JPH0896125A - Medical image display device - Google Patents

Abstract

(57) [Summary] [Purpose] Obtain the window window and window level that reduce the influence of the background, and appropriately adjust the contrast of the part to be diagnosed by the doctor. A pixel extraction unit 2 compares each density value of a plurality of pixels arbitrarily extracted from an image captured by an image diagnostic device 1 with a predetermined threshold value and determines a density value above or below the threshold value. Take out the pixels it has and create the density histogram creation unit 3
Creates a density histogram for the extracted pixels. The learning unit 6 learns and determines the weight of the neural network 4 based on the density histogram created as described above with respect to various types of images, and the neural network 4 uses the weight determined as a result of the learning to capture the imaged image. With respect to the density histogram data created as described above, the weighted sum is calculated, and the window width and the window level are output.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a medical image display apparatus for displaying an image of a subject imaged by using an image diagnostic apparatus such as an X-ray CT apparatus or a magnetic resonance tomography apparatus (MRI apparatus). In particular, it relates to a technique for adjusting the contrast of a displayed image.

[0002]

2. Description of the Related Art As a conventional medical image display device of this type, there is, for example, one proposed by the applicant of the present application in Japanese Patent Laid-Open No. 5-61973. This conventional apparatus is a means for creating a density histogram of an image picked up by an image diagnostic device, a weighted sum calculation of the created density histogram data, and a window window (hereinafter abbreviated as WW) and a window level (abbreviated as WW). Hereinafter, a neural network for outputting (hereinafter, abbreviated as WL) and a learning unit for learning and determining the weight of the neural network based on density histograms of various images.

Now, WW and WL will be described with reference to FIG. FIG. 17 exemplifies a density histogram of the image of the subject imaged by the image diagnostic apparatus. As is well known, a density histogram is created by extracting the density value of each pixel of an image and counting the density value. When a CRT display is used as a device for displaying an image represented by such a density histogram, since the display gradation value of the CRT display is generally 256 gradations, the width of the density value on the horizontal axis of the density histogram is It is divided into 256 from the minimum value to the maximum value. The above WW is a parameter for arbitrarily setting the width of this density value, and WL indicates the central value of WW.

The contrast adjustment is performed in the direction of clarifying the difference in brightness between the site to be diagnosed (for example, a lesion site) and the surrounding site. Now, it is assumed that the difference in brightness between areas A in the density histogram shown in FIG. 17 is to be made clear. As described above, the density value of the density histogram is divided into 256 as a whole. However, if the density value of the area A is divided into 256, the density difference in the area A becomes remarkable and the light-dark difference becomes clear. That is, WW may be set to the area A, and WL may be set to the center value.

In the above-mentioned conventional apparatus, first, the learning means preliminarily determines the weight of the neural network by learning based on the density histograms of various images.
Then, in the production process, the neural network uses the determined weight and calculates the weight sum of the density histogram data of the created image to obtain WW and WL and outputs them. With this, it is configured to obtain appropriate WW and WL for various images.

[0006]

However, the conventional example having such a structure has the following problems. That is, the histogram creating means of the conventional device is
A plurality of (for example, 4096) pixels (hereinafter referred to as sampling pixels) are arbitrarily extracted from various images,
A density histogram is created for the density values of these sampling pixels, the learning means determines the weight, and the neural network calculates WW and WL. However, in the conventional device, since the density histogram of the image captured by changing the image capturing condition or the image capturing site is created,
It is unavoidable that the sampling pixels include pixels forming a background portion (a region where nothing is imaged) unrelated to a site (for example, a lesion site) observed by a doctor during diagnosis. Based on the density histogram created for such sampling pixels, the weight of the neural network is determined, and WW and WL are calculated by the neural network based on the weight.
Is added with the density value of the pixels forming the background portion. As a result, the contrast adjustment of the image by the conventional device does not properly adjust the contrast of the part diagnosed by the doctor.

In the conventional device, the sampling pixel is
For example, selecting a pixel that corresponds to the intersection of each vertical line and each horizontal line when a plurality of vertical lines and horizontal lines are virtually drawn in a grid pattern on the image is extracted so that the entire image is selected evenly. ing. With such an extraction method, there is a possibility that the ratio of the pixels forming the background part included in the sampling pixels extracted from each image can be lowered to some extent, but the background part is formed on the sampling pixels extracted from each image. Pixels are likely to be included evenly.

