JP2021094219A - Image processing method, image processing device, and program - Google Patents

Abstract

To provide an image processing method, an image processing device, and a program capable of executing correction processing for properly reducing or eliminating a structure component even on a structure whose contrast is especially high for an image acquired by photographing.SOLUTION: When executing photographing of a subject M by arranging a plurality of radiation detectors 10 in a state that part of them is overlapped, a position of a pixel indicating a predetermined structure such as a screw is specified from image information on an image region of an overlapped part of the radiation detectors 10, which is a correction region ArR for which correction is required, and the image information is subjected to correction processing for reducing/eliminating a structure component Cs of the predetermined structure by using machine learning for the correction region ArR including the pixel indicating the predetermined structure.SELECTED DRAWING: Figure 6

Description

The present invention relates to an image processing method, an image processing apparatus and a program.

In recent years, there have been known imaging devices that use a radiation detector (Flat Panel Detector) to radiograph a relatively wide range, such as the entire spinal column and all lower limbs of a patient, that exceeds the imaging range of a single radiation detector.
For example, Patent Document 1 and Patent Document 2 disclose a radiation imaging system that performs imaging using an imaging table having a holder capable of loading three radiation detecting devices in a state of being arranged in the body axis direction. There is.

Further, when such an imaging table is used, it is also possible to perform one-shot imaging in which a wide range of imaging is performed by irradiating radiation only once.
In one-shot photography, the burden on the patient who is the subject can be reduced by reducing the number of irradiations, and the patient moves while taking a wide range of images compared to the case where the image is repeated in multiple times. There is also the merit of being able to avoid the problem of physical movement.

However, when using a plurality of radiation detection devices as described above, a part of the radiation detection devices (for example, upper and lower ends) should be overlapped and arranged in the holder so as not to cause omission in the radiation detection area. It has become.
Therefore, in this superposed portion, the structure of the radiation detection device arranged on the front side close to the radiation irradiation device is reflected in the image obtained by the radiation detection device arranged on the back side far from the radiation irradiation device. There's a problem.

In this regard, in Patent Document 1 and Patent Document 2, it is possible to remove the structural component of the structure reflected in the image from the photographed image by performing the calibration correction using the image for calibration. Proposed.

Japanese Unexamined Patent Publication No. 2016-198424 Japanese Unexamined Patent Publication No. 2017-225724

However, some structures and the like included in the radiation detection device show high contrast.
For such a structure having a particularly high contrast, it may not be possible to reduce or remove the structure component from the image even if a conventional correction process such as calibration correction is performed.

The present invention has been made in view of the above circumstances, and it is necessary to perform a correction process on an image obtained by photographing to appropriately reduce or remove structural components even for a structure having a particularly high contrast. It is an object of the present invention to provide an image processing method, an image processing apparatus, and a program capable of performing the above.

The invention according to claim 1 is an image processing method.
When multiple radiation detection devices that detect radiation are arranged in a state where some of them are superimposed and a subject is photographed.
The position of the pixel indicating the predetermined structure is specified from the image information of the image area of the superimposed portion of the radiation detection device, which is the correction area requiring correction, and the correction area including the pixel indicating the predetermined structure is provided. It is characterized in that a correction process for reducing a structural component of the predetermined structure is performed on the image information by using machine learning.

The invention according to claim 7 is an image processing apparatus.
A control unit that performs image processing by the image processing method according to any one of claims 1 to 6 is provided.

The invention according to claim 8 is a program.
It is characterized in that a computer of an image processing apparatus is made to execute image processing by the image processing method according to any one of claims 1 to 6.

By performing image processing as in the present invention, it is possible to perform correction processing for appropriately reducing or removing structural components even for a structure having a particularly high contrast with respect to the image obtained by photographing.

It is a figure which shows the whole structure of the radiation imaging system in this embodiment. It is a schematic cross-sectional view of a main part which shows the front radiation detection device, the rear radiation detection device, etc. in the holder of the photographing table. It is a block diagram of the main part of the console in this embodiment. It is a figure which shows the example of the image obtained by the front radiation detection apparatus shown in FIG. 2 and the image obtained by the rear radiation detection apparatus. It is a flowchart which shows the image correction processing in this embodiment. It is a flowchart which shows the detail of the reduction / removal process of the predetermined structural component shown in FIG. It is explanatory drawing explaining the calibration correction. It is explanatory drawing explaining the correction using machine learning. It is explanatory drawing explaining the machine learning to design an image estimation filter. It is explanatory drawing explaining the blending process.

Hereinafter, an embodiment of the image processing method and the image processing apparatus according to the present invention will be described with reference to the drawings.

[Overall configuration of radiation imaging system]
FIG. 1 is a diagram showing an overall configuration of a radiation imaging system to which a console as an image processing device according to the present embodiment is applied.
As shown in FIG. 1, in the radiation image photographing system 100, the photographing device 1 arranged in the photographing room R and the console 3 arranged outside the photographing room R such as the front room (not shown) are connected by a communication cable or the like. The console 3 and the image analysis device 5 are connected to each other via a communication network NT such as a LAN (Local Area Network). Each device constituting the radiation imaging system 100 conforms to the DICOM (Digital Image and Communications in Medicine) standard, and communication between the devices is performed according to DICOM.
In the following embodiment, the case where the radiographic image capturing system 100 has a 1: 1 correspondence between the imaging device 1 and the console 3 in one imaging room R will be described as shown in FIG. , The configuration of the radiation imaging system is not limited to the illustrated example. For example, it is possible to have a plurality of photographing rooms R, each of which is arranged with a single or a plurality of photographing devices, and the single or a plurality of consoles 3 are associated with each other via a network NT or the like. is there.

[Configuration of imaging device]
The photographing device 1 includes a photographing table 11 including the holder 11A and a radiation irradiating device 12.
The holder 11A is for loading the radiation detection device 10 inside, and the photographing table 11 is arranged so that the radiation detection device 10 in the holder 11A faces the radiation irradiation device 12 with the patient who is the subject M sandwiched between them.
In the present embodiment, the photographing table 11 is for standing photography in which the patient who is the subject M is made to stand in front of the photographing table 11 and photographed, and the patient's cervical spine to the femur area is photographed in one shot. It is a shooting stand for one-shot long shooting.
Here, one-shot photography means taking a picture by irradiating radiation once, and one-shot long-length photography means taking a picture with a single radiation detection device 10 by combining a plurality of radiation detection devices 10. It means taking a picture of a wide range that cannot be done with one irradiation. According to this, since a wide range can be photographed with one irradiation, the burden on the patient who is the subject M can be minimized. In addition, it is possible to avoid the problem of body movement that causes the patient to move and shift or blur the image during a wide range of imaging, as compared with the case where the imaging is repeated in a plurality of times.

