Radiation image acquiring device, radiation image acquiring system, and radiation image acquiring method

Abstract

To effectively improve an S / N ratio while enhancing a resolution in a radiation image.SOLUTION: An image acquisition device 1 comprises: a camera 10 which scans and images radiation passing through an object in one direction to acquire an X-ray image; a scintillator layer 11 which is provided on the camera 10 to convert the X-ray into light; and a control device 20 which executes noise removal processing of removing noise from the X-ray image. The camera 10 comprises: N (N is an integer equal to or greater than 2) pixels 72 which are arranged in the direction orthogonal to the one direction, detect the light and output detection signals; and a readout circuit 73 which outputs the X-ray image by outputting the detection signal of each of the N pixels 72. The scintillator layer 11 comprises: P (P is an integer equal to or greater than 2) scintillator parts 11a which are arranged so as to be apart from each other correspondingly to the N pixels 72; and a separation part 11b which is arranged between the P scintillator parts 11a.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] One aspect of the embodiment relates to a radiation image acquiring device, a radiation image acquiring system, and a radiation image acquiring method. [Background technology]

[0002] Conventionally, a device has been used that has multiple rows of line sensors arranged perpendicular to the transport direction of the object, and adds up the detection data output from the multiple rows of line sensors to obtain the distribution of electromagnetic waves such as X-rays that have passed through the object as image data. With such a device, it is possible to obtain an integral exposure effect in the image data that detects the electromagnetic waves that have passed through the object. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Republished Patent Publication WO2019 / 082276 [Patent Document 2] JP 2019-158663 A Summary of the Invention [Problem to be solved by the invention]

[0004] In the conventional device described above, by adding the detection data obtained from multiple rows of line sensors, the signal value in the addition result tends to increase, but the noise value also tends to increase. Therefore, the S / N ratio of the image data may not be sufficiently improved. In addition, the resolution of the image data obtained by using a scintillator may be reduced.

[0005] Therefore, one aspect of the embodiments has been made in consideration of such problems, and has an objective of providing a radiological image acquisition device, a radiological image acquisition system, and a radiological image acquisition method that can effectively improve the S / N ratio while increasing the resolution in a radiological image. [Means for solving the problem]

[0006] A radiological image acquisition device according to one aspect of an embodiment includes an imaging device that scans and captures radiation that has passed through an object in one direction to acquire a radiological image, a scintillator layer provided on the imaging device to convert radiation into light, and an image processing module that performs noise removal processing to remove noise from the radiological image, wherein the imaging device includes N (N is an integer of 2 or more) detection elements arranged in a direction perpendicular to the one direction, which detect light and output detection signals, and a readout circuit that outputs a radiological image by outputting detection signals for each of the N detection elements, and the scintillator layer includes P (P is an integer of 2 or more) scintillator sections arranged separately corresponding to the N detection elements, and a separation section arranged between the P scintillator sections.

[0007] Alternatively, a radiographic image acquisition system according to another aspect of the embodiment includes the radiographic image acquisition device described above, a radiation source that irradiates an object with radiation, and a transport device that transports the object in one direction relative to the imaging device.

[0008] Alternatively, a radiological image acquisition method according to another aspect of the embodiment includes the steps of acquiring a radiological image by scanning and imaging scintillation light corresponding to radiation that has passed through an object in one direction, and performing a noise removal process to remove noise from the radiological image, in which the acquisition step outputs a detection signal for each of the N detection elements using an imaging device including N (N is an integer of 2 or more) detection elements arranged in a direction perpendicular to the one direction and detecting light and outputting a detection signal, and a scintillator layer that converts radiation into light and includes P (P is an integer of 2 or more) scintillator sections arranged separately corresponding to the N detection elements and a separation section arranged between the P scintillator sections, to output a radiological image.

[0009] According to the above aspect or any of the other aspects, scintillation light corresponding to radiation transmitted through the object is detected by N detection elements arranged in a direction perpendicular to the direction of scanning the object, and detection signals from the N detection elements are output, thereby outputting a radiological image. At this time, the radiation transmitted through the object is converted into scintillation light by P scintillator sections arranged separately, and the scintillation light from the P scintillator sections is detected by the N detection elements, respectively. In addition, the output radiological image is subjected to a noise removal process to remove noise. In this way, by using scintillator sections separated corresponding to the detection elements, it is possible to increase the resolution of the radiological image and effectively improve the S / N ratio in the radiological image.

[0010] Here, in the above-mentioned one aspect, it is preferable that the imaging device includes a pixel line having M (M is an integer of 2 or more) pixels arranged along one direction, arranged in N rows (N is an integer of 2 or more) in a direction perpendicular to the one direction, a detection element that outputs a detection signal related to light for each pixel, and a readout circuit that adds detection signals output from at least two of the M pixels for each of the N rows of pixel lines of the detection element and outputs the N detection signals obtained by the addition process, thereby outputting a radiographic image. Also, in the above-mentioned other aspect, it is preferable that the imaging device includes a pixel line having M (M is an integer of 2 or more) pixels arranged along one direction, arranged in N rows (N is an integer of 2 or more) in a direction perpendicular to the one direction, and includes a detection element that outputs a detection signal related to light for each pixel, and the acquiring step adds detection signals output from at least two of the M pixels for each of the N rows of pixel lines of the detection element and outputs the N detection signals obtained by the addition process, thereby outputting a radiographic image.

[0011] In such a configuration, scintillation light corresponding to radiation transmitted through the object is detected by detection elements arranged in N rows of pixel lines, each pixel line having M pixels arranged in the scanning direction of the object, detection signals of at least two pixels among the M pixel detection signals output for each pixel line are added, and the N detection signals obtained by the addition are sequentially output, thereby outputting a radiological image. This makes it possible to effectively improve the S / N ratio in the radiological image by increasing the signal components in the radiological image.

[0012] Furthermore, it is also preferable that the image processing module inputs the radiographic image to a trained model constructed by machine learning using the image data in advance, and executes a noise removal process to remove noise from the radiographic image. Furthermore, it is also preferable that in the executing step, the radiographic image is input to a trained model constructed by machine learning using the image data in advance, and executes a noise removal process to remove noise from the radiographic image. In this case, it is possible to appropriately remove noise components from the radiographic image, and it is possible to further improve the S / N ratio in the radiographic image.

[0013] In the above aspect or the other aspect, it is preferable that the trained model is constructed by machine learning using image data obtained by adding a noise value along a normal distribution to a radiation image of a predetermined structure as training data. This makes it easy to prepare image data that is training data used to construct the trained model, and enables the trained model to be constructed efficiently.

[0014] In addition, it is also preferable that the trained model is constructed by machine learning using a radiation image obtained using the scintillator layer as training data. In this way, it becomes easy to reflect the conditions at the time of image acquisition in the image data, which is the training data used to construct the trained model, and a trained model that realizes appropriate noise removal can be constructed.

[0015] It is also preferable that the image processing module has a noise map generating unit that derives an evaluation value from the pixel value of each pixel of the radiographic image based on the relationship data representing the relationship between the pixel value and the evaluation value evaluating the spread of the noise value, and generates a noise map that is data in which the derived evaluation value is associated with each pixel of the radiographic image, and a processing unit that inputs the radiographic image and the noise map into a trained model and executes a noise removal process to remove noise from the radiographic image. It is also preferable that the executing step derives an evaluation value from the pixel value of each pixel of the radiographic image based on the relationship data representing the relationship between the pixel value and the evaluation value evaluating the spread of the noise value, generates a noise map that is data in which the derived evaluation value is associated with each pixel of the radiographic image, and inputs the radiographic image and the noise map into the trained model to execute a noise removal process to remove noise from the radiographic image. In this case, an evaluation value is derived from the pixel value of each image of the radiographic image based on the relationship data representing the relationship between the pixel value and the evaluation value evaluating the spread of the noise value, and generates a noise map that is data in which the derived evaluation value is associated with each pixel of the radiographic image. Then, the radiographic image and the noise map are input to a trained model previously constructed by machine learning, and a noise removal process is executed to remove noise from the radiographic image. As a result, the spread of noise values ​​evaluated from the pixel values ​​of each pixel of the radiographic image is taken into consideration, and noise in each pixel of the radiographic image is removed by machine learning, and noise removal corresponding to the relationship between the pixel values ​​and the spread of noise in the radiographic image can be realized using the trained model. As a result, noise in the radiographic image can be effectively removed.

[0016] Furthermore, it is preferable that the image processing module has an input unit that accepts an input of condition information indicating either the condition of the radiation source or the imaging condition when the object is imaged by irradiating radiation, a calculation unit that calculates the average energy of the radiation that has passed through the object based on the condition information, and a narrowing down unit that narrows down the trained models to be used for the noise removal process from among a plurality of trained models each constructed in advance by machine learning using image data based on the average energy. It is also preferable that the executing step accepts an input of condition information indicating either the condition of the radiation source or the imaging condition when the object is imaged by irradiating radiation, calculates the average energy of the radiation that has passed through the object based on the condition information, and narrows down the trained models to be used for the noise removal process from among a plurality of trained models each constructed in advance by machine learning using image data based on the average energy. In this case, the average energy of the radiation that has passed through the object is calculated based on the condition of the radiation source or the imaging condition when acquiring a radiation image of the object. Then, based on the average energy, candidates for the trained models to be used for the noise removal are narrowed down from among the trained models constructed in advance. In this way, a trained model corresponding to the average energy of radiation of the imaging target is used for noise removal, so that noise removal corresponding to the relationship between luminance and noise in the radiographic image can be realized, and as a result, noise in the radiographic image can be effectively removed.

[0017] Furthermore, it is also preferable that the image processing module has an identifying unit that identifies image characteristics of a radiographic image acquired by an imaging device with the jig as the target, a selecting unit that selects a learned model from among a plurality of learned models each constructed in advance by machine learning using image data based on the image characteristics, and a processing unit that executes a noise removal process using the selected learned model. It is also preferable that in the executing step, image characteristics of a radiographic image acquired with the jig as the target are identified, a learned model is selected from a plurality of learned models each constructed in advance by machine learning using image data based on the image characteristics, and a noise removal process is executed using the selected learned model. According to this configuration, image characteristics of a radiographic image of the jig are identified, and a learned model to be used for noise removal is selected from among learned models constructed in advance based on the image characteristics. As a result, it is possible to estimate the characteristics of a radiographic image that change depending on the conditions of a radiation source in the system, and the learned model selected according to the estimation result is used for noise removal, so that noise removal corresponding to the relationship between luminance and noise in a radiographic image can be realized. As a result, noise in a radiographic image can be effectively removed.

[0018] It is also preferable that the image processing module performs a filter process on the radiographic image to perform a noise removal process to remove noise from the radiographic image. It is also preferable that the performing step performs a filter process on the radiographic image to perform a noise removal process to remove noise from the radiographic image. With this configuration, it is possible to appropriately remove noise components from the radiographic image, and it is possible to further improve the S / N ratio of the radiographic image.

[0019] Furthermore, it is also preferable that the image processing module performs edge enhancement processing on the radiation image in addition to the filtering processing. Furthermore, it is also preferable that the executing step performs edge enhancement processing on the radiation image in addition to the filtering processing. In this case, in addition to improving the S / N ratio, the contrast of the radiation image may be improved. Effect of the Invention

[0020] According to the embodiment, it is possible to effectively improve the S / N ratio while increasing the resolution of a radiological image. [Brief description of the drawings]

[0021] [Figure 1] 1 is a schematic configuration diagram of an image acquisition device 1 according to a first embodiment. [Diagram 2] 2 is a plan view showing the configuration of the scan camera 12 in FIG. [Diagram 3] 2 is a diagram showing the configuration of a scintillator layer 11 disposed on a scan camera 12. FIG. [Figure 4] 2 is a block diagram showing an example of a hardware configuration of a control device 20 in FIG. 1. [Diagram 5] 2 is a block diagram showing a functional configuration of a control device 20 in FIG. 1. [Figure 6] FIG. 6 is a diagram showing an example of an X-ray image acquired by the image acquisition unit 203 in FIG. 5. [Figure 7] 6 is a diagram showing an example of a noise standard deviation map generated by the noise map generating unit 204 in FIG. 5. FIG. [Figure 8] FIG. 6 is a diagram showing an example of input / output data of the trained model 207 in FIG. [Figure 9] A figure showing an example of a training image, which is one of the training data used to construct the trained model 207. [Figure 10] 10 is a flowchart showing a procedure for creating image data that is teacher data (training data) used by the construction unit 206 to construct the trained model 207. [Figure 11] 4 is a flowchart showing the procedure of an observation process by the image acquisition device 1. [Figure 12] 13 is a block diagram showing a functional configuration of a control device 20A according to a modified example of the present disclosure. FIG. [Figure 13] 13 is a flowchart showing the procedure of an observation process by an image acquisition device 1 according to a modified example of the present disclosure. [Figure 14] 13 is a graph showing an example of a simulation calculation result of an energy spectrum of a transmitted X-ray by the calculation unit 202A of FIG. 12. [Figure 15] 13 is a table showing an example of a simulation calculation result of the relationship between the thickness of the object and the average energy and transmittance, performed by the calculation unit 202A of FIG. 12. [Figure 16] 13 is a graph showing an example of a simulation calculation result of the relationship between the thickness of an object and the X-ray transmittance, performed by the calculation unit 202A of FIG. [Figure 17] 13 is a graph showing an example of a simulation calculation result of the relationship between the thickness of the object and the average energy of the transmitted X-rays, performed by the calculation unit 202A of FIG. [Figure 18] 13 is a graph showing an example of a simulation calculation result of the relationship between pixel values ​​and average energy of an X-ray image, performed by the calculation unit 202A of FIG. 12. [Figure 19] 11 is a graph showing an example of a simulation calculation result of a relationship between a pixel value of an X-ray image and a standard deviation of a noise value. [Figure 20] 13 is a graph showing an example of the relationship between the pixel value and the standard deviation of the noise value when the material of the object changes, which is derived in calculation section 202A of FIG. 12. [Figure 21] FIG. 13 is a block diagram showing a functional configuration of a control device 20B according to another modified example of the present disclosure. [Figure 22] 13 is a flowchart showing the procedure of an observation process by an image acquisition device 1 according to another modified example of the present disclosure. [Diagram 23] 22 is a diagram showing an example of a noise standard deviation map generated by the noise map generating unit 204B in FIG. 21. FIG. [Figure 24] 13 is a perspective view showing an example of the structure of a jig used for imaging in an image acquisition device 1 according to another modified example of the present disclosure. FIG. [Diagram 25] FIG. 25 is a diagram showing an example of a captured image of the jig in FIG. 24. [Figure 26] FIG. 11 is a block diagram showing a functional configuration of a control device 20C according to a second embodiment. [Figure 27]FIG. 27 is a diagram showing an example of image data that is training data used to construct the trained model 206C in FIG. 26. [Figure 28] FIG. 27 is a diagram showing an example of an X-ray transmission image that is an analysis target of a selection unit 204C in FIG. 26. [Figure 29] FIG. 27 is a diagram showing an example of a thickness-brightness characteristic graph acquired by the selection unit 204C of FIG. 26. [Diagram 30] FIG. 27 is a diagram showing an example of a luminance-SNR characteristics graph acquired by the selection unit 204C of FIG. 26. [Diagram 31] FIG. 27 is a diagram for explaining the function of selecting a trained model based on image characteristics by the selection unit 204C in FIG. 26. [Diagram 32] FIG. 27 is a diagram showing an example of an X-ray transmission image used for evaluating the resolution by the selection unit 204C in FIG. 26. [Diagram 33] FIG. 27 is a perspective view showing an example of the structure of a jig used in evaluating the brightness-to-noise ratio by the selection unit 204C in FIG. 26. [Diagram 34] FIG. 34 is a diagram showing an X-ray transmission image obtained for the jig in FIG. 33 after noise removal processing. [Diagram 35] 10 is a flowchart showing the procedure of an observation process using the image acquisition device 1 according to the second embodiment. [Diagram 36] FIG. 13 is a block diagram showing a functional configuration of a control device 20D according to a modified example of the second embodiment. [Figure 37] 13 is a flowchart showing the procedure of an observation process using the image acquisition device 1 according to a modified example of the second embodiment. [Figure 38] FIG. 13 is a block diagram showing a functional configuration of a control device 20E according to a third embodiment. [Figure 39] A figure showing an example of image data that is training data used to construct the trained model 206E of Figure 38. [Diagram 40] 39 is a diagram showing an example of an X-ray transmission image of an analysis target of a specific portion 202E in FIG. 38. FIG. [Diagram 41] FIG. 39 is a diagram showing an example of a thickness-brightness characteristic graph acquired by the specifying unit 202E of FIG. 38. [Diagram 42]FIG. 39 is a diagram showing an example of a luminance-SNR characteristics graph acquired by the specification unit 202E of FIG. 38. [Diagram 43] FIG. 39 is a diagram showing an example of an X-ray transmission image used for evaluating the resolution by the specifying unit 202E in FIG. 38. [Diagram 44] FIG. 38 is a diagram for explaining the function of selecting a trained model based on image characteristics by the selection unit 204E of FIG. 38. [Diagram 45] 39 is a perspective view showing an example of the structure of a jig used in evaluating the brightness-to-noise ratio by the selection unit 204E in FIG. 38. FIG. [Diagram 46] FIG. 46 is a diagram showing an X-ray transmission image obtained for the jig in FIG. 45 after noise removal processing. [Figure 47] 13 is a flowchart showing the procedure of an observation process using the image acquisition device 1 according to the third embodiment. [Figure 48] FIG. 13 is a plan view showing the configuration of a scan camera 12A according to a modified example. [Figure 49] FIG. 2 is a diagram showing the configuration of a scintillator layer 11A disposed on a scan camera 12A. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings. In the description, the same elements or elements having the same functions will be denoted by the same reference numerals, and duplicated description will be omitted. [First embodiment]

[0023] FIG. 1 is a configuration diagram of an image acquisition device 1 which is a radiation image acquisition device and a radiation image acquisition system according to the present embodiment. As shown in FIG. 1, the image acquisition device 1 is a device that irradiates an object F conveyed in a conveying direction TD with X-rays (radiation) and acquires an X-ray transmission image (radiation image) of the object F based on the X-rays that have penetrated the object F. The image acquisition device 1 performs foreign body inspection, weight inspection, and inspection of the object F using the X-ray transmission image, and examples of applications include food inspection, baggage inspection, board inspection, battery inspection, and material inspection. The image acquisition device 1 is configured to include a belt conveyor (conveyor device) 60, an X-ray irradiator (radiation generation source) 50, an X-ray detection camera (imaging device) 10, a control device (image processing module) 20, a display device 30, and an input device 40 for performing various inputs. Note that the radiation image in the embodiment of the present disclosure is not limited to an X-ray image, but also includes an image generated by ionizing radiation other than X-rays, such as gamma rays.

[0024] The belt conveyor 60 has a belt portion on which the object F is placed, and the belt portion is moved in a conveying direction (one direction) TD to convey the object F in the conveying direction TD at a predetermined conveying speed. The conveying speed of the object F is, for example, 48 m / min. The belt conveyor 60 can change the conveying speed to, for example, 24 m / min, 96 m / min, etc., as necessary. The belt conveyor 60 can also change the height position of the belt portion as appropriate to change the distance between the X-ray irradiator 50 and the object F. Note that the object F conveyed by the belt conveyor 60 can include various items such as, for example, meat, seafood, agricultural products, confectionery, and other foods, rubber products such as tires, resin products, metal products, mineral, and other resource materials, waste, and electronic parts, electronic boards, etc. The X-ray irradiator 50 is a device that irradiates (outputs) X-rays as an X-ray source to the object F. The X-ray irradiator 50 is a point light source, and irradiates X-rays in a certain irradiation direction by diffusing them within a predetermined angle range. The X-ray irradiator 50 is disposed above the belt conveyor 60 at a predetermined distance from the belt conveyor 60 so that the X-ray irradiation direction is directed toward the belt conveyor 60 and the diffusing X-rays cover the entire width direction (direction intersecting with the conveying direction TD) of the object F. The X-ray irradiator 50 irradiates a predetermined divided range in the length direction of the object F (direction parallel to the conveying direction TD), and irradiates the entire length direction of the object F with X-rays by conveying the object F in the conveying direction TD on the belt conveyor 60. The tube voltage and tube current of the X-ray irradiator 50 are set by the control device 20, and the X-ray irradiator 50 irradiates X-rays with a predetermined energy and radiation amount according to the set tube voltage and tube current toward the belt conveyor 60. A filter 51 that transmits a predetermined wavelength range of X-rays is provided near the belt conveyor 60 side of the X-ray irradiator 50.

[0025] The X-ray detection camera 10 detects X-rays that have passed through the object F among the X-rays irradiated to the object F by the X-ray irradiator 50, and acquires and outputs a detection signal based on the X-rays. The image acquisition device 1 according to this embodiment sequentially or simultaneously outputs detection signals based on the X-rays that have passed through the object F transported by the belt conveyor 60, thereby scanning the X-ray transmission image in the transport direction TD and outputting the captured X-ray transmission image.

[0026] The X-ray detection camera 10 includes a filter 19, a scintillator layer 11, a scan camera 12, a sensor control unit 13, an amplifier 14, an AD converter 15, a correction circuit 16, an output interface 17, and an amplifier control unit 18. The scan camera 12, the amplifier 14, the AD converter 15, the correction circuit 16, and the output interface 17 are electrically connected to each other.

[0027] Scintillator layer 11 is fixed onto scan camera 12 by adhesive or the like, and converts X-rays transmitted through object F into scintillation light (the detailed configuration will be described later). Scintillator layer 11 outputs the scintillation light to scan camera 12. Filter 19 transmits a predetermined wavelength range of X-rays toward scintillator layer 11.

[0028] Scan camera 12 detects scintillation light from scintillator layer 11, converts it into an electric charge, and outputs it as a detection signal (electrical signal) to amplifier 14. Fig. 2 is a plan view showing the configuration of scan camera 12. As shown in Fig. 2, scan camera 12 includes a plurality of pixels (detection elements) 72, which are photodiodes (photoelectric conversion elements) two-dimensionally arranged on a substrate 71, a readout circuit 73 that outputs to the outside the detection signals output by the plurality of pixels 72 after photoelectrically converting the scintillation light, and wiring sections W that electrically connect between readout circuit 73 and each of the plurality of pixels 72.

[0029] In detail, the scan camera 12 has a configuration in which pixel lines (pixel groups) 74, each consisting of M pixels 72 (M is an integer of 2 or more) arranged along the transport direction TD, are arranged in N rows (N is an integer of 2 or more) in a direction substantially perpendicular to the transport direction TD on a substrate 71. For example, the number of pixels M is 4, and the number of pixel lines N is an arbitrary integer between 200 and 30,000.

[0030] The readout circuit 73 sequentially receives detection signals output from the M pixels 72 for each pixel line 74 at intervals of a predetermined detection period (details will be described later) in response to control by the sensor control unit 13, performs a process (addition process) of adding (adding together) the detection signals of at least two pixels 72 among the detection signals from the M pixels 72, combines the detection signals subjected to the addition process for each pixel line 74, and outputs to the outside as a detection signal of one line of the object F perpendicular to the conveying direction TD. Note that the addition process in this embodiment also includes an averaging process of adding the detection signals of at least two pixels 72 and obtaining an average value. In this embodiment, the readout circuit 73 performs an addition process on all M detection signals. In addition, the readout circuit 73 performs an addition process on the detection signals sequentially output from the M pixels 72 by shifting the predetermined detection period, thereby outputting a detection signal of the next one line of the object F perpendicular to the conveying direction TD. In the same manner, the readout circuit 73 sequentially or simultaneously outputs detection signals of multiple lines of the object F perpendicular to the conveying direction TD.

[0031] The configuration of the scintillator layer 11 arranged on the scan camera 12 having the above configuration will be described. FIG. 3 is a diagram showing the configuration of the scintillator layer 11, with a cross-sectional view along the transport direction TD shown above and a plan view seen from the pixel 72 side of the substrate 71 shown below. In this way, the scintillator layer 11 is formed with M×P (P is an integer of 2 or more) rectangular scintillator sections 11a arranged separately corresponding to the M×N pixels 72 arranged along the transport direction TD and a direction perpendicular to the transport direction TD, and separation sections 11b located between the scintillator sections 11a. The number P may be 2 or more and N or less. Furthermore, the number P may be an integer of 2 or more and an integer obtained by dividing N by an integer. In this case, blurring due to the spread of light can be suppressed according to the interval between the separation sections 11b of the scintillator layer 11. The number P may also be an integer greater than N. In this case, the interval between the separation portions 11b of the scintillator layer 11 becomes smaller than the interval between the multiple pixels 72, but it becomes easier to align the scintillator layer 11 with the multiple pixels 72. In this embodiment, for example, P=N, but is not limited to this.

[0032] The M×P scintillator sections 11a are made of a scintillator material capable of converting incident X-rays into scintillation light, and are arranged to cover the entirety of the corresponding pixels 72. The separation sections 11b are formed in a mesh shape to separate the M×P scintillator sections 11a, and are made of a material capable of blocking scintillation light. The separation sections 11b may also contain a material that reflects scintillation light. Furthermore, the separation sections 11b may also be made of a material capable of blocking radiation. The material constituting such a scintillator layer 11 and the manufacturing method of the scintillator layer 11 may be, for example, the material and manufacturing method described in JP-A-2001-99941 or JP-A-2003-167060. However, the material and manufacturing method of the scintillator layer 11 are not limited to those described in the above documents.

[0033] The sensor control unit 13 controls the scan camera 12 to repeatedly capture images at a predetermined detection period so that all pixels 72 in a pixel line 74 in the scan camera 12 can capture X-rays that have passed through the same region of the object F. The predetermined detection period may be set based on the pixel width of the pixels 72 in the pixel line 74 in the scan camera 12. The predetermined detection period may be set based on a deviation (delay time) in the detection timing of the pixels 72 in the pixel line 74 in the scan camera 12, for example, based on the distance between the pixels 72 in the pixel line 74 in the scan camera 12, the speed of the belt conveyor 60, the distance between the X-ray irradiator 50 and the object F on the belt conveyor 60 (FOD (Focus Object Distance: radiation source-object distance)), and the distance between the X-ray irradiator 50 and the scan camera 12 (FDD (Focus Detector Distance: radiation source-sensor distance)).

[0034] The amplifier 14 amplifies the detection signal at a predetermined set amplification factor to generate an amplified signal, and outputs the amplified signal to the AD converter 15. The set amplification factor is an amplification factor that is set by the amplifier control unit 18. The amplifier control unit 18 sets the set amplification factor of the amplifier 14 based on predetermined imaging conditions.

[0035] The AD converter 15 converts the amplified signal (voltage signal) output by the amplifier 14 into a digital signal and outputs it to the correction circuit 16. The correction circuit 16 performs predetermined corrections, such as signal amplification, on the digital signal and outputs the corrected digital signal to the output interface 17. The output interface 17 outputs the digital signal to the outside of the X-ray detection camera 10.

[0036] The control device 20 is a computer such as a PC (Personal Computer). The control device 20 generates an X-ray transmission image based on digital signals (amplified signals) corresponding to detection signals of multiple lines outputted sequentially or simultaneously from the X-ray detection camera 10 (more specifically, the output interface 17). In this embodiment, the control device 20 generates one X-ray transmission image based on digital signals of 128 lines outputted from the output interface 17. The generated X-ray transmission image is outputted to the display device 30 after noise removal processing described later, and is displayed by the display device 30. The control device 20 also controls the X-ray irradiator 50, the amplifier control unit 18, and the sensor control unit 13. The control device 20 in this embodiment is a device independently provided outside the X-ray detection camera 10, but may be integrated inside the X-ray detection camera 10.

[0037] 4 shows a hardware configuration of the control device 20. As shown in FIG. 4, the control device 20 is physically a computer or the like including a processor such as a CPU (Central Processing Unit) 101 and a GPU (Graphic Processing Unit) 105, a recording medium such as a RAM (Random Access Memory) 102 and a ROM (Read Only Memory) 103, a communication module 104, and an input / output module 106, each of which is electrically connected. The control device 20 may include a display, a keyboard, a mouse, a touch panel display, or the like as the input device 40 and the display device 30, or may include a data recording device such as a hard disk drive or a semiconductor memory. The control device 20 may also be composed of a plurality of computers.

[0038] FIG. 5 is a block diagram showing the functional configuration of the control device 20. The control device 20 includes an input unit 201, a calculation unit 202, an image acquisition unit 203, a noise map generation unit 204, a processing unit 205, and a construction unit 206. Each functional unit of the control device 20 shown in FIG. 5 is realized by reading a program (radiation image processing program of the first embodiment) onto hardware such as the CPU 101, the GPU 105, and the RAM 102, and operating the communication module 104 and the input / output module 106 under the control of the CPU 101 and the GPU 105, and reading and writing data in the RAM 102. The CPU 101 and the GPU 105 of the control device 20 execute the computer program to make the control device 20 function as each functional unit of FIG. 5, and sequentially execute processes corresponding to a radiation image acquisition processing method described later. The CPU 101 and the GPU 105 may be a single piece of hardware, or only one of them may be used. Moreover, the CPU 101 and the GPU 105 may be implemented in a programmable logic such as an FPGA as a software processor. The RAM and the ROM may be standalone hardware or may be built into a programmable logic such as an FPGA. All of the various data required for the execution of this computer program and the various data generated by the execution of this computer program are stored in an internal memory such as the ROM 103 and the RAM 102, or a storage medium such as a hard disk drive. In addition, the internal memory or storage medium in the control device 20 stores in advance a trained model 207 that is read by the CPU 101 and the GPU 105 to cause the CPU 101 and the GPU 105 to perform noise removal processing on an X-ray image (X-ray transmission image) (described later).

[0039] The functions of each functional unit of the control device 20 will be described in detail below.

[0040] The input unit 201 accepts an input of condition information indicating either the conditions of the radiation source or the imaging conditions when irradiating radiation to image the object F. Specifically, the input unit 201 accepts an input of condition information indicating the operating conditions of the X-ray irradiator (radiation source) 50 when capturing an X-ray image of the object F, or the imaging conditions of the X-ray detection camera 10, from a user of the image acquisition device 1. The operating conditions include all or some of the tube voltage, target angle, target material, etc. The condition information indicating the imaging condition includes all or a part of the following: the material and thickness of the filters 51, 19 arranged between the X-ray irradiator 50 and the X-ray detection camera 10, the distance (FDD) between the X-ray irradiator 50 and the X-ray detection camera 10, the type of the window material of the X-ray detection camera 10, and information on the material and thickness of the scintillator layer 11 of the X-ray detection camera 10, X-ray detection camera information (for example, gain setting value, circuit noise value, saturation charge amount, conversion coefficient value (e- / count), camera line rate (Hz) or line speed (m / min)), and information on the object F. The input unit 201 may accept the input of the condition information as a direct input of information such as a numerical value, or may accept it as a selection input for information such as a numerical value set in advance in an internal memory. The input unit 201 accepts the input of the above condition information from the user, but may acquire some of the condition information (tube voltage, etc.) according to the detection result of the control state by the control device 20.

[0041] The calculation unit 202 calculates the average energy of X-rays (radiation) transmitted through the object F based on the condition information. The condition information includes at least one of the following: the tube voltage of the source, information on the object F, information on a filter included in a camera used to capture the object F, information on a scintillator included in the camera, and information on a filter included in the X-ray source. Specifically, the calculation unit 202 calculates the average energy value of X-rays transmitted through the object F using the image acquisition device 1 and detected by the X-ray detection camera 10 based on the condition information input by the input unit 201. For example, the calculation unit 202 calculates the spectrum of X-rays detected by the X-ray detection camera 10 using, for example, a known approximation formula such as Tucker's, based on information included in the condition information, such as the tube voltage, the target angle, the material of the target, the material and thickness of the filters 51, 19 and the presence or absence thereof, the type of window material of the X-ray detection camera 10 and the presence or absence thereof, and the material and thickness of the scintillator layer 11 of the X-ray detection camera 10. Then, the calculation unit 202 further calculates a spectral intensity integral value and a photon number integral value from the X-ray spectrum, and calculates the value of the average energy of the X-rays by dividing the spectral intensity integral value by the photon number integral value.

[0042] A calculation method using the known Tucker approximation formula will be described. For example, when the calculation unit 202 specifies the target as tungsten and the target angle as 25°, it can determine Em: kinetic energy at the time of collision of electrons with the target, T: electron kinetic energy in the target, A: proportionality constant determined by the atomic number of the target material, ρ: target density, μ(E): linear attenuation coefficient of the target material, B: function of Z and T that change slowly, C: Thomson-Whiddington constant, θ: target angle, and c: speed of light in a vacuum. Furthermore, the calculation unit 202 can calculate the irradiated X-ray spectrum by calculating the following formula (1) based on these.

[0043]

number

[0044] Next, the calculation unit 202 can calculate the energy spectrum of X-rays transmitted through the filter and the object F and absorbed by the scintillator using the X-ray attenuation equation given by the following equation (2).

[0045]

number

[0046] The image acquiring unit 203 acquires a radiological image in which radiation is irradiated onto the object F and the radiation that has passed through the object F is captured. Specifically, the image acquiring unit 203 generates an X-ray image based on a digital signal (amplified signal) output from the X-ray detection camera 10 (more specifically, the output interface 17). The image acquiring unit 203 generates one X-ray image based on multiple lines of digital signals output from the output interface 17. FIG. 6 is a diagram showing an example of an X-ray image acquired by the image acquiring unit 203.

[0047] The noise map generating unit 204 derives an evaluation value from the pixel value of each pixel of the radiographic image based on the relational data indicating the relationship between the pixel value and the evaluation value evaluating the spread of the noise value, and generates a noise map which is data in which the derived evaluation value is associated with each pixel of the radiographic image. At this time, the noise map generating unit 204 derives the evaluation value from the average energy of the radiation transmitted through the object F and the pixel value of each pixel of the radiographic image. Specifically, the noise map generating unit 204 derives the standard deviation of the noise value from the average energy of the X-rays calculated by the calculating unit 202 and the pixel value of each pixel of the X-ray image (radiographic image) acquired by the image acquiring unit 203, using a relational expression (relational data) between the pixel value and the standard deviation of the noise value (evaluation value evaluating the spread of the noise value). The noise map generating unit 204 generates a noise standard deviation map (noise map) by associating the derived standard deviation of the noise value with each pixel of the X-ray image.

[0048] The relational expression between the pixel value and average energy used by the noise map generating unit 204 and the standard deviation of the noise value is expressed by the following formula (4).

[0049]

number

[0050] FIG. 7 is a diagram showing an example of a noise standard deviation map generated by the noise map generating unit 204. The noise map generating unit 204 derives a relationship graph G3 representing the correspondence between pixel values ​​and the standard deviation of noise values ​​by substituting various pixel values ​​into the variable Signal using the relational expression (4) between pixel values ​​and the standard deviation of noise values ​​to obtain the correspondence between pixel values ​​and the variable Noise. Then, the noise map generating unit 204 derives relationship data G2 representing the correspondence between each pixel position and pixel value from the X-ray image G1 acquired by the image acquiring unit 203. Furthermore, the noise map generating unit 204 derives the standard deviation of noise values ​​corresponding to the pixels at each pixel position in the X-ray image by applying the correspondence shown in the relationship graph G3 to each pixel value in the relationship data G2. As a result, the noise map generating unit 204 associates the derived standard deviation of noise with each pixel position, and derives relationship data G4 indicating the correspondence between each pixel position and the standard deviation of noise. Then, the noise map generating unit 204 generates a noise standard deviation map G5 based on the derived relational data G4.

[0051] The processing unit 205 inputs the radiographic image and the noise map to a trained model 207 constructed in advance by machine learning, and executes image processing to remove noise from the radiographic image. That is, as shown in FIG. 8, the processing unit 205 acquires a trained model 207 (described later) constructed by the constructing unit 206 from an internal memory or a storage medium in the control device 20. The processing unit 205 inputs the X-ray image G1 acquired by the image acquiring unit 203 and the noise standard deviation map G5 generated by the noise map generating unit 204 to the trained model 207. As a result, the processing unit 205 generates an output image G6 by executing image processing to remove noise from the X-ray image G1 using the trained model 207. Then, the processing unit 205 outputs the generated output image G6 to the display device 30 or the like.

[0052] The construction unit 206 constructs a trained model 207 that outputs noise-removed image data based on the training image and the noise map by machine learning using a training image, which is a radiation image, a noise map generated from the training image based on a relational expression between pixel values ​​and standard deviations of noise values, and noise-removed image data, which is data obtained by removing noise from the training image, as training data. The construction unit 206 stores the constructed trained model 207 in an internal memory or a storage medium in the control device 20. Machine learning includes supervised learning, unsupervised learning, and reinforcement learning, and among these learning, there are deep learning and neural network learning. In the first embodiment, a two-dimensional convolutional neural network described in the paper "Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising" by Kai Zhang et al. is adopted as an example of a deep learning algorithm. In addition to being constructed by the construction unit 206, the trained model 207 may be generated by an external computer or the like and downloaded to the control device 20. The radiological images used in machine learning include radiological images captured of known structures, or images that reproduce such radiological images.

[0053] FIG. 9 is an example of a training image, which is one of the training data used to construct the trained model 207. As the training image, X-ray images of patterns of various thicknesses, various materials, and various resolutions may be used. The example shown in FIG. 9 is a training image G7 generated for chicken. The training image G7 may be an X-ray image actually generated for a plurality of types of known structures using an image acquisition device 1 including a scintillator layer 11 and a scan camera 12, or an image generated by simulation calculation. The X-ray image may be acquired using a device other than the image acquisition device 1.

[0054] As a pre-processing for machine learning, the construction unit 206 derives an evaluation value from the pixel value of each pixel of the radiographic image based on the relational data indicating the relationship between the pixel value and the evaluation value that evaluates the spread of the noise value, and generates a noise map, which is data in which the derived evaluation value is associated with each pixel of the radiographic image. Specifically, when constructing the trained model 207, the construction unit 206 acquires training images generated by actual imaging or simulation calculation from the image acquisition unit 203, etc. Then, the construction unit 206 sets, for example, the operating conditions of the X-ray irradiator 50 of the image acquisition device 1 or the imaging conditions of the image acquisition device 1. Alternatively, the construction unit 206 sets the operating conditions or imaging conditions of the X-ray irradiator 50 at the time of the simulation calculation. The construction unit 206 calculates the average energy of X-rays based on the above operating conditions or imaging conditions using a method similar to that of the calculation unit 202. Furthermore, the construction unit 206 generates a noise standard deviation map based on the average energy of X-rays and the training image using a method similar to that of the noise map generation unit 204 shown in FIG. 7. In other words, the preprocessing method for the machine learning method includes a noise map generation step of deriving an evaluation value from the pixel values ​​of each pixel of the radiological image based on relational data representing the relationship between pixel values ​​and an evaluation value that evaluates the spread of noise values, and generating a noise map, which is data that associates the derived evaluation value with each pixel of the radiological image.

[0055] The construction unit 206 constructs the trained model 207 by machine learning using the training image, the noise map generated from the training image, and the noise-removed image data, which is data obtained by removing noise from the training image in advance, as training data (teacher data). Specifically, the construction unit 206 acquires in advance the noise-removed image data obtained by removing noise from the training image. When the training image is an X-ray image generated by simulation calculation, the construction unit 206 sets the image before noise is added in the training image generation process as the noise-removed image data. On the other hand, when the training image is an X-ray image actually generated using the image acquisition device 1 for a plurality of types of known structures, the construction unit 206 sets the image obtained by removing noise from the X-ray image using image processing such as an average filter, a median filter, a bilateral filter, or an NLM filter as the noise-removed image data. The construction unit 206 executes training by machine learning to construct the trained model 207 that outputs noise-removed image data based on the training image and the noise standard deviation map.

[0056] FIG. 10 is a flowchart showing a procedure for creating image data that is teacher data (training data) used by the construction unit 206 to construct the trained model 207.

[0057] Image data (also called teacher image data) as teacher data is created by a computer in the following procedure. First, an image of a structure having a predetermined structure (structure image) is created (step S301). For example, an image of a structure having a predetermined structure may be created by simulation calculation. Alternatively, an X-ray image of a structure such as a chart having a predetermined structure may be acquired to create a structure image. Next, a sigma value, which is a standard deviation of pixel values, is calculated for one pixel selected from a plurality of pixels constituting the structure image (step S302). Then, a normal distribution (Poisson distribution) indicating a noise distribution is set based on the sigma value obtained in step S302 (step S303). In this way, teacher data of various noise conditions can be generated by setting a normal distribution based on the sigma value. Next, a noise value randomly set according to the normal distribution set based on the sigma value in step S303 is calculated (step S304). Furthermore, a pixel value constituting the image data as teacher data is generated by adding the noise value obtained in step S304 to the pixel value of the one pixel (step S305). The process from step S302 to step S305 is performed for each of the multiple pixels constituting the structure image (step S306), and teacher image data that will become teacher data is generated (step S307). If further teacher image data is required, it is determined that the process from step S301 to step S307 will be performed for another structure image (step S308), and another teacher image data that will become teacher data is generated. Note that the other structure image may be an image of a structure having the same structure, or may be an image of a structure having a different structure.

[0058] It is necessary to prepare a large amount of image data, which is training data used to construct the trained model 207. In addition, it is preferable that the structure image has little noise, and ideally, it is preferable that the structure image has no noise. Therefore, when the structure image is generated by simulation calculation, it is possible to generate a large number of images without noise, and therefore it is effective to generate the structure image by simulation calculation.

[0059] Next, a procedure for observing an X-ray transmission image of the object F using the image acquisition device 1 according to the first embodiment, i.e., a flow of the radiographic image acquisition method according to the first embodiment, will be described. Fig. 11 is a flowchart showing the procedure for the observation process by the image acquisition device 1.

[0060] First, the construction unit 206 uses a training image, a noise standard deviation map generated from the training image based on the relational expression, and noise-removed image data as training data to construct a trained model 207 that outputs noise-removed image data based on the training image and the noise standard deviation map through machine learning (step S100). Next, the input unit 201 accepts input of condition information indicating the operating conditions of the X-ray irradiator 50 or the imaging conditions of the X-ray detection camera 10 from an operator (user) of the image acquisition device 1 (step S101). Then, the calculation unit 202 calculates the average energy value of the X-rays detected by the X-ray detection camera 10 based on the condition information (step S102).

[0061] Next, the object F is set in the image acquisition device 1, the object F is imaged, and an X-ray image of the object F is acquired by the control device 20 (step S103). Furthermore, the control device 20 derives the standard deviation of the noise values ​​from the average energy of the X-rays and the pixel values ​​of each pixel of the X-ray image based on the relational expression between the pixel values ​​and the standard deviation of the noise values, and generates a noise standard deviation map by associating the derived noise standard deviation with each pixel value (step S104).

[0062] Next, the X-ray image of the object F and the noise standard deviation map are input to the trained model 207 constructed and stored in advance by the processing unit 205, and noise removal processing is performed on the X-ray image (step S105). Furthermore, the processing unit 205 outputs an output image, which is the X-ray image that has been subjected to the noise removal processing, to the display device 30 (step S106).

[0063] According to the image acquisition device 1 described above, the scintillation light corresponding to the X-rays transmitted through the object F is detected by the scan camera 12 in which the pixel line 74 having M pixels 72 arranged in the scanning direction TD of the object F is arranged in N rows, and the detection signals of at least two pixels 72 among the detection signals of the M pixels 72 output for each pixel line 74 are added, and the N detection signals obtained by the addition process are output sequentially or simultaneously, thereby outputting an X-ray image. At this time, the X-rays transmitted through the object F are converted into scintillation light by the M×P scintillator units 11a arranged separately, and the scintillation light from the P scintillator units 11a is detected by the M×N pixels 72, respectively. In addition, the output X-ray image is input to the trained model 207 constructed in advance by machine learning using image data, and the X-ray image is subjected to noise removal processing. This makes it possible to remove noise components while increasing the signal components in the X-ray image, and effectively improve the S / N ratio in the X-ray image. Furthermore, the resolution of the X-ray image can be improved by using the scintillator section 11a separated in correspondence with the pixel 72. In general, by forming the separation section 11b that blocks or reflects the scintillation light, the scintillation light can be prevented from entering the adjacent scintillator section, and a decrease in resolution can be prevented. However, since the portion where the separation section 11b exists does not contribute to the conversion of the incident X-rays to the scintillation light, there is also a problem that the detection signal becomes smaller (decreased) according to the size (area) of the separation section 11b. The decrease in the detection signal means a decrease in the S / N ratio. Therefore, it has been considered difficult to achieve both an improvement in the resolution and an improvement in the S / N ratio by using a scintillator having the separation section 11b. In contrast, the image acquisition device 1 can effectively improve the S / N ratio while increasing the resolution in the X-ray image.Specifically, in the present embodiment, in which an X-ray image is acquired using the scintillator layer 11 and noise removal processing is performed using the trained model 207, it has been found that the CNR (Contrast to Noise Ratio) is improved by about 20 times compared to the case in which an X-ray image is acquired using a scintillator that integrally covers the entire scan camera 12 and noise removal processing is not performed, and that false signals are sufficiently removed in the binarized image generated by performing threshold processing. As a result, according to this embodiment, in the application of foreign body inspection, it is possible to prevent foreign bodies from overlapping and being detected.

[0064] Furthermore, in the image acquisition device 1, the trained model 207 is constructed by machine learning using image data obtained by adding a noise value along a normal distribution to an X-ray image of a predetermined structure as training data. This makes it easy to prepare image data that is training data used to construct the trained model 207, and allows the trained model 207 to be constructed efficiently.

[0065] Moreover, in the image acquisition device 1, the trained model 207 is constructed by machine learning using the X-ray image obtained using the scintillator layer 11 as training data. In an X-ray image using the scintillator layer 11 in which the scintillator portion 11a is separated into a plurality of parts, noise is not blurred and high-frequency noise tends to occur more easily than in an X-ray image using an integrated scintillator. In the above case, it becomes easy to reflect the conditions at the time of image acquisition in the image data, which is the training data used to construct the trained model 207, and a trained model 207 that realizes appropriate noise removal corresponding to the tendency for high-frequency noise to occur can be constructed.

[0066] Moreover, according to the image acquisition device 1, the standard deviation of the noise value is derived from the pixel value of each image of the X-ray image using a relational expression between the pixel value and the standard deviation of the noise value, and a noise standard deviation map is generated, which is data in which the standard deviation of the noise value derived is associated with each pixel of the X-ray image. Then, the X-ray image and the noise standard deviation map are input to a trained model 207 constructed in advance by machine learning, and image processing is performed to remove noise from the X-ray image. According to this configuration, the standard deviation of the noise value derived from the pixel value of each pixel of the X-ray image is taken into consideration, and the noise in each pixel of the X-ray image is removed by machine learning. As a result, noise removal corresponding to the relationship between the pixel value in the X-ray image and the standard deviation of the noise value can be realized using the trained model 207. As a result, noise in the X-ray image can be effectively removed.

[0067] In particular, the noise state of X-ray images changes depending on the tube voltage, filter, scintillator, X-ray detection camera conditions (gain setting value, circuit noise value, saturation charge amount, conversion coefficient value (e- / count), camera line rate), object, etc. Therefore, when trying to realize noise removal by machine learning, it is considered to prepare learning models trained under various conditions. That is, as a comparative example, a method can be adopted in which multiple learning models are constructed according to the conditions when the X-ray image is measured, a learning model is selected for each condition, and the noise removal process is executed. In such a comparative example, for example, a learning model must be constructed for each noise condition such as the average energy of X-rays, the gain of the X-ray detection camera, and the type of X-ray camera, and a huge number of learning models must be generated, which may take a lot of time for construction. As an example, when there are 10 types of average energy of X-rays, 8 types of gain of the X-ray detection camera, and 3 types of products, 240 learned models are required, but if it takes one day to construct one learned model, it will take as long as 240 days for machine learning. In this regard, according to the present embodiment, a noise map is generated from an X-ray image and the noise map is used as input data for machine learning, thereby reducing the noise conditions necessary to generate a trained model, and significantly reducing the learning time required to construct the trained model 207. [Modification of the control device 20 of the first embodiment]

[0068] FIG. 12 is a block diagram showing the functional configuration of the control device 20A in a modified example of the first embodiment. The control device 20A is different from the first embodiment described above in that the calculation unit 202A has a function of deriving the average energy of X-rays from the pixel values ​​of the X-ray image, and the noise map generation unit 204A has a function of deriving a noise standard deviation map based on the pixel values ​​of the X-ray image and the average energy of X-rays derived from the X-ray image. FIG. 13 is a flowchart showing the procedure of the observation process by the image acquisition device 1 including the control device 20A of FIG. 12. As shown in FIG. 13, in the control device 20A, the process shown in step S103 of the control device 20 according to the first embodiment shown in FIG. 11 is performed immediately after step S100. Then, in the control device 20A, the processes shown in S102A and S104A are executed by replacing the processes of steps S102 and S104 of the control device 20.

[0069] The calculation unit 202A calculates the average energy from the pixel value of each pixel of the radiation image (step S102A). Specifically, the calculation unit 202A derives in advance the relationship between the pixel value and the average energy for each piece of condition information by a simulation calculation of the X-ray spectrum or the like. The calculation unit 202A acquires condition information including at least the tube voltage acquired by the input unit 201 and information on the scintillator included in the X-ray detection camera 10. Then, based on the condition information, the calculation unit 202A selects a relationship corresponding to the condition information from among the relationships between the pixel value and the average energy that have been derived in advance. Furthermore, the calculation unit 202A derives the average energy for each pixel from the pixel value of each pixel of the X-ray image acquired by the image acquisition unit 203 based on the selected relationship.

[0070] Hereinafter, the derivation of the relationship between the pixel value and the average energy for each piece of condition information by the calculation unit 202A will be described with reference to FIGS.

[0071] First, the calculation unit 202A derives a graph G18 showing the relationship between the thickness of the object F and the transmittance of X-rays, and a graph G19 showing the relationship between the thickness of the object F and the average energy of X-rays, based on the condition information. Specifically, as shown in parts (a) to (d) of FIG. 14, the calculation unit 202A calculates, by simulation calculation, the energy spectra G14 to G17 of X-rays transmitted through the object F when the thickness is variously changed, based on the condition information including at least the tube voltage and the information of the scintillator included in the X-ray detection camera 10. FIG. 14 is a graph showing an example of a simulation calculation result of the energy spectrum of X-rays transmitted through the object F by the calculation unit 202A. Here, the energy spectra G14 to G17 of transmitted X-rays when the simulation calculation is performed by gradually increasing the thickness of the object F made of water are illustrated. Furthermore, the calculation unit 202A calculates the average energy of X-rays transmitted through the object F when the thickness is variously changed, based on the calculated energy spectra G14 to G17. Note that, other than the simulation calculation, the calculation unit 202A may obtain the relationship between the thickness of the object F and the average energy based on an X-ray image obtained by imaging a structure having a known thickness.

[0072] Furthermore, the calculation unit 202A also derives the relationship between the thickness of the object F and the transmittance of X-rays based on the above simulation results. Fig. 15 is a chart showing an example of the relationship between the thickness of the object F and the average energy and the transmittance derived by the calculation unit 202A. As shown in Fig. 15, the average energy of the transmitted X-rays and the transmittance of X-rays are derived corresponding to each of the energy spectra G14 to G17 calculated for each thickness of the object F.

[0073] Next, the calculation unit 202A derives a graph G18 showing the relationship between the thickness of the object F and the transmittance of the X-rays from the transmittance of the X-rays derived for the object F of various thicknesses. Fig. 16 is a graph showing the relationship between the thickness of the object F and the transmittance of the X-rays for the object F derived by the calculation unit 202A. In addition, the calculation unit 202A derives a graph G19 showing the relationship between the thickness of the object F and the average energy of the X-rays from the average energy of the X-rays derived for the object F of various thicknesses. Fig. 17 is a graph showing the relationship between the thickness of the object F and the average energy of the X-rays transmitted through the object F derived by the calculation unit 202A.

[0074] Then, the calculation unit 202A derives a graph G20 showing the relationship between the pixel value and the average energy of the X-ray image as shown in FIG. 18 for each of various pieces of condition information based on two graphs G18 and G19 derived for each of various pieces of condition information. FIG. 18 is a graph showing the relationship between the pixel value and the average energy of the X-ray image derived by the calculation unit 202A. Specifically, the calculation unit 202A derives a pixel value I0 of the X-ray transmission image when the object F does not exist based on the condition information. Then, the calculation unit 202A sets a pixel value I of the X-ray image when the object F exists, and calculates I / I0, which is the transmittance of the X-ray. Furthermore, the calculation unit 202A derives the thickness of the object F from I / I0, which is the calculated transmittance of the X-ray, based on the graph G18 of the thickness of the object F and the transmittance of the X-ray for the object F. Finally, the calculation unit 202A derives the average energy of the transmitted X-rays corresponding to the thickness based on the derived thickness of the object F and the graph G19 of the thickness of the object F and the average energy of the transmitted X-rays. Next, the calculation unit 202A performs the above derivation for each of various pieces of condition information while variously changing the pixel value I of the X-ray image, thereby deriving a graph G20 showing the relationship between the pixel value of the X-ray image and the average energy of the transmitted X-rays for each piece of condition information.

[0075] Here, an example of deriving the average energy based on the pixel value by the calculation unit 202A will be described. For example, it is assumed that the calculation unit 202A derives the pixel value of the X-ray image when the object F does not exist as I0=5000 based on the condition information, and sets the pixel value of the X-ray image when the object F exists as I=500. In this case, the calculation unit 202A calculates the X-ray transmittance as I / I0=0.1. Next, the calculation unit 202A derives that the thickness corresponding to the X-ray transmittance of 0.1 is 30 mm based on the graph G18 showing the relationship between the thickness of the object F and the transmittance of the X-ray for the object F. Furthermore, the calculation unit 202A derives that the average energy corresponding to the pixel value of 500 is 27 keV based on the graph G19 showing the relationship between the thickness of the object F and the average energy of the transmitted X-ray. Finally, the calculation unit 202A repeats the derivation of the average energy of X-rays for each pixel value, and derives a graph G20 that indicates the relationship between the pixel value and the average energy of the X-ray image.

[0076] Furthermore, the calculation unit 202A selects a graph G20 corresponding to the condition information acquired by the input unit 201 from among the multiple graphs G20 derived in advance by the above procedure. The calculation unit 202A derives the average energy of the transmitted X-rays corresponding to the pixel value of each pixel of the X-ray image acquired by the image acquisition unit 203 based on the selected graph G20.

[0077] The calculation unit 202A may derive the average energy of X-rays from the condition information acquired by the input unit 201 and the pixel value of each pixel of the X-ray image by referring to the graphs G18 and G19, instead of deriving the relationship between the pixel value and the average energy of X-rays for each piece of condition information in advance. Specifically, the calculation unit 202A derives the pixel value I0 of the X-ray image when the object does not exist based on the condition information. Then, the calculation unit 202A calculates the transmittance by calculating the ratio of each pixel value I of each pixel of the X-ray image acquired by the image acquisition unit 203 to the pixel value I0. Furthermore, the calculation unit 202A derives the thickness based on the graph G18 showing the relationship between the thickness and the transmittance of X-rays and the calculated transmittance. Then, the calculation unit 202A derives the average energy based on the graph G19 showing the relationship between the thickness and the average energy and the derived thickness, thereby deriving the average energy for each pixel value of each pixel of the X-ray image.

[0078] The noise map generating unit 204A generates a noise standard deviation map from the X-ray image acquired by the image acquiring unit 203 and the average energy of the X-ray corresponding to each pixel of the X-ray image derived by the calculating unit 202A (step S104A). Specifically, the noise map generating unit 204A derives the standard deviation of the noise value for each pixel taking into account the thickness of the object by substituting the pixel value of each pixel of the X-ray image acquired by the image acquiring unit 203 and the average energy derived for each pixel by the calculating unit 202A into the relational expression (4). The noise map generating unit 204A generates the standard deviation of the noise value corresponding to each pixel of the X-ray image as a noise standard deviation map.

[0079] FIG. 19 is a graph showing an example of the relationship between pixel values ​​and the standard deviation of noise values. This graph shows the relationship between the standard deviation of noise values ​​derived from the pixel values ​​of an X-ray image by the calculation unit 202A and the noise map generation unit 204A according to this modification and the pixel values ​​of the X-ray image. In this modification, the standard deviation of noise values ​​is derived taking into account the thickness of the object, so that the thickness of the object decreases as the pixel value increases, and the average energy in the pixel decreases. Therefore, as can be estimated from the relational expression (4), the first embodiment and this modification have different changes in the standard deviation of noise values ​​when the pixel value increases. In the example shown in FIG. 19, the degree of increase in the standard deviation of noise values ​​when the pixel value increases is smaller in the graph G22 of this modification than in the graph G21 of the first embodiment.

[0080] In the control device 20A of the modified example of the first embodiment, the average energy is calculated from the pixel value of each pixel of the X-ray image. Here, for example, when multiple objects with different thicknesses or materials are present in the X-ray image, the average energy differs greatly for each object, and noise cannot be sufficiently removed from the X-ray image. According to this configuration, the average energy of the X-rays passing through the object F is calculated for each pixel value of each pixel of the X-ray image, so that noise removal corresponding to the relationship between the pixel value of each pixel of the X-ray image and noise can be realized, taking into account, for example, differences in thickness and material. As a result, noise in the X-ray image can be effectively removed.

[0081] The control device 20A according to this modification derives the average energy from the pixel values ​​of the X-ray image using the graph G20 derived for each of various condition information. At this time, the average energy may be derived from the pixel values ​​while ignoring the difference in the material of the object F. FIG. 20 is a graph showing the relationship between the pixel values ​​of the X-ray image derived by the calculation unit 202A and the standard deviation of the noise value. Here, the relationship is derived taking into consideration the change in the material of the object F as condition information, and the graph G24 shows an example of derivation when the material is aluminum, the graph G23 shows an example of derivation when the material is PET (Polyethylene terephthalate), and the graph G25 shows an example of derivation when the material is copper. In this way, even if the material of the object F changes, if the tube voltage of the X-ray irradiator 50 and the information of the scintillator provided in the X-ray detection camera 10 used to image the object F are the same, the relationship between the pixel value and the average energy of the transmitted X-ray does not change significantly, and therefore the relationship between the pixel value and the standard deviation of the noise value does not change significantly either. Considering such a property, the control device 20A can derive the average energy from the pixel values ​​of the X-ray image, ignoring the difference in the material of the object F as the condition information. Even in such a case, the control device 20A of this modification can realize noise removal corresponding to the relationship between the pixel values ​​and the standard deviation of the noise. As a result, the noise in the X-ray image can be removed more effectively. [Another Modification of the Control Device 20 of the First Embodiment]

[0082] FIG. 21 is a block diagram showing the functional configuration of a control device 20B according to another modified example of the first embodiment. The control device 20B is different from the first embodiment described above in that the image acquisition unit 203B has a function of acquiring an X-ray image of the jig, and the noise map generation unit 204B has a function of deriving a graph showing the relationship between the pixel value and the standard deviation of the noise value from the X-ray image of the jig. FIG. 22 is a flowchart showing the procedure of the observation process by the image acquisition device 1 including the control device 20B of FIG. 21. As shown in FIG. 22, in the control device 20B according to this modified example, the processes shown in steps S201 and S202 are replaced with the processes of steps S101, S102, and S104 by the control device 20 according to the first embodiment shown in FIG. 11 and executed.

[0083] The image acquiring unit 203B acquires a radiological image of the jig in which the jig is irradiated with radiation and the radiation that has passed through the jig is captured (step S201). Specifically, the image acquiring unit 203B acquires an X-ray image captured by irradiating the jig and the object F with X-rays using the image acquiring device 1. The jig is a flat plate-like member or the like whose thickness and material are known. That is, the image acquiring unit 203B acquires an X-ray image of the jig captured using the image acquiring device 1 prior to the observation process of the object F. Then, the image acquiring unit 203B acquires an X-ray image of the object F captured using the image acquiring device 1. However, the timing of acquiring the X-ray images of the jig and the object F is not limited to the above, and may be simultaneous or reversed (step S103). Similarly to the image acquiring unit 203, the image acquiring unit 203B acquires an X-ray image in which the object F is irradiated with X-rays and the X-rays that have passed through the object F are captured.

[0084] The jig is set in the image acquisition device 1 and an image of the jig is taken, and the noise map generation unit 204B derives, from the resulting radiographic image of the jig, relationship data representing the relationship between pixel values ​​and evaluation values ​​that evaluate the spread of noise values ​​(step S202). Specifically, the noise map generation unit 204B derives a noise standard deviation map representing the relationship between pixel values ​​and standard deviations of noise values ​​from the X-ray image of the jig.

[0085] FIG. 23 is a diagram showing an example of a noise standard deviation map generated by the noise map generating unit 204B. The noise map generating unit 204B derives a relationship graph G27 representing the correspondence between pixel values ​​and the standard deviation of noise values ​​from an X-ray image G26 of the jig. Then, the noise map generating unit 204B derives relationship data G2 representing the correspondence between each pixel position and pixel value from an X-ray image G1 acquired by the image acquiring unit 203B in the same manner as in the first embodiment. Furthermore, the noise map generating unit 204 derives the standard deviation of noise values ​​corresponding to the pixels at each pixel position in the X-ray image by applying the correspondence shown in the relationship graph G27 to each pixel in the relationship data G2. As a result, the noise map generating unit 204 associates the derived standard deviation of noise with each pixel position, and derives relationship data G4 indicating the correspondence between each pixel position and the standard deviation of noise. Then, the noise map generating unit 204 generates a noise standard deviation map G5 based on the derived relationship data G4.

[0086] The derivation of the relationship graph G27 representing the relationship between the pixel value and the standard deviation of the noise value from the X-ray image G26 of the jig by the noise map generating unit 204B will be described. FIG. 24 shows an example of the structure of the jig used for imaging in this modified example. For example, a member P1 whose thickness changes stepwise in one direction may be used for the jig. FIG. 25 shows an example of the X-ray image of the jig in FIG. 24. First, the noise map generating unit 204B derives pixel values ​​(hereinafter, referred to as true pixel values) in the X-ray image G26 of the jig when there is no noise for each step of the jig, and derives the standard deviation of the noise value based on the true pixel values. Specifically, the noise map generating unit 204B derives the average value of the pixel values ​​at a certain step of the jig. Then, the noise map generating unit 204B sets the derived average value of the pixel values ​​as the true pixel value at that step. The noise map generating unit 204B derives the difference between each pixel value and the true pixel value at that step as the noise value. The noise map generating unit 204B derives the standard deviation of the noise values ​​from the derived noise values ​​for each pixel value.

[0087] Then, the noise map generating unit 204B derives the relationship between the true pixel value and the standard deviation of the noise value as a relationship graph G27 between the pixel value and the standard deviation of the noise value. Specifically, the noise map generating unit 204B derives the true pixel value and the standard deviation of the noise value for each step of the jig. The noise map generating unit 204B plots the derived relationship between the true pixel value and the standard deviation of the noise value on a graph and draws an approximation curve to derive the relationship graph G27 representing the relationship between the pixel value and the standard deviation of the noise value. Note that for the approximation curve, exponential approximation, linear approximation, logarithmic approximation, polynomial approximation, power approximation, or the like is used.

[0088] In the control device 20B according to this modified example, relational data is generated based on a radiographic image obtained by capturing an image of an actual jig. This provides relational data that is optimal for removing noise from the radiographic image of the object F. As a result, noise in the radiographic image can be removed more effectively.

[0089] The noise map generating unit 204B may derive the relationship between pixel values ​​and standard deviation of noise values ​​from captured images when the tube current or exposure time is changed without using a jig and without an object. With this configuration, relationship data is generated based on radiographic images obtained by actually capturing images, and a noise map is generated, so that noise removal corresponding to the relationship between pixel values ​​and noise spread can be realized. As a result, noise in radiographic images can be removed more effectively.

[0090] Specifically, the image acquisition unit 203B acquires a plurality of radiation images captured in the absence of an object (step S201), and the noise map generation unit 204B may derive a relationship between pixel values ​​and standard deviations of noise values ​​from the radiation images acquired by the image acquisition unit 203B (step S202). The plurality of radiation images are a plurality of images in which at least one of the conditions of the radiation source or the imaging conditions is different from each other. As an example, the image acquisition unit 203B acquires a plurality of X-ray images captured using the image acquisition device 1 in the absence of the object F prior to the observation process of the object F while changing the tube current or the exposure time. Then, the noise map generation unit 204B derives true pixel values ​​for each X-ray image, and derives the standard deviation of noise based on the true pixel values ​​in the same manner as in this modified example. Furthermore, the noise map generation unit 204B derives a relationship graph showing the relationship between pixel values ​​and standard deviations of noise values ​​by plotting the relationship between the true pixel values ​​and standard deviations of noise on a graph and drawing an approximation curve in the same manner as in this modified example. Finally, the noise map generating unit 204B generates a noise standard deviation map from the X-ray image acquired by the image acquiring unit 203B based on the derived relationship graph, in the same manner as in the first embodiment. [Second embodiment] 26 is a block diagram showing the functional configuration of a control device 20C according to the second embodiment. The control device 20C includes an input unit 201C, a calculation unit 202C, a narrowing down unit 203C, a selection unit 204C, and a processing unit 205C.

[0091] In addition, the control device 20C stores a plurality of trained models 206C that perform noise removal processing on X-ray transmission images. Each of the trained models 206C is a trained model by machine learning that is constructed in advance using image data as teacher data. Machine learning includes supervised learning, deep learning, reinforcement learning, neural network learning, and the like. In this embodiment, a two-dimensional convolutional neural network described in the paper "Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising" by Kai Zhang et al. is adopted as an example of a deep learning algorithm. The trained models 206C may be generated by an external computer or the like and downloaded to the control device 20C, or may be generated within the control device 20C.

[0092] FIG. 27 shows an example of image data that is training data used to construct the trained model 206C. As training data, X-ray transmission images of patterns of various thicknesses, various materials, and various resolutions may be used. The example shown in FIG. 27 is an example of an X-ray transmission image generated for chicken. This image data may be an X-ray transmission image actually generated for a plurality of types of objects using the image acquisition device 1, or image data generated by simulation calculation. The X-ray transmission image may be acquired using a device other than the image acquisition device 1. In addition, the X-ray transmission image and image data generated by simulation calculation may be used in combination. The trained models 206C are each image data obtained for transmitted X-rays with different average energies, and are constructed in advance using image data with known noise distribution. The average energy of the X-rays in the image data is set to different values ​​in advance by setting the operating conditions of the X-ray irradiator (radiation source) 50 of the image acquisition device 1 or the imaging conditions of the image acquisition device 1, or by setting the operating conditions or imaging conditions of the X-ray irradiator 50 during simulation calculation (the method of setting the average energy by the operating conditions or imaging conditions will be described later). That is, the multiple trained models 206C are constructed by machine learning using training images, which are X-ray images corresponding to the average energy of the X-rays transmitted through the object F calculated based on the operating conditions of the X-ray irradiator (radiation source) 50 when capturing an X-ray transmission image of the object F, or the condition information indicating the imaging conditions by the X-ray detection camera 10, as training data (construction step). For example, in this embodiment, the multiple trained models 206C are constructed using multiple frames (e.g., 20,000 frames) of multiple types of image data in which the average energy is set to values ​​in increments of 10 keV, such as 10 keV, 20 keV, 30 keV, ...

[0093] Image data, which is training data used to construct the trained model 206C, is generated by a creation procedure similar to that in the first embodiment described above.

[0094] Returning to FIG. 26, the functions of the functional units of the control device 20C will now be described in detail.

[0095] The input unit 201C receives input of condition information indicating the operating conditions of the X-ray irradiator (radiation source) 50 when capturing an X-ray transmission image of the object F, or the imaging conditions by the X-ray detection camera 10, from the user of the image acquisition device 1. Examples of the operating conditions include all or part of the tube voltage, the target angle, the material of the target, etc. Examples of the condition information indicating the imaging conditions include all or part of the material and thickness of the filters 51, 19 (filters provided in the camera used to capture the image of the object or filters provided in the source) arranged between the X-ray irradiator 50 and the X-ray detection camera 10, the distance (FDD) between the X-ray irradiator 50 and the X-ray detection camera 10, the type of window material of the X-ray detection camera 10, and information on the material and thickness of the scintillator layer 11 of the X-ray detection camera 10, X-ray detection camera information (for example, gain setting value, circuit noise value, saturation charge amount, conversion coefficient value (e- / count), camera line rate (Hz) or line speed (m / min)), information on the object, etc. The input unit 201C may accept the input of the condition information as a direct input of information such as a numerical value, or may accept the input as a selection input for information such as a numerical value previously set in an internal memory. The input unit 201C accepts the input of the above condition information from a user, and may acquire a part of the condition information (tube voltage, etc.) according to the detection result of the control state by the control device 20C.

[0096] The calculation unit 202C calculates the average energy value of X-rays (radiation) that are transmitted through the object F using the image acquisition device 1 and detected by the X-ray detection camera 10 based on the condition information received by the input unit 201C. For example, the calculation unit 202C calculates the spectrum of X-rays detected by the X-ray detection camera 10 using, for example, a known approximation formula such as Tucker's, based on information included in the condition information, such as tube voltage, target angle, target material, material and thickness of a filter and its presence or absence, type of window material and its presence or absence, and material and thickness of the scintillator layer 11 of the X-ray detection camera 10. Then, the calculation unit 202C further calculates a spectral intensity integral value and a photon number integral value from the X-ray spectrum, and calculates the average energy value of the X-ray by dividing the spectral intensity integral value by the photon number integral value.

[0097] A calculation method using the known Tucker approximation formula will be described. For example, when the calculation unit 202C specifies the target as tungsten and the target angle as 25°, it can determine Em: kinetic energy at the time of electron target collision, T: electron kinetic energy in the target, A: proportionality constant determined by the atomic number of the target material, ρ: target density, μ(E): linear attenuation coefficient of the target material, B: function of Z and T that change slowly, C: Thomson-Whiddington constant, θ: target angle, and c: speed of light in vacuum. Furthermore, the calculation unit 202C can calculate the irradiated X-ray spectrum by calculating the above formula (1) based on them.

[0098] Next, the calculation unit 202C can calculate the X-ray energy spectrum that is transmitted through the filter and the object F and absorbed by the scintillator using the X-ray attenuation formula of formula (2) above. The X-ray photon number spectrum is obtained by dividing this X-ray energy spectrum by the energy of each X-ray. The calculation unit 202C calculates the average energy of the X-rays using the above formula (3) by dividing the integral value of the energy intensity by the integral value of the photon number. Through the above calculation process, the calculation unit 202C calculates the average energy of the X-rays. It should be noted that the calculation of the X-ray spectrum may use known approximation formulas by Kramers or Birch et al.

[0099] The narrowing unit 203C narrows down candidates of the trained models 206C constructed in advance based on the value of the average energy calculated by the calculation unit 202C. That is, the narrowing unit 203C compares the calculated average energy value with the value of the average energy of X-rays in the image data used to construct the trained models 206C, and narrows down the trained models 206C constructed by image data having similar average energy values ​​as candidates. More specifically, when the average energy value calculated by the calculation unit 202C is 53 keV, the narrowing unit 203C narrows down the trained models 206C constructed by image data having average energy values ​​of 40 keV, 50 keV, and 60 keV, the difference between which is less than a predetermined threshold value (e.g., 15 keV), as candidates of the trained models.

[0100] The selection unit 204C selects the trained model 206C to be finally used for noise removal processing of the X-ray transmission image of the object F from among the candidates narrowed down by the narrowing down unit 203C. In detail, the selection unit 204C acquires an X-ray transmission image captured by irradiating the jig with X-rays in the image acquisition device 1, and selects the trained model 206C to be finally used based on the image characteristics of the X-ray transmission image. At this time, the selection unit 204C analyzes the energy characteristics, noise characteristics, resolution characteristics, etc. as the image characteristics of the X-ray transmission image, and selects the trained model 206C based on the analysis results.

[0101] More specifically, the selection unit 204C acquires an X-ray transmission image of a flat plate-like member whose thickness and material are known as a jig and whose relationship between the average energy of X-rays and the X-ray transmittance is known, and calculates the X-ray transmittance of one point (or the average of multiple points) in the jig by comparing the brightness of the X-ray image transmitted through the jig with the brightness of the X-ray image transmitted through air. For example, when the brightness of the X-ray image transmitted through the jig is 5,550 and the brightness of the X-ray image transmitted through air is 15,000, the transmittance is calculated as 37%. Then, the selection unit 204C specifies the average energy of the transmitted X-rays (e.g., 50 keV) estimated from the transmittance of 37% as the energy characteristic of the X-ray transmission image of the jig. The selection unit 204C selects one of the trained models 206C constructed by image data of the average energy closest to the specified average energy value.

[0102] The selection unit 204C may also analyze the characteristics at multiple points of a jig whose thickness or material changes as the energy characteristics of the X-ray transmission image of the jig. FIG. 28 is a diagram showing an example of an X-ray transmission image to be analyzed by the selection unit 204C. FIG. 28 shows an X-ray transmission image of a jig whose thickness changes stepwise. The selection unit 204C selects multiple measurement regions (ROIs: Regions of Interest) with different thicknesses from such an X-ray transmission image, analyzes the average brightness value for each of the multiple measurement regions, and obtains a thickness-brightness characteristic graph as the energy characteristics. FIG. 29 shows an example of the thickness-brightness characteristic graph obtained by the selection unit 204C.

[0103] Further, the selection unit 204C similarly acquires a thickness-brightness characteristic graph for the image data used to construct the trained model 206C narrowed down by the narrowing down unit 203C, and selects the trained model 206C constructed by the image data having characteristics closest to the characteristic graph acquired for the jig as the final trained model 206C. However, the image characteristics of the image data used to construct the trained model 206C may refer to those calculated in advance outside the control device 20C. In this way, by setting multiple measurement regions, it is possible to select a trained model that is optimal for removing noise from the X-ray transmission image of the object F. In particular, it is possible to accurately estimate the difference in the X-ray spectrum or the effect of the filter when measuring the X-ray transmission image.

[0104] The selection unit 204C can also analyze the luminance value and noise for each of a plurality of measurement regions as the noise characteristics of the X-ray transmission image of the jig, and acquire a characteristic graph of the luminance-noise ratio as the noise characteristics. That is, the selection unit 204C selects a plurality of measurement regions ROIs having different thicknesses or materials from the X-ray transmission image, analyzes the standard deviation of the luminance values ​​and the average value of the luminance values ​​of the plurality of measurement regions ROIs, and acquires a characteristic graph of the luminance-SNR (S / N ratio) as the noise characteristics. At this time, the selection unit 204C calculates the SNR for each measurement region ROI by SNR=(average value of the luminance values)÷(standard deviation of the luminance values). FIG. 30 shows an example of the characteristic graph of the luminance-SNR acquired by the selection unit 204C. Then, the selection unit 204C selects the trained model 206C constructed by image data having noise characteristics closest to the acquired characteristic graph as the final trained model 206C.

[0105] Here, the selection unit 204C may obtain a characteristic graph in which the vertical axis represents noise calculated from the standard deviation of the luminance values, instead of the above-mentioned luminance-SNR characteristic graph, as the noise characteristic. By using such a luminance-noise characteristic graph, it is possible to identify the dominant noise factor (shot noise, read noise, etc.) for each signal amount detected by the X-ray detection camera 10 from the slope of the graph of each signal amount region, and to select the trained model 206C based on the identification result.

[0106] 31 is a diagram for explaining the function of selecting a trained model based on image characteristics by the selection unit 204C. In FIG. 31, (a) shows characteristic graphs G1, G2, and G3 of the brightness-SNR of each image data used to construct a plurality of trained models 206C, and (b) shows, in addition to these characteristic graphs G1, G2, and G3, a characteristic graph G1 of the brightness-SNR of an X-ray transmission image of an image of a jig. T The graph shows the characteristics of G1, G2, G3, and G T In the case of targeting, the selection unit 204C selects the characteristic graph G T The selection unit 204C selects a trained model 206C constructed by image data of the characteristic graph G2 that is closest to the characteristic of the characteristic graph G1. T The SNR error for each luminance value at a fixed interval between and is calculated, the root mean squared error (RMSE) of the errors is calculated, and the trained model 206C corresponding to the characteristic graphs G1, G2, and G3 with the smallest mean squared error is selected. The selection unit 204C can also select the trained model 206C in the same manner when selecting using the energy characteristic.

[0107] The selection unit 204C can also select the trained model 206C based on the characteristics of the image after applying multiple trained models to an X-ray transmission image of a jig and performing noise removal processing.

[0108] For example, the selection unit 204C applies multiple trained models 206C to an X-ray transmission image of a jig having charts of various resolutions, and evaluates the resulting image after noise removal. The selection unit 204C then selects the trained model 206C used for the image with the smallest change in resolution before and after the noise removal process. FIG. 32 shows an example of an X-ray transmission image used for evaluating the resolution. In this X-ray transmission image, a chart whose resolution changes stepwise in one direction is taken as the image target. The resolution of the X-ray transmission image can be measured using MTF (Modulation Transfer Function) or CTF (Contrast Transfer Function).

[0109] In addition to evaluating the change in resolution, the selection unit 204C may evaluate the characteristics of the brightness-to-noise ratio of the image after noise removal, and select the trained model 206C used to generate the image with the highest characteristics. FIG. 33 shows an example of the structure of a jig used to evaluate the brightness-to-noise ratio. For example, a jig may be used in which foreign bodies P2 of various materials and sizes are scattered in a member P1 whose thickness changes stepwise in one direction. FIG. 34 shows an X-ray transmission image after noise removal processing obtained for the jig of FIG. 33. The selection unit 204C selects an image region R1 including an image of the foreign body P2 in the X-ray transmission image and an image region R2 not including an image of the foreign body P2 near the region R1, and calculates the minimum brightness L in the image region R1. MIN and the average luminance value L in the image region R2 AVE and the standard deviation of luminance in image region R2, L SD Then, the selection unit 204C calculates the following formula: CNR=(L AVE -L MIN ) / L SD Further, the selection unit 204C calculates the brightness-to-noise ratio CNR for each of the X-ray transmission images after application of the multiple trained models 206C, and selects the trained model 206C used to generate the X-ray transmission image with the highest brightness-to-noise ratio CNR.

[0110] Alternatively, the selection unit 204C selects the average luminance L in the image region R1. AVE_R1 and the average luminance value L in the image region R2 AVE_R2 and the standard deviation of the luminance in the image region R2, L SD Based on this, the calculation may be performed according to the following formula: CNR=(L AVE_R1 -L MIN_R2 ) / L SD

[0111] The processing unit 205C applies the trained model 206C selected by the selection unit 204C to the X-ray transmission image acquired of the object F, and executes image processing to remove noise, thereby generating an output image. Then, the processing unit 205C outputs the generated output image to the display device 30 or the like.

[0112] Next, a procedure for observing an X-ray transmission image of an object F using the image acquisition device 1 according to the second embodiment, i.e., a flow of a radiographic image acquisition method according to the second embodiment, will be described. Fig. 35 is a flow chart showing the procedure for the observation process by the image acquisition device 1.

[0113] First, the control device 20C accepts input of condition information indicating the operating conditions of the X-ray irradiator 50 or the imaging conditions of the X-ray detection camera 10 from an operator (user) of the image acquisition device 1 (step S1). Next, the control device 20C calculates the average energy value of the X-rays detected by the X-ray detection camera 10 based on the condition information (step S2).

[0114] Furthermore, the control device 20C identifies the value of the average energy of X-rays in the image data used to construct the trained model 206C stored in the control device 20C (step S3). After that, the identification of the value of the average energy of X-rays is repeated for all trained models 206C stored in the control device 20C (step S4).

[0115] Next, the control device 20C compares the calculated values ​​of the average energy of the X-rays to narrow down the candidates of the trained model 206C (step S5). Furthermore, the jig is set in the image acquisition device 1 and an image of the jig is captured to acquire an X-ray transmission image of the jig (step S6).

[0116] Thereafter, the control device 20C acquires image characteristics (average X-ray energy value, thickness-brightness characteristics, brightness-noise ratio characteristics, brightness-noise characteristics, resolution change characteristics, etc.) of the X-ray transmission image of the jig (step S7). Then, the control device 20C selects the final trained model 206C based on the acquired image characteristics (step S8).

[0117] Furthermore, the object F is set in the image acquisition device 1 and an X-ray transmission image of the object F is acquired by capturing an image of the object F (step S9). Next, the control device 20C applies the finally selected trained model 206C to the X-ray transmission image of the object F, thereby executing a noise removal process on the X-ray transmission image (step S10). Finally, the control device 20C outputs an output image, which is the X-ray transmission image that has been subjected to the noise removal process, to the display device 30 (step S11).

[0118] The image acquisition device 1 described above can also remove noise components while increasing signal components in an X-ray transmission image, and can effectively improve the S / N ratio in the X-ray transmission image. In addition, the average energy of the X-rays transmitted through the object F is calculated based on the operating conditions of the X-ray source when acquiring the X-ray transmission image of the object F or the imaging conditions of the X-ray transmission image. Then, based on the average energy, candidates for the trained model 206C used for noise removal are narrowed down from the trained models 206C constructed in advance. As a result, the trained model 206C corresponding to the average energy of the X-rays of the imaging object is used for noise removal, so that noise removal corresponding to the relationship between luminance and noise in the X-ray transmission image can be realized. As a result, noise in the X-ray transmission image can be effectively removed, and for example, foreign object detection performance can be improved. In particular, the noise state of the X-ray transmission image changes depending on the tube voltage, filter, scintillator, conditions of the X-ray detection camera (gain setting value, circuit noise value, saturation charge amount, conversion coefficient value (e- / count), camera line rate), object, etc. Therefore, when trying to realize noise removal by machine learning, it is necessary to prepare multiple learning models trained under various conditions. Conventionally, it has not been possible to select a learning model that matches the noise state from multiple learning models according to the conditions when measuring an X-ray transmission image. According to this embodiment, the learned model 206C corresponding to the average energy of the X-rays of the imaging target is selected, thereby realizing the selection of a learning model that always matches the noise state.

[0119] Generally, X-ray transmission images contain noise due to X-ray generation. In order to improve the S / N ratio of X-ray transmission images, it is possible to increase the X-ray dose, but in that case, there is a problem that increasing the X-ray dose increases the amount of exposure to the sensor, shortening the life of the sensor, and shortening the life of the X-ray generation source, making it difficult to improve the S / N ratio and extend the life at the same time. In this embodiment, there is no need to increase the X-ray dose, so it is possible to improve the S / N ratio and extend the life at the same time.

[0120] In addition, the control device 20C of this embodiment has a function of executing image processing to remove noise from the X-ray transmission image of the object F using the selected trained model 206C. With such a function, noise removal corresponding to the relationship between the luminance and noise in the X-ray transmission image can be realized, and the noise in the X-ray transmission image can be effectively removed.

[0121] In addition, the control device 20C of this embodiment has a function of narrowing down candidates for the trained model by comparing the value of the average energy of the X-rays calculated from the selection information with the value of the average energy specified from the image data used to construct the trained model 206C. With this function, it is possible to reliably realize noise removal corresponding to the relationship between the brightness and noise in the X-ray transmission image.

[0122] Furthermore, the control device 20C of this embodiment has a function of selecting the trained model 206C from the candidates based on the image characteristics of the X-ray transmission image of the jig. With this function, it is possible to select the trained model 206C that is optimal for removing noise from the X-ray transmission image of the object F. As a result, it is possible to more reliably achieve noise removal that corresponds to the relationship between brightness and noise in the X-ray transmission image.

[0123] [Modification of the second embodiment] The control device 20C in the second embodiment described above selected a candidate for the trained model 206C based on the value of the average energy of X-rays calculated from the condition information, but may also have a function to respond to performance deterioration of the X-ray detection camera 10, output fluctuations or performance deterioration of the X-ray irradiator 50.

[0124] 36 is a block diagram showing a functional configuration of a control device 20D according to a modification of the second embodiment. The control device 20D is different from the control device 20C according to the second embodiment in that it has a measurement unit 207C and in the functions of a calculation unit 202D and a narrowing down unit 203D.

[0125] In the control device 20C, the trained model 206C is narrowed down on the premise that the relationship between brightness and noise in an X-ray transmission image can be estimated from the average energy of X-rays, assuming that there is no performance degradation of the X-ray detection camera 10 and no output fluctuation or performance degradation of the X-ray irradiator 50. In contrast, the control device 20D according to this modification has a function of calculating an X-ray conversion coefficient in consideration of the performance degradation of the X-ray detection camera 10, the output fluctuation of the X-ray irradiator 50, or the performance degradation thereof, and narrowing down the trained model 206C based on the X-ray conversion coefficient. The X-ray conversion coefficient is a parameter indicating the efficiency of X-rays being converted into visible light by the scintillator and then converted into electrons (electrical signals) by the camera sensor.

[0126] In general, the X-ray conversion factor F T is expressed by the following formula, where E [keV] is the average energy of X-rays, EM [photon / keV] is the amount of scintillator light emitted, C is the coupling efficiency in the sensor, and QE is the quantum efficiency of the sensor; F T = E × EM × C × QE The signal-to-noise ratio (SNR) in an X-ray image can be calculated by the X-ray conversion coefficient F T and the number of X-ray photons, N P and the read noise Nr of the camera are used to obtain the following formula: SNR=F T N P / {(F T N P +Nr 2 ) 1 / 2} Therefore, the X-ray conversion factor F T Based on this, the relationship between brightness and noise in X-ray transmission images can be estimated, taking into account the degradation of camera performance.

[0127] The measurement unit 207C of the control device 20D has a function of measuring the amount of decrease in the amount of light emission EM as the performance deterioration of the scintillator layer 11, the amount of decrease in the quantum efficiency QE of the sensor as the performance deterioration of the scan camera 12, and the output fluctuation and the amount of change in the average energy E as the performance deterioration of the X-ray irradiator 50. For example, the measurement unit 207C measures the amount of decrease in the amount of light emission between the state without performance deterioration of the scintillator layer 11 (state when new) and the current scintillator layer 11, and estimates the current amount of light emission EM from the amount of decrease. In addition, the measurement unit 207C measures the amount of decrease in luminance between the state without performance deterioration of the scan camera 12 (state when new) and the current scan camera 12, and estimates the current quantum efficiency QE from the amount of decrease. In addition, the measurement unit 207C estimates the current average energy E from the amount of change in the average energy between the state without performance deterioration of the X-ray irradiator 50 (state when new) and the current X-ray irradiator 50. The average energy E may be determined from imaging data of a flat member whose thickness and material are known and whose relationship between the average X-ray energy and the X-ray transmittance is known, or may be determined from imaging data at multiple points on a jig whose thickness or material varies.

[0128] The calculation unit 202D of the control device 20D calculates an X-ray conversion coefficient F using the calculated average energy E of the X-rays and the light emission amount E and quantum efficiency QE estimated by the measurement unit 207C. T The narrowing unit 203D of the control device 20D calculates the calculated X-ray conversion coefficient F T and the X-ray conversion coefficient F in the image data used to build the trained model 206C T By comparing with the trained model 206C, it has the function of narrowing down the candidates for the trained model 206C.

[0129] In addition, the control device 20D of the above modification narrows down the candidates for the trained model and then selects the trained model based on the image characteristics obtained by imaging the jig, but it may also perform noise removal processing on the X-ray transmission image of the object without imaging the jig. Fig. 37 is a flowchart showing the procedure of the observation processing by the image acquisition device 1 according to another modification. In this way, it is also possible to omit the processing of steps S6 to S8 in Fig. 35 and perform noise removal processing using the trained model narrowed down based on the average energy. [Third embodiment]

[0130] 38 is a block diagram showing the functional configuration of a control device 20E according to the third embodiment. The control device 20E includes an acquisition unit 201E, an identification unit 202E, a selection unit 204E, and a processing unit 205E.

[0131] In addition, the control device 20E stores a plurality of trained models 206E in advance, which perform noise removal processing on X-ray transmission images. Each of the trained models 206E is a trained model by machine learning that is constructed in advance using image data as teacher data. Machine learning includes supervised learning, deep learning, reinforcement learning, neural network learning, and the like. In this embodiment, a two-dimensional convolutional neural network described in the paper "Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising" by Kai Zhang et al. is adopted as an example of a deep learning algorithm. The trained models 206E may be generated by an external computer or the like and downloaded to the control device 20E, or may be generated within the control device 20E.

[0132] FIG. 39 shows an example of image data that is training data used to construct the trained model 206E. As the training data, X-ray transmission images of patterns of various thicknesses, various materials, and various resolutions may be used. The example shown in FIG. 39 is an example of an X-ray transmission image generated for chicken. This image data may be an X-ray transmission image actually generated for a plurality of types of objects using the image acquisition device 1, or image data generated by simulation calculation. The X-ray transmission image may be acquired using a device other than the image acquisition device 1. In addition, the X-ray transmission image and image data generated by simulation calculation may be used in combination. The trained models 206E are each image data obtained for transmitted X-rays with different average energies, and are constructed in advance using image data with known noise distribution. The average energy of X-rays in the image data is set to different values ​​in advance by setting the operating conditions of the X-ray irradiator (radiation source) 50 of the image acquisition device 1 or the imaging conditions of the image acquisition device 1, or by setting the operating conditions or imaging conditions of the X-ray irradiator 50 during the simulation calculation. That is, the plurality of trained models 206E are constructed by machine learning using, as training data, training images that are X-ray images corresponding to the average energy of X-rays transmitted through the object F calculated based on condition information indicating the operating conditions of the X-ray irradiator (radiation source) 50 when capturing an X-ray transmission image of the object F or the imaging conditions of the X-ray detection camera 10 (construction step). For example, in this embodiment, the plurality of trained models 206E are constructed using a plurality of frames (e.g., 20,000 frames) of a plurality of types of image data in which the average energy is set to 10 keV increments, such as 10 keV, 20 keV, 30 keV, ...

[0133] Image data, which is training data used to construct the trained model 206E, is generated by a creation procedure similar to that in the first embodiment described above.

[0134] Returning to FIG. 38, the functions of the functional units of the control device 20E will now be described in detail.

[0135] The acquisition unit 201E acquires an X-ray transmission image captured by irradiating the jig and the object F with X-rays using the image acquisition device 1. As the jig, a flat plate member whose thickness and material are known and whose relationship between the average energy of X-rays and the X-ray transmittance is known, or a jig having a chart imaged at various resolutions is used. That is, the acquisition unit 201E acquires an X-ray transmission image of the jig captured using the image acquisition device 1 prior to the observation process of the object F. Then, the acquisition unit 201E acquires an X-ray transmission image of the object F captured using the image acquisition device 1 at a timing after the trained model 206E is selected based on the X-ray transmission image of the jig. However, the timing of acquiring the X-ray transmission images of the jig and the object F is not limited to the above, and may be simultaneous or reversed.

[0136] The identifying unit 202E identifies image characteristics of the X-ray transmission image of the jig acquired by the acquiring unit 201 E. Specifically, the selecting unit 204E identifies energy characteristics, noise characteristics, resolution characteristics, frequency characteristics, or the like as the image characteristics of the X-ray transmission image.

[0137] For example, when a flat plate member with known thickness and material is used as the jig, the identifying unit 202E compares the brightness of the X-ray image transmitted through the jig with the brightness of the X-ray image transmitted through air to calculate the transmittance of X-rays at one point (or the average of multiple points) on the jig. For example, when the brightness of the X-ray image transmitted through the jig is 5,550 and the brightness of the X-ray image transmitted through air is 15,000, the identifying unit 202E calculates the transmittance to be 37%. The identifying unit 202E then identifies the average energy of the transmitted X-rays (e.g., 50 keV) estimated from the transmittance of 37% as the energy characteristic of the X-ray transmission image of the jig.

[0138] Furthermore, the identification unit 202E may analyze the characteristics at multiple points of a jig whose thickness or material changes as the energy characteristics of the X-ray transmission image of the jig. FIG. 40 is a diagram showing an example of an X-ray transmission image to be analyzed by the identification unit 202E. FIG. 40 is an X-ray transmission image of a jig whose thickness changes stepwise. The identification unit 202E selects multiple measurement regions (ROI: Region Of Interest) with different thicknesses from such an X-ray transmission image, analyzes the average brightness value for each of the multiple measurement regions, and acquires a thickness-brightness characteristic graph as the energy characteristics. FIG. 41 shows an example of the thickness-brightness characteristic graph acquired by the identification unit 202E.

[0139] The determination unit 202E can also analyze the luminance value and noise for each of a plurality of measurement regions as the noise characteristics of the X-ray transmission image of the jig, and obtain a characteristic graph of the luminance-noise ratio as the noise characteristics. That is, the determination unit 202E selects a plurality of measurement regions ROI having different thicknesses or materials from the X-ray transmission image, analyzes the standard deviation of the luminance values ​​and the average value of the luminance values ​​of the plurality of measurement regions ROI, and obtains a characteristic graph of the luminance-SNR (S / N ratio) as the noise characteristics. At this time, the determination unit 202E calculates the SNR for each measurement region ROI by SNR=(average value of the luminance values)÷(standard deviation of the luminance values). FIG. 42 shows an example of the luminance-SNR characteristic graph obtained by the determination unit 202E. Here, the determination unit 202E may obtain a characteristic graph in which the vertical axis represents the noise calculated from the standard deviation of the luminance values, instead of the above-mentioned luminance-SNR characteristic graph, as the noise characteristics.

[0140] In addition, when a jig having a chart is used, the specifying unit 202E can also acquire the distribution of resolution in the X-ray transmission image of the jig as the resolution characteristic. Furthermore, the specifying unit 202E has a function of acquiring the resolution characteristic for an image after applying a plurality of trained models 206E to the X-ray transmission image of the jig and performing a noise removal process. FIG. 43 shows an example of an X-ray transmission image used for evaluating the resolution. In this X-ray transmission image, a chart whose resolution changes stepwise in one direction is taken as the imaging target. The resolution of the X-ray transmission image can be measured using MTF (Modulation Transfer Function) or CTF (Contrast Transfer Function).

[0141] 38 again, the selection unit 204E selects, based on the image characteristics acquired by the identification unit 202E, from among the multiple trained models 206E stored in the control device 20E, a trained model 206E to be finally used for noise removal processing of the X-ray transmission image of the object F. That is, the selection unit 204E compares the image characteristics identified by the identification unit 202E with image characteristics identified from the image data used to construct the multiple trained models 206E, and selects a trained model 206E in which the two are similar.

[0142] For example, the selection unit 204E selects one of the trained models 206E constructed using image data of an average energy closest to the value of the average energy of the transmitted X-rays identified by the identification unit 202E.

[0143] In addition, the selection unit 204E acquires a thickness-brightness characteristic graph for the image data used to construct the multiple trained models 206E in the same manner as the identification method by the identification unit 202E, and selects the trained model 206E constructed by the image data having the closest characteristics to the thickness-brightness characteristic graph acquired for the jig as the final trained model 206E. However, the image characteristics of the image data used to construct the trained model 206E may be calculated in advance outside the control device 20E. In this way, by using the image characteristics obtained by setting multiple measurement areas, it is possible to select a trained model that is optimal for removing noise from the X-ray transmission image of the object F. In particular, it is possible to accurately estimate the difference in the X-ray spectrum or the effect of the filter when measuring the X-ray transmission image.

[0144] The selection unit 204E may select the trained model 206E constructed by image data having the brightness-noise ratio characteristic closest to the brightness-noise ratio characteristic acquired by the identification unit 202E as the final trained model 206E. However, the image characteristics of the image data used to construct the trained model 206E may be acquired by the selection unit 204E from the image data, or may be calculated in advance outside the control device 20E. Here, the selection unit 204E may select the trained model 206E using the brightness-noise characteristic instead of the brightness-noise ratio characteristic as the noise characteristic. By using such brightness-noise characteristic, it is possible to identify the dominant noise factor (shot noise, read noise, etc.) for each signal amount detected by the X-ray detection camera 10 from the slope of the graph of each signal amount area, and select the trained model 206E based on the identification result.

[0145] Fig. 44 is a diagram for explaining the function of selecting a trained model based on image characteristics by the selection unit 204E. In Fig. 44, part (a) shows characteristic graphs G1, G2, and G3 of the brightness-SNR of each image data used to construct a plurality of trained models 206E, and part (b) shows, in addition to these characteristic graphs G1, G2, and G3, a characteristic graph G1 of the brightness-SNR of an X-ray transmission image of an image of a jig. T The graph shows the characteristics of G1, G2, G3, and G T In the case of the target, the selection unit 204E selects the characteristic graph G T The selection unit 204E selects a trained model 206E constructed by image data of the characteristic graph G2 that is closest to the characteristic of the characteristic graph G1. T The SNR error for each luminance value at a fixed interval between and is calculated, the root mean squared error (RMSE) of the errors is calculated, and the trained model 206E corresponding to the characteristic graphs G1, G2, and G3 with the smallest mean squared error is selected. The selection unit 204E can also select the trained model 206E in the same manner when selecting using energy characteristics.

[0146] The selection unit 204E can also select a trained model 206E used to generate an image with relatively superior characteristics based on the characteristics of the image after noise removal processing is performed by applying multiple trained models to an X-ray transmission image of a jig.

[0147] For example, the selection unit 204E applies multiple trained models 206E to an X-ray transmission image of a jig having charts of various resolutions, and evaluates the resolution characteristics of the resulting noise-removed image. The selection unit 204E then selects the trained model 206E used for the image in which the change in resolution of each distribution before and after the noise removal process is the smallest.

[0148] In addition to evaluating the change in resolution, the selection unit 204E may evaluate the characteristics of the brightness-to-noise ratio of the image after noise removal, and select the trained model 206E used to generate the image with the highest characteristics. FIG. 45 shows an example of the structure of a jig used to evaluate the brightness-to-noise ratio. For example, a jig may be used in which foreign bodies P2 of various materials and sizes are scattered in a member P1 whose thickness changes stepwise in one direction. FIG. 46 shows an X-ray transmission image after noise removal processing obtained for the jig of FIG. 45. The selection unit 204E selects an image region R1 including an image of the foreign body P2 in the X-ray transmission image and an image region R2 not including an image of the foreign body P2 near the region R1, and selects the minimum brightness value L in the image region R1. MIN and the average luminance value L in the image region R2 AVE and the standard deviation of the luminance in the image region R2, L SD Then, the selection unit 204E calculates the following formula: CNR=(L AVE -L MIN ) / L SD Further, the selection unit 204E calculates the brightness-to-noise ratio CNR for each of the X-ray transmission images after application of the multiple trained models 206E, and selects the trained model 206E used to generate the X-ray transmission image with the highest brightness-to-noise ratio CNR.

[0149] Alternatively, the selection unit 204E selects the average luminance L in the image region R1. AVE_R1 and the average luminance value L in the image region R2 AVE_R2 and the standard deviation of the luminance in the image region R2, L SD Based on this, the calculation may be performed according to the following formula: CNR=(L AVE_R1 -L MIN_R2 ) / L SD

[0150] The processing unit 205E applies the trained model 206E selected by the selection unit 204E to the X-ray transmission image acquired of the object F, and executes image processing to remove noise, thereby generating an output image. Then, the processing unit 205E outputs the generated output image to the display device 30 or the like.

[0151] Next, the procedure of observation processing of an X-ray transmission image of the object F using the image acquisition device 1 according to the third embodiment, i.e., the flow of the radiographic image acquisition method according to the third embodiment, will be described. Fig. 47 is a flow chart showing the procedure of observation processing by the image acquisition device 1.

[0152] First, an operator (user) of the image acquisition device 1 sets imaging conditions in the image acquisition device 1, such as the tube voltage of the X-ray irradiator 50 or the gain of the X-ray detection camera 10 (step S1E). Next, a jig is set in the image acquisition device 1, and an X-ray transmission image of the jig is acquired by the control device 20E (step S2E). At this time, X-ray transmission images of multiple types of jigs may be acquired sequentially.

[0153] In response to this, the control device 20E specifies image characteristics (energy characteristics, noise characteristics, and resolution characteristics) of the X-ray transmission image of the jig (step S3E). Furthermore, the control device 20E applies a plurality of trained models 206E to the X-ray transmission image of the jig, and specifies image characteristics (resolution characteristics or brightness-to-noise ratio value, etc.) of each of the X-ray transmission images after the application of the plurality of trained models 206E (step S4E).

[0154] Next, the control device 20E selects the learned model 206E based on the comparison result between the energy characteristics of the X-ray transmission image of the jig and the energy characteristics of the image data used to construct the learned model 206E, and the degree of change in the resolution characteristics of the X-ray transmission image of the jig before and after application of the learned model (step S5E). Here, the learned model 206E may be selected based on the comparison result between the noise characteristics of the X-ray transmission image of the jig and the noise characteristics of the image data used to construct the learned model 206E, and the change state of the resolution characteristics of the X-ray transmission image of the jig before and after application of the learned model. Also, in step S5E, instead of the above processing, the learned model 206E having the highest brightness-to-noise ratio CNR after application of the learned model of the X-ray transmission image of the jig may be selected.

[0155] Furthermore, the object F is set in the image acquisition device 1 and an X-ray transmission image of the object F is acquired by capturing an image of the object F (step S7E). Next, the control device 20E applies the finally selected trained model 206E to the X-ray transmission image of the object F, thereby executing a noise removal process on the X-ray transmission image (step S8E). Finally, the control device 20E outputs an output image, which is the X-ray transmission image subjected to the noise removal process, to the display device 30 (step S9E).

[0156] The image acquisition device 1 described above can also remove noise components while increasing signal components in an X-ray transmission image, and can effectively improve the S / N ratio in the X-ray transmission image. In addition, the image characteristics of the X-ray transmission image of the jig are specified, and a trained model used for noise removal is selected from trained models constructed in advance based on the image characteristics. This makes it possible to estimate the characteristics of the X-ray transmission image that change depending on the operating conditions of the X-ray irradiator 50 in the image acquisition device 1, and the trained model 206E selected according to this estimation result is used for noise removal, so that noise removal corresponding to the relationship between brightness and noise in the X-ray transmission image can be realized. As a result, noise in the X-ray transmission image can be effectively removed.

[0157] Generally, X-ray transmission images contain noise due to X-ray generation. In order to improve the S / N ratio of X-ray transmission images, it is possible to increase the X-ray dose, but in that case, there is a problem that increasing the X-ray dose increases the amount of exposure to the sensor, shortening the life of the sensor, and shortening the life of the X-ray generation source, making it difficult to improve the S / N ratio and extend the life at the same time. In this embodiment, there is no need to increase the X-ray dose, so it is possible to improve the S / N ratio and extend the life at the same time.

[0158] In the present embodiment, in selecting the trained model, the image characteristics of the X-ray transmission image of the jig are compared with the image characteristics of the image data used to construct the trained model. This allows the trained model 206E constructed using image data corresponding to the image characteristics of the X-ray transmission image of the jig to be selected, so that noise in the X-ray transmission image of the object F can be effectively removed.

[0159] In addition, in this embodiment, a trained model is selected using image characteristics of an image in which a plurality of trained models 206E are applied to an X-ray transmission image of a jig. In this case, the trained model 206E is selected based on image characteristics of an X-ray transmission image of a jig to which a plurality of trained models 206E are actually applied, so that noise in the X-ray transmission image of the object F can be effectively removed.

[0160] In particular, in this embodiment, energy characteristics or noise characteristics are used as the image characteristics. In this case, a trained model 206E constructed by an image having characteristics similar to the energy characteristics or noise characteristics of the X-ray transmission image of the jig that change depending on the imaging conditions of the image acquisition device 1 is selected. As a result, it is possible to remove noise in the X-ray transmission image of the object F that corresponds to the change in the conditions of the image acquisition device 1.

[0161] In this embodiment, the resolution characteristic or the brightness-to-noise ratio is also used as the image characteristic. With this configuration, by applying the selected trained model 206E, it becomes possible to obtain an X-ray transmission image with a good resolution characteristic or brightness-to-noise ratio. As a result, it becomes possible to remove noise from the X-ray transmission image of the object corresponding to the change in the conditions of the image acquisition device 1. [Other Modifications of the First to Third Embodiments]

[0162] The image acquisition device 1 according to the first to third embodiments may include a scan camera 12A and a scintillator layer 11A having the following configurations. Fig. 48 is a plan view showing the configuration of a scan camera 12A according to a modified example. Fig. 49 is a diagram showing the configuration of a scintillator layer 11A according to a modified example that is arranged on the scan camera 12A, showing a cross-sectional view along the transport direction TD at the top and a plan view seen from the surface of the substrate 71 on the pixel 72 side at the bottom.

[0163] Scan camera 12A has N pixels 72 (N is an integer of 2 or more) arranged on substrate 71 along a direction substantially perpendicular to transport direction TD. For example, the number of pixels N is an arbitrary integer between 200 and 30,000. Scintillator layer 11 is formed with P (P is an integer of 2 or more) rectangular scintillator portions 11a that are separated and arranged corresponding to the N pixels 72 arranged along a direction substantially perpendicular to transport direction TD, and separation portions 11b located between these scintillator portions 11a.

[0164] The readout circuit 73 provided in the scan camera 12A configured as described above sequentially or simultaneously receives detection signals output from the N pixels 72 at intervals of a predetermined detection period in response to control by the sensor control unit 13, combines the detection signals from the N pixels 72, and outputs to the outside as a detection signal of one line of the object F perpendicular to the transport direction TD. In addition, the readout circuit 73 outputs a detection signal of the next line of the object F perpendicular to the transport direction TD by processing the detection signals output sequentially or simultaneously from the N pixels 72 with a shift in the predetermined detection period. In the same manner, the readout circuit 73 sequentially or simultaneously outputs detection signals of multiple lines of the object F perpendicular to the transport direction TD.

[0165] The above-mentioned specified detection period may be set based on, for example, the speed of the conveyor belt 60, the distance between the X-ray irradiator 50 and the object F on the conveyor belt 60 (FOD (Focus Object Distance)), the distance between the X-ray irradiator 50 and the scan camera 12 (FDD (Focus Detector Distance)), and the detection interval of pixels 72 on the object F.

[0166] Generally, by forming a separation portion 11b that blocks or reflects scintillation light, it is possible to prevent the scintillation light from entering the adjacent scintillator portion, and to prevent a decrease in resolution. However, since the portion where the separation portion 11b exists does not contribute to the conversion of the incident X-rays to scintillation light, there is also a problem that the detection signal becomes smaller (decreased) according to the size (area) of the separation portion 11b. The decrease in the detection signal means a decrease in the S / N ratio. Therefore, it has been considered difficult to achieve both an improvement in resolution and an improvement in the S / N ratio by using a scintillator having the separation portion 11b. In contrast, according to this modification, the resolution of the X-ray image can be increased by using the scintillator portion 11a separated corresponding to the pixel 72, and the S / N ratio in the X-ray image can be effectively improved.

[0167] As another modified example, the control device 20 may perform a noise removal process by applying a filter process to the X-ray image instead of inputting the X-ray image to the learned model and performing the noise removal process. For example, the control device 20 performs an edge-preserving filter process as such a filter process. More specifically, the control device 20 performs a process using a median filter, a bilateral filter, an NLM (Non Local Mean) filter, a total variation filter, BM3D, or the like as the edge-preserving filter process. At this time, the control device 20 may perform a combination of multiple types of filter processes.

[0168] According to the above-mentioned another modified example, noise components can be appropriately removed from the X-ray image, and the S / N ratio of the radiation image can be further improved. Specifically, when an X-ray image is acquired using the scan camera 12 and the scintillator layer 11, and a noise removal process using edge-preserving filter processing is performed, it has been found that the CNR is improved by about two times compared to when an X-ray image is acquired using a scintillator that covers the entire scan camera 12 as a whole and no noise removal process is performed, and false signals are sufficiently removed from the binarized image generated by performing threshold processing. As a result, according to this modified example, in the application of foreign body inspection, it is possible to prevent foreign bodies from overlapping and being detected.

[0169] Moreover, the control device 20 according to the above-mentioned other modified example may perform edge enhancement processing on the X-ray image in addition to the filter processing. For example, the control device 20 performs processing using an edge enhancement filter, unsharp mask processing, etc., as such edge enhancement processing. At this time, the control device 20 may perform a combination of a plurality of types of processing. In this case, in addition to improving the S / N ratio, the contrast of the X-ray image can be improved. Specifically, when an X-ray image is acquired using the scan camera 12 and the scintillator layer 11 and both edge-preserving filter processing and edge enhancement processing are performed, it has been found that the CNR is improved by about four times compared to when an X-ray image is acquired using a scintillator that covers the entire scan camera 12 as a whole and noise removal processing is not performed, and false signals are sufficiently removed in the binarized image generated by performing threshold processing. As a result, even with this modified example, it is possible to prevent foreign objects from being detected as overlapping with each other in the application of foreign object inspection. [Explanation of symbols]

[0170] 1...image acquisition device (radiation image acquisition device, radiation image acquisition system), 10...X-ray detection camera (imaging device), 11, 11A...scintillator layer, 11a...scintillator section, 11b...separation section, 12, 12A...scan camera, 20, 20A to 20E...control device (image processing module), 50...X-ray irradiator (radiation source), 60...belt conveyor (transport device), 72...pixel (detection element), 74...image Elementary line (pixel group), 73...readout circuit, 201, 201C...input section, 202, 202A, 202C, 202D...calculation section, 202E...identification section, 203C, 203D...narrowing section, 204, 204A, 204B...noise map generation section, 204C, 204E...selection section, 205, 205C, 205E...processing section, 206C, 206E, 207...trained model, F...target object, TD...transport direction (one direction).

Claims

1. an imaging device that scans and captures radiation that has passed through an object in one direction to obtain a radiation image; a scintillator layer provided on the imaging device and configured to convert the radiation into light; an image processing module that executes a noise reduction process to remove noise from the radiation image; Equipped with The imaging device includes: A detection element configured to be arranged in a direction perpendicular to the one direction and to detect the light and output a detection signal, wherein N pieces of the detection element (N is an integer of 2 or more) are arranged in the direction perpendicular to the one direction; a readout circuit that outputs the detection signals for each of the N detection elements to output the radiation image; Including, The scintillator layer comprises: P scintillator sections (P is an integer of 2 or more) arranged separately corresponding to the N detection elements; A separator disposed between the P scintillator sections; Including, The image processing module includes: inputting the radiographic image into a trained model constructed by machine learning using image data in advance, and executing a noise removal process to remove noise from the radiographic image; a noise map generating unit that derives an evaluation value from the pixel values ​​of each pixel of the radiographic image based on relationship data indicating the relationship between pixel values ​​and an evaluation value that evaluates a spread of noise values, and generates a noise map that is data correlating the derived evaluation value with each pixel of the radiographic image; a processing unit that inputs the radiographic image and the noise map into the trained model and executes a noise removal process that removes noise from the radiographic image; having Radiography image acquisition device.

2. The imaging device includes: a pixel line having M pixels (M is an integer of 2 or more) arranged along the one direction, the pixel line being arranged in N rows (N is an integer of 2 or more) in a direction perpendicular to the one direction, the detection element outputting a detection signal related to the light for each pixel; the readout circuit adding up the detection signals output from at least two of the M pixels for each pixel line of the N columns of the detection elements and outputting the N detection signals obtained by the addition processing, thereby outputting the radiation image; Including, The radiation image acquisition device according to claim 1 .

3. The trained model is constructed by machine learning using image data obtained by adding a noise value along a normal distribution to a radiation image of a predetermined structure as training data. The radiation image acquisition device according to claim 1 .

4. The trained model is constructed by machine learning using a radiation image obtained by using the scintillator layer as training data. The radiation image acquisition device according to claim 1 .

5. The image processing module includes: an input unit that receives input of condition information indicating either a condition of a radiation source or an imaging condition when an object is imaged by irradiating the object with radiation; a calculation unit that calculates an average energy of the radiation that has passed through the object based on the condition information; a narrowing down unit that narrows down a trained model to be used in the noise removal process from among a plurality of trained models each constructed in advance by machine learning using image data, based on the average energy; The radiation image acquiring device according to claim 1 , further comprising:

6. The image processing module includes: an identification unit that identifies image characteristics of a radiation image acquired by the imaging device with the jig as a target; A selection unit that selects a trained model from a plurality of trained models each constructed by machine learning using image data in advance based on the image characteristics; A processing unit that executes the noise removal process using the selected trained model; The radiation image acquiring device according to claim 1 , further comprising:

7. the image processing module performs a filter process on the radiographic image to perform a noise removal process to remove noise from the radiographic image; The radiation image acquiring device according to claim 1 .

8. the image processing module performs edge enhancement processing on the radiation image in addition to the filtering processing; The radiation image acquiring device according to claim 7 .

9. A radiation image acquisition device according to any one of claims 1 to 8, a source for irradiating the object with radiation; a conveying device that conveys the object in the one direction relative to the imaging device; A radiation image acquisition system comprising:

10. a step of acquiring a radiation image by scanning the object in one direction with light corresponding to the radiation transmitted through the object; and performing a noise removal process to remove noise from the radiation image. In the step of acquiring, an imaging device including N (N is an integer of 2 or more) detector elements arranged in a direction perpendicular to the one direction and which detect the light and output detection signals; and a scintillator layer which converts the radiation into light and which includes P (P is an integer of 2 or more) scintillator sections arranged separately corresponding to the N detector elements and separation sections arranged between the P scintillator sections, and outputs the detection signals for each of the N detector elements, thereby outputting the radiation image; In the step of performing, inputting the radiographic image into a trained model constructed by machine learning using image data in advance, and executing a noise removal process to remove noise from the radiographic image; deriving an evaluation value from the pixel values ​​of each pixel of the radiographic image based on relationship data indicating the relationship between pixel values ​​and an evaluation value evaluating the spread of noise values, generating a noise map which is data associating the derived evaluation value with each pixel of the radiographic image, inputting the radiographic image and the noise map into the trained model, and executing a noise removal process for removing noise from the radiographic image; Radiographic image acquisition method.

11. the imaging device includes a pixel line having M pixels (M is an integer of 2 or more) arranged along the one direction, the pixel line being arranged in N rows (N is an integer of 2 or more) in a direction perpendicular to the one direction, the detection element outputting a detection signal related to the light for each of the pixels, In the acquiring step, the detection signals output from at least two of the M pixels are added for each pixel line of the N columns of the detection elements, and the N detection signals obtained by the addition are output, thereby outputting the radiation image. The method for acquiring a radiation image according to claim 10.

12. The trained model is constructed by machine learning using image data obtained by adding a noise value along a normal distribution to a radiation image of a predetermined structure as training data. The method for acquiring a radiation image according to claim 10.

13. The trained model is constructed by machine learning using a radiation image obtained by using the scintillator layer as training data. The method for acquiring a radiation image according to claim 10.

14. In the executing step, an input of condition information indicating either a condition of a radiation source or an imaging condition when an object is imaged by irradiating the object with radiation is accepted, an average energy of the radiation transmitted through the object is calculated based on the condition information, and a trained model to be used for the noise removal process is narrowed down from a plurality of trained models each constructed in advance by machine learning using image data based on the average energy. The method for obtaining a radiological image according to claim 10 .

15. In the executing step, image characteristics of a radiographic image acquired of a jig are identified, a trained model is selected from a plurality of trained models each constructed in advance by machine learning using image data based on the image characteristics, and the noise removal process is performed using the selected trained model. The method for obtaining a radiological image according to claim 10 .

16. In the performing step, a noise removal process is performed by filtering the radiation image to remove noise from the radiation image. The method for obtaining a radiation image according to claim 10 or 11.

17. In the performing step, an edge enhancement process is performed on the radiation image in addition to the filtering process. The method of claim 16 .

18. irradiating the object with radiation; conveying the object in the one direction relative to the detection element; The radiation image acquiring method according to any one of claims 10 to 17, further comprising:

19. An imaging device that scans and captures radiation that has passed through an object in one direction to obtain a radiation image; a scintillator layer provided on the imaging device and configured to convert the radiation into light; an image processing module that executes a noise reduction process to remove noise from the radiation image; Equipped with The imaging device includes: A detection element configured to be arranged in a direction perpendicular to the one direction and to detect the light and output a detection signal, wherein N pieces of the detection element (N is an integer of 2 or more) are arranged in the direction perpendicular to the one direction; a readout circuit that outputs the detection signals for each of the N detection elements to output the radiation image; Including, The scintillator layer comprises: P scintillator sections (P is an integer of 2 or more) arranged separately corresponding to the N detection elements; A separator disposed between the P scintillator sections; Including, The image processing module includes: inputting the radiographic image into a trained model constructed by machine learning using image data in advance, and executing a noise removal process to remove noise from the radiographic image; an input unit that receives input of condition information indicating either a condition of a radiation source or an imaging condition when an object is imaged by irradiating the object with radiation; a calculation unit that calculates an average energy of the radiation that has passed through the object based on the condition information; a narrowing down unit that narrows down a trained model to be used in the noise removal process from among a plurality of trained models each constructed in advance by machine learning using image data, based on the average energy; having Radiography image acquisition device.

20. An imaging device for scanning radiation transmitted through an object in one direction to capture an image and obtain a radiation image; a scintillator layer provided on the imaging device and configured to convert the radiation into light; an image processing module that executes a noise reduction process to remove noise from the radiation image; Equipped with The imaging device includes: A detection element configured to be arranged in a direction perpendicular to the one direction and to detect the light and output a detection signal, wherein N pieces of the detection element (N is an integer of 2 or more) are arranged in the direction perpendicular to the one direction; a readout circuit that outputs the detection signals for each of the N detection elements to output the radiation image; Including, The scintillator layer comprises: P scintillator sections (P is an integer of 2 or more) arranged separately corresponding to the N detection elements; A separator disposed between the P scintillator sections; Including, The image processing module includes: inputting the radiographic image into a trained model constructed by machine learning using image data in advance, and executing a noise removal process to remove noise from the radiographic image; an identification unit that identifies image characteristics of a radiation image acquired by the imaging device with the jig as a target; A selection unit that selects a trained model from a plurality of trained models each constructed by machine learning using image data in advance based on the image characteristics; A processing unit that executes the noise removal process using the selected trained model; having Radiography image acquisition device.

21. A step of acquiring a radiation image by scanning and capturing an image of a target object in one direction with light corresponding to radiation transmitted through the target object; and performing a noise removal process to remove noise from the radiation image. In the step of acquiring, an imaging device including N (N is an integer of 2 or more) detector elements arranged in a direction perpendicular to the one direction and which detect the light and output detection signals; and a scintillator layer which converts the radiation into light and which includes P (P is an integer of 2 or more) scintillator sections arranged separately corresponding to the N detector elements and separation sections arranged between the P scintillator sections, and outputs the detection signals for each of the N detector elements, thereby outputting the radiation image; In the step of performing, inputting the radiographic image into a trained model constructed by machine learning using image data in advance, and executing a noise removal process to remove noise from the radiographic image; accepts input of condition information indicating either a condition of a radiation source or an imaging condition when an object is imaged by irradiating the object with radiation, calculates an average energy of the radiation that has passed through the object based on the condition information, and narrows down a trained model to be used in the noise removal process from among a plurality of trained models each constructed in advance by machine learning using image data based on the average energy; Radiographic image acquisition method.

22. A step of acquiring a radiation image by scanning and capturing an image of a target object in one direction with light corresponding to radiation transmitted through the target object; and performing a noise removal process to remove noise from the radiation image. In the step of acquiring, an imaging device including N (N is an integer of 2 or more) detector elements arranged in a direction perpendicular to the one direction and which detect the light and output detection signals; and a scintillator layer which converts the radiation into light and which includes P (P is an integer of 2 or more) scintillator sections arranged separately corresponding to the N detector elements and separation sections arranged between the P scintillator sections, and outputs the detection signals for each of the N detector elements, thereby outputting the radiation image; In the step of performing, inputting the radiographic image into a trained model constructed by machine learning using image data in advance, and executing a noise removal process to remove noise from the radiographic image; Identifying image characteristics of a radiographic image acquired of the jig as a target, selecting a trained model from among a plurality of trained models each constructed in advance by machine learning using image data based on the image characteristics, and executing the noise removal process using the selected trained model. Radiographic image acquisition method.

23. An imaging device for scanning radiation transmitted through an object in one direction to capture an image and obtain a radiation image; a scintillator layer provided on the imaging device and configured to convert the radiation into light; an image processing module that executes a noise reduction process to remove noise from the radiation image; Equipped with The imaging device includes: A detection element configured to be arranged in a direction perpendicular to the one direction and to detect the light and output a detection signal, wherein N pieces of the detection element (N is an integer of 2 or more) are arranged in the direction perpendicular to the one direction; a readout circuit that outputs the detection signals for each of the N detection elements to output the radiation image; Including, The scintillator layer comprises: P scintillator sections (P is an integer of 2 or more) arranged separately corresponding to the N detection elements; A separator disposed between the P scintillator sections; Including, The image processing module includes: inputting the radiographic image into a trained model constructed by machine learning using image data in advance, and executing a noise removal process to remove noise from the radiographic image; The trained model is constructed by machine learning using image data obtained by adding a noise value to the pixel value of each pixel of a radiation image of a structure obtained by simulation calculation as training data. Radiography image acquisition device.

24. A step of acquiring a radiation image by scanning and capturing an image of a target object in one direction with light corresponding to radiation transmitted through the target object; and performing a noise removal process to remove noise from the radiation image. In the step of acquiring, an imaging device including N (N is an integer of 2 or more) detector elements arranged in a direction perpendicular to the one direction and which detect the light and output detection signals; and a scintillator layer which converts the radiation into light and which includes P (P is an integer of 2 or more) scintillator sections arranged separately corresponding to the N detector elements and separation sections arranged between the P scintillator sections, and outputs the detection signals for each of the N detector elements, thereby outputting the radiation image; In the step of performing, inputting the radiographic image into a trained model constructed by machine learning using image data in advance, and executing a noise removal process to remove noise from the radiographic image; The trained model is constructed by machine learning using image data obtained by adding a noise value to the pixel value of each pixel of a radiation image of a structure obtained by simulation calculation as training data. Radiographic image acquisition method.

Images