Therefore, for example, in a tomographic image of the brain, etc., the tissue portion including the diagnosis target portion is projected in the vicinity of the center of the image. Therefore, when sampling pixels are extracted from the vicinity of the center of the image, sampling is performed from the image of the brain or the like. Although the ratio of the pixels forming the background portion included in the pixels can be made extremely low, for example, in the case of images of both knees, the background portion is projected in the vicinity of the center of the image, so it is extracted from such an image. The ratio of the pixels forming the background portion included in the sampling pixels is high.

Further, since the images handled by the conventional apparatus are various images with different image pickup conditions and image pickup parts, it is automatically determined which part of each image is the background portion according to these image pickup conditions and image pickup parts. It is difficult to recognize and automatically extract the pixels excluding the pixels forming the background portion from each image.

Therefore, with the configuration of the conventional device, it is difficult to eliminate the influence of the background portion from the required WW and WL, and it is difficult to properly adjust the contrast of the diagnosis target portion of the doctor.

The present invention has been made in view of the above circumstances, and seeks a window window and window level in which the influence of the background portion is reduced,
An object of the present invention is to provide a medical image display device capable of appropriately adjusting the contrast of a diagnosis target site of a doctor.

[0012]

The present invention has the following constitution in order to achieve such an object. That is, the present invention provides two parameters for adjusting the contrast of an image captured by an image diagnostic apparatus, that is, a window window for arbitrarily setting the width of the density value when the image is displayed, and the window. A medical image display device for setting a window level for setting the central value of Widus and a display image,
(A) Each density value of a plurality of pixels arbitrarily extracted from the image captured by the image diagnostic apparatus, or the density value of each section obtained by dividing the image is compared with a predetermined threshold value,
Pixels / section extracting means for extracting pixels or sections having density values above or below the threshold value, and (b)
Density histogram creating means for creating a density histogram of the image with respect to the pixels or sections extracted by the pixel / section extracting means; and (c) a weighted sum calculation of the created density histogram data A neural network for outputting the window and the window level, and (d) learning means for learning and determining the weight of the neural network based on the density histograms of various images.

[0013]

The operation of the present invention is as follows. The pixel / section extracting means compares each density value of a plurality of pixels arbitrarily extracted from the image picked up by the image diagnostic apparatus, or the density value of each section into which the image is divided, with a predetermined threshold value. Then, pixels or sections having density values above or below the threshold value are extracted. This extracted pixel, or
For the classification, the density histogram creating means creates a density histogram of the image. Then, the learning means learns and determines the weight of the neural network based on the density histogram created as described above for various types of images, and the neural network uses the weight determined as a result of the learning to capture an image. The density histogram data created as described above is weighted and summed for the created image, and the window width and the window level are output.

Here, by appropriately determining the threshold value, the pixel / section extracting means can eliminate the pixels or the sections constituting the background portion unrelated to the diagnosis, and the density histogram creating means, A density histogram can be created for pixels or sections that constitute a tissue part including a diagnosis target site.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. 1 is a block diagram showing a schematic configuration of a medical image display apparatus according to a first embodiment of the present invention.

The apparatus of this embodiment is an X-ray CT apparatus and an MRI.
Each density value of a plurality of pixels arbitrarily extracted from the image of the subject imaged by the image diagnostic device 1 such as a device is compared with a predetermined threshold value, and the density value above or below the threshold value is determined. A pixel extraction unit 2 that extracts the pixels that it has, a density histogram creation unit 3 that obtains a density histogram for the pixels extracted by the pixel extraction unit 2, a weighted sum calculation of the created density histogram data, and the values of WW and WL are calculated. A desired neural computer 4 (hereinafter referred to as a neural network 4), a random number generator 5,
A learning unit 6 that determines the weight of the neural network 4 by learning, a normalization processing unit 9, and a denormalization processing unit 10 that denormalizes the output from the neural network 5.
Consists of a contrast setting unit 7 for adjusting the contrast of the image using the values of WW and WL obtained by the neural network 4 and denormalized by the denormalization processing unit 10, and a CRT display 8 for displaying the image. ing.

