CN116188619B - Method, apparatus and medium for generating X-ray image pair for training - Google Patents
Method, apparatus and medium for generating X-ray image pair for training Download PDFInfo
- Publication number
- CN116188619B CN116188619B CN202310459477.5A CN202310459477A CN116188619B CN 116188619 B CN116188619 B CN 116188619B CN 202310459477 A CN202310459477 A CN 202310459477A CN 116188619 B CN116188619 B CN 116188619B
- Authority
- CN
- China
- Prior art keywords
- ray
- image
- imaging
- real
- ideal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
- G06T11/005—Specific pre-processing for tomographic reconstruction, e.g. calibration, source positioning, rebinning, scatter correction, retrospective gating
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10116—X-ray image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
The present application relates to a method, apparatus and medium for generating an X-ray image pair for training. The method comprises the steps of forming a Monte Carlo model of a target X-ray imaging system comprising a point source and a component by using Monte Carlo simulation software, and setting a first configuration parameter which is related to the component and does not change along with the change of imaging conditions; setting second configuration parameters based on the Monte Carlo model and the ideal imaging condition and the real imaging condition respectively to obtain an ideal imaging model and a real imaging model respectively; respectively performing simulated imaging based on the ideal imaging model and the real imaging model to obtain an ideal image and a simulated real image, wherein the simulated real image is used for simulating a real image of the target X-ray imaging system; the ideal image and the simulated real image are taken as an X-ray image pair for training. The patient can thus be kept free of radiation risk by the method of generating an X-ray image pair for training; the obtained ideal image has high image definition and low image distortion.
Description
Technical Field
The present application relates to the field of X-ray imaging technology, and more particularly, to a method, apparatus and medium for generating an X-ray image pair for training.
Background
In recent years, deep learning has been rapidly developed in the field of computer vision, and as an important branch of computer vision, the application of deep learning in medical images has been increasingly widespread and deep. When image processing is performed in deep learning, it is often necessary to generate a pair of "good" and "bad" images, and the deep learning neural network learns the relationship between "good" and "bad" so that the "bad" image can be restored to the "good" image. Therefore, deep learning is actually required to solve two problems, namely, how to obtain a "good" and "bad" image pair, and how to construct a network to describe the mapping relationship between "good" and "bad". Obtaining high quality clinical medical images (i.e. "good" images) is difficult for three reasons: firstly, if a low noise image needs to be acquired, the X-ray dose needs to be increased, which increases the radiation risk of the patient; secondly, the image with smaller pixels cannot be obtained due to the limitation of hardware; and thirdly, any one system is a non-ideal system, and the image after passing through the non-ideal system has a certain degree of distortion.
The monte carlo method is used as a statistical simulation method which uses random numbers to simulate statistical distribution to solve the relevant calculation problem. Because there is a deterministic mathematical statistical distribution of interactions between energetic particles and substances, the Monte Carlo method can be used to model interactions between energetic particles and substances. Many Monte Carlo simulation tools that perform particle-to-substance interactions, such as Geant4, EGSnrc, MCNP, etc., have also been derived therefrom.
Disclosure of Invention
The present application provides a method for solving the above-described deficiencies in the prior art by using a Meng Ka simulation tool. There is a need for a method, apparatus and medium for generating X-ray image pairs for training that do not rely on high doses of X-ray radiation, do not expose the patient to radiation risk, and that are capable of flexibly and rapidly generating as many "good" X-ray images with high signal-to-noise ratio, high definition and low distortion as required and corresponding "bad" X-ray images that introduce noise and distortion due to real imaging conditions, the differences in the paired "good" and "bad" X-ray images being strictly limited to differences in imaging conditions. The paired 'good' and 'bad' X-ray images can be used for training the deep learning neural network, so that the deep learning neural network can accurately learn the mapping relation between the 'good' image and the 'bad' image, and the image quality of the 'good' image obtained by the deep learning neural network from the 'bad' image processing can be improved.
