is neural-coded. Input to the network, the position of the edge candidate point from the neural network, that is, the distance from the center O of the stimulable phosphor sheet 103 to the edge portion 103a of the sheet 103 in the Di direction to the edge candidate point. Is output.

Next, the second method of the present invention will be described. The configuration of the radiation image information recording / reproducing system utilizing the second method of the present invention is the same as that of the radiation image information recording / reproducing system utilizing the first method described above, and thus detailed description thereof will be omitted and the neural network will be omitted. Describe the process after it is entered in.

Similar to the first method described above, the profile of the preread image signal Sp is uniformly thinned out along the direction Di in which the profile is extracted, and the thinned profile of the preread image signal Sp is input to the neural network. To be done. Then, a threshold value for detecting the edge candidate points is output from the neural network, and the edge candidate point detecting means obtains the edge candidate points based on the threshold value. This edge candidate point detecting means is a computer system 22 of a radiation image information recording / reproducing system using the first method of the present invention.
It is stored in 2.

As a method for detecting the edge candidate points, a known method as disclosed in JP-A-63-259538 can be used. That is, a method of differentiating the profile of the pre-read image signal Sp and detecting all points where the differential value exceeds the threshold value output from the neural network as edge candidate points, or a point where the differential value exceeds the threshold value Among them, a method of detecting a point closest to the center O of the stimulable phosphor sheet as an edge candidate point can be used.

Next, the learning function of the neural network will be described.

FIG. 6 is a block diagram showing the basic concept of the neural network. In the embodiment of the present invention, the input section receives the image signal Sp and the output section outputs the position of the edge candidate point. If the output unit outputs an incorrect value when the neural network is trained, the correction value is input from the correction input unit to perform learning.

FIG. 7 is a diagram showing an example of a neural network having a backpropagation learning (backpropagation) function. Error back-propagation learning (backpropagation) is, as described above, comparing the output of the neural network with the correct answer (teaching signal) to sequentially connect the weights from the output side to the input side (synaptic connection weights). A "learning" algorithm that modifies

As shown in the figure, the first layer (input layer), the second layer (intermediate layer), and the third layer (output layer) of this neural network are n ₁ , n ₂ , and ₁ units, respectively. (Neurons). The signals F ₁ , F ₂ , ..., F _n1 input to the first layer (input layer) are signals in which the profile of the preread image signal Sp is thinned out uniformly along the Di direction. The output Y3,1 from the layer (output layer) is used as an edge candidate point (ratio of the distance from the center O to the edge candidate point to the distance to the center O of the sheet 103) or a threshold value for detecting the edge candidate point. Corresponding signal. The unit i of the k-th layer is Uk, i,
Each input to k, i is Xk, i, each output is Yk, i, and the coupling weight from Uk, i to Uk + 1, j is Wk, i; k + 1, j, and each unit Uk, j Are the same characteristic functions

[0040]

[Equation 1]

It is assumed that At this time, the input Xk, j and the output Yk, j of each unit Uk, j are

[0042]

[Equation 2]

[0043]

[Equation 3]

It becomes However the units U1, i (i = 1,2, ..., n 1) constituting the input layer the input F ₁ to, F _2,
…, F _n1 is not weighted and each unit U1, i is directly
It is input to (i = 1,2, ..., n ₁ ). The input n ₁ signals F ₁ , F ₂ , ..., F _n1 have weights Wk, i; k + 1,
It is transmitted to the final output Y3,1 while being weighted by j.
The ratio from the center O to the edge candidate point with respect to the distance to 103a) or the threshold value for detecting the edge candidate point is obtained.

Here, a method of determining the weights Wk, i; k + 1, j of the respective connections will be described. First, an initial value of the weight Wk, i; k + 1, j of each connection is given by a random number. In this case, even if the input F ₁ to F _n1 is varied to maximize the output Y3,1 such that a value or a value close thereto within a predetermined range, it is preferable to limit the scope of the random number.

A large number of stimulable phosphor sheets on which X-ray images having optimum threshold edge points or threshold values for detecting edge candidate points are recorded are read as described above, and the obtained points are obtained. The read image signal Sp is thinned out to obtain the n ₁ inputs F ₁ , F ₂ , ..., F _n1 . The n ₁ inputs F ₁ , F ₂ , ..., F _n1 are input to the neural network shown in FIG. 7, and each unit Uk,
The output Yk, i of i is monitored.

When each output Yk, i is obtained, a final output Y3,1 and a teacher signal d indicating a correct edge candidate point or a threshold value for detecting a correct edge candidate point for this image are obtained. Squared error of

[0048]

[Equation 4]

Is required. In order to minimize this squared error E, the weights Wk, i; k + 1, j of each connection are set as follows.
Is fixed.

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

[0051]

[Equation 5]

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

Here,

[0054]

[Equation 6]

From equation (2),

[0056]

[Equation 7]

Therefore, the equation (6) is

[0058]

[Equation 8]

It becomes

From equation (4),

[0061]

[Equation 9]

When this equation (9) is transformed using the equation (3),

[0063]

[Equation 10]

Here, from the equation (1),

[0065]

[Equation 11]

Therefore,

[0067]

[Equation 12]

It becomes

In equation (8), k = 2 is set, and equation (10) and (1
Substituting equation (1) into equation (8),

[0070]

[Equation 13]

Substituting equation (13) into equation (5),

[0072]

[Equation 14]

It becomes According to this equation (14), W2, i; 3,1
The weight of each connection (i = 1, 2, ..., N ₁ ) is modified.

Next,

[0075]

[Equation 15]

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

[0077]

[Equation 16]

From equation (11),

[0079]

[Equation 17]

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

[0081]

[Equation 18]

In equation (8), k = 1 is set, and equation (18) is replaced by
Substituting into equation (8),

[0083]

[Formula 19]

Substituting equation (19) into equation (5), k = 1
And put

[0085]

[Equation 20]

Then, W2, i; 3,1 (i =
1,2, ..., n ₁ ) are substituted into this equation (20), and W1, i; 2, j (i =
, 1, ..., N ₁ ; j = 1, 2, ..., N ₂ ) are modified.

Theoretically, the equations (14) and (20) are used, and the learning coefficient η is set to be sufficiently small to increase the number of times of learning, whereby the weight Wk, i; k + 1 of each connection is calculated. , j can be converged to a predetermined value, but it is not realistic to make the learning coefficient η too small because the learning progress is delayed. On the other hand, if the learning coefficient η is set to be large, learning may oscillate (the weight of the above coupling does not converge to a predetermined value). Therefore, in practice, an inertial term such as the following equation is added to the correction amount of the coupling weight to suppress the vibration, and the learning coefficient η is set to a relatively large value. (For example, DE Rumelhart, GEHinton and RJWil
liams: Learning internal representations by error pr
opagation In Parallel Distributed Processing, Volum
e 1, JLMcClelland, DERumelhart and The PDP Resea
rch Group, MIT Press, 1986b '')

[0088]

[Equation 21]

However, ΔWk, i; k + 1, j (t) is the weight of the connection Wk, i; k + 1, j after modification from the weight Wk, i; k + 1, j after modification in the t-th learning. It represents the correction amount obtained by subtracting k + 1, j. Further, α is a coefficient called an inertial term.

As the inertia term α and the learning coefficient η, for example, α =
The weights Wk, i; k + 1, j of each connection are corrected (learned) by using, for example, 0.9 and η = 0.25, for example, 200,000 times, and then the weights Wk, i; k + 1, j of each connection are It is fixed to the final value. In addition, here, after the system including the irradiation field recognizing device for obtaining the irradiation field using this neural network is installed in the user, learning is continued as described below, so the final value here Is, for example, the final value of the initial startup stage of the user. At the end of this learning, the output Y3,1 becomes a signal that almost correctly represents the position of the edge candidate point or the threshold value for detecting the edge candidate point.

The neural network is not limited to the one having a three-layer structure, and needless to say, it may have more layers. Also, the number of yuts in each layer,
The number is not limited to the number in the above embodiment, but may be determined according to the number of profiles of the input image signal, the edge candidate point, or the required accuracy of the signal representing the threshold value for detecting the edge candidate point. Of course, each layer can be composed of any number of units.

After obtaining the edge candidate points by the above-mentioned first and second methods, the computer system 222 extracts the pre-reading image signal Sp for the edge candidate points, and the pixel corresponding to each of the extracted pre-reading image signals Sp. The position is obtained, and information Se indicating the pixel position is sent to the arithmetic unit 223. The pre-reading image signal Sp extracted as described above becomes an image signal which mostly bears an edge portion of the radiation irradiation field B (see FIG. 2) on the stimulable phosphor sheet 103, that is, an edge candidate point signal. In this example, the pixel position is represented by an xy orthogonal coordinate system on the stimulable phosphor sheet 103, as shown in FIG.

As described above, if the edge candidate points are obtained as described above and then the lines along these points are obtained, the lines become the contours of the irradiation field. The line along this edge candidate point is, for example, a method of connecting the remaining points after smoothing the points, a method of locally applying the least squares method to obtain a plurality of straight lines, and a method of connecting them.
It can be obtained by a method of applying a spline curve or the like.
It is configured to obtain a plurality of straight lines along the edge candidate points by using conversion. Hereinafter, the process of obtaining this straight line will be described in detail.

When the coordinates of the pixel position (edge candidate point) indicated by the above information Se are (x ₀ , y ₀ )
With these x ₀ and y ₀ as constants, a curve represented by ◆ ρ = x ₀ cos θ + y ₀ sin θ ◆ is obtained for all edge candidate point coordinates (x ₀ , y ₀ ). This curve is as shown in FIG. 8 and exists as many as the edge candidate point coordinates (x ₀ , y ₀ ).

Next, the computing unit 223 makes the intersection point (ρ ₀ , ρ ₀ , at which a predetermined number Q or more of the plurality of curves described above intersect each other).
θ ₀ ) is obtained. Note that the edge candidate point coordinates (x ₀ , y ₀ )
Since many curves do not intersect exactly at one point due to the error of, etc., in reality, for example, when the intersections of two curves exist at intervals of a minute predetermined value or less, the centers of the intersections are Let the intersection point be (ρ ₀ , θ ₀ ). Next is the calculation section
223 obtains a straight line defined by the following formula: ρ ₀ = x cos θ ₀ + y sin θ ₀ ◆ in the xy Cartesian coordinate system from the intersection point (ρ ₀ , θ ₀ ). This straight line is a straight line extending along a plurality of edge candidate point coordinates (x ₀ , y ₀ ).

When the edge candidate points are distributed as shown in FIG. 2, the above straight line becomes as shown in FIG.
The calculation unit 223 then determines the plurality of straight lines L ₁ and L thus obtained.
A region surrounded by ₂ , L ₃ ... Ln is obtained, and this region is recognized as a radiation field B. This area is recognized in detail as follows, for example. The calculation unit 223 stores the line segments M ₁ , M ₂ , M ₃ ... Mm (four if the stimulable phosphor sheet 103 is rectangular) connecting the corners and the center G of the stimulable phosphor sheet 103. And each of these line segments M _{1 to} M
Existence of an intersection between m and each of the straight lines L _{1 to} Ln is checked. When this intersection exists, the arithmetic unit 223 cuts off the plane on the side including the sheet corner among the planes bisected by the straight line. This operation applies to all straight lines L _{1 to} Ln and line segment M
_By carrying out for _{1 to} Mm, the region surrounded by the straight lines L _{1 to} Ln is left. This remaining area is
That is, the radiation field B.

The calculation unit 223 sends the information St indicating the radiation irradiation field B recognized as described above to the signal extraction unit 350 of the control circuit 314. The signal extraction unit 350 uses the prefetch image signal Sp.
Then, only the signal for the area indicated by this information St is extracted and sent to the histogram analysis section 351. Therefore, the histogram analysis in the histogram analysis unit 351 is performed only on the area of the stimulable phosphor sheet 103 that is actually irradiated with radiation, so that the above-mentioned set values a, b and c are the actual accumulation values. It is optimal for recorded information.

In the embodiment described above, the point in the irradiation field, which is the starting point in the direction of extracting the profile of the image signal Sp, is the center O of the sheet, but this point is not limited to the center point of the sheet, but may be in the irradiation field. Any existing point may be used. For example, when the radiation irradiation field is narrowed down to an extremely small value, the sheet center point may be located outside the irradiation field. It is desirable to use the point that will always exist in the irradiation field, such as the center of gravity of the high-concentration region when the value is converted.

In the above embodiment, the directions D _{1 to} Dn for extracting the profile of the image signal Sp are set at equal angular intervals around the sheet center O, but these directions are set at equal angular intervals. It doesn't matter. That is, for example, as shown in FIG. 10, a plurality of equidistant points are set on the sides of the stimulable phosphor sheet 103, and the respective directions D _{1 to} Dn from the point P in the irradiation field B toward these points are image signals. You may make it the direction of extraction of the profile of Sp.

Further, as shown in FIG. 11, the point P in the irradiation field B is
When the distance g from the edge candidate point E to the edge candidate point E does not change so much, the extraction direction Di of the profile of the image signal Sp is set relatively coarse (in the range of h _{1 in} the figure), and when the distance g becomes considerably changed. The extraction direction Di of the profile of the image signal Sp may be set relatively finely (in the range of h _{2 in} the drawing).

In the present embodiment, only one profile of the image signal Sp is input to the neural network along the certain direction Di, but the profile of several neighboring image signals Sp along the certain direction Di is input. May be input to the neural network, or the average profile of the profiles of several neighboring image signals Sp along a certain direction Di may be input. The number of units in the first layer of the neural network will increase if several neighboring profiles are input.

Further, in the case of recognizing the irradiation field B having the shape as shown in FIG. 2, normally only one edge candidate point of the irradiation field contour is obtained by one input of the neural network. When recognizing the irradiation field B having the shape as shown in FIG. 13, a plurality of edge candidate points of the irradiation field contour portion may be detected. Even in such a case, if the neural network is trained to detect a plurality of edge candidate points, all edge candidate points in the irradiation field contour portion can be detected, and the irradiation field B having a complicated shape can be correctly recognized. Become.

When the radiation field on the stimulable phosphor sheet 103 is not narrowed down, the information St output from the calculation unit 223 naturally indicates the entire area of the stimulable phosphor sheet 103. Also, the set values a, b and c are properly set. However, in this case, the irradiation field recognition circuit 220 is, so to speak, useless. Therefore, in order to avoid such a situation, a switch for turning on and off the operation of the irradiation field recognition circuit 220 is provided. When the irradiation field recognition circuit 220 is in the OFF state, the signal extraction unit 350
It is preferable to let all pre-read image signals Sp pass. Then, if it is known in advance that the reading is from the stimulable phosphor sheet 103 to which the irradiation aperture has not been applied, the whole prefetch image signal Sp can be quickly input to the histogram analysis unit 351 by a manual operation or the like. Like

Further, the reading area in the main reading reading section 40 may be controlled based on the information St indicating the radiation irradiation field B obtained by the irradiation field recognition circuit 220. By doing so, the main reading is performed only for the radiation irradiation field on the stimulable phosphor sheet 103, and the reading processing can be speeded up.

The apparatus shown in FIG. 1 has a reading section for main reading and a reading section for pre-reading separately. For example, as shown in JP-A-58-67242, the reading system for main reading and the pre-reading section are provided. When the pre-reading is completed, the sheet transfer means returns the stimulable phosphor sheet to the reading system for the main reading, and the excitation light energy adjusting means at the pre-reading causes the excitation light energy to be the same as that at the main reading. However, the method of the present invention can be applied even when the radiation image information is read by such an apparatus.

Further, in the above embodiment, the radiation irradiation field is recognized from the look-ahead image signal.
It is of course possible to recognize the radiation field by the method of the present invention using the read image signal. In such a case, the recognized irradiation field information can be used to properly set the image processing condition setting value c, for example.

Further, in the above embodiment, the reading gain setting value a, the recording scale factor setting value b, and the reproduced image processing condition setting value c in the actual reading are performed by histogram analysis. The recognition device as described in 151040 may be used to obtain the optimum reading condition for the main reading from the preread image signal by a neural network.

In the above embodiment, the learning of the neural network is completed at the stage when the system for recognizing the irradiation field is installed in the user. However, during the use of the user, re-learning is performed according to the use of the user. The accuracy of outputting the edge candidate points of the irradiation field or the threshold value for detecting the edge candidate points suitable for the user may be improved.

[0109]

As described in detail above, according to the radiation field recognition method of the present invention, the influence of the portion outside the radiation field is eliminated in the pre-reading, the accumulated record information regarding the subject is correctly grasped, and the main reading is performed. It is possible to optimally set the reading condition and the image processing condition. Therefore, according to the method of the present invention, it is possible to always reproduce a radiation image having excellent observation and interpretation suitability.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an apparatus for recognizing an irradiation field and reading radiation image information according to the present invention.

FIG. 2 is an explanatory view showing a state of recording radiation image information on a stimulable phosphor sheet according to the present invention.

FIG. 3 is an explanatory diagram illustrating a direction of extracting an image signal according to the present invention.

FIG. 4 is a diagram showing an example of a distribution state of image signals and a neural network according to the present invention.

FIG. 5 is a block diagram showing in detail a part of the apparatus shown in FIG.

FIG. 6 is a block diagram showing the basic concept of a neural network.

FIG. 7 is a diagram showing an example of a neural network adopted by the irradiation field recognition means.

FIG. 8 is a graph for explaining a method of detecting a straight line along an edge candidate point.

FIG. 9 is an explanatory diagram for explaining a method of extracting a region surrounded by a straight line along an edge candidate point.

FIG. 10 is an explanatory view showing another example of the direction setting of the image signal extraction in the present invention.

FIG. 11 is an explanatory view showing still another example of the direction setting of the image signal extraction in the present invention.

FIG. 12 is an explanatory view showing another example of recording state of radiation image information on the stimulable phosphor sheet according to the present invention.

FIG. 13 is an explanatory view showing still another example of recording state of radiation image information on the stimulable phosphor sheet according to the present invention.

[Explanation of symbols]

20 Radiation image capturing unit 30 Pre-reading reading unit 40 Main-reading reading unit 100 Radiation source 101 Subject 102 Radiation 103 Accumulative phosphor sheet 104 Aperture 201 Pre-reading laser light source 202 Pre-reading laser light 204 Pre-reading light deflector 208 Pre-reading light Detector 210 Pre-reading sheet transfer means 220 Irradiation field recognition circuit 301 Main reading laser light source 302 Main reading laser light 305 Main reading optical deflector 310 Main reading photo detector 311 Amplifier 312 A / D converter 313 Signal processing circuit 314 Control circuit 320 Sheet reading sheet transfer means B Radiation irradiation field a Reading gain setting value b Recording scale factor setting value c Reproduction image processing condition setting value D _{1 to} Dn Image signal profile extraction direction O, P Image field profile extraction starting point Point So Main reading image signal Sp Pre-reading image signal Se Information indicating edge candidate point

Claims (4)

[Claims] 1. A stimulable luminescence emitted from the sheet by irradiating the stimulable phosphor sheet on which radiation image is accumulated and recorded by irradiating radiation with a narrowed irradiation field and irradiating the stimulating light. When obtaining an image signal that carries the radiation image information by photoelectrically reading light with a light detection unit, digital image data at each position on the stimulable phosphor sheet is obtained from the image signal, and the radiation image is stored in the radiation field. Based on image data along a plurality of radial directions from a predetermined point to the sheet end, edge candidate points considered to be the edge portion of the radiation field on the stimulable phosphor sheet are neural network based. A method for recognizing a radiation field, characterized in that a region surrounded by a line along these edge candidate points is recognized as a radiation field. 2. A stimulable luminescence emitted from the sheet by irradiating the stimulable phosphor sheet on which the radiation image is accumulated and recorded by irradiating radiation with a narrowed irradiation field and irradiating the stimulating light. Light detection means for photoelectrically reading light to obtain an image signal carrying the radiation image information, signal conversion means for obtaining digital image data at each position on the stimulable phosphor sheet from the image signal, and radiation irradiation An edge candidate point that is considered to be an edge portion of the radiation field on the stimulable phosphor sheet, using the image data as input along a plurality of radial directions from a predetermined point included in the field toward the sheet end. And a recognition means for recognizing a region surrounded by a line along these edge candidate points as a radiation field. Radiation field recognition device. 3. A stimulable luminescence emitted from the sheet by irradiating the stimulable phosphor sheet, which is irradiated with radiation by narrowing down the irradiation field to record and record the radiation image information, with the stimulating light irradiation. When obtaining an image signal that carries the radiation image information by photoelectrically reading light with a light detection unit, digital image data at each position on the stimulable phosphor sheet is obtained from the image signal, and the radiation image is stored in the radiation field. For edge candidate point detection that is considered to be an edge portion of the radiation field on the stimulable phosphor sheet based on image data along a plurality of radial directions from a predetermined point toward the sheet end portion A threshold value is obtained by a neural network, the edge candidate points are obtained based on the threshold value, and an area surrounded by a line along these edge candidate points is released. The radiation field recognition method and recognizes a line irradiation field. 4. A stimulable luminescence emitted from the sheet by irradiating excitation light onto a stimulable phosphor sheet on which radiation image is accumulated and recorded by irradiating radiation with an irradiation field narrowed. Light detection means for photoelectrically reading light to obtain an image signal carrying the radiation image information, signal conversion means for obtaining digital image data at each position on the stimulable phosphor sheet from the image signal, and radiation irradiation An edge candidate inspection that is considered to be an edge portion of the radiation field on the stimulable phosphor sheet using the image data as input along a plurality of radial directions from a predetermined point included in the field toward the sheet end. A neural network that outputs a threshold value for output, an edge candidate point detection unit that obtains an edge candidate point based on the threshold value, Irradiation field recognition apparatus characterized by comprising a region surrounded by a line along the candidate point and a recognition means for recognizing the irradiation field.