Image generation device, image generation method, and program

Abstract

To provide an image generation device, an image generation method and program that can generate a virtual scintillation image, can reproduce integral images in simultaneous parallel manner in photon count imaging, can predict an image obtained without fabricating a prototype panel in scintillation imaging, and thus can facilitate diagnosis, shorten a development period and reduce costs.SOLUTION: A photon count image is generated using actual measurement or random numbers, and a virtual scintillation image is generated by applying a convolution filter that reflects light diffusion in a scintillator on the photon count image. This makes it possible to predict imaging performance in transmitted X-ray imaging without actually fabricating a prototype panel. Or, an image similar to normal integral imaging is generated in a photon count imaging device.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention relates to radiation imaging, for example, medical transmission X-ray imaging, and particularly relates to an image generation apparatus, an image generation method, and a program. Background Art

[0002] As an imaging method for flying particles such as transmission X-rays, represented by medical X-ray transmission imaging, integral scintillation imaging using a scintillator and an optical imaging element is the mainstream.

[0003] In addition, photon counting imaging has been proposed in recent years. This method counts the number of particles incident on each pixel of an imaging element to form an image, and a semiconductor material that generates an electrical signal in response to incident particles is used for the imaging element.

[0004] Furthermore, indirect photon counting imaging using a scintillator and an optical imaging element has also been proposed (for example, Patent Document 1). Prior Art Literature Patent Literature

[0005] Patent Document 1 Japanese Unexamined Patent Publication No. 2017-020912 Summary of the Invention Problems to be Solved by the Invention

[0006] However, although the above-mentioned photon counting imaging provides high-contrast clear images, it contains large Poisson noise on the other hand, and the properties of the image are greatly different from those of scintillation images that clinicians are usually accustomed to. Therefore, there is a disadvantage that it becomes difficult for clinicians to make a diagnosis.

[0007] On the other hand, with integral scintillation imaging, the generated image is greatly influenced by the characteristics of the scintillator and image sensor, and predicting this is difficult without actually fabricating an imaging panel by combining these elements and evaluating it by irradiating it with X-rays. This results in disadvantages such as a longer development period and higher costs.

[0008] The present invention aims to provide an image generation device, an image generation method, and a program that can generate virtual scintillation images, reproduce integrated images simultaneously and in parallel in photon count imaging, and predict images obtained in scintillation imaging without the need for panel prototyping, thereby facilitating diagnosis, shortening development time, and reducing costs. [Means for solving the problem]

[0009] An image generation apparatus according to a first aspect of the present invention includes: a photon count image generation unit that generates a photon count image of a flying particle using measured values ​​or random numbers; a filter generation unit that generates a convolution filter that reflects the internal diffusion state of scintillation light in an arbitrary scintillator; and a scintillation image generation unit that generates a virtual scintillation image of a flying particle by applying the convolution filter to the photon count image.

[0010] A second aspect of the present invention relates to an image generation method which includes: a photon count image generation step of generating a photon count image of a flying particle using measured values ​​or random numbers; a filter generation step of generating a convolution filter that reflects the internal diffusion state of scintillation light in an arbitrary scintillator; and a scintillation image generation step of generating a virtual scintillation image of a flying particle by applying the convolution filter to the photon count image.

[0011] A third aspect of the present invention is a program for causing a computer to execute each step of an image generation method. [Effects of the Invention]

[0012] According to the present invention, it is possible to generate virtual scintillation images, reproduce integrated images simultaneously and in parallel in photon count imaging, and predict images obtained in scintillation imaging without the need for panel prototyping. This ultimately leads to easier diagnosis, shorter development times, and reduced costs. [Brief explanation of the drawing]

[0013] [Figure 1] This is a block diagram showing a basic configuration example of an image generation device according to an embodiment of the present invention. [Figure 2] This is a flowchart illustrating the basic concept and image generation procedure of the present invention as adopted in this embodiment. [Figure 3] This is an explanatory diagram of the imaging configuration assumed for generating photon count images. [Figure 4] This is a diagram illustrating the procedure for generating a virtual photon count image. [Figure 5] This is an explanatory diagram (cross-sectional view) of an imaging configuration for imaging a single X-ray photon emission spot. [Figure 6] This figure shows an example of a filter kernel obtained by imaging a single X-ray photon emission spot. [Figure 7] This diagram illustrates the basic procedure for generating a filter kernel for larger pixels from a filter kernel for smaller pixels, simplified to one dimension. [Figure 8] This diagram shows a typical Gaussian filter kernel. [Figure 9] This is a flowchart illustrating the second image generation procedure employed in this embodiment. [Modes for carrying out the invention]

[0014] Hereinafter, embodiments of the present invention will be described in relation to the drawings. Embodiments of the present invention relate to radiation imaging, for example, medical transmission X-ray imaging, and in particular, the present embodiment provides an image generation apparatus and an image generation method for generating a virtual scintillation image obtained by an arbitrary scintillator from a photon-counted image.

[0015] With the use of the present invention, in photon-counting imaging, a clinician can obtain not only a clear photon-counted image but also a scintillation image that they are commonly accustomed to viewing. On the other hand, in scintillation imaging, by generating a virtual photon-counted image and generating a scintillation image therefrom, it becomes possible to predict an obtained image without prototyping an imaging panel. This greatly improves the development efficiency of imaging panels.

[0016] Note that although X-rays are exemplified as radiation in the embodiments of the present invention, the radiation in the present invention may include, in addition to α-rays, β-rays, γ-rays and the like, which are beams formed by particles (including photons) emitted by radioactive decay, beams having energy equal to or higher than that of the foregoing, such as X-rays, particle beams, cosmic rays and the like.

[0017] <Example of Basic Configuration of Image Generating Apparatus> FIG. 1 is a block diagram showing an example of a basic configuration of an image generating apparatus according to an embodiment of the present invention.

[0018] As shown in FIG. 1, the image generating apparatus 100 according to the present embodiment includes a photon-counted image generating unit 110, a filter generating unit 120, and a scintillation image generating unit 130.

[0019] The photon-counted image generating unit 110 generates a photon-counted image of flying particles such as X-rays using actually measured values or random numbers.

[0020] The filter generating unit 120 generates a convolution filter that reflects the internal diffusion state of scintillation light in an arbitrary scintillator.

[0021] The scintillation image generation unit 130 generates a virtual scintillation image of flying particles by processing the generated photon count image with a convolution filter.

[0022] In the example shown in Figure 1, the scintillation image generation unit 130 has a function to generate a realistic scintillation image by adding system noise to the generated virtual scintillation image when the photon count image is generated using random numbers.

[0023] Furthermore, in the image generation device 100, if the photon count image is generated using random numbers, the photon count image generation unit 110 can be configured to generate a realistic photon count image by adding scintillation noise to the generated virtual photon count.

[0024] The basic concepts of the present invention and the image generation procedure adopted in this embodiment will be explained below in relation to the flowchart in Figure 2. Figure 2 is a flowchart illustrating the basic concept and image generation procedure of the present invention as adopted in this embodiment.

[0025] <Basic Concepts and Flowcharts of Invention> This invention virtually generates a scintillation image equivalent to an image obtained by actual integral imaging, without performing integral imaging, using a photon count image and a convolution filter. Figure 2 shows the flowchart. The flow is divided into two cases, CS1 and CS2, depending on the application.

[0026] The first case, CS1, is compatible with a photon count type imaging device. The photon count type imaging device using the present invention simultaneously generates photon count images and integral type images, which are commonly seen by clinicians, and provides them as a set.

[0027] The second case, CS2, primarily contributes to the development of imaging panels for integral imaging. The imaging panel is realized by combining an arbitrary scintillator with an optical imaging element. However, the resulting X-ray transmission images varied greatly depending on the X-ray dose, beam quality, scintillator emission performance, light diffusion within the scintillator, pixel size of the optical image sensor, and noise performance of the optical image sensor, making prior prediction difficult. This invention enables the accurate and simple generation of a virtual integral image that comprehensively reflects these elements, thereby allowing for the prediction of panel performance.

[0028] The first step (ST1-ST4) in the flow chart of Figure 2 is the process related to the generation of a photon count image in the photon count image generation unit 110. Here, a photon count image is an image or image data composed of the count of flying particles incident on each pixel region divided at predetermined intervals.

[0029] In the first case, CS1, the photon count image is acquired through actual measurements using a photon count type imaging device. On the other hand, in the second case, CS2, it is virtually generated using the calculated transmitted dose and random numbers. The detailed procedure will be explained later.

[0030] The second step in the flow (ST5, ST6) is the generation of a convolution filter that reflects the light diffusion within the scintillator in the filter generation unit 120 and the scintillation image generation unit 130, and the application of the convolution filter to the generated photon count image (ST6). These convolution filters are generated based on actual measurements for each scintillator, and they will differ depending on the assumed pixel size of the optical image sensor. The detailed procedure for generating such filters and their application will be described later.

[0031] The scintillation image obtained at this point (ST7) is an ideal scintillation image free from system noise associated with the readout of the optical image sensor, etc. In this case, CS1, the first case, provides clinicians with high-contrast photon count images as well as low-noise integrated images with the same pixel size.

[0032] On the other hand, when predicting the performance of the imaging panel in the second case, CS2, system noise, including the readout noise of the optical image sensor, is uniformly added to each pixel of this image. This generates a realistic scintillation image.

[0033] For example, if a scintillator manufacturer provides a convolution filter for each scintillator they sell, an imaging panel manufacturer can virtually generate scintillation images for any combination of optical imaging sensors and predict their performance.

[0034] <Example of generating a virtual photon count image> Here, we will explain the procedure for generating a virtual photon count image using random numbers in the first step ST1 to ST5 of the flowchart in Figure 2. Figure 3 is an explanatory diagram of the imaging configuration assumed for generating photon count images. Figure 4 is a diagram illustrating the procedure for generating a virtual photon count image.

[0035] Figure 3 shows an example of an imaging configuration for a hypothetical photon count imaging. In Figure 3, 1 represents the image sensor, 2, 2b_1, and 2b_2 represent pixels, 3 represents the subject, 4 and 4b represent the test pattern region, 5 represents the background region, and 6 represents the X-rays. Figure 3 shows how X-rays 6 pass through the subject 3 and reach the image sensor 1. Although actual X-rays have a broadened beam, for simplicity, they are described here as parallel light.

[0036] Subject 3 is, for example, a performance test phantom, in which a disk-shaped test pattern 4 with a transmittance of 40% is formed on a background 5 with an X-ray transmittance of 30%. Multiple pixels 2 are formed in an array on the image sensor 1, and the number of X-ray photons incident on each pixel is counted. X-rays 6 are irradiated onto subject 3 at an average density of 1000 photons per pixel.

[0037] Figure 4 shows the data values ​​of the photon count image generated in the configuration shown in Figure 3, along with the projection diagram 4b of test pattern 4 onto the image sensor. F4_1 represents the average pixel output when an infinite number of irradiations are performed. The average output of pixel 2b_1, which is completely contained within projection 4b of the test pattern, is 300, which is 30% of 1000 photons. The output of pixel 2b_2, which is outside the projection 4b of the test pattern and completely contained within the background, is 400, which is 40% of 1000 photons. Pixels that partially include the projection of the test pattern produce a linear output between 300 and 400, depending on their area occupancy.

[0038] On the other hand, the actual photon count image F4_2 has Poisson noise added to it. This can be achieved, for example, by generating a random number from a Poisson distribution for each pixel output in image F4_1, with the average occurrence count λ being the output of that pixel. For simplicity, one can generate random numbers following a normal distribution with mean λ and standard deviation √λ. Since the image quality of photon count images is dominated by poison noise, reproducing this noise using random numbers is key to generating a virtual image.

[0039] The X-ray transmittance of subject 3 can be easily calculated if the X-ray quality (intensity spectrum) and the material composition of subject 3 are known. Also, since X-rays are usually emitted from a point source, their intensity is inversely proportional to the square of the distance from the source. Therefore, the number of X-rays incident on image sensor 1 may be corrected to account for this.

[0040] <Example of generating a convolution filter> In an embodiment of the present invention, an example of creating a convolution filter corresponding to the light diffusion of an arbitrary scintillator by measurement is described below.

[0041] Figure 5 is an explanatory diagram (cross-sectional view) of the imaging configuration for imaging a single X-ray photon emission spot. Figure 6 shows an example of a filter kernel obtained by imaging a single X-ray photon emission spot.

[0042] The measurement is performed by imaging the emission spot using a single-photon X-ray scintillator. Figure 5 shows a cross-sectional view of the imaging device used for the measurement. In Figure 5, 7 represents the optical image sensor, 8 the imaging surface (pixel array), 9 the adhesive layer, 10 the scintillator plate, 11 the scintillator, 12 the FOP (fiber optic plate), 13 the X-ray 1 photon, and 14 the scintillation emission of the X-ray 1 photon.

[0043] The imaging device consists of a light image sensor (image sensor) 7 with a scintillator plate 10 attached via an adhesive layer 9. The light image sensor 7 has an imaging surface 8 made up of a pixel array. The scintillator plate 10 is made up of a scintillator 11 and a fiber optic plate (FOP) which protects the light image sensor from direct X-ray incidence. When a single X-ray photon 13 enters the scintillator 11, it generates scintillation light 14. The scintillation light 14 is imaged by the optical imaging sensor 7 after light diffusion within the scintillator plate 10. This is recorded as an image of a single X-ray photon emission spot.

[0044] The emission intensity and light diffusion of the scintillation light 14 vary considerably depending on the intensity spectral distribution of the X-rays and the depth at which the X-rays react with the scintillator 11. Furthermore, noise from the optical image sensor 7 and the system, as well as shot noise of the scintillation light, contribute to the fact that each emission spot image will be different. However, by collecting and averaging many such light spot images, a single light intensity profile reflecting the light diffusion of the scintillator plate 10 can be obtained. This is the pixelated light diffusion profile of the scintillator, and a filter kernel is generated using this.

[0045] Figure 6 shows an example of a filter kernel generated by actual measurement. Each filter kernel value (coefficient) is set corresponding to a 71x71 XY pixel arrangement with the center coordinates (0,0). This was generated through the following process. 1. Using an image sensor with 15-micron square pixels and a scintillator plate for mammography, numerous single-photon X-ray emission spots are imaged. 2. The pixel output within the concentric circles at a distance of 35 pixels from the light-emitting center pixel of the bright spot is obtained. The image is then analyzed by taking the difference between the image and the average of multiple dark images taken with the same exposure time to remove the offset due to dark current. Bright spots are automatically extracted by the program, and bright spots whose concentric circles overlap or whose concentric circles extend beyond the frame are removed from the evaluation. 3. For approximately 18,000 bright spots, the pixel output within concentric circles is acquired, and the average value is obtained for each pixel whose relative coordinates from the light source are equal. 4. Divide the output value of each pixel within the averaged concentric circles by the sum of the outputs. This generates a 71x71 filter kernel. (All values ​​outside the concentric circles are set to 0.)

[0046] Such filter kernels are unique to the scintillator plate 10 and the pixel size of the optical image sensor 7 in the imaging device shown in Figure 5. When applying this as a convolution filter to a photon count image, the expected pixel size of the image sensor must match. The filter is applied in the same way as a typical discrete convolution filter. That is, for each pixel in the photon count image, the output of the neighboring pixels is multiplied with the kernel coefficient value according to the relative coordinate from the center of the filter kernel, and the sum of these multipliers is used as the pixel output of the output image.

[0047] That is, if the coefficients of the 71x71 filter kernel are Wi and j, and the pixel output value of the photon count image is I(x,y), then the pixel output value J(x,y) after applying the filter is given by the following equation.

[0048]

number

[0049] <Regarding the generation of additional filter kernels that support large pixels> Incidentally, filter kernels corresponding to larger pixels can be generated numerically from filter kernels corresponding to smaller pixels. Therefore, for example, using the measured results of filter generation described above, it is possible to generate a filter kernel for any optical image sensor with pixels of 15 microns square or larger that correspond to the same scintillator, and to generate a virtual scintillation image.

[0050] Figure 7 shows, for simplicity, an example of generating a filter kernel with a 25-micron spacing from a filter kernel with a 15-micron spacing in one dimension. F7_1 is a one-dimensional filter kernel with a 15-micron spacing. In F7_2, five kernels are arranged evenly from the center, each shifted horizontally by 5 microns to correspond to the central 25-micron region. That is, five kernels are arranged with shifts of -10 microns, -5 microns, 0 microns, 5 microns, and 10 microns. In F7_3, the above kernel was added to regions with 5-micron intervals. F7_4 is further obtained by taking the average of F7_3 at 25-micron intervals from the center and rounding it. If necessary, normalization is performed so that the sum of the coefficients equals 1.

[0051] The key point here is that instead of simply rounding the filter, multiple filter kernels shifted at equal intervals are placed before rounding. Simple rounding would only reflect the case where X-rays were incident only within the central 15-micron region of a 25-micron pixel, degrading the accuracy of the kernel. Shifting the kernels is implemented to address this. This type of processing can be easily achieved through numerical calculations.

[0052] For example, a scintillator manufacturer could obtain the basic kernel of its scintillator from the above-mentioned measurement evaluations and provide an application on the web that generates a filter kernel according to the input of scintillator selection and pixel size. This would allow panel manufacturers to predict the scintillation images obtained with any combination of the above-mentioned scintillator and optical image sensor without having to prototype the panel.

[0053] <Regarding filters applicable to photon-count type imaging devices> Incidentally, when applying a filter to a photon count image obtained with a photon count imaging device, a specific scintillator is not necessarily identified. For example, a general scintillator used in integral imaging for the same purpose can be arbitrarily assumed, and a filter that reflects the light diffusion inside it and is adapted to the pixel size of the imaging device can be applied. In this case, actual measurements are not necessarily required; existing data obtained elsewhere can be reused, for example. Alternatively, it is possible to generate a filter by mimicking such data. The key point here to reproduce the integral images that clinicians are accustomed to seeing is to reflect the general properties of light diffusion inside the scintillator.

[0054] A common filter used for denoising images is the Gaussian filter. Figure 8 shows an example of a Gaussian filter kernel with a standard deviation of three times the pixel width. The denoising effect of this filter is approximately equivalent to that of the scintillator filter kernel shown in Figure 6. The scintillator filter kernel shown in Figure 6 has the following distinct characteristics: Compared to a Gaussian filter with equivalent noise reduction effect, the profile slope is steep near the center of the bright spot, while the slope is gentler farther away, resulting in a long tail. As a result, compared to a Gaussian filter, the values ​​are large near and far from the center of the bright spot, and small in the intermediate range. These filter properties are unique to columnar crystalline scintillators used in high-resolution medical imaging such as mammography.

[0055] A desirable form of the photon-count imaging system using the present invention is to provide a virtual integrated image in addition to the photon-count image, by applying a filter that reflects the light diffusion within the scintillator as described above. Furthermore, an image with a Gaussian filter applied may be provided as an intermediate step. For example, by gradually increasing the standard deviation of the Gaussian filter from a small value, an image with gradually reduced noise is provided from a clear but noisy photon-count image. This then leads to a virtual integrated image that clinicians are familiar with. This provides more information for lesion detection.

[0056] Figure 9 also shows a flowchart illustrating the second image generation procedure used in this embodiment. In the embodiment described above, the first image generation procedure associated with Figure 2 was explained, but it is also possible to use a second image generation procedure, such as the one shown in Figure 9. In other words, since variations in scintillator emission can generate noise, this element may be added as step ST11 (addition of scintillation noise) after the photon count image. For example, if the pixel count value is A, the standard deviation σ of the noise caused by a 20% variation in light emission is σ=(0.2 2 xA) 0.5 This is the result. Therefore, for example, random numbers following a normal distribution with mean 0 and standard deviation σ are generated and added as noise to each pixel value of the photon count image. Note that the output value of the photon count image generated by Poison random numbers is an integer, but the noise added here and the pixel values ​​after noise addition are decimals or floating-point numbers.

[0057] Furthermore, the present invention can also be realized by performing the following process: that is, supplying software (program) that realizes the functions of the above-described embodiment to a system or device via a network or various storage media, and having the computer (or CPU, MPU, etc.) of that system or device read and execute the program.

[0058] By using this invention, in the development of X-ray imaging panels, the performance of any combination of scintillator and optical image sensor can be predicted in advance without the need to fabricate a prototype panel. This makes it possible to significantly reduce development costs and time.

[0059] On the other hand, photon-count type medical imaging devices can provide both photon-count images and integrated scintillation images, which clinicians are familiar with, thus providing more information for lesion detection. [Explanation of symbols]

[0060] 1 Image sensor 2,2b_1,2b_2 pixels 3 Subject 4,4b Test pattern region 5 Background area 6 X-ray 7. Optical Image Sensor 8. Imaging surface (pixel array) 9 Adhesive layer 10 Scintillator Plate 11 Scintillator 12 FOP (Fiber Optic Plate) 13 X-rays, 1 photon 14 X-ray 1-photon scintillation emission 100 Image Generator 110 Photon count image generation unit 120 Filter generation unit 130 Scintillation Image Generation Unit

Claims

1. A photon count image generation unit that generates a photon count image of flying particles using measured values ​​or random numbers, A filter generation unit that generates a convolution filter that reflects the internal diffusion state of scintillation light in any scintillator, A scintillation image generation unit generates a virtual scintillation image of flying particles by applying the above filter to the above photon count image, An image generation device including [specific components].

2. The above-mentioned scintillation image generation unit is: If the above photon count image is generated using random numbers, a realistic scintillation image is generated by adding system noise to the generated virtual scintillation image. The image generation apparatus according to claim 1.

3. The above photon count image generation unit is: If the above photon count image is generated using random numbers, scintillation noise is added to the generated virtual photon count to create a realistic photon count image. The image generation apparatus according to claim 1.

4. The above photon count image generation unit is: If the above photon count image is generated using random numbers, scintillation noise is added to the generated virtual photon count to create a realistic photon count image. The above-mentioned scintillation image generation unit is: Since the above photon count image is generated using random numbers, a realistic scintillation image is generated by adding system noise to the generated virtual scintillation image. The image generation apparatus according to claim 1.

5. The above fill generation unit is, It is possible to generate a filter kernel for larger pixels from a filter kernel for smaller pixels using numerical calculations. After placing multiple filter kernels that are shifted at equal intervals, rounding is performed. The image generation apparatus according to claim 1.

6. A photon count image generation process that generates a photon count image of a flying particle using measured values ​​or random numbers, A filter generation step that generates a convolution filter that reflects the internal diffusion state of scintillation light in an arbitrary scintillator, A scintillation image generation process is performed by applying the above-mentioned convolution filter to the above-mentioned photon count image to generate a virtual scintillation image of flying particles, An image generation method that includes [a specific feature / method].

7. In the above scintillation image generation process, If the above photon count image is generated using random numbers, a realistic scintillation image is generated by adding system noise to the generated virtual scintillation image. The image generation method according to claim 6.

8. The above photon count image generation process is as follows: If the above photon count image is generated using random numbers, scintillation noise is added to the generated virtual photon count to create a more realistic photon count image. The image generation method according to claim 6.

9. In the above fill generation process, It is possible to generate a filter kernel for larger pixels from a filter kernel for smaller pixels using numerical calculations. After placing multiple filter kernels that are shifted at equal intervals, rounding is performed. The image generation method according to claim 6.

10. A program for causing a computer to perform each step of the image generation method described in any one of claims 6 to 9.

Images