According to a first aspect of the present application there is provided a method of generating an X-ray image pair for training comprising: forming a Monte Carlo model of a target X-ray imaging system comprising a point source and a component by using Monte Carlo simulation software, wherein the Monte Carlo model is provided with a first configuration parameter related to the component, and the first configuration parameter is not changed along with the change of imaging conditions; setting second configuration parameters based on the Monte Carlo model and the ideal imaging condition and the real imaging condition of the target X-ray imaging system respectively to obtain an ideal imaging model and a real imaging model respectively; respectively carrying out simulated imaging on the same subject based on the ideal imaging model and the real imaging model to respectively obtain an ideal image and a simulated real image, wherein the simulated real image is used for simulating a real image of a target X-ray imaging system; and taking the ideal image and the simulated real image with the signal-to-noise ratio related parameter higher than the first threshold as an X-ray image pair for training.
According to a second aspect of the present application there is provided an apparatus for generating an X-ray image pair for training, comprising: an interface configured to receive a configuration of the second configuration parameter by a user based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system; and a processor configured to perform a method of generating an X-ray image pair for training as described in the first aspect of the application.
According to a third aspect of the present application there is provided a non-transitory computer readable medium having instructions stored thereon, wherein the instructions, when executed by a processor, perform the steps of the method of generating an X-ray image pair for training as described in the first aspect of the present application.
The method, the device and the medium for generating the X-ray image pair for training, provided by the various embodiments of the application, utilize Monte Carlo simulation software to form a Monte Carlo model containing a target X-ray imaging system, and the Monte Carlo model sets first configuration parameters which do not change along with imaging condition changes, so that differences between ideal imaging conditions and real imaging conditions are endowed with corresponding imaging models through the setting of various second configuration parameters. The ideal imaging model and the real imaging model which are basically matched in aspects except the imaging condition can be conveniently obtained through the setting of the second configuration parameters based on the Monte Carlo model, the ideal image generated by the ideal imaging model is a high-quality image, and the image quality can be higher than that of a clinical medical image obtained by an X-ray device; compared with the method of obtaining high-quality clinical medical images through X-ray equipment, the method of obtaining the 'good' images does not need to irradiate the patient with X-rays, and can avoid the use of high-dose X-rays, so that the patient does not have radiation risks; the real image of the simulation target X-ray imaging system, namely the 'bad' image, is obtained based on the real imaging condition, and compared with a mode of clinically obtaining the 'bad' image, the radiation risk of a patient can be reduced. Therefore, a large number of paired ideal images and simulated real images can be obtained rapidly and efficiently, the difference of the ideal images and the simulated real images is strictly limited to the difference of imaging conditions, the method is more suitable for training of a deep learning neural network, the deep learning neural network can accurately learn the mapping relation between a 'good' X-ray image and a corresponding 'bad' X-ray image which introduces noise and distortion due to the real imaging condition, and the image quality of the 'good' image obtained by processing the 'bad' image by the deep learning neural network can be improved.
Drawings
In the drawings, which are not necessarily drawn to scale, like numerals may describe similar components in different views. The same reference numerals with letter suffixes or different letter suffixes may represent different instances of similar components. The accompanying drawings illustrate various embodiments by way of example in general and not by way of limitation, and together with the description and claims serve to explain the disclosed embodiments. Wherever possible, the same reference numbers will be used throughout the drawings to refer to the same or like parts. Such embodiments are illustrative and not intended to be exhaustive or exclusive of the present apparatus or method.
FIG. 1 illustrates a flow chart of a method of generating an X-ray image pair for training in accordance with an embodiment of the present application;
FIG. 2 shows a schematic diagram of the basic construction of a target X-ray imaging system forming a Monte Carlo model according to an embodiment of the application;
FIG. 3 illustrates simulation of X-ray imaging of a cylinder as a subject using an ideal imaging model, resulting in an ideal image, in accordance with an embodiment of the present application;
fig. 4 shows a simulated real image obtained by simulating X-ray imaging of the cylinder as the subject using a real imaging model according to an embodiment of the present application;
FIG. 5 shows a continuous energy spectrum of a Monte Carlo model according to an embodiment of the application; and
fig. 6 shows a block diagram of an apparatus for generating an X-ray image pair for training according to an embodiment of the present application.
Detailed Description
The present application will be described in detail below with reference to the drawings and detailed description to enable those skilled in the art to better understand the technical scheme of the present application. Embodiments of the present application will be described in further detail below with reference to the drawings and specific examples, but not by way of limitation.
