JP2561158B2 - Pattern recognizer - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Industrial field of application) The present invention is based on image data representing a radiation image of a subject, and a tumor shadow corresponding to the tumor of the subject specified on the radiation image is a malignant tumor. The present invention relates to a pattern recognition device that seeks the probability of being a tumor shadow.

(Prior Art) Read the recorded radiation image to obtain image data,
It has been performed in various fields to reproduce and record an image after performing appropriate image processing on the image data. For example, an X-ray image is recorded using an X-ray film having a low gamma value designed to be compatible with later image processing, and this X-ray image is recorded.
Contrast and sharpness are obtained by reading an X-ray image from a film on which a line image is recorded, converting it into an electric signal (image data), performing image processing on this image data, and then reproducing it as a visible image on a copy photograph or the like. A reproduced image having good image quality performance such as graininess has been obtained (see Japanese Patent Laid-Open No. 61-5193).

In addition, according to the applicant, radiation (X-ray, α-ray, β-ray, γ
Ray, electron beam, ultraviolet ray, etc.) causes a part of this radiation energy to be accumulated, and then irradiation with excitation light such as visible light causes stimulated emission depending on the accumulated energy. The radiation image of a subject such as a human body is temporarily photographed and recorded on a sheet-shaped stimulable phosphor using a fluorescent substance), and the stimulable luminescent light is scanned by scanning the stimulable phosphor sheet with excitation light such as laser light. Then, the resulting stimulated emission light is photoelectrically read to obtain image data, and the radiation image of the subject is recorded on the basis of this image data, a recording material such as a photographic light-sensitive material, or a CR.
A radiation recording / reproducing system for outputting a visible image on a T display device or the like has already been proposed (JP-A-55-55).
12429, 56-11395, 55-0163472, 56-164
No. 645, No. 55-116340, etc.).

This system has the practical advantage of being able to record images over a very wide radiation exposure area compared to conventional radiographic systems using silver halide photography. In other words, it has been recognized that the amount of light emitted by stimulated emission after storage is proportional to the amount of radiation exposure over a very wide range. Therefore, even if the amount of radiation exposure varies considerably under various imaging conditions, The amount of stimulated emission light emitted from the stimulable phosphor sheet is read and the gain is set to an appropriate value to be read by the photoelectric conversion means to be converted into an electric signal (image data), and this image data is used for photographic exposure. By outputting a radiation image as a visible image on a display device such as a material or a CRT, it is possible to obtain a radiation image that is not affected by variations in radiation exposure dose.

In a system using the above X-ray film, a stimulable phosphor sheet, etc., particularly in a system configured for medical diagnosis of the human body, recently, a reproduced image having good image quality performance suitable for observation (diagnosis) is obtained. In addition to this, automatic image recognition has been carried out (for example, Japanese Patent Laid-Open No. 62-1254).
No. 81). Here, automatic image recognition refers to an operation of extracting a target pattern from a complicated radiographic image by performing various processes on the image data, for example, various lines such as a chest X-ray image of a human body. It refers to an operation to extract a shadow corresponding to a tumor from a very complicated image in which circular and circular patterns are mixed.

Such a complicated radiographic image (for example, chest X of a human body)
A target pattern (for example, tumor shadow) is extracted from a line image or a breast X-ray image, and a visible image that clearly shows the extracted pattern is reproduced and displayed to assist the observer in observing (for example, a doctor's diagnosis). Assistance).

(Problems to be Solved by the Invention) Japanese Patent Application Laid-Open No. 62-125481 discloses, for example,
There is described an apparatus that scans a line image using a specific real space filter, extracts a circular pattern and a linear pattern, and displays the extracted circular pattern as a tumor shadow and the linear pattern as a blood vessel shadow.

By the way, the radiation image of the human body is usually very complicated, and the tumor shadows appearing in, for example, a chest X-ray image or a breast X-ray image also have different sizes, and the tumor shadows are different. A tumor shadow that is variously deformed, such as the shape and density of which differs depending on the position of the X-ray image in which appears. Therefore, the tumor shadow is determined, for example, in the above-mentioned JP-A-62.
As the circular pattern extraction filter as described in Japanese Patent No. -125481, it is necessary to use a filter having a certain degree of tolerance so that even a tumor shadow that is variously deformed can be obtained.

However, on the other hand, if a tumor shadow extraction filter with a large tolerance is used, not only the tumor shadow of the malignant tumor that originally needs to be extracted but also the tumor shadow of the crying benign tumor that needs to be extracted will be extracted. Become.

In view of the above circumstances, the present invention provides a pattern recognition device that seeks a probability that a tumor shadow obtained by scanning on a radiation image using a real space filter is a tumor shadow of a malignant tumor. It is intended.

(Means for Solving the Problem) FIG. 1 is a block diagram clearly showing the configuration of the pattern recognition apparatus of the present invention.

First, the image data representing the radiation image of the subject in the spectrum calculation means 1 is polar coordinates (r, θ) with the approximate center of the tumor shadow corresponding to the tumor of the subject specified on the radiation image as the origin (where r is The radius vector, θ represents the tilt angle.), And the image data D (r, θ) is subjected to frequency analysis using r and θ as variables to obtain the data D.
The spectral distribution of (r, θ) is obtained.

Here, “frequency analysis using r and θ as variables” means that r and θ are coordinate axes orthogonal to each other.
It means obtaining the spectral distributions in the r direction and the θ direction, and is not limited to a specific calculation method, but for example, one-dimensional Fourier transform is performed in each of the r direction and the θ direction, and the maximum entropy method (MEM method) (For example, refer to "Statistical Library Spectral Analysis" by Mikio Hino, published by Asakura Shoten Co., Ltd.) to obtain spectra in the r and θ directions, and to perform two-dimensional Fourier transform on r and θ. .

Next, this spectral distribution is input to the characteristic amount computing means 2, and the characteristic amount computing means 2 obtains a characteristic amount representing this spectral distribution based on this spectral distribution.

Here, the “feature amount” refers to an amount represented by the spectral distribution so that the tumor shadow corresponding to this spectral distribution is a benign tumor or a malignant tumor. Refers to the centroid (first moment), second moment, etc. of the spectral distribution in space.

When the feature quantity is calculated by the feature quantity calculation means 2, the feature quantity is input to the determination means 3. In the judging means 3,
Probability that the tumor shadow is a tumor shadow of a malignant tumor is required based on this feature amount.

(Effect) The tumor shadow of a benign tumor typically has a shape close to a circle that is approximately point-symmetrical, and the density in the tumor shadow is uniform, that is, the value of image data changes smoothly. The tumor shadow of a malignant tumor is characterized in that the density in the tumor shadow is inhomogeneous, that is, the value of image data changes drastically and its shape is irregular. The present invention takes advantage of this feature.

The pattern recognition device of the present invention is a spectrum calculation means 1
First, the image data of the tumor shadow is represented by polar coordinates (r, θ) with the origin approximately at the center of the tumor shadow, and the image of the tumor shadow is considered by assuming that r and θ are mutually orthogonal coordinate axes. The data D (r, θ) was subjected to frequency analysis using r and θ as variables, and its spectral distribution was determined. Therefore, the density of the tumor shadow is uniform or inhomogeneous, the shape of the tumor shadow is not shaped, and The degree of shaping will be noticeable in this spectral distribution.

Further, in the pattern recognition apparatus of the present invention, the feature amount calculating unit 2 obtains the feature amount representing the spectral distribution, and the determining unit 3 obtains the probability that the tumor shadow is the tumor shadow of the malignant tumor based on the feature amount. Therefore, the feature of whether the tumor is a benign tumor or a malignant tumor, which is shown in the spectral distribution, is summarized in the above feature amount, and the probability that the target tumor shadow is the tumor shadow of the malignant tumor can be easily obtained.

(Examples) Examples of the present invention will be described below with reference to the drawings. Here, an example will be described in which the above-described stimulable phosphor sheet is used to detect the shadow of a tumor that typically occurs in the human breast in a substantially spherical shape. This tumor appears as a generally circular tumor shadow that is whitish (low in density) on the visible image based on the image data S1 to be described later.

 FIG. 2 is a schematic diagram of an example of an X-ray imaging apparatus.

The X-ray source 11 of the X-ray imaging apparatus 10 emits X-rays 12 toward the breast 13a of the human body 13, and the X-rays 12a that have passed through the human body 13 are emitted to the stimulable phosphor sheet 14. A transmission X-ray image of the breast 13a is accumulated and recorded on the sheet 14.

FIG. 3 is a perspective view showing an example of the X-ray image reading apparatus and a computer system which is an example of the pattern recognition apparatus of the present invention.

The stimulable phosphor sheet 14 on which the breast X-ray image is recorded is X
The line image reading device 20 is set at a predetermined position. The stimulable phosphor sheet 14 set at this predetermined position is the motor 21
Sheet conveying means such as endless belt driven by
By 22 the sheet is conveyed (sub-scanned) in the direction of arrow Y. on the other hand,
A light beam 24 emitted from a laser light source 23 is reflected and deflected by a rotating polygon mirror 26 driven by a motor 25 and rotating at a high speed in the arrow direction, and after passing through a focusing lens 27 such as an fθ lens, the optical path is changed by a mirror 28. The light is incident on the sheet 14 and the main scanning is performed in the arrow X direction substantially perpendicular to the sub-scanning direction (arrow Y direction). From the portion of the sheet 14 irradiated with the excitation light 24, the stimulated emission light 29 of a light amount corresponding to the accumulated and recorded X-ray image information is emitted, and this stimulated emission light 29 is used as a light guide.
It is guided by 30 and is detected photoelectrically by a photomultiplier (photomultiplier tube) 31. The light guide 30 is formed by molding a light guide material such as an acrylic plate, and the incident end face 30a having a linear shape has a stimulable phosphor sheet 14.
The light receiving surface of the photomultiplier 31 is coupled to the emission end face 30b which is arranged so as to extend along the upper main scanning line and is formed in an annular shape. The photostimulated luminescent light 29 that has entered the light guide 30 from the incident end face 30a proceeds by repeating total reflection inside the light guide 30, is emitted from the emitting end face 30b, is received by the photomultiplier 31, and is an X-ray image. Photostimulated luminescent light 29 representing the light is converted into an electric signal by the photomultiplier 31.

The analog output signal S0 output from the photomultiplier 31 is logarithmically amplified by a logarithmic amplifier 32, and the A / D converter 3
Image data S1 digitized by 3 and as an electrical signal
Is obtained.

The obtained image signal S1 is input to the computer system 40. The computer system 40 constitutes an example of the pattern recognition device of the present invention, and includes a main body 41 having a CPU and an internal memory built therein, a drive unit 42 into which a floppy disk as an auxiliary memory is inserted and driven, and an operator. Is composed of a keyboard 43 for inputting necessary instructions and the like to the computer system 40 and a CRT display 44 for displaying necessary information.

FIG. 4 is a block diagram showing processing based on the image data S1 in the computer system 40 shown in FIG. Here, in the present embodiment, each means shown in each block 40a-40h is performed using software in the computer system 40, but hardware for satisfying the function of each means shown in each block 40a-40h, A combination of software is considered as each means shown in each block 40a to 40h.

The image data S1 is first input to the noise removing means 40a of the computer system 40, and high frequency noise unnecessary for representing an X-ray image is removed. Process of removal includes, for example, image data S1 ₀ of the predetermined point P _0, the image data of the pixel points around the said predetermined fixed point _{P 0 S1 i (i = 0,1,2} ,
..., n) is averaged Then, the X-ray image is scanned by using a filter that sets the image data after the averaging process as the new image data S2 ₀ at the predetermined point P ₀ .

The image data S2 from which the high frequency noise has been removed is then input to the trend removing means 40b. The trend removing means 40b removes low frequency noise of the input image data S2. To remove this low-frequency noise, if the tumor shadow exists in an area with a density gradient, the density gradient can be eliminated and the image data can be converted as if the tumor shadow exists in a flat density area. included. Process of removal include, for example, image data S2 ₀ of the predetermined point P _0, the image data S2 _j of a number of pixel points over a fairly wide range around the said predetermined fixed point P ₀ (j = 0,1,2, ... , m) averaging process The performed, subtracting the value obtained by the averaging processing from the image data S2 ₀ of the predetermined point P _0, the image data after the subtraction as a new image data S3 ₀ of a predetermined point P ₀ filter, i.e., This is performed by scanning the X-ray image using a filter that performs the calculation of.

The image data S3 from which the high-frequency noise and the low-frequency noise have been removed as described above is then processed by the tumor shadow extraction means 40c.
Is input to The tumor shadow extraction means 40a extracts the tumor shadow as follows.

FIG. 5 is a view virtually drawn on a predetermined pixel P ₀ on the X-ray image centering on the image in order to explain an example of the real space filter for extracting the tumor shadow. It is determined whether the predetermined pixel P ₀ is a pixel in the tumor shadow. In the tumor shadow extraction means 1, the tumor shadow on the X-ray image is extracted by scanning the X-ray image using the filter shown here.

As shown in FIG. 5, a plurality of (here, eight) line segments extending from a predetermined pixel P _{0 in the} X-ray image to the periphery of the X-ray image.
Assuming L _i (i = 1, 2,..., 8), a circle R _j (j = ₁ , ₂ , _{3) with} a radius r ₁ , r ₂ , and r ₃ centered on a predetermined pixel P ₀ )
Assume Image data of a predetermined pixel P ₀ is set to f _0, and each pixel P _ij (P ₁₁ , P ₁₂ , P ₁₃ , P _{51 in} FIG. 5) located at each intersection of each line segment L _i and each circle R _j . , the image data of are shown symbols for P _52, P _53.) and f _ij.

Here, the difference delta _ij between the image data f ₀ of a given pixel P ₀ and the image data f _ij of each pixel P _ij is calculated according to the following equation (2).

_{Δ ij = f ij -f 0 ......} (2) (i = 1,2, ..., 8; j = 1,2,3) then for each line segment L _i, difference calculated by the equation (2) The maximum value of Δ _ij is obtained. That is, when an example line segments L _1, L _5, for the line segment L _1, pixel P _11, P _12, P each difference corresponding to _{13 Δ 11 = f 11 -f 0} Δ 12 = f 12 - The maximum value of f ₀ Δ ₁₃ = f ₁₃ −f ₀ is obtained. In this example, for example, Δ ₁₁
Is the maximum value. For line segment L ₅ , pixels P ₅₁ , P ₅₂ ,
The maximum value of each difference corresponding to P ₅₃ Δ ₅₁ = f ₅₁ −f ₀ Δ ₅₂ = f ₅₂ −f ₀ Δ ₅₃ = f ₅₃ −f ₀ , for example, Δ ₅₃ is obtained.

Thus the predetermined pixels P _0, by to obtain the maximum value of the difference between the plurality of pixels for each line segment L _i, it is possible to cope with the tumor shadow of various sizes.

Next, the person sets the two line segments extending in opposite directions from a given pixel P _0, i.e. the line segment L ₁ and the line segment L _5, a line segment L ₂ and the line segment L _6, a line segment L ₃ With the line segment L ₇ and each of the line segment L ₄ and the line segment L ₈ as one set, the average value of the two maximum values (M ₁₅ , M ₂₆ , M ₃₇ , and M ₄₈ , respectively) is obtained for each set. . For the set of line segment L ₁ and line segment L ₅ , the average value M ₁₅ is Is required.

In this way, by treating two line segments extending in the opposite directions from the predetermined pixel P ₀ as one set, the tumor shadow is present in a region where the density is too high to be removed by the trend removing means 40b. Even if the distribution of image data is asymmetric, the tumor shadow can be reliably detected.

If the average value _{M 15, M 26, M 37} , M 48 as described above is determined, on the basis of these average values _{M 15, M 26, M 37} , M 48 , as described below, predetermined characteristic value C ₁ of the pixel P ₀ is used for determining whether or not a pixel of tumor Kagenai is required.

FIG. 6 is a diagram for explaining how to obtain the characteristic value C ₁ . The horizontal axis represents the average values M ₁₅ , M ₂₆ ,
M ₃₇ , M ₄₈ , the vertical axis is each evaluation value corresponding to these average values
C ₁₅ , C ₂₆ , C ₃₇ , and C ₄₈ .

If the average value M ₁₅ , M ₂₆ , M ₃₇ , M ₄₈ is smaller than a certain value M _{1, the} evaluation value is zero, and if it is larger than a certain value M _{2, the} evaluation value is 1.0, M _{1 to} M _1
In the middle of _2, the evaluation value is a value between 0.0 and 1.0 depending on the magnitude of the value. In this way, the average values M ₁₅ , M ₂₆ ,
Evaluation values C ₁₅ , C ₂₆ , C ₃₇ , C ₄₈ corresponding to M ₃₇ , M ₄₈ , respectively, are obtained, and the sum of these evaluation values C ₁₅ , C ₂₆ , C ₃₇ , C ₄₈ C ₁ = C ₁₅ + C ₂₆ ＋ C ₃₇ ＋ C ₄₈ …… (4) is regarded as the characteristic value C ₁ . That is, this characteristic value C ₁ is the minimum value 0.0
And any value between 4.0 and the maximum value of 4.0.

This characteristic value C ₁ is compared with a predetermined threshold value Th ₁ , and C ₁ ≧ Th ₁
Or C ₁ <Th ₁ , it is determined whether or not the predetermined pixel P ₀ is a pixel in the tumor shadow.

The tumor shadow appearing in the X-ray image is extracted by scanning the X-ray image using the above real space filter. When the tumor shadow is obtained in this way, the image data S3 having the smallest value in each tumor shadow is searched, and the pixel corresponding to this image data S3 is set as the center of the tumor shadow.

FIGS. 7A and 7B are diagrams showing typical shapes of a benign tumor and a malignant tumor, respectively, obtained by using the real space filter. However, at this stage, it has not been determined whether the tumor is a benign tumor or a malignant tumor. In each figure, the origin 0 where the x axis and the y axis intersect is the center point obtained as described above.

As described above, a benign tumor typically has a nearly circular shape centered on the center point 0 as shown in FIG. 7A, and the image data S3 in the tumor shadow is smooth. Is changing. Further, the malignant tumor has an irregular shape as shown in FIG. 7B, and the image data S3
Has changed significantly.

When the tumor shadow is extracted by the tumor shadow extraction unit 40c and the center point thereof is obtained, the image data S3 is input to the polar coordinate conversion unit 40d together with the position information of the tumor shadow. In the polar coordinate conversion means 40d, at least the image data S near the tumor shadow
The polar coordinates of 3 are displayed.

FIGS. 8A and 8B show the tumor shadows shown in FIGS. 7A and 7B in polar coordinates (r, θ), and the variables r and θ in the polar coordinates are shown as coordinate axes orthogonal to each other. is there.

In the case of the image data S3 of a benign tumor, it is typically almost constant irrespective of the value of θ, so that it has a shape that extends substantially uniformly in the θ direction as shown in FIG. Image data of malignant tumor
S3 also changes in the θ direction, and thus has a shape that changes along the θ axis as shown in FIG. 8B.

The image data S3 displayed in polar coordinates (the image data S3 in the vicinity of each tumor shadow) is then input to the frequency analysis means 40e. In the present embodiment, the frequency analysis means 40e subjects the image data S3 to a two-dimensional Fourier transform with r and θ as variables, and obtains the spectral distribution F (u, v) of the image data S3 for each tumor shadow. To be In the present embodiment, the combined function of the polar coordinate conversion means 40d and the frequency analysis means 40e is regarded as the spectrum calculation means in the present invention.

As described above, the spectral distribution F (u,
When v) is obtained, this spectral distribution F (u, v) is input to the feature amount calculation means 40f, and the following equations (5) and (6) are obtained. On the u-v space of the spectral distribution F (u, v) according to
The respective second moments U and V in the u direction and the v direction are obtained. In this embodiment, the second moment (U, V) corresponds to the feature amount of the present invention.

9A and 9B are benign tumors (see FIGS. 7A and 8A) and malignant tumors (see FIGS. 7A and 8A) plotted on U-V coordinates.
FIG. 7B is a diagram showing the coordinates (U, V) of the second moment of FIG. 7B and FIG. 8B).

In the pattern of the benign tumor shown in FIG. 8A, there is a gradual change in the r direction and almost no change in the θ direction, so that the position is near the origin 0 on the U axis on the UV coordinate. There are coordinates (U, V) of the second moment. In the malignant tumor pattern shown in FIG. 8B, the image data S3 changes drastically in both the r direction and the θ direction, and therefore the coordinates of the second moment (U,
V) exists at a position away from the origin 0.

When the secondary moment (U, V) is obtained by the characteristic amount calculating means 40f, information on the secondary moment (U, V) is input to the characteristic amount determining means 40g. Based on the information of the second moments (U, V), the feature amount discriminating means 40g calculates the probability that the target tumor shadow is a tumor shadow of a malignant tumor, thresholds it, and the tumor shadow becomes malignant. It is determined whether the tumor shadow is a tumor shadow or a benign tumor shadow.

Figure 10 shows the malignant tumor that appeared in the X-ray image of the human breast.
It is a figure showing the experimental data which calculated | required the 2nd moment (U, V) by the same process as the above about 14 cases and 15 benign tumors, and was plotted on the UV coordinate. The circles are the coordinate points (U, V) of the second moment of the tumor shadow of a benign tumor, and ●
The mark indicates the coordinate point (U,
V).

As shown in this figure, the coordinate points (U,
V) is located near the origin O, and the coordinate point (U, V) of the tumor shadow of the malignant tumor is located away from the origin O.

Here, in most cases, it is a benign tumor or a malignant tumor depending on whether or not the position of the coordinate point (U, V) exists within an arc of a predetermined radius r shown by a broken line from the origin O. Can be determined.

Here, in order to determine whether each tumor is a benign tumor or a malignant tumor, the characteristic value C ₂ is calculated from each coordinate point (second moment) U, V of the coordinate point (U, V). Sought according to. This C ₂ is a coordinate point (U,
It represents the distance to V) and has a large value as it deviates from the origin O.

In the characteristic amount discriminating means 40g, next, the characteristic value C ₂ obtained as described above is compared with a predetermined threshold value Th2 (radius r shown in FIG. 10), and when C ₂ ≧ Th2, this characteristic The tumor corresponding to the value C ₂ is determined to be a malignant tumor, and when C ₂ <Th2, it is considered to be a benign tumor. Note that the characteristic value C ₂ according to equation (7)
And the comparison with the threshold Th2, the benign tumor and the malignant tumor are determined depending on which side of the arc (U, V) of the second moment of the target tumor shadow is on the arc shown in FIG. It is equivalent to making a distinction.

The position information of the tumor shadow determined as a malignant tumor is input to the display unit 40h (CRT display 44 shown in FIG. 3) together with the image data S, and a visible image clearly indicating the position of the malignant tumor is reproduced and displayed. And is used for observation.

FIG. 11 shows the same tumor shadow as that obtained in FIG. 10 in order to verify the efficacy of the present invention (14 malignant tumors, benign tumors).
FIG. 14 is a diagram showing experimental data obtained by plotting the coordinates (U, V) of the second moment similar to FIG. 10 when the polar coordinate conversion means 40d is bypassed as shown by the broken line in FIG. is there. Similar to the case of FIG. 10, the circles represent the coordinate points of the tumor shadow of the benign tumor, and the circles represent the coordinate points of the tumor shadow of the malignant tumor.

Bypassing the polar coordinate conversion means 40d here means that
The two-dimensional Fourier transform was performed in the state shown in FIGS. 7A and 7B without the polar coordinate transformation of the tumor shadow shown in FIGS. 7A and 7B as described above as shown in FIGS. 8A and 8B. Similarly, it means to obtain the second moment in the frequency space and plot the coordinate points. The method without polar coordinate conversion is an application of the method described in, for example, Japanese Patent Laid-Open No. 125675/1989.

Here, in order to confirm the effectiveness of the polar coordinate conversion,
For each of FIG. 10 and FIG. 11, the distance from the origin O to each coordinate point is calculated, the average value and the variance of the distance are calculated for benign tumor and malignant tumor, and the origin of the coordinate point of the benign tumor is calculated. The average value of the distance from O and the variance are ₁ , and
Let σ ₁ ^{2 be} the mean value and the variance of the distance from the origin O of the coordinate point of the malignant tumor to be ₂ and σ ₂ ² , respectively, and then the degree of separation D,
The following formula Is defined. According to this definition, the smaller the separation degree D, the better the separation between benign tumor and malignant tumor.

When the above-mentioned degree of separation D is calculated in the case of polar coordinate conversion shown in FIG. 10, it is D = 1.66 (9), while in the case of not performing polar coordinate conversion shown in FIG. When the degree D was obtained, it was D = 5.28 (10).

By comparing the equations (9) and (10), it can be seen that the polar coordinate conversion greatly contributes to the discrimination between benign tumor and malignant tumor. This is because the tumor shadow of the target malignant tumor is typically a so-called whiskers shadow (spicula), and the characteristics of the whiskers shadow (spicula) are utilized by performing the polar coordinate transformation. is there.

In the above embodiment, the noise removing means 40a and the trend removing means 40b remove high-frequency noise and low-frequency noise, but such processing may be appropriately performed as necessary. The present invention is not an essential constituent element. Further, in the above embodiment, the tumor shadow is extracted by scanning the X-ray image using the real space filter in the tumor shadow extracting means 40c, but the filter for extracting the tumor shadow is limited to the filter described in the above embodiment. In addition, the visible image is first reproduced and displayed on the CRT display 44, and the observer inputs the position information of the tumor shadow shown in the visible image on the CRT display 44 using the keyboard 43. Then, it is possible to determine whether the tumor shadow designated in this way is a benign tumor shadow or a malignant tumor shadow. Further, in the above embodiment, the feature quantity calculating means 40f causes the second moment (U, V) of the spectral distribution F (u, v) to be calculated.
Was calculated, the following equations (11), (12) The center of gravity U ′, V ′ is calculated in accordance with
This center of gravity (U ′, V ′) may be used instead of V),
Various other characteristic quantities representing the spectral distribution F (u, V) may be obtained by various other calculations.

Further, in the above embodiment, the spectral distribution F (u, v) of the image data S3 for each tumor shadow was obtained by performing a two-dimensional Fourier transform with r and θ as variables.
Instead of performing the three-dimensional Fourier transform, the image data S3
Is expressed in polar coordinates (r, θ), and each variable in the polar coordinates
When r and θ are regarded as mutually orthogonal coordinate axes (Fig. 8A,
8B), to obtain the spectrum for each row and each column using the maximum entropy method or the one-dimensional Fourier transform method described above for each row along the r direction and each column along the θ direction of the image. You may In this case, the subsequent processing may be performed as follows, for example. A feature amount representing the spectral distribution of each row and each column (for example,
The first moment (center of gravity), the second moment, the total sum of spectra (total area (sum of power of each frequency) in a graph in which the horizontal axis represents frequency and the vertical axis represents power of each frequency), etc.) are obtained, and the above is obtained. The average value of the feature amounts for each row is the feature amount U ″ in the r direction, and the average value of the feature amounts for each column is θ.
The coordinate point (U ″, V ″) is obtained as the directional feature amount V ″, and then the coordinate point (U ″, V ′) in place of the coordinate point (U, V) or (U ′, V ′) in the above-described embodiment. By using V ″), a benign tumor and a malignant tumor can be discriminated in the same manner as in the above-mentioned embodiment.

Furthermore, in each of the above-mentioned examples, it is determined whether the target tumor shadow is a tumor shadow of a malignant tumor or a tumor shadow of a benign tumor, but before the determination, that is, the characteristic value C ₂ It is possible to display a visible image in which the numerical value itself of (the larger this value is, the more likely it is to be a malignant tumor) is, and to leave the subsequent judgment to the observer.

In addition, the above-mentioned feature amount (U,
V), etc., or after automatically determining whether or not it is a malignant tumor, instead of displaying a visible image, for example, the feature amount or the determination result obtained in this way is used as image data. It may be stored in the storage means together with S3.

Further, although the above-mentioned embodiment is an example of extracting a tumor shadow appearing in a breast X-ray image of a human body obtained by using the stimulable phosphor, the present invention is not limited to the breast X-ray image and the chest X-ray image is not limited. It can be applied to line images and the like, and is not limited to a system using a stimulable phosphor, and the tumor shadow specified on the radiation image based on the image data representing the radiation image of the subject is a malignant tumor. It can be widely used to determine the probability of the tumor shadow.

(Effect of the Invention) As described in detail above, the pattern recognition device of the present invention, in the spectrum calculation means, represents the image data of the tumor shadow in polar coordinates (r, θ) with the approximate center of the tumor shadow as the origin, The spectral distribution of the image data D (r, θ) of the tumor shadow is obtained by performing frequency analysis with r and θ as variables, the characteristic amount calculating means obtains the characteristic amount representative of the spectral distribution, and the determining means uses this. Since the target tumor shadow based on the feature amount is the probability of being a tumor shadow of a malignant tumor, the feature indicating whether it is a benign tumor or a malignant tumor represented in the spectral distribution is aggregated in the feature amount, The probability that the target tumor shadow is a tumor shadow of a malignant tumor can be easily obtained.

[Brief description of drawings]

FIG. 1 is a block diagram clearly showing the configuration of a pattern recognition apparatus of the present invention, FIG. 2 is a schematic view of an example of an X-ray image capturing apparatus, and FIG. 3 is an example of an X-ray image reading apparatus and a book. FIG. 4 is a perspective view showing a computer system which is an embodiment of the pattern recognition apparatus of the invention, FIG. 4 is a block diagram showing processing based on image data in the computer system shown in FIG. 3, and FIG. In order to explain an example of a real space filter for extracting a tumor shadow, a diagram virtually drawn on a predetermined pixel P ₀ on the X-ray image, and FIG. 6 shows a predetermined pixel P _0. Figures 7A and 7B are diagrams for explaining how to determine the characteristic value used to determine whether or not a pixel is in a tumor shadow. Figures 7A and 7B show typical shapes of benign tumor and malignant tumor, respectively. Figures 8A and 8B show the tumor shadows shown in Figures 7A and 7B, respectively. Figures showing the tumor shadows with the coordinates (r, θ) and the variables r and θ in the polar coordinates as mutually orthogonal coordinate axes. FIGS. 9A and 9B are benign tumors plotted on the UV coordinates. (See Figures 7A and 8B), Malignant tumor (Figure 7B)
Fig., Fig. 8B) showing the second moment, Fig. 10 shows the malignant tumor appearing in the X-ray image of the human breast.
Fig. 11 showing the experimental data in which the second moments were obtained and plotted on the U-V coordinates for the 15 examples and the benign tumor. Fig. 11 was obtained to verify the effectiveness of the present invention. It is a figure showing the experimental data which calculated | required the 2nd moment without performing polar coordinate conversion about the same tumor shadow as the case, and was plotted on the UV coordinate. 1 ... Spectrum calculating means 2 ... feature amount calculating means 3 ... judging means, 10 ... X-ray imaging device 14 ... accumulative phosphor sheet 20 ... X-ray image reading device 23 ... laser light source, 26 ... … Rotating polygon mirror 29 …… stimulated emission light, 30 …… light guide 31 …… photomultiplier 40 …… computer system

Claims (1)

(57) [Claims] 1. A pattern recognition device for determining the probability that a tumor shadow corresponding to a tumor of the subject specified on the radiation image is a tumor shadow of a malignant tumor, based on image data representing a radiation image of the subject. Then, in the image data D (r, θ) of the tumor shadow represented by polar coordinates (r, θ) (where r is the radius vector and θ is the tilt angle) with the origin approximately at the center of the tumor shadow. Spectrum calculation means for obtaining a spectral distribution of the image data D (r, θ) by performing frequency analysis with .theta. And .theta. As variables, and representing the spectral distribution based on the spectral distribution obtained by the spectrum calculation means. A pattern recognition apparatus comprising: a feature amount calculating unit that obtains a feature amount; and a determination unit that obtains a probability that the tumor shadow is a tumor shadow of a malignant tumor based on the feature amount.