In the present embodiment, in order to perform such one-shot long imaging, the holder 11A of the imaging table 11 is configured to be capable of loading a plurality of radiation detection devices 10.
In the example shown in FIG. 1, three radiation detection devices 10 (10a, 10b, 10c) are loaded into the holder 11A so as to be aligned in the body axis direction of the patient who is the subject M in a state where a part of the radiation detection devices 10 (10a, 10b, 10c) is superimposed. ing.
Specifically, the radiation detection device 10 (10b, 10c) arranged in the holder 11A on the lower side in the body axis direction is attached to the radiation irradiation device 12 than the radiation detection device 10 (10a) on the upper side. The plurality of radiation detection devices 10 are loaded in a state where the ends in the vertical direction (vertical direction in FIG. 1) are overlapped so as to be close to each other.

FIG. 2 is an enlarged view of the portion II surrounded by the alternate long and short dash line in FIG. 1, and is a schematic showing a main part of the radiation detection device 10 (a part of 10a and 10b is shown in FIG. 2) arranged in the holder 11A. Cross-sectional view.
As shown in FIG. 2, the radiation detection device 10 is a panel-shaped detection device such as an FPD (Flat Panel Detector) in which the detection unit 10s is housed in the housing 101. The detection unit 10s is composed of a semiconductor image sensor, for example, has a glass substrate or the like, and detects at least the radiation emitted from the radiation irradiation device 12 and transmitted through the subject M at a predetermined position on the substrate according to the intensity thereof. , A plurality of detection elements (pixels) that convert the detected radiation into an electric signal and store it are arranged in a matrix. Each pixel is configured to include a switching unit such as a TFT (Thin Film Transistor). The FPD has an indirect conversion type that converts X-rays into an electric signal by a photoelectric conversion element via a scintillator and a direct conversion type that directly converts X-rays into an electric signal, and either of them may be used.

In each radiation detection device 10 (10a and 10b in FIG. 2), the lower end portion of the radiation detection device 10a and the upper end portion of the radiation detection device 10b overlap each other. Therefore, in the image detected by the detection unit 10s of the radiation detection device 10a, the housing 101 of the radiation detection device 10b and the detection unit 10s are reflected by the region of the width La.
When synthesizing an image obtained by one-shot long shooting into one image, this overlapping portion (the portion is a superimposition region and a correction region that needs to be corrected, as will be described later). ) Is reduced or removed by correction to make the joints between the radiation detection devices 10 inconspicuous, and it is desired to make the finish as if it was taken as one image.
This correction process (image processing) will be described later.

The holder 11A of the photographing table 11 is not limited to one that can be loaded with the three radiation detection devices 10. For example, it may be configured so that two or four or more radiation detection devices 10 can be loaded.
Further, the method of arranging the plurality of radiation detection devices 10 is not limited to the examples shown in FIGS. 1 and 2. For example, a plurality of radiation detection devices 10 (10a, 10b, 10c) may be loaded in the holder 11A so as to be alternately arranged on the side closer to the radiation irradiation device 12 and the side far from the radiation irradiation device 12. ..
Further, FIG. 1 illustrates a case where the shooting table 11 for one-shot long shooting is for standing shooting in which the patient who is the subject M stands in front of the shooting table 11 for shooting. The stand 11 is not limited to this. For example, in the photographing table 11 for one-shot long photographing, the holder 11A loaded with a plurality of radiation detection devices 10 (10a, 10b, 10c) is arranged substantially parallel to the floor surface, and is a plate on the upper side of the holder 11A. It may be for recumbent radiography in which the patient lies down or sits on top of the radiography.

Further, the photographing table 11 is provided with a reading control device 15.
The reading control device 15 controls the switching unit of each pixel of the radiation detection device 10 based on various image reading conditions input from the console 3 to switch the reading of the electric signal stored in each pixel. Then, the image data is acquired by reading the electric signal stored in the radiation detection device 10.
Here, the reading control device 15 is connected to a radiation irradiation control device 13 described later, and exchanges synchronization signals with each other to synchronize the radiation irradiation operation and the image reading operation.
The reading control device 15 is connected to the console 3 and outputs the acquired image to the console 3.

The radiation irradiation device 12 is arranged at a position facing the photographing table 11 on which the radiation detection device 10 (10a, 10b, 10c) is set with the subject M interposed therebetween, and is a radiation source (not shown) according to the control of the radiation irradiation control device 13. Irradiates the target portion of the subject M with radiation (X-rays).
Specifically, when an operator such as a radiologist operates an exposure switch (not shown) for instructing the radiation irradiation device 12 to start irradiating radiation, the console 3 responds to the pressing. Controls the radiation irradiation control device 13 to perform radiography.
The radiation irradiation control device 13 is connected to the console 3 and controls the radiation source based on various radiation irradiation conditions input from the console 3 to perform radiography. The irradiation conditions input from the console 3 are, for example, an X-ray tube current value, an X-ray tube voltage value, an additional filter type, and the like.
In order to perform long-length imaging in the present embodiment, the radiation irradiation device 12 is attached to a plurality of radiation detection devices 10 (10a, 10b, 10c) loaded in the holder 11A of the imaging table 11 via the patient who is the subject M. A so-called wide-angle irradiation type is applied so that radiation can be irradiated once (that is, one shot) at the same time.

[Console configuration]
The console 3 outputs radiation irradiation conditions and image reading conditions to the photographing device 1, and functions as a console for photographing and controlling the radiation photographing and the reading operation of the radiation image by the photographing device 1.
In addition, the console 3 of the present embodiment also functions as an image processing device that performs various corrections and the like on the image acquired by the photographing device 1.
The console 3 transmits the image data to the image analysis device 5 after performing necessary corrections and the like.

As shown in FIG. 3, the console 3 includes a control unit 31, a storage unit 32, an operation unit 33, a display unit 34, a communication unit 35, and the like, and each unit is connected by a bus 36.

The control unit 31 is a computer composed of a CPU (Central Processing Unit), a RAM (Random Access Memory), and the like. The CPU of the control unit 31 reads out the system program and various processing programs stored in the storage unit 32 in response to the operation of the operation unit 33 and develops them in the work area of the RAM, and will be described later according to the expanded programs. Various processes such as image processing are executed to centrally control the operation of each part of the console 3.
The control unit 31 functions as various condition setting means for setting the shooting conditions, reading conditions, and the like of the shooting device 1, a display control means for controlling the display unit 34, a communication control means for controlling the communication unit 35, and the like.
Further, particularly in the present embodiment, the control unit 31 is an image area of the superimposed portion of the radiation detection device 10 which is a correction area ArR (see FIG. 4) that needs to be corrected in the image data acquired by the photographing device 1. From the image information, the position of the pixel indicating the predetermined structure is specified, and the structure component of the predetermined structure is reduced / removed from the image information by using machine learning for the correction region including the pixel indicating the predetermined structure. It functions as a correction means for performing correction processing.
This correction process (image processing) will be described later.