The terms "first," "second," and the like, as used herein do not denote any order, quantity, or importance, but rather are used to distinguish one element from another. The word "comprising" or "comprises" and the like means that elements preceding the word encompass the elements recited after the word, and not exclude the possibility of also encompassing other elements.
FIG. 1 illustrates a flow chart of a method of generating an X-ray image pair for training in accordance with an embodiment of the present application. Comprising the following steps: in step 101, a Monte Carlo model of a target X-ray imaging system comprising a point source and a component is formed using Monte Carlo simulation software, the Monte Carlo model having first configuration parameters associated with the component set, and the first configuration parameters not changing with changes in imaging conditions. The Monte Carlo simulation software can use Geant4, MCNP, EGS and other software or modified versions thereof, and the like, and can also be a Monte Carlo algorithm software tool which is researched and developed by oneself, and the application is not limited in detail. The target X-ray imaging system comprises a point source and a member capable of generating an X-ray image, and a Monte Carlo model comprising the target X-ray imaging system can be obtained by designing the target X-ray imaging system in Monte Carlo simulation software.
Taking Geant4 software as an example, the model modeling method is a tool for simulating the transport process of particles in materials based on the Monte Carlo method, is open and open to all users, and can build and customize the model by using C++ as a bottom language. Specifically, geant4 may receive various configurations as input items, such as, but not limited to, a particle source (X-ray source), a material with which the particles interact (air on the path of the particles and a subject to be imaged), a detector, the number of simulated particles, etc., and automatically calculate various information of the particles on interacting with the material, such as a reaction time, a reaction trajectory, and a reaction deposition energy of the particles, etc., based on the input items. Further, to form a Monte Carlo model of a target X-ray imaging system comprising a point source and components, such tools may be utilized to construct a model comprising an X-ray source (e.g., a bulb), a detector, and a subject. In the imaging of the subject, GATE physical modules may be used, involving physical processes such as photoelectric effect, compton effect, rayleigh scattering, electron ionization, bremsstrahlung, and multiple scattering. With the monte carlo model constructed by the monte carlo simulation software, the generation process of the X-ray image under various imaging conditions (i.e., conditions such as an ideal imaging condition and a real imaging condition, and even a deteriorated imaging condition) can be rapidly and accurately simulated.
In some embodiments, the means for X-ray imaging comprises a bulb and a detector. As shown in fig. 2, a filter may be provided for the bulb, and the radiation emitted from the bulb is emitted after passing through the filter. The process of generating X-ray images of the target X-ray imaging system may include: the X-ray photons emitted by the bulb tube pass through the subject and then reach the detector, which obtains an X-ray image based on the received photons. Thus, the Monte Carlo model, which contains a point source, bulb and detector, can obtain an X-ray image by simulating the process of X-ray imaging.
The first configuration parameters may be basic parameters of the process of constructing the target X-ray imaging system, which are no longer changed during the subsequent imaging process, i.e. parameters which do not change with the change of the imaging conditions. In some embodiments, the first configuration parameters may include the number, geometric properties, materials, composition, and positional relationship between the components. Wherein the geometric attributes may include the apparent geometric dimensions of the bulb and the probe, the materials may include the materials of the bulb and the probe, the compositions may include the structures or the compositions of the internal elements of the bulb and the probe, and the like. As shown in fig. 2, the first configuration parameters are as follows: the material 201 of the bulb, the spacing 202 between the bulb and the filter, and the material 203 of the filter. The first configuration parameter is the same basic parameter in the ideal imaging model and the real imaging model and may not change with the change of imaging conditions. That is, the first configuration parameters are the same for both the configuration of the target X-ray imaging system forming the ideal imaging model and the configuration of the target X-ray imaging system forming the real imaging model.