First, a plurality of pixel extraction units 2 corresponding to the pixel / section extraction unit, which is a main part of the present invention, are extracted from the image of the subject imaged by the image diagnostic apparatus 1 (for example,
(4096 pixels) are arbitrarily extracted. There is no particular limitation on the extraction method, but here, for example, a plurality of vertical lines and horizontal lines are selected on the image to select the pixels corresponding to the intersections of each vertical line and each horizontal line when virtually drawn in a grid pattern, Extract so that pixels are evenly selected from the entire image.

Next, with respect to each of the extracted pixels, each density value is compared with a predetermined threshold value, and pixels having a density value above or below the threshold value are extracted. Here, in a normal image, the density value of the pixels forming the part to be diagnosed (lesion site, etc.) is larger than the density value of the pixels forming the background part irrelevant to the diagnosis. The lower limit value that can eliminate the density value of the pixels constituting the above is determined, and with this lower limit value as the threshold value, only the pixels having the density value equal to or higher than the threshold value are extracted.

In this case, for example, as shown in FIG. 2, the maximum density value and the minimum density value of the extracted 4096 pixels are searched for, and the density width between the maximum density value MAX and the minimum density value MIN is set to 100%. , The density value SL corresponding to n% (for example, n = 10) from the minimum density value MIN.
Is determined as the threshold value. And extracted 4096
Pixels having a density value less than the threshold value (SL) are excluded from the individual pixels, and only the other pixels are extracted. It should be noted that the above-mentioned n may be obtained by, for example, experimentally obtaining a value capable of effectively excluding the pixels forming the background portion. However, as described above, by setting n = 10, Pixels forming the background portion can be almost eliminated, and as will be described later, the influence of the background portion on WW and WL obtained by the neural network 4 can be reduced.

By the way, depending on the imaging conditions, there are cases where the density values of the pixels forming the part to be diagnosed and the density values of the pixels forming the background part are opposite. In this case, an upper limit value that can eliminate the density value of the pixels forming the background portion is determined, and with this upper limit value as a threshold value, only pixels having a density value below the threshold value are extracted.

In this case, contrary to the above case, the extracted 4
The density width between the maximum density value and the minimum density value of 096 pixels is set to 100%, and the maximum density value MAX to m%
A density value corresponding to (for example, m = 10) is determined as a threshold value, and among the extracted 4096 pixels, pixels having a density value exceeding this threshold value are excluded, and only other pixels are excluded. Take out. It should be noted that for the above-mentioned m as well, a value that can effectively exclude the pixels forming the background portion may be obtained experimentally, for example.

For example, in the image obtained by the X-ray CT apparatus, the background portion is excluded based on the CT value corresponding to the density value. In this case, the CT value of the background portion is known in advance. Therefore, the CT value that can eliminate the CT value in the background portion may be determined as the threshold value.

Next, in the density histogram creating section 3 corresponding to the density histogram creating means, the tissue extracted from the pixel extracting section 2, that is, the tissue including the site to be diagnosed, which does not include the pixels forming the background section, is included. Pixels forming a copy are received, pixels having the same density value are counted, and a density histogram as shown in FIG. 17, for example, is created.

Then, the width from the minimum value to the maximum value of the density value of the density histogram is divided into, for example, 30. This is to correspond to the number of neurons NAn (see FIG. 5) of the input layer A of the neural network 4 described later, and the average of the count numbers for each divided region is calculated as the counted number of divided regions.

Next, the density histogram creating section 3 normalizes the created density histogram so that it can be easily learned by the neural network 4. The purpose of learning is to uniformly optimize the W
It is to create a neural network that can set W and WL. Since the density histograms of all the patterns are learned, only the patterns are learned by defining the maximum and minimum values of the density values of the density histogram and the maximum and minimum values of the count number. The normalization of the density histogram is a process for that. For example:

Each count number is converted so that the maximum count number Y1 on the vertical axis of the density histogram as shown in FIG. 3 (a) becomes "1". This figure is standardized in this way.
It is a density histogram as shown in (b). When the density histogram shown in FIG. 4 (a) is standardized as described above, it becomes as shown in FIG. 4 (b).