The storage unit 32 is composed of a non-volatile semiconductor memory, a hard disk (HDD: Hard Disk Drive), or the like. The storage unit 32 stores data such as parameters or processing results required for executing the processing by various programs and programs including a program for executing the image processing (correction processing and the like) in the control unit 31. These various programs are stored in the form of a readable program code, and the control unit 31 sequentially executes an operation according to the program code.

The operation unit 33 is configured to include a keyboard equipped with cursor keys, number input keys, various function keys, and a pointing device such as a mouse, and controls an instruction signal input by key operation on the keyboard or mouse operation. Output to 31. Further, the operation unit 33 may include a touch panel on the display screen of the display unit 34, and in this case, the operation unit 33 outputs an instruction signal input via the touch panel to the control unit 31.

The display unit 34 is composed of a monitor such as an LCD (Liquid Crystal Display) or a CRT (Cathode Ray Tube), and performs various displays according to instructions of a display signal input from the control unit 31.

The communication unit 35 includes a LAN adapter, a modem, a TA, and the like, and controls data transmission / reception with each device connected to the communication network NT.
Although not shown, the console 3 has HIS (Hospital Information System), RIS (Radiology Information System), PACS (Picture Archiving and Communication System), etc. via the communication network NT or the like. Is connected.

[Configuration of image analyzer]
The image analysis device 5 is, for example, a computer device used by a doctor or the like who performs diagnosis using images.
The image analysis device 5 includes a control unit, a storage unit, an operation unit, a display unit, a communication unit, and the like (not shown), and performs various analyzes and the like on the image transmitted from the console 3 to perform various analyzes and the like to each part of the subject M (patient). When the doctor determines whether or not there is a lesion or the like in (the part to be imaged), this is assisted, and the diagnosis result is shown to the patient and displayed for explanation.

In the present embodiment, the console 3 gives various instructions regarding imaging to the imaging device 1, and various corrections are made to the image data acquired by the imaging device 1, so that the plurality of radiation detection devices 10 (10a, 10a, The case where the image data acquired by 10b and 10c) also functions as an image processing device for performing image processing such as synthesizing into one image will be described as an example, but the radiation imaging system is not limited to such a configuration. ..
For example, the console 3 may be used as a console for shooting, and various instructions and the like related to shooting may be given only to the shooting device 1.
In this case, the console receives the image data acquired by the plurality of radiation detection devices 10 (10a, 10b, 10c) of the photographing device 1 and transmits the image data to an image processing device separate from the console, and the image processing device. Various corrections are made to the image data on the side.
An image analysis device that performs image analysis may function as an image processing device. In this case, the image analysis device may perform various corrections on the image data, perform image processing such as synthesizing the corrected image data into one image, and then perform various image analysis. Good.

[About each process performed by the radiation imaging system when taking a long shot of one shot]
At the time of one-shot long shooting, each process at the time of shooting (that is, from the radiation irradiation device 12) in each of the radiation detection devices 10 (10a, 10b, 10c) loaded in the console 3 and the holder 11A of the shooting table 11 and the like. The processing performed before and after the irradiation of radiation and the subsequent processing such as reading out image data) are basically the same as in the case of simple photographing, and since they are known processing, the description thereof will be omitted.

The console 3 corresponds to the image data D from the radiation detection device 10 (10a, 10b, 10c) and the offset amount caused by the dark charge (also referred to as dark current) generated in each radiation detection element. When the offset data O is transferred, the offset data O is subtracted from the image data D for each radiation detection element of the radiation detection device 10 (10a, 10b, 10c) according to the following equation (1) to be true. The image data D ^{* of the above} is calculated, and the calculated true image data D ^{* is} subjected to precise image processing such as gain correction, defect pixel correction, and gradation processing according to the imaged portion, and the radiation detection device 10 (10a) is performed. , 10b, 10c), respectively, to generate an image.
D ^* = DO ... (1)

Hereinafter, the image generated based on the image data D or the like obtained by the radiation detection device 10 as described above is referred to as an image P obtained by the radiation detection device 10 (note that, in FIG. 4, the radiation detection device). The image obtained by 10a is referred to as "image P1", and the image obtained by the radiation detection device 10b is referred to as "image P2").
Further, at the above time point, as shown in FIG. 4, of the images P1, P2, etc. obtained by the radiation detection devices 10a, 10b, etc. generated as described above, the rear side (radiation irradiation) in the holder 11A. In the image P1 obtained by the radiation detection device 10a (farther from the device 12), the correction area ArR (see FIG. 4), which is an overlapping area between the radiation detection devices 10, is in front (from the radiation irradiation device 12). For the housing 101 of the radiation detection device 10b (closer), the horizontal streak component caused by the linear structure such as the edge portion of the internal structure, and various structures inside the housing 101 of the front radiation detection device 10b. The resulting structural component C and the like are reflected.

That is, in the holder 11A of the photographing table 11 according to the present embodiment, as shown in FIG. 2, each radiation detection device 10 has, for example, the upper end portion of the front lower side radiation detection device 10b and the rear upper radiation detection device 10a. The lower end of the is overlapped in the front-back direction.
Therefore, in the lower end portion of the image P1 obtained by the radiation detection device 10a on the rear upper side, a streak component caused by the upper end portion of the housing 101 of the front radiation detection device 10b, the upper end side portion of the detection portion 10s, and the like, and the streak component, etc. The structure component C caused by the structure inside the housing 101 such as various electronic components mounted on the circuit board (not shown) attached to the detection unit 10s of the front radiation detection device 10b is reflected.
The structure component C appears in the image P when various structures such as electronic components mounted inside the housing 101 affect the transmission of radiation, and is reflected in white in FIG. 4 and the like. The one that is out.
Such an overlapping region between the radiation detection devices 10 (a region having a width indicated by the width La in FIG. 2) is a correction region ArR that needs to be corrected.

Further, the structure that has moved into the correction region ArR may include a "predetermined structure" that exhibits a high contrast of a predetermined value or more. In FIG. 4 and the like, the structural component of this predetermined structure is referred to as “structural component Cs”.
Here, the predetermined structure is, for example, a metal screw or the like.
In the case of high contrast of a predetermined value or more, how much value should be set as a predetermined value is a matter that can be appropriately set. The predetermined value may be preset by default, or may be set or changed by the user or the like after the fact.
As described above, the "structure component Cs" appears in the image P when the "predetermined structure" showing a high contrast of a predetermined value or more, such as a screw provided in the housing 101, affects the transmission of radiation. It refers to a thing that is reflected in white in FIG. 4 and the like.

[About the image processing (image correction) method in this embodiment]
In the radiation imaging system 100 according to the present embodiment, the structure component C and the predetermined structure are obtained from the image P obtained by each of the radiation detection devices 10 (10a, 10b, 10c) loaded in the holder 11A of the imaging table 11. The image correction process for reducing / removing the structural component Cs and the like from the image information is performed by the console 3 (control unit 31 of the console 3) that functions as an image processing device.

Hereinafter, the image correction process will be described with reference to FIGS. 5 and 6 and the like.
FIG. 5 is a flowchart showing an outline of the image correction process in the present embodiment. Further, FIG. 6 is a flowchart showing the contents of step S3 “reduction / removal processing of structural components of a predetermined structure” in FIG. 5 in more detail.

First, in the console 3 as an image processing device, calibration correction is performed on the image P obtained by each radiation detection device 10 (10a, 10b, 10c) (see steps S1 and S2 in FIG. 5).
The calibration correction is performed using a pre-acquired calibration image Pcal (see FIG. 7).
In this calibration image Pcal, each radiation detection device 10 (10a, 10b, 10c) is loaded in the holder 11A of the photographing table 11 and the radiation is irradiated from the radiation irradiation device 12 without the intervention of the patient who is the subject M. It was taken after.
The calibration image Pcal is generated by the same process as the above-mentioned image P generation process based on the image data or the like obtained by each radiation detection device 10 (10a, 10b, 10c).

For example, when the radiation detection device 10 is shipped from a factory or when the radiation detection device 10 is introduced into a facility such as a hospital, a calibration image Pcal related to the radiation detection device 10 is acquired in advance. It should be noted that the calibration image Pcal can be acquired for each shooting before shooting, or can be configured to be acquired periodically.
The calibration image Pcal is associated with the radiation detection device 10 by writing the acquired information of the radiation detection device 10 as header information, and is associated with the storage unit 32 of the console 3 or a server (not shown). Etc. are stored in advance in a database of storage means such as.

In the present embodiment, the radiation detection device 10a is arranged at the rearmost position (a position away from the radiation irradiation device 12) in the holder 11A, and the radiation detection devices 10b and 10c are sequentially overlapped with each other at the front position (a position away from the radiation irradiation device 12). One-shot long shooting is performed in a state (see FIG. 1) in a state (position close to the radiation irradiation device 12), and the same arrangement is used when acquiring the calibration image Pcal. As described above, each radiation detection device 10 (10a, 10b, 10c) is loaded into the holder 11A of the imaging table 11 to perform imaging.
As a result, the structure component C, the structure component Cs of a predetermined structure, and the like are reflected in the calibration image Pcal obtained by the radiation detection device 10a arranged behind (see FIG. 7).

The components reflected in the image P (also the calibration image Pcal) differ depending on the type and structure of the radiation detection device 10. Further, in what arrangement the plurality of radiation detection devices 10 (10a, 10b, 10c) are loaded in the holder 11A of the photographing table 11, the superimposition position of each radiation detection device 10 and the context of superimposition (radiation irradiation device). The components and the way of being reflected in the image P (also the calibration image Pcal) are different depending on the anteroposterior relationship in the irradiation direction of the radiation from 12.
In order to properly perform calibration correction, a calibration image Pcal shot in the same arrangement (superimposition method, etc.) by the radiation detection device 10 of the same type as when the subject M was actually shot is displayed. It is important to use.

Therefore, when performing calibration correction on the image P (image Pa of the correction area ArR) obtained in the main shooting (shooting with the subject M arranged), for example, in the holder 11A of the shooting table 11. An operator such as a radiologist inputs information indicating the type of the loaded radiation detection device 10 and the loading position of each radiation detection device 10 into the console 3 as an image processing device, and at the time of the actual shooting, the relevant information is obtained. It is preferable to adjust so that various information at the time of shooting and the header information of the calibration image Pcal match.

Specifically, as shown in FIG. 5, first, the control unit 31 of the console 3 reads out the calibration image Pcal used for correction from the storage unit 32 or the like, and writes it in the header or the like of the calibration image Pcal. Information such as the degree of overlap of each radiation detection device 10 during calibration imaging (width La, etc. in FIG. 2) and the degree of overlap of each radiation detection device 10 when actually performing one-shot long imaging (FIG. 2). The position of the calibration image Pcal is adjusted so as to match the position of the image P (for example, the image P1 acquired by the radiation detection device 10a) based on the information of the width La and the like.
Further, the control unit 31 of the console 3 increases the magnification of the calibration image Pcal based on various information at the time of shooting the calibration image Pcal and various information at the time of actual one-shot long shooting (main shooting). Adjusts the enlargement ratio so as to match the enlargement ratio of the image P (image P1) (step S1).

After the position and magnification of the calibration image Pcal are adjusted, each radiation detection device 10 (here, in particular, radiation) is used when the one-shot long image is actually taken using the adjusted calibration image Pcal. Calibration correction is performed on the image P (for example, the image P1 acquired by the radiation detection device 10a) acquired by the detection device 10a) (step S2).

FIG. 7 is an explanatory diagram schematically showing the calibration correction.
In FIG. 7, the image Pa corresponding to the correction region ArR (correction region ArR in FIG. 4, that is, the region corresponding to the portion of the width La in FIG. 2), which is a superposition region between the radiation detection devices 10 in the image P. And, only the area corresponding to the image Pa in the calibration image Pcal is shown in the figure. The calibration correction is not limited to the correction area ArR, which is a superposition region between the radiation detection devices 10, and may generate an image P full-correction image PC, but the radiation detection devices 10 that require special correction are used. It is more effective and efficient and desirable to specify the superposed region (correction region ArR) of the above and perform correction in this range.

In the calibration correction, the control unit 31 of the console 3 reflects the structure (for example, IC on the substrate) of the radiation detection device 10b arranged in front of the image P1 (image Pa of the correction area ArR). Image P1 (image Pa in the correction area ArR) is corrected by appropriately raising the pixel value f of the pixel portion (that is, the portion in which the structure component C is reflected) whose pixel value is small. ) The structural component C is reduced / removed from the inside.

Specifically, the pixel value of each pixel (x, y) of the image P before correction is f (x, y), the contrast value of the pixel value of the calibration image Pcal is k (x, y), and the correction coefficient. Is A (x, y), and the pixel value of the corrected image P is g (x, y). Then, the control unit 31 of the console 3 follows the following equation (2) for each pixel (x, y). The pixel value g (x, y) is calculated. The correction coefficient A is calculated and estimated for each pixel by a predetermined method.
g (x, y) = f (x, y) + A (x, y) x k (x, y) ... (2)
As a result, a corrected image PC as a correction result of the calibration correction is generated in which the structure component C and the like reflected in the image P are reduced / removed from the image information (see FIG. 7).
In this way, the reduction / removal process of reducing / removing the structure component C or the like is the pixel value of the pixel portion where the pixel value is reduced due to the influence of various structures such as electronic components on the transmission of radiation. By appropriately raising f and correcting the image information of the image P, it means that the reflection that appears white in FIG. 4 and the like is reduced from the image P. It should be noted that the case of "reduction / removal" and the case of simply "reduction" also include the case of completely removing (removing / erasing) from the level that makes the image inconspicuous.
It should be noted that each process related to the calibration correction is basically the same as the process performed in the simple photographing, and since it is a known process, detailed description thereof will be omitted.

However, for a predetermined structure (for example, a screw or the like) having a high contrast value equal to or higher than a predetermined value, the structure component Cs of the structure can be effectively obtained only by performing the usual calibration correction as described above. Difficult to reduce / remove.
That is, as described above, in the existing calibration correction, the pixel value f of the image P before correction is added by multiplying the contrast value k of the pixel value of the calibration image Pcal by the correction coefficient A. Finally, the corrected image PC is obtained.

Here, when the contrast value k of the correction portion is small, even if there is some error in the estimated correction coefficient A, there is not much difference in the value of the final corrected pixel value g. Therefore, it is possible to obtain an image PC (image as a result of calibration correction) having a finish that is not noticeable as a diagnostic image.
However, when the contrast value k of the correction portion is large (in the case of high contrast), if there is even a slight error in the correction coefficient A, a large difference will occur in the final corrected pixel value g value. High accuracy is required for the correction coefficient A.
Therefore, depending on the estimation accuracy of the correction coefficient A, the structure component Cs of a predetermined structure (for example, a screw or the like) showing high contrast to the corrected pixel value g can be effectively obtained from the final image PC (image information). It may not be possible to reduce or remove it, and it may remain.

Therefore, the control unit 31 of the console 3 which is the image processing device in the present embodiment further reduces / removes the structural component Cs of the predetermined structure (step S3).
The removal process for removing the structure component Cs is the position of a pixel whose pixel value is reduced due to the influence of a "predetermined structure" such as a screw showing a high contrast of a predetermined value or more on the transmission of radiation. By specifying and correcting the image information of the image P, it means that the reflection that appears white in FIG. 4 and the like is reduced from the image P. It should be noted that the case of "reduction / removal" and the case of simply "reduction" also include the case of completely removing (removing / erasing) from the level of making the image inconspicuous.
Specifically, the following processing is performed.

That is, in the removal process of the structural component Cs of the predetermined structure, first, as shown in FIGS. 6 and 8, first, in the image P, the correction region ArR which is the overlapping portion between the radiation detection devices 10 corresponds to. The image is set as the image Pa, and the high frequency component is extracted from the image Pa in the correction region ArR (step S11).
Specifically, the image Pa is subjected to a filter process in which a Gaussian filter is applied in the horizontal direction to obtain an image Pb after the filter process from which low-frequency components are extracted.
Then, the high-frequency component is extracted by subtracting the filtered image Pb (image showing the low-frequency component) from the original image Pa to obtain the difference (in FIG. 8, the image showing the high-frequency component is imaged. Let it be Pc).

Next, the position of a predetermined structure (for example, a screw or the like) showing a particularly high contrast value is specified and detected from the image Pc (step S12), and an image of the predetermined structure (for example, a screw or the like) and its surroundings is obtained. Extract (in FIG. 8, the extracted image is referred to as image Pd).

The method for specifying and detecting the position of a predetermined structure is not particularly limited, but for example, a pixel showing a predetermined structure based on a calibration image (image PC in FIG. 7) obtained by performing calibration correction. The position of may be specified.
In this case, a Gaussian filter is applied to the calibrated image (image PC), and the difference between the filtered image (image showing low frequency components) and the original calibrated image (image PC) is taken to obtain a high frequency. Extract the ingredients.

Here, the predetermined structure such as a screw, which is the target of the reduction / removal treatment in the "reduction / removal treatment of the structural component of the predetermined structure", is a structure having a high contrast value equal to or higher than the predetermined value. .. Therefore, the control unit 31 performs threshold processing for determining whether or not the structure specified by extracting the high-frequency component corresponds to the "predetermined structure" in light of a predetermined constant threshold value. ..
Thereby, it is specified at which position in the calibration image (image PC) the structural component Cs of the predetermined structure is reflected. Similarly, a Gaussian filter is applied to the actual captured image (image P), and the high frequency component is obtained by taking the difference between the filtered image (image showing the low frequency component) and the original captured image (image P). Extract. Further, by superimposing the calibration image (image PC) on the actual photographed image (image P), it is possible to specify the approximate position of the position where the structure component Cs is reflected.

However, when the superposition position is deviated, even if the deviating position is about 1 mm, the deviation is about 5 pixels. Further, the position of the structural component Cs reflected in the photographed image (image P) may be different depending on various conditions such as the irradiation angle of radiation. Therefore, template matching is performed in the peripheral area of the position where the structure component Cs is reflected, and the position of the structure component Cs detected in the calibration image (image PC) is the actual photographed image (image PC). It is preferable to accurately identify where in the image P). The pseudo degree in the case of template matching can be calculated by, for example, a normalization correlation function.
The filter applied to extract the low frequency component as a premise for extracting the high frequency component is not limited to the Gaussian filter. Further, the method of accurately specifying the position of the structure component Cs in the actual photographed image (image P) is not limited to template matching.
In addition, the threshold value for determining whether or not the structure corresponds to a predetermined structure is a matter that can be appropriately set. The threshold value may be appropriately changed according to the structure to be reduced / removed.

Next, an image estimation filter F (see FIG. 9) is applied to generate an image of a predetermined structure (step S13). In the present embodiment, the image estimation filter F is obtained by machine learning.
Here, machine learning for obtaining the image estimation filter F will be described with reference to FIG.
FIG. 9 illustrates a case where learning data is created using a phantom such as a human body model. The learning data is not limited to the one created by using the phantom, and may be created by actually arranging a person as a subject and taking a picture.

In FIG. 9, the image PF is an image obtained by photographing a phantom with a plurality of radiation detection devices 10 loaded in the holder 11A of the photographing table 11, and is a correction portion in the image PF which is an overlapping portion of the radiation detection devices 10. The image corresponding to the region ArR is referred to as an image PFa.
Further, the image PFb is an image PFa to which a Gaussian filter is applied in the horizontal direction, and the image PFb (an image showing a low frequency component) after the filter processing is subtracted from the original image PFa to obtain a high frequency. Extract the components (image PFc showing high frequency components). A predetermined structure (for example, a screw or the like) showing a particularly high contrast value and its surroundings are extracted from the image PFc (the extracted image is referred to as an image PFd).

When such an extracted image is used, normalization is usually performed by the standard deviation of the entire image extracted before input. However, in the image PFd of the present embodiment (the same applies to the image Pd in FIG. 8), in order to give priority to the contrast of the structure component Cs of a predetermined structure (for example, a screw) showing high contrast, it is not the entire image. , It is preferable to perform normalization using the standard deviation obtained from the central portion of the image PFd (the same applies to the image Pd), that is, a small region around a predetermined structure such as a screw.
In addition, although the case where the standard deviation is obtained from the periphery of a predetermined structure such as a screw and the normalization is performed is shown here, in some cases, it is more appropriate to perform the normalization with the standard deviation obtained from the entire image. It may be possible. Therefore, the normalization method is not limited to the one illustrated here.

On the other hand, the image NPF is an image obtained by photographing the phantom with the single radiation detection device 10 loaded in the holder 11A of the photographing table 11, and the image corresponding to the correction region ArR in the image NPF is referred to as the image NPFa. ..
Further, the image NPFb is an image NPFa to which a Gaussian filter is applied in the horizontal direction, and the high frequency is obtained by subtracting the image NPFb (an image showing a low frequency component) after this filter processing from the original image NPFa to obtain a difference. Extract the components (image NPFc showing high frequency components). An image at a position corresponding to the image PFd is extracted from the image NPFc (the extracted image is referred to as an image NPFd).

Then, the extracted image (image NPFd) extracted from the image NPF without the overlapping portion was subtracted from the extracted image (image PFd) showing the predetermined structure extracted from the image PF having the overlapping portion between the radiation detection devices 10. The resulting image (image PFR) is used as correct answer data indicating image information corresponding to a predetermined structure.
Further, an extracted image (image PFd) showing a predetermined structure extracted from an image PF having a superposed portion is used as input data.

In the present embodiment, in machine learning, a plurality of sets of such input data and correct answer data are prepared as learning data. The set of training data is, for example, when shooting the front of the human body, when shooting the back of the human body, when shooting the side of the human body, etc. with different positioning, or when shooting the legs instead of the chest etc. It is preferable to shoot in various patterns and prepare a large amount, such as those with different doses, those with different doses of radiation, those with different shooting conditions such as those with different radiation angles. .. By preparing a large amount of learning data for various patterns, more accurate learning can be expected.
In the present embodiment, an image estimation filter F that estimates an image of a predetermined structure (for example, a screw or the like) showing a high contrast value by giving learning data prepared in this way is designed in advance, and the console 3 It is stored in the storage unit 32 or the like.

Learning (machine learning) for designing the image estimation filter F is realized by deep learning, and is implemented based on a so-called convolutional network such as U-NET or Seg-NET, for example.
The network applied to learning (machine learning) for designing the image estimation filter F is not limited to the one illustrated here.
The accuracy of the image estimation filter F obtained as a learning result is appropriately verified, and if the accuracy is insufficient, further learning is performed by appropriately changing the network configuration or adding learning data to estimate the image. The filter F may be updated to a more accurate one.

The structural component Cs of a predetermined structure is reduced by using machine learning (image estimation filter F as a learning result) for the actually captured image P (for example, the image P1 acquired by the radiation detection device 10a). When performing the correction process for removing, the image estimation filter F obtained by machine learning is applied to the image Pd detected and extracted in step S12, and an image of a predetermined structure (for example, a screw or the like) showing a high contrast value is applied. Obtain Pf (see step S13, FIG. 8). At this time, the image Pd may be trimmed to a size suitable for applying the image estimation filter F.

After that, the image Pd and the image Pf are added to obtain an image Pg in which the structural component Cs of a predetermined structure is canceled (step S14).
On the other hand, the image Pe at the position corresponding to the image Pd is cut out from the image Pb showing the low frequency component, and the image Pg is added to the image Pe showing the low frequency component to obtain the image Ph (step S15).
The image Ph is an image obtained by adding an image in which the structural component Cs of a predetermined structure is canceled and an image showing a low frequency component corresponding to the portion, and is corrected by using machine learning in the present embodiment. It is an image as a correction result of processing.

In the present embodiment, when the correction result by the calibration correction and the correction result of the correction process using the machine learning (image estimation filter F) are acquired, the control unit 31 of the console 3 then sets both correction results. Based on this, the final correction result is obtained. Specifically, a blending process for blending both correction results is performed to obtain the final correction result. In the blending process, after calculating the blending ratio for each pixel, the adoption ratio of both correction results for each pixel is determined according to this blending ratio, and the blending is performed. The method of obtaining the final correction result based on both correction results is not limited to the method shown here.

Specifically, first, from the calibration correction result image (image PC of FIGS. 7 and 10) to the correction result image (image Ph of FIGS. 9 and 10) using machine learning (image estimation filter F). The corresponding region is extracted (step S16, extracted image PCa in FIG. 10).
Then, the blend ratio is calculated from the difference value between the image Ph (correction result using the image estimation filter F) and the corresponding image PCa (calibration correction result) (step S17).
Then, the control unit 31 of the console 3 performs a blending process based on the calculated blending ratio (step S18). As a result, it is possible to obtain a final processing result in which the structural component Cs of a predetermined structure is reduced / removed from the image information.

The blend ratio is the difference value (for example, pixel value, brightness value) between the processing result by calibration correction (image PCa in FIG. 10) and the processing result by machine learning (image estimation filter F) (image Ph in FIGS. 9 and 10). , The difference value of the density value) is calculated for each pixel. Specifically, after obtaining the difference value, a predetermined function conversion is performed to determine the blend ratio in the range of 0 to 1.0.
As for the tendency of blending, if there is no difference between the two results, the existing calibration correction processing result is adopted, and the larger the difference between the two, the more the processing result by machine learning is adopted.
It is preferable to set a certain threshold value in advance as to what the absolute value of the difference is and at what rate the processing result by machine learning is adopted. As a result, for example, if the absolute value of the difference is 25 or more, the processing result by machine learning is 100% adopted, and if the absolute value of the difference is about 12, the processing result by calibration correction and the processing result by machine learning are obtained. A blending ratio for each pixel is required, such as adopting a ratio of 50% each.

The threshold value for determining the blend ratio may be set by default, or may be appropriately changed by an operator (user) or the like.
In addition, there are multiple types of threshold values depending on various conditions such as the location where the structural component Cs to be reduced / removed is reflected, the type of image to be corrected, and the type of structure to be reduced / removed. It may be prepared in advance and selected automatically or manually according to each condition.
Furthermore, after the blending process is performed once, when the disappearance condition (remaining condition) of the structural component Cs is confirmed, the setting can be adjusted / changed as appropriate, such as adjusting the threshold value, so that the blending process can be redone. You may.

After synthesizing the images, only the difference value is used in determining the blend ratio so that the boundary between the rectangular region containing a predetermined structure (for example, a screw or the like) and the other correction region ArR is not conspicuous. Rather than considering, for the outer region of a given structure, the outer side (ie, the farther away from the given structure) the processing result of the calibration correction is adopted, and the inner side (that is, the closer it is to the given structure). ) It is preferable that the processing result by machine learning is adopted.

The blend ratio is the processing result of reduction / removal of the structure component C and the structure component Cs of a predetermined structure (for example, a screw) by the calibration correction as described above, and the machine learning (image estimation filter F). Since it is calculated from the difference value from the processing result of reduction / removal of the structure component Cs of a predetermined structure (for example, screw) by, most of the cases where the average concentration values of the two results are significantly different. The processing result by machine learning is adopted in the area.
The processing result by machine learning is the processing using a filter, and the processing result is liable to be blurred. Therefore, if the processing result by machine learning is adopted more than necessary, the image may have a conspicuous blur.
Therefore, before calculating the blend ratio, it is preferable to perform density correction on the processing result by machine learning and perform pretreatment to match the average density value with the average density value of the processing result by calibration correction.

As described above, the processing result by machine learning (image estimation filter F) tends to cause blurring of the image, and the image estimation filter F obtained by machine learning in the present embodiment is a predetermined structure showing high contrast. Since it is designed specifically for reducing / removing (for example, screws), it is not possible to widely apply correction (reduction / removal processing of structural component Cs) by machine learning (image estimation filter F) to the entire image P. Not preferred.
Therefore, in the present embodiment, the image region of the correction region ArR that does not include the pixel indicating the predetermined structure is calibrated without performing the reduction / removal processing by machine learning (image estimation filter F). It is decided that only the reduction / removal processing by the motion correction is performed and the result (image PC of FIGS. 7 and 10) is 100% adopted, and machine learning is performed only for the image region including the pixel indicating the predetermined structure. The correction (reduction / removal processing) by (image estimation filter F) and the reduction / removal processing by calibration correction are performed, and the blend processing as described above is performed.

Further, for the same reason, when the radiation detection device 10 that does not include a predetermined structure (for example, a screw or the like) is used for photographing, the reduction / removal by the machine learning (image estimation filter F) shown above is performed. It is preferable that no processing is performed and only the processing result obtained by the conventional calibration correction is adopted.
For this reason, it is preferable to provide a processing unit for determining what kind of radiation detection device 10 was used for imaging before image processing or after image processing.
The configuration of the processing unit in this case is not particularly limited, but for example, a reading unit that reads an identification mark (for example, a bar code or various tags provided on the housing 101 or the like) of the radiation detection device 10 used for photographing is contained in the holder 11A. The control unit 31 of the console 3 may determine the type of the radiation detection device 10 or the like based on the information from the reading unit as the processing unit. Then, when the processing unit determines that the radiation detection device 10 is not provided with a predetermined structure (for example, a screw or the like), the correction processing by machine learning (image estimation filter F) is not performed. ..

The radiation detection device 10 used for imaging is provided with a predetermined structure (for example, a screw), but the predetermined structure (that is, that is, a predetermined structure (that is, a screw or the like) is provided at a position where it should be due to a part thereof being missing or the like. When a structure exhibiting a high contrast exceeding a certain threshold value) is not detected, it is preferable to perform correction processing by machine learning (image estimation filter F) only on the portion where a predetermined structure is detected. As a result, it is possible to prevent the image quality of the image from being deteriorated by performing unnecessary processing.

Returning to FIG. 5, when the correction process (reduction / removal process) for reducing / removing the structural component Cs of the predetermined structure as described above from the image information is completed, various other basic processes are performed (step S4). ..
The specific processing to be performed as the basic processing is not particularly limited.
For example, when the images P (for example, P1, P2, etc. in FIG. 4) acquired by a plurality of radiation detection devices 10 (10a, 10b, 10c) are combined into one long image, each of them is used. The density of any image P or all images P is corrected so that the image densities match, the positional relationship between the images P and the enlargement ratio are adjusted, and the same part of the subject M is displayed in each image P. The captured parts are overlapped and adjusted so that the images P are smoothly connected to each other.
It should be noted that such density correction, position, and magnification adjustment are known processes, and detailed description thereof will be omitted.
Further, in addition to the removal treatment of the structure component C and the structure component Cs of a predetermined structure, the streak component and the like appearing in the image in the form of streaks may be reduced or removed by various correction treatments.

By synthesizing the images P obtained by the plurality of radiation detection devices 10 after making various adjustments and corrections as described above, a composite image having a finish as if it was taken as one image from the beginning can be obtained. Can be generated.
It is not essential to synthesize the images P acquired by the plurality of radiation detection devices 10 (10a, 10b, 10c), and the images P may be used as they are divided. Further, it may be used after being synthesized once and then divided. In the one-shot shooting, even if the image P is not combined after that, a wide range of shooting can be performed by one irradiation, and the patient who is the subject M is compared with the case where the shooting is repeated for each part. The burden can be reduced.
The image generated on the console 3 is sent to the image analysis device 5, displayed on the display unit and used for diagnosis and explanation to the patient, and various analyzes and analyzes are performed to support the diagnosis as appropriate. Used.

〔effect〕
As described above, according to the present embodiment, when a plurality of radiation detection devices 10 are arranged in a state where a part thereof is superimposed and an image is taken for the subject M, the radiation detection device is a correction region ArR that requires correction. From the image information (image Pa) of the image area (correction area ArR) of the superimposed portion of 10, the position of the pixel indicating the predetermined structure such as a screw is specified, and the correction area ArR including the pixel indicating the predetermined structure is specified. A correction process for reducing / removing the structural component Cs of a predetermined structure is performed on the image information by using machine learning.
Therefore, it is possible to perform correction specialized in reducing / removing the structural component Cs of a predetermined structure from the image information with high accuracy by using machine learning. As a result, it is possible to obtain an image that has been appropriately corrected (image processed) even when a wide area is photographed in one image.

Further, in the correction process using machine learning in the present embodiment, the image estimation filter F for estimating the image information corresponding to the predetermined structure among the image information in the correction region ArR is designed in advance by machine learning, and the predetermined structure is used. The image estimation filter F is applied to an unknown image region (image Pa) including pixels indicating the above, and a predetermined image is determined based on the image information corresponding to the estimated predetermined structure and the original image information. It reduces / removes the structural component Cs of the structure.
In this way, in order to reduce / remove the structural component Cs of a predetermined structure from the image information by using the image estimation filter F prepared in advance by machine learning, the correction is performed efficiently and with high accuracy. Can be done.
In addition, by accumulating learning results by machine learning, even when an unknown image Pa is input, the structural component Cs of a predetermined structure can be appropriately reduced / removed from the image information, resulting in high quality. You can get an image.

Further, in the present embodiment, the existing calibration correction is performed on the correction area ArR based on the calibration image taken without the subject M intervening, and the correction area ArR does not include a pixel indicating a predetermined structure. For the image area, the correction result by calibration correction is adopted, while for the image area including pixels showing a predetermined structure, the correction result by calibration correction and the correction using machine learning are adopted. After acquiring both the correction results of the processing, a blending process for blending both correction results is performed to obtain the final correction result.
By combining the existing calibration correction and the correction process using machine learning in this way, it is possible to appropriately reduce / remove the structural component Cs that cannot be completely reduced / removed by the calibration correction from the image information. ..
Further, when the correction process using machine learning uses the image estimation filter F, the high-contrast structural component Cs can be effectively reduced / removed, but the image tends to be blurred, but the calibration correction In combination with, an appropriate correction result can be obtained by giving priority to the result of the calibration correction other than the reduction / removal of the high-contrast structural component Cs.

Further, in the present embodiment, the positions of pixels indicating a predetermined structure in the correction process using machine learning are specified based on the calibration image.
By identifying the target to be corrected using machine learning from the result of the calibration correction, it is possible to efficiently identify the correction target accurately and precisely.

Further, depending on whether or not the radiation detection device 10 used for photographing includes a predetermined structure, it is switched whether or not to perform a correction process for reducing or removing the structure component Cs of the predetermined structure from the image information (radiation). In the case of the configuration (providing a processing unit for determining the type of the detection device 10), it is possible to save the trouble of performing unnecessary processing when a good result cannot be obtained even if the correction processing using machine learning is performed. be able to. As a result, appropriate image processing (correction processing) according to the radiation detection device 10 can be performed.

Further, in the present embodiment, a plurality of images P (radiation images) acquired from a plurality of radiation detection devices 10 by one-shot photography are combined to generate a composite image.
As a result, for example, when the image to be imaged is a medical image, it is possible to photograph the whole body of the patient who is the subject M by one irradiation, and the burden on the patient can be reduced.
Further, by synthesizing the images taken by the plurality of radiation detection devices 10, it is possible to obtain an image that is easy for doctors and patients to see and contributes to accurate diagnosis.

[Modification example]
Although the embodiments of the present invention have been described above, it goes without saying that the present invention is not limited to such embodiments and can be variously modified without departing from the gist thereof.

For example, in the above embodiment, the case where the patient is photographed as the subject M to obtain a medical image has been described, but the image processing method of the present invention is not limited to the case where it is applied to the image processing (correction processing) of the medical image.
The present invention is applicable when a wide range of objects that cannot be photographed by a single radiation detection device 10 are photographed by using a plurality of radiation detection devices 10.
For example, even in a non-destructive inspection in which an object that exceeds the imageable range of a single radiation detection device 10 in the vertical direction, the horizontal direction, or the vertical and horizontal directions is photographed as the subject M, the image processing method of the present invention is used for the photographed image. Can be applied.

Further, in the above embodiment, a case where a plurality of radiation detection devices 10 (10a, 10b, 10c) are arranged side by side in the body axis direction of the patient who is the subject M and a long image is taken in one shot is illustrated. However, the direction in which the plurality of radiation detection devices 10 are arranged is not particularly limited.
For example, in the case of photographing in the above non-destructive inspection or the like, a plurality of radiation detection devices 10 may be arranged side by side in the vertical direction, the horizontal direction, or both the vertical and horizontal directions so as to cover the entire subject.
Also in this case, the image processing method of the present invention can be applied to correct the image of the overlapping portion between the radiation detection devices 10.

Further, in the above embodiment, the case of performing one-shot photography is illustrated, but a plurality of radiation detection devices 10 (10a, 10b, 10c, etc.) are arranged in a state in which some of them are superimposed to photograph the subject. In some cases, the image processing method of the present invention can be applied to the captured image.
For example, the image processing method of the present invention can be applied to correct the image of the overlapping portion of the radiation detection devices 10 even when the entire subject such as the whole body of the patient is photographed in a plurality of times.

Further, in the above embodiment, the case where one image estimation filter F is generated by machine learning is illustrated, but there may be a plurality of image estimation filters F prepared in advance for image processing (image correction).
For example, a plurality of image estimation filters F may be prepared depending on the type of a predetermined structure showing high contrast. In this case, the type of the radiation detection device 10 used for photographing and the type of a predetermined structure included in the radiation detection device 10 may be specified, and the image estimation filter F to be applied may be switched according to the specific result.
For example, even if a predetermined structure is a “screw”, if the type is changed, the same image estimation filter F may not be able to appropriately reduce / remove the structure component Cs. Therefore, for example, the learning coefficient of the image estimation filter F may be prepared for each type of screw.
Then, when actually performing image processing (correction processing), first, preprocessing for determining the type of screw included in the radiation detection device 10 used for photographing is performed, and then learning according to the type of the screw is performed. An image generation filter F having a coefficient is applied to perform image processing for reducing / removing the structural component Cs of the screw from the image information.

Needless to say, the present invention is not limited to the above-described embodiments and modifications, and can be appropriately modified as long as the gist of the present invention is not deviated.

1 Imaging device 10 Radiation detection device 11 Imaging table 11A Holder 12 Radiation irradiation device 3 Console (image processing device)
31 Control unit 32 Storage unit 5 Image analyzer 100 Radiation imaging system ArR Correction area Cs Structure component F Image estimation filter P Image

Claims (8)

When multiple radiation detection devices that detect radiation are arranged in a state where some of them are superimposed and a subject is photographed.
The position of the pixel indicating the predetermined structure is specified from the image information of the image area of the superimposed portion of the radiation detection device, which is the correction area requiring correction, and the correction area including the pixel indicating the predetermined structure is provided. An image processing method characterized in that a correction process for reducing a structural component of the predetermined structure is performed on the image information by using machine learning. The correction process using the machine learning
An image estimation filter that estimates the image information corresponding to the predetermined structure among the image information in the correction region is designed in advance by the machine learning.
The image estimation filter is applied to an unknown image region including pixels indicating the predetermined structure, and the image information corresponding to the estimated predetermined structure and the original image information are used as the basis. The image processing method according to claim 1, wherein the correction is performed to reduce the structural components of a predetermined structure. Based on the calibration image taken without the subject intervening, the existing calibration correction is performed on the correction area.
For an image region in the correction region that does not include pixels indicating the predetermined structure, the correction result by the calibration correction is adopted, while an image including pixels indicating the predetermined structure is adopted. For the region, after acquiring both the correction result by the calibration correction and the correction result of the correction process using the machine learning, the final correction result is obtained based on both correction results. The image processing method according to claim 1 or 2. The image processing method according to claim 3, wherein the position of a pixel indicating the predetermined structure in the correction process using the machine learning is specified based on the calibration image. 2. The second aspect of the present invention is that the radiation detection device used for photographing switches whether or not a correction process for reducing a structural component of the predetermined structure is performed depending on whether or not the radiation detection device includes the predetermined structure. The image processing method according to any one of claims 4. The image processing method according to any one of claims 1 to 5, wherein a plurality of radiation images acquired from the plurality of radiation detection devices by photographing are combined to generate a composite image. An image processing apparatus including a control unit that performs image processing by the image processing method according to any one of claims 1 to 6. A computer-readable program comprising causing a computer of an image processing apparatus to perform image processing by the image processing method according to any one of claims 1 to 6.

Images