In step 102, based on the Monte Carlo model, setting second configuration parameters based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system, respectively, to obtain an ideal imaging model and a real imaging model, respectively. The image quality of the X-ray image generated by the monte carlo model may change as the imaging conditions change. Under the condition of configuring imaging conditions under ideal imaging conditions, an image with more ideal image quality can be obtained, and a high-quality image which is not limited by the imaging conditions of an actual X-ray device, such as a high-dose condition of the actual X-ray device, can be obtained, and a Monte Carlo model with second configuration parameters under ideal imaging conditions is set, namely the ideal imaging model. Under the condition of configuring the imaging condition under the real imaging condition, a real image which is more close to reality (for example, noise and/or distortion is introduced to a certain extent) can be obtained, the real image can be an X-ray image obtained by an X-ray device under the conventional condition instead of the high-dose condition, and a Monte Carlo model for setting the second configuration parameter under the real imaging condition is the real imaging model.
In step 103, the same subject is subjected to simulated imaging based on the ideal imaging model and the real imaging model, so as to obtain an ideal image and a simulated real image, respectively, wherein the simulated real image is used for simulating the real image of the target X-ray imaging system. Therefore, after various imaging conditions are configured, the ideal imaging model can obtain a high-quality ideal image of the detected body through simulating the corresponding X-ray imaging process under various imaging conditions, and the real imaging model can obtain a simulated real image of the detected body through simulating the X-ray imaging process, wherein the high-quality image has the advantages of high signal-to-noise ratio and high definition, and the simulated real image has the characteristics of higher noise and lower signal-to-noise ratio. The ideal image and the simulated real image can be used as a 'good' image and a 'bad' image for training the deep learning neural network respectively. The ideal image and the simulated real image obtained by taking the cylinder as an example are respectively shown in fig. 3 and fig. 4, the signal-to-noise ratio and the contrast ratio of the ideal image in fig. 3 are higher, and the signal-to-noise ratio and the contrast ratio of the simulated real image in fig. 4 are lower. It can be seen that there is a large difference in image quality between the ideal image and the simulated real image.
Generating a high quality image in accordance with imaging conditions of an ideal imaging situation based on an actual X-ray imaging device is limited, whereas generating an ideal image in accordance with imaging conditions of an ideal imaging situation based on a monte carlo model is less constrained. Therefore, compared with the actual X-ray imaging equipment, the Monte Carlo model is easier to obtain high-quality images, and the obtained ideal image has the advantages of low noise and high signal to noise ratio, so that the obtained ideal image can be used as a 'good' image for training a deep learning neural network. Moreover, the manner of acquiring a "good" image does not require the use of X-rays for the patient, and can avoid the use of high doses of X-rays, as compared to the manner of acquiring a high quality clinical medical image with an actual X-ray device, so that the patient is not at risk for radiation. The ideal image has higher image quality, can have high reproducibility, and has a low distortion degree compared with the real image. Based on imaging under ideal imaging conditions, the problem of hardware condition limitation of the existing detector can be broken through, smaller pixel images can be obtained, and the definition of the whole image is improved.
At step 104, the ideal image and the simulated real image with the signal-to-noise ratio related parameter above the first threshold are taken as an X-ray image pair for training. The ideal image is used as a 'good' image, the simulated real image is used as a 'bad' image, and the deep learning neural network can accurately learn the mapping relation between the 'good' image and the 'bad' image, so that the trained deep learning neural network can restore the X-ray image detected by a patient under the conventional X-ray imaging condition into a 'good' image with higher quality, and the difficulty of obtaining a clinical image with high quality can be reduced. The first threshold value can be set according to a signal-to-noise ratio standard which is required to be achieved by the ideal image or a signal-to-noise ratio difference between the ideal image and the simulated real image.
In some embodiments, as described above with respect to Geant4, the first configuration parameters and the second configuration parameters are configured by writing to corresponding configuration files in the monte carlo simulation software, the second configuration parameters include a focus shape and a focus size, and setting the second configuration parameters based on the ideal imaging condition specifically includes: setting the focal spot shape as a point source and the focal spot size as a minimum focal spot size allowed in the monte carlo simulation software based on the ideal imaging conditions. The smaller the focal spot size, the higher the resolution of the image, the sharper the image, for example, the focal spot size can be set to 0.001mm, and the focal spot size under ideal imaging conditions can be set to be closer to 0 based on the monte carlo model, so the focal spot size can be much smaller than that of a real X-ray apparatus. Therefore, the problem of image blurring caused by the focal point size condition of the X-ray device can be solved, wherein the problem cannot be solved even if the X-ray is set to be high in dosage. The focus shape setting and the focus size setting of the ideal imaging condition contribute to obtaining an ideal image with higher definition.