First, based on the density histogram (the result of dividing the density value width into 30 and standardizing it) created as described above for various types of images, the learning section 6 is used to use a neural network 4 which will be described later. The weight of the synaptic connection of is determined, and then based on the density histogram (the result of dividing the density value width into 30 and standardizing) created as described above for the actual image, The weight is set in the neural network 4 to adjust the actual image contrast.

As shown in the conceptual model of FIG. 5, the neural network 4 is provided with a plurality of neurons N (electronic components that perform the weighted sum processing similar to nerve cells) in layers, and each neuron N is a synapse. It is configured by network connection with a bond called bond C. In this example, a three-layer model of an input layer A, an intermediate layer B, and an output layer C is used. The input layer A includes 30 neurons NA1 to NAn, and the intermediate layer B includes 20 neurons NB1 to NBm. The output layer C is provided with two neurons NC1 and NC2. The number of layers and the number of neurons are not specified,
Any number may be set. The reason why the number of neurons in the output layer C is two is that these neurons NC1 and N
This is because the output value of C2 corresponds to WW and WL. The neural network 4 may be configured by hardware or software.

Next, the process of determining the weight of the synaptic connection C of the neural network 4 using the learning unit 6 will be described with reference to the flowcharts of FIGS. 6 and 7. These flowcharts show the flow of one learning process.

First, in step S1 of FIG. 6, a random number is generated from the random number generator 5 shown in FIG. 1 toward the neural network 3, and the random number value is weighted for the synaptic connection C connecting each neuron N. And With the weights set here as initial values, the following learning process is executed to determine the final weights. For simplification of description, attention is paid to the i-th neuron in each layer as shown in FIG.
Here, i of the neuron NAi in the A layer changes from 1 to n, i of the neuron NBi in the B layer changes from 1 to m,
The i of the C layer neuron NCi changes to 1, 2.

The weight of the synaptic connection connecting the neuron NAi of the input layer A and the neuron NBi of the intermediate layer B set by the random number generator 5 is represented by Wii, and the neuron NBi of the intermediate layer B and the neuron NCi of the output layer C are represented. The weight of the synaptic connection that connects and is represented by Wji.

Each histogram data (count number) of the density histogram of a certain image, which is standardized by the density histogram generation unit 3 of FIG. 1, is input to the neuron NAi of the input layer A. The value of this histogram data is XAi,
This is set as the neuron value of the neuron NAi (step S2).

In step S3, the neuron value XBi of each neuron NBi of the intermediate layer B is calculated by the following formulas (1) and (2). XBi ′ = Σ (XAi × Wii) (1) Σ on the right side of the equation (1) is for obtaining the total sum when i is changed from 1 to n. First, the weight sum with the weight Wi of the synaptic connection is calculated. XBi = f (XBi ′) = 1 / (1 + Exp (XBi ′)) (2) Next, using this equation (2), the sum of weights calculated by equation (1) is converted into a sigmoid function f. Substituting into (X), the neuron value XBi of the neuron NBi is calculated. The sigmoid function is f
It is represented by (X) = 1 / (1 + Exp (X)).

In step S4, as in step S3, the neuron value X of each neuron NCi in the output layer C is calculated.
Calculate Ci. That is, (3), which is similar to Eqs. (1) and (2),
The neuron value NCi is calculated using the equation (4). XCi ′ = Σ (XBi × Wji) (3) XCi = f (XCi ′) = 1 / (1 + Exp (XCi ′)) (4)

Each of these calculated neuron values, XA
i, XBi, and XCi are stored in a memory (not shown) (step S5).

In the neural network 4, the neuron values NCi (i = 1, 2) propagated from the input layer A and appearing in the output layer C as described above correspond to WW and WL, respectively. Since the weights Wii and Wji of the synaptic connection at the time of turning are values set by random numbers, the neuron value XCi output here does not correspond to the optimum WW and WL. In the following process, the weight of synaptic connection is determined by learning based on WW and WL determined by an expert. This learning method is based on a generally used back vacation method (back propagation method).

In step S6, the contrast of the image represented by the density histogram input in step S2 is adjusted by some experts (doctors, operators, etc.), and the WW set by each person at that time is set. Average value of WL ww, wl
Ask for. The standardization processing unit 9 calculates X based on the following equation.
1 and X2 are standardized. This is because the value that the neuron value can take is 0 to 1. NWW = ww / (2 (X2-X1)) NWL = (wl-X1) / (X2-X1) The symbols X1 and X2 are the minimum and maximum density values shown in FIG. 3 (a). Is. The reason why X2-X1 is multiplied by 2 on the right side of the equation for calculating NWW is that NWW takes a value of 0 to 1 even when ww is larger than X2-X1. The obtained NWW and NWL are input to the neuron NCi of the output layer C.

At step S7, the neuron N of the output layer C is
The learning signal DCi is calculated by calculating the difference Eci of the neuron NCi by calculating the difference between the value XCi of Ci and NWW, NWL. Eci = NCi-NWW (NWL). In the case of the neuron NCi of the output layer C, the error Eci calculated here is calculated.
Becomes the learning signal DCi as it is (DCi = Eci)

In step S8, the learning signal DBi is calculated by obtaining the error Ebi that each neuron NBi of the intermediate layer B should bear. The error Ebi is calculated using the following equation (5). Ebi = Σ (Wji × DCi) (5) As shown in FIG. 9, the equation (5) is applied to the neuron NBi of the intermediate layer B and the neuron NCi of the output layer C (i = 1, 1,
2) and weights Wj1 and Wj2 of the synaptic connection connecting
This is an equation for calculating the weighted sum with the learning signal DCi of the neuron NCi of the output layer C. This becomes the error Ebi that the neuron NBi should bear.

Next, the neuron N is calculated using the following equation (6).
A learning signal DBi for Bi is calculated. DBi = Ebi × f ′ (XBi ′) (6) f ′ (XBi ′) in the expression (6) is obtained by adding the differential expression f ′ (X ′) of the sigmoid function in the step S3. The equation is obtained by substituting the obtained weight sum XBi '= Σ (XAi × Wii) of the neuron NBi and multiplying the obtained result by the error Ebi obtained in step S8. Thus, the learning signal DBi of each neuron NBi of the intermediate layer B is calculated.

In step S9, the learning signal DAi is calculated by similarly obtaining the error Eai to be borne by each neuron NAi of the input layer A. That is, the following equations (7) and (8) similar to the equations (5) and (6) are used. Eai = Σ (Wji × DBi) (7) DAi = Eai × f ′ (XAi) (8)

Referring to FIG. The error Eai to be borne by the neuron NAi of the input layer A is the weight Wii (i = 1 to m of the second letter) of the synaptic connection connecting the neuron NAi and the neuron NBi (i = 1 to m) of the intermediate layer B, Learning signal DB of neuron NBi calculated in step S8
It is obtained by the sum of loads with i (i = 1 to m). The learning signal DAi is obtained by multiplying the obtained error Eai by the derivative of the sigmoid function f ′ (XAi), where XAi is multiplied by the value of the histogram data (count number) input to the neuron NAi.

In step S10, the weight Wi of the synaptic connection C is calculated using the above-calculated learning signal and neuron value.
Change i and Wji. The following equations (9) and (10) are used to change the weights. ΔWii (n + 1) = α (DBi × XAi) + β (ΔWii (n)) (9) In the expression (9), the symbol n indicates the number of changes, α is a learning constant set arbitrarily, β indicates an arbitrarily set stabilization constant. ΔWii (n + 1) indicates the current weight correction amount, and ΔWii (n) indicates the previous weight correction amount. Now, since it is the first learning, there is no weight value changed last time, and ΔWii (n) is zero. That is, the weight Wii connecting the neuron NAi of the input layer A and the neuron NBi of the intermediate layer B is changed by the value calculated by α (DBi × XAi).

ΔWji (n + 1) = α (DCi × XBi) + β (ΔWji (n)) (10) This equation (10) is also the same, and the neuron NCi of the intermediate layer B and the neuron NCi of the output layer C are the same. The weight Wji of the synaptic connection connecting and is changed by the value calculated by α (DCi × XBi).

As described above, the weight of the synaptic connection connecting the neurons of each layer is changed once to obtain 1
Learning is completed. In the second learning, the weight correction amount obtained by the above processing is ΔWii (n), (9), (10)
It is calculated by substituting for ΔWji (n). In the process of repeating this many times, the value of the error Eci is monitored, and if all the learning images have sufficiently small errors, learning ends.

As described above, the purpose of this learning is to uniformly set the optimum WW and WL for the density histograms of all patterns. Therefore, a plurality of images represented by different density histograms are prepared, these images are learned many times, and final weights Wii and Wji of synaptic connection are determined. At this time, it is desirable to learn by changing the input order of different images. For example, when 10 images are input one after another and one learning is completed, the arrangement of 10 images is changed and input at the time of the second learning. This is because the neural network 4 learns the sequence of images and determines the synaptic connection weights accordingly.

When the contrast adjustment of the image is actually performed using the learned neural network 4,
It must be assumed that images having different density values are input at random. For this purpose, it is preferable to perform learning by changing the arrangement of images.

When the learning process is completed and the weights Wii, Wji of the synapse connection of the neural network 3 are determined, the contrast adjustment of the actual image is performed using the neural network 3. First, the display image data is input to the density histogram creation unit 3 shown in FIG. 1 to create a standardized density histogram, and this is input to the neural network 4.

The neural network 4 executes the processing from step S2 to step S4 of the flow chart shown in FIG. 6 to obtain the values XC1 and XC2 of the neurons NC1 and NC2 of the output layer C. This neuron value XC
1, XC2 are the values of NWW and NWL.

However, the NWW and NWL obtained here are
Are WW and WL for the standard density histogram. Since the image actually displayed on the CRT display 8 is not standardized in such a density histogram, the denormalization processing unit 10 performs the following (in order to correspond the obtained NWW and NWL to it): Convert using Eqs. 11) and 12). WW '= 2NWW (X2-X1) ... (11) WL' = NWL (X2-X1) + X1 ... (12)

The symbols X1 and X2 in the above equation represent the minimum and maximum density values shown in FIG. The obtained values WW ', WL' are the window width and window level for the actual image.

The obtained WW 'and WL' are sent to the contrast setting section 7 shown in FIG. The contrast setting unit 7 sets the WW 'and WL' of the input image and sets C
The output is displayed on the RT display 8.

In the case of this embodiment, the extracted 409
Since the pixels forming the background portion are excluded from the six pixels, the number of pixels when creating the density histogram differs for each image, but the density histogram creation unit 3 normalizes the density histogram. So there is no particular problem.

According to the apparatus having the above-described structure, the density histogram is created only for the pixels forming the tissue portion including the diagnosis target region, excluding the pixels forming the background portion, and the density histogram is formed for various images. Since the weight of the neural network 4 is learned and determined based on the density histogram created as described above, the background portion is not added to the determined weight. Then, in the production process, using the load in which the background portion is not added as described above, based on the density histogram created as described above from the captured image, the window
Since the window and window level (WW 'and WL') are output, the obtained window window and window level are those in which the influence of the background portion is reduced. The contrast is adjusted according to the level, and the image displayed on the CRT display 8 is one in which the contrast of the diagnosis target portion is properly adjusted.

Next, the configuration of the second embodiment device of the present invention will be described with reference to FIG. FIG. 11 is a block diagram showing a schematic configuration of the second embodiment device. In the second embodiment apparatus, a section extracting section 11 is provided in place of the pixel extracting section 2 of the first embodiment apparatus, and in the density histogram creating section 3, the image extracted by the section extracting section 11 is described later. For example, the density histogram is created for each of the categories.

In this embodiment, the segment extraction unit 11 corresponds to the pixel / segment extraction unit of the present invention, and the segment extraction unit 11 performs the following processing.

First, as shown in FIG. 12, a plurality of images G (64 × 64 in the figure) taken by the diagnostic imaging device 1 are taken.
= 4096). For example, image G
Is 512 pixels × 512 pixels, each section PA has 64 pixels ((512 ÷ 64) × (512 ÷ 6)
4)).

Next, for each section PA, the average value of the density values of the pixels forming the section PA is obtained, and this average value is used as the density value of each section PA. Here, each section PA is a plurality (4096) extracted in the apparatus of the first embodiment.
The density value of each section PA (the average value of the density values of the pixels forming the section PA) corresponds to the density values of a plurality of (4096) pixels extracted by the apparatus of the first embodiment. Correspond.

Then, with respect to each of these sections PA, each density value is compared with a predetermined threshold value, and the section PA above or below the threshold value is taken out. This allows
The section PA forming the background portion is excluded. The threshold value may be determined by the same method as in the first embodiment.

Next, the density histogram creating section 3 makes one partition P for the partition PA extracted as described above.
A density histogram is created with A as one pixel, and the density width is divided into, for example, 30 and standardized and output.

Since the subsequent steps are the same as those of the apparatus of the first embodiment described above, the duplicated description will be omitted. Even with this configuration, as in the first embodiment, WW and WL in which the influence of the background portion is reduced are required, and an image in which the contrast of the diagnosis target site is properly adjusted is displayed on the CRT display 8. In addition to the above, according to the second embodiment, the following effects can be obtained. That is, in the second embodiment, the image is segmented, and the average value of the pixel density values in the segment PA unit is taken as the density value of the segment PA, so even if there is noise in the image,
The noise is removed. Therefore, in the second embodiment, it is possible to create a density histogram in which the influence of noise is eliminated, and as a result, it is possible to obtain more appropriate WW and WL in which the influence of noise is eliminated.

In the second embodiment described above, each section P
Although the density value of A is the average value of the density values of the pixels forming the section PA, for example, when the density values of the pixels forming the section PA are sorted in descending order, The density value of the section PA may be used.

Next, a third embodiment of the present invention will be described. The basic configuration of the device of the third embodiment is the same as that of the second embodiment.
This is the same as the apparatus of the embodiment (see FIG. 11), but in the division extracting unit 11 of the third embodiment, the density value of each division PA is compared with a predetermined threshold value, and then described below.
The section PA remaining as a result of the point neighborhood expansion processing and the four point neighborhood reduction processing is given to the density histogram creation unit 3.

Specifically, first, as shown in FIG.
Each section PA (PA _{1 to} PA ₄₀₉₆ ) is made the same as the arrangement on the image G, and is expanded in the first memory M1 to obtain each section PA ₁
As a result of comparing the density value of each of PA _{4096 with} a predetermined threshold value, the density value on the first memory of the section PA which is the section PA to be excluded is stored as, for example, “0”, and the other values are stored. The density value on the first memory of the section PA (section PA extracted as a result of comparison with the threshold value) is stored as it is.

Next, from the first memory, one section PA _i is sequentially set as a target section, and the target section PA _i and eight surrounding sections PA _i-65 , PA _i-64 , PA _i-63 , PA _{i- 1} , PA _{i + 1} ,
Take out PA _{i + 63} , PA _{i + 64} , and PA _{i + 65} ,
Among segments, if there is division density value other than "0" in a section, the density value of the target segment PA _i, prior to comparison with a threshold, nuclear constituting the target segment PA _i If there is no classification other than "0" by replacing the average value of the original density values, the density value of the target pixel PA _i is unchanged and the second memory same as the target pixel PA _i of the first memory M1 is used. M
2 Store in the place above. This processing is performed by setting i to 1 to 4096
Do about. As a result, the second memory M2 has 8
The section PA after the point vicinity expansion processing is stored in the same manner as the array on the image G. As shown in FIG. 14A, when the section of the corner of the image G is set as the target section PA _i , the same processing as described above is performed for the three sections indicated by diagonal lines in the vicinity thereof, and FIG. As shown in FIG. 5, when the side section of the image G is the target section PA _i , the same processing as described above is performed for the five sections indicated by the shaded areas in the vicinity.

Next, as shown in FIG. 15, each of the sections PA ₁ to PA ₁ after the 8-point neighborhood expansion processing stored in the second memory M2 is processed.
From PA ₄₀₉₆ , one segment PA _i is sequentially set as a segment of interest, and the segment of interest PA _i and its four segments PA _i-64 , PA _i-1 , P
Take out A _{i + 1} and PA _{i + 64} , and select ₁ out of 4 surrounding sections.
If there is a segment of the density value is also "0" in the division, replacing the density value of the attention category PA _i to "0", if there is no classification of "0", the density value of the attention category PA _i As it is, it is stored in the same location on the third memory M3 as the pixel of interest PA _i of the second memory M1. This process, i is 1 to
4096. Thereby, the third memory M2
In, the section PA after the 4-point neighborhood reduction processing is stored in the same manner as the array on the image G. Note that, as shown in FIG. 16A, when the section of the corner of the image G is set as the target section PA _i , the same processing as described above is performed for the two sections indicated by diagonal lines in the vicinity thereof, and FIG. As shown in image G
When the section of the side of is the target section PA _i , the same processing as described above is performed for the three sections shown by the diagonal lines in the vicinity thereof.

Then, out of the sections PA _{1 to} PA ₄₀₉₆ stored in the third memory M3, the sections other than "0" are taken out and given to the density histogram creating section 3. Subsequent processing is the same as that of the apparatus of the second embodiment.

As described above, by performing the 8-point neighborhood expansion processing and the 4-point neighborhood reduction processing, the contour of the tissue portion including the diagnosis target portion can be smoothly image-processed and formed like islands in the tissue portion. The background portion that should be recognized as the tissue portion can be recognized as the tissue portion, and the density histogram faithful to the image can be created.

[0069]

As is apparent from the above description, according to the present invention, the density histogram used in the learning means and the neural network is created for the pixels or the sections constituting the tissue portion including the diagnosis target portion. Since the weight of the neural network determined by the learning means excludes the influence of the background portion, the neural network can appropriately adjust the contrast of the diagnosis target portion. The window width and window level can be determined.

Further, according to the present invention, since the density histogram is created after the pixels arbitrarily extracted from the image and the unnecessary pixels and the sections are excluded from the sections, substantially,
It is also possible to target an image captured by an imaging region that covers a wider range than that of the conventional device, or an image captured by widely varying the imaging conditions.

[Brief description of drawings]

FIG. 1 is a block diagram showing a schematic configuration of a medical image display apparatus according to a first embodiment of the present invention.

FIG. 2 is a diagram for explaining a threshold value.

FIG. 3 is a diagram illustrating a density histogram normalization process.

FIG. 4 is a diagram for explaining a density histogram standardization process.

FIG. 5 is a diagram conceptually modeling the configuration of a neural network.

FIG. 6 is a flowchart showing the flow of one learning process.

FIG. 7 is a flowchart following FIG.

FIG. 8 is a diagram conceptually modeling a part of the configuration of the neural network.

FIG. 9 is a model diagram of a neural network showing a state of learning by the back-propagation method.

FIG. 10 is a model diagram of a neural network similarly showing a state of learning by the back propulsion method.

FIG. 11 is a block diagram showing a schematic configuration of second and third embodiment devices.

FIG. 12 is a diagram showing a state in which an image is divided.

FIG. 13 is a diagram for explaining an 8-point vicinity expansion process.

FIG. 14 is a diagram for explaining an 8-point vicinity expansion process.

FIG. 15 is a diagram for explaining a 4-point neighborhood reduction process.

FIG. 16 is a diagram for explaining a 4-point neighborhood reduction process.

FIG. 17 is a density histogram for explaining a window width and a window level which are parameters for image contrast adjustment in the conventional art.

[Explanation of symbols]

 1 ... Image diagnostic device 2 ... Pixel extraction part 3 ... Density histogram creation part 4 ... Neural network 5 ... Random number generator 6 ... Learning part 11 ... Classification extraction part PA ... Classification

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification number Internal reference number FI Technical display location G06F 19/00 G06T 1/00 G09G 5/10 B 9377-5H G06F 15/64 400 400 L

Claims (1)

[Claims] 1. A window parameter for arbitrarily setting two parameters for contrast adjustment of an image captured by an image diagnostic device, that is, a window width for setting a width of a density value when the image is displayed, and the window A medical image display device for setting a window level for setting a central value of Widus and a display image,
(A) Each density value of a plurality of pixels arbitrarily extracted from the image captured by the image diagnostic apparatus, or the density value of each section obtained by dividing the image is compared with a predetermined threshold value,
Pixels / section extracting means for extracting pixels or sections having density values above or below the threshold value, and (b)
Density histogram creating means for creating a density histogram of the image with respect to the pixels or sections extracted by the pixel / section extracting means; and (c) a weighted sum calculation of the created density histogram data A medical image, comprising: a neural network that outputs a window and a window level; and (d) a learning unit that learns and determines the weight of the neural network based on density histograms of various images. Display device.