Setting a second configuration parameter based on a true imaging condition of the target X-ray imaging system further comprises: setting the focal spot shape to be the shape of a point source used by the target X-ray imaging system based on the real imaging condition of the target X-ray imaging system, and setting the focal spot size to be the same as the focal spot size of the point source used by the target X-ray imaging system. The real image condition of the target X-ray imaging system may be set according to the configuration of the shape of the focal spot and the size of the focal spot of the X-ray apparatus actually used. For example, the shape of the focal spot of an X-ray apparatus for clinical practical use includes a rectangle or the like, and the focal spot size is 0.3mm. Thus, the simulated real image obtained by the real imaging model is more similar to the real image of the target X-ray imaging system. Therefore, the deep learning neural network trained based on the ideal image and the simulated real image can be processed into a high-quality X-ray image based on the X-ray image detected by the patient under the conventional X-ray imaging condition, and the generated high-quality X-ray image does not have the problem of image blurring caused by the focal point size imaging condition of the X-ray device.
In some embodiments, the second configuration parameters further include an X-ray dose, and setting the second configuration parameters based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system, respectively, specifically includes: based on the ideal imaging conditions, the X-ray dose is set to be 10 times or more of a conventional X-ray dose used by the target X-ray imaging system. The conventional X-ray dose can be the X-ray dose commonly used by clinical X-ray equipment, and the X-ray dose is 10 times and more than that under the ideal imaging condition, so that compared with the conventional X-ray dose, the noise of an X-ray image can be greatly reduced under the ideal imaging condition, and the definition of the image is improved. Analog imaging under ideal imaging conditions can obtain high quality images that may not be available with clinical X-ray devices.
In some embodiments, the method further comprises: the X-ray dose is set to a conventional X-ray dose used by the target X-ray imaging system based on a real imaging condition of the target X-ray imaging system. The conventional X-ray dose used by the subject X-ray imaging system may correspond to the X-ray dose typically used by clinical X-ray equipment and the resulting X-ray image may have some noise. Thus, the acquisition of "good" and "bad" images for deep learning neural network training based on ideal and real imaging models may be replaced with the use of X-ray equipment. And compared with the mode of acquiring high-quality images through an X-ray device, the X-ray irradiation is not needed to be used for a patient, and the use of high-dose X-rays can be avoided. In addition, the simulated real image obtained based on the conventional X-ray dose can better simulate the image of the real situation, and compared with the mode of obtaining the real image through the clinical X-ray equipment, the X-ray irradiation is not needed to be used for a patient. On the basis, after the deep learning neural network is trained based on the ideal image and the simulated real image, an X-ray image with higher definition can be generated based on the real image of the patient, so that medical observation is facilitated.
In some embodiments, the second configuration parameters further comprise scattered photons, and setting the second configuration parameters based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system, respectively, specifically comprises: based on the ideal imaging conditions, it is arranged to remove scattered photons reaching the detector. Further comprises: based on the real imaging conditions of the target X-ray imaging system, scattered photons reaching the detector are set to be included in the imaging photons. The ideal imaging model and the real imaging model can be respectively configured with scattered photon parameters through a configuration file of photon attributes. The target X-ray imaging system receives direct photons and scattered photons during the imaging process in accordance with real imaging conditions, whereas conventional X-ray devices may account for scattered photons in the imaging photons, in which case the resulting X-ray image may have problems with image blurring. In the Monte Carlo model, the direct photons and the scattered photons (the state parameters of the direct photons and the scattered photons are different) can be distinguished, so that the scattered photons reaching the detector can be removed through configuration based on the Monte Carlo model, and the detector can only receive the direct photons and generate an X-ray image based on the direct photons. The ideal image obtained based on the ideal imaging model can have higher signal-to-noise ratio and higher image definition, and can avoid the condition of special image blurring of X-ray imaging caused by scattered photons, wherein the blurring caused by X-ray scattering is an image problem which can not be described by mathematical models such as Gaussian blurring and the like simply. Therefore, the deep learning neural network trained based on the ideal image and the simulated real image can generate a high-quality X-ray image based on the real image of the patient, and the generated high-quality X-ray image does not have the problem of blurring on the image due to scattering.
In some embodiments, the second configuration parameters further include an X-ray energy spectrum, and setting the second configuration parameters based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system, respectively, specifically includes: and setting the X-ray energy spectrum as a continuous energy spectrum based on the real imaging condition of the target X-ray imaging system. Voltage values, typically in KV (kilovolts), can be set in the monte carlo model to simulate a real continuous spectrum. The simulated X-ray continuous spectrum is shown in fig. 5, the X-ray energy is shown on the abscissa, the X-ray value is given in KeV (kiloelectron volts), and the X-ray particle count is shown on the ordinate, and a plurality of characteristic peaks are seen in fig. 5. The real imaging model is based on an X-ray continuous energy spectrum, and can cause the problem of beam hardening due to a plurality of different photon energies, so that the quality of a real image can be influenced, and the obtained simulated real image can be closer to the real image.
In some embodiments, the method further comprises: the X-ray energy spectrum is set to a single energy spectrum based on the ideal imaging condition. A single photon energy simulation single energy spectrum can be set to directly act on the subject based on the monte carlo model. Therefore, the ideal imaging model is based on an ideal image obtained by a single energy spectrum, and the problem of beam hardening can be avoided. Therefore, the deep learning neural network after training based on the ideal image and the simulated real image can generate a high-quality X-ray image without the problem of beam hardening due to the X-ray continuous energy spectrum based on the input real image of the patient.
In some embodiments, the second configuration parameters further comprise a size of a single pixel of the detector, and setting the second configuration parameters based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system, respectively, specifically comprises: the size of the individual pixels of the detector is set to the size of the individual pixels of a conventional detector used by the target X-ray imaging system based on the real imaging conditions of the target X-ray imaging system. The sizes of the single pixels of the detectors in the X-ray devices with different specifications are generally fixedly arranged, so that the sizes of the single pixels of the detectors in the real imaging conditions can be selected according to the requirements for setting.
In some embodiments, the method further comprises: based on the ideal imaging conditions, the size of the individual pixels of the detector is set to be less than or equal to 10 times the size of the individual pixels of a conventional detector used in the target X-ray imaging system. The size of a single pixel of the detector can be reduced based on the Monte Carlo model, and the magnitude of the reduction can be larger. The sharpness of the corresponding resulting ideal image is greatly improved over the simulated real image while the size of the individual pixels of the detector is reduced. The manner in which an X-ray image is acquired with an X-ray device may not be able to improve the sharpness of the image with a reduced size of individual pixels or may not be able to be further reduced in size due to the detector's specification limitations. Generating a "good" image for deep learning neural network training based on the monte carlo model can overcome the problem of detector hardware limitations that do not allow imaging based on the smaller single pixel size. Therefore, the deep learning neural network trained based on the ideal image and the simulated real image can generate a high-quality clinical image based on the real image of the patient.
In some embodiments, the method further includes comparing a Signal-to-Noise Ratio (PSNR) of the simulated real image with a real image obtained by the target X-ray imaging system, and adjusting a second configuration parameter of the monte carlo model according to a result of the comparison until a difference between the Signal-to-Noise ratios of the simulated real image and the real image is below a second threshold. The parameters in which the simulated real image is compared with the real image may also be Structural similarity (Structural SIMilarity, SSIM) or the like. Taking peak signal-to-noise ratio as an example, if the difference between the signal-to-noise ratio of the simulated real image and the real image is below a second threshold, the simulated real image can reliably replace the real image as a "bad" image of the training deep learning depth network. And if the difference value of the signal to noise ratio of the simulated real image and the real image is larger than a second threshold value, adjusting the second configuration parameters of the Monte Carlo model to change the parameters such as the definition, the contrast and the like of the simulated real image until the difference value of the signal to noise ratio of the simulated real image and the real image is lower than the second threshold value. This process of simulating the comparison of the real images and determining the adjusted parameters may be implemented using a deep learning neural network, which may be trained using a plurality of simulated real images and real image pairs, which may include simulated real images and real image pairs having peak signal to noise ratio differences below a second threshold and simulated real images and real image pairs having peak signal to noise ratio differences above a second threshold.
In some embodiments, the method further comprises: the attribute parameters of the subject are imported into the Monte Carlo simulation software, or the mold body is processed and then the attribute parameters of the mold body are imported into the Monte Carlo simulation software. In a Monte Carlo simulation software capable of simulating the transmission of X-rays in a subject, the construction of the subject may include: the attribute parameters of the patient are imported into the Monte Carlo simulation software, so that the Monte Carlo simulation software knows the characteristics of the object, and the transmission process of X-rays in the object can be simulated. Alternatively, a processing phantom (such as a simulation phantom) may be simulated based on the design image, and then the properties of the simulation phantom may be imported into Monte Carlo simulation software to simulate the transmission of X-rays in the subject. In this way, the true subject can be better simulated, and the accuracy of the obtained X-ray image can be improved.
There is also provided, in accordance with an embodiment of the present application, an apparatus for generating an X-ray image pair for training. As shown in fig. 6, the apparatus 600 includes: an interface 601 configured to receive a configuration of the second configuration parameter by a user based on the ideal imaging conditions and the real imaging conditions of the target X-ray imaging system; and a processor 602 configured to perform a method of generating an X-ray image pair for training as described in any of the embodiments of the application. This way of acquiring a "good" image does not have to use X-ray radiation for the patient, compared to the way of acquiring a high quality clinical medical image by means of an actual X-ray device, and can avoid the use of high doses of X-rays, leaving the patient without radiation risk. Compared with a real image, the ideal image has higher image quality and high reducibility, and the problem of image distortion is avoided. Based on imaging under ideal imaging conditions, the problem that existing detectors cannot further reduce the size of individual pixels can be changed, increasing the sharpness of the overall image by a smaller individual pixel.
In some embodiments, the processor may be a processing device including more than one general purpose processing device, such as a microprocessor, central Processing Unit (CPU), graphics Processing Unit (GPU), or the like. More specifically, the processor may be a Complex Instruction Set Computing (CISC) microprocessor, a Reduced Instruction Set Computing (RISC) microprocessor, a Very Long Instruction Word (VLIW) microprocessor, a processor running other instruction sets, or a processor running a combination of instruction sets. The processor may also be one or more special purpose processing devices such as an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), a Digital Signal Processor (DSP), a system on a chip (SoC), or the like.
There is also provided, in accordance with an embodiment of the present application, a non-transitory computer readable medium having instructions stored thereon, which when executed by a processor, perform the steps of a method of generating an X-ray image pair for training as described in any of the embodiments of the present application. The ideal image and simulated real image may be obtained after execution of the instructions to generate an X-ray image pair for training. This way of acquiring a "good" image does not have to use X-ray radiation for the patient, compared to the way of acquiring a high quality clinical medical image by means of an actual X-ray device, and can avoid the use of high doses of X-rays, leaving the patient without radiation risk. Compared with a real image, the ideal image has higher image quality and high reducibility, and the problem of image distortion is avoided. Based on imaging under ideal imaging conditions, the problem that existing detectors cannot further reduce the size of individual pixels can be changed, increasing the sharpness of the overall image by a smaller individual pixel.
Furthermore, although exemplary embodiments have been described herein, the scope thereof includes any and all embodiments having equivalent elements, modifications, omissions, combinations (e.g., of the various embodiments across), adaptations or alterations as pertains to the present application. The elements in the claims are to be construed broadly based on the language employed in the claims and are not limited to examples described in the present specification or during the practice of the application, which examples are to be construed as non-exclusive. It is intended, therefore, that the specification and examples be considered as exemplary only, with a true scope and spirit being indicated by the following claims and their full scope of equivalents.
The above description is intended to be illustrative and not restrictive. For example, the above-described examples (or one or more aspects thereof) may be used in combination with each other. For example, other embodiments may be used by those of ordinary skill in the art upon reading the above description. In addition, in the above detailed description, various features may be grouped together to streamline the application. This is not to be interpreted as an intention that the features of the non-claimed application are essential to any claim. Rather, the inventive subject matter may lie in less than all features of a particular disclosed embodiment. Thus, the claims are hereby incorporated into the detailed description as examples or embodiments, with each claim standing on its own as a separate embodiment, and it is contemplated that these embodiments may be combined with one another in various combinations or permutations. The scope of the application should be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.
The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application, the scope of which is defined by the claims. Various modifications and equivalent arrangements of this application will occur to those skilled in the art, and are intended to be within the spirit and scope of the application.
Claims (9)
1. A method of generating an X-ray image pair for training, comprising:
forming a Monte Carlo model of a target X-ray imaging system comprising a point source and a component by using Monte Carlo simulation software, wherein the Monte Carlo model is provided with a first configuration parameter related to the component, and the first configuration parameter is not changed along with the change of imaging conditions;
setting second configuration parameters based on the Monte Carlo model and the ideal imaging condition and the real imaging condition of the target X-ray imaging system respectively to obtain an ideal imaging model and a real imaging model respectively;
the first configuration parameters and the second configuration parameters are configured by writing corresponding configuration files in the Monte Carlo simulation software;
setting the second configuration parameters based on the ideal imaging conditions and the real imaging conditions of the target X-ray imaging system, respectively, comprises:
the second configuration parameters comprise a focus shape and a size, the focus shape is set to be a point source based on the ideal imaging condition, and the focus size is set to be the minimum focus size allowed in the Monte Carlo simulation software; setting the focal spot shape as the shape of a point source used by the target X-ray imaging system based on the real imaging condition of the target X-ray imaging system, and setting the focal spot size to be the same as the focal spot size of the point source used by the target X-ray imaging system;
the second configuration parameters further include scattered photons, based on the ideal imaging conditions, setting to remove scattered photons reaching the detector; setting scattered photons reaching a detector to be counted into imaging photons based on the real imaging condition of the target X-ray imaging system;
respectively carrying out simulated imaging on the same subject based on the ideal imaging model and the real imaging model to respectively obtain an ideal image and a simulated real image, wherein the simulated real image is used for simulating a real image of a target X-ray imaging system; and
and taking the simulated real image and an ideal image with the signal-to-noise ratio related parameter higher than a first threshold value as an X-ray image pair for training, wherein the first threshold value is set according to a signal-to-noise ratio standard which the ideal image needs to reach or a signal-to-noise ratio difference between the ideal image and the simulated real image.
2. The method of claim 1, wherein the components of the X-ray imaging include a bulb and a detector, and the first configuration parameters include a number, geometric properties, materials, composition, and positional relationship between the components.
3. The method of claim 1, wherein the second configuration parameters further comprise an X-ray dose, and wherein setting the second configuration parameters based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system, respectively, comprises:
setting the X-ray dose to be greater than or equal to 10 times a conventional X-ray dose used by the target X-ray imaging system based on the ideal imaging condition;
the X-ray dose is set to a conventional X-ray dose used by the target X-ray imaging system based on a real imaging condition of the target X-ray imaging system.
4. The method of claim 1, wherein the second configuration parameters further comprise an X-ray energy spectrum, and wherein setting the second configuration parameters based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system, respectively, comprises:
setting the X-ray energy spectrum to a single energy spectrum based on the ideal imaging condition;
and setting the X-ray energy spectrum as a continuous energy spectrum based on the real imaging condition of the target X-ray imaging system.
5. The method of claim 2, wherein the second configuration parameters further comprise a size of a single pixel of the detector, and wherein setting the second configuration parameters based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system, respectively, specifically comprises:
setting a size of a single pixel of the detector to be equal to or less than 10 times a size of a single pixel of a conventional detector used in the target X-ray imaging system based on the ideal imaging condition;
the size of the individual pixels of the detector is set to the size of the individual pixels of a conventional detector used by the target X-ray imaging system based on the real imaging conditions of the target X-ray imaging system.
6. The method of claim 1, further comprising comparing the simulated real image with a signal-to-noise ratio of a real image obtained by the target X-ray imaging system, and adjusting a second configuration parameter of the monte carlo model based on the result of the comparison until a difference in the signal-to-noise ratio of the simulated real image and the real image is below a second threshold.
7. The method as recited in claim 1, further comprising: the attribute parameters of the subject are imported into the Monte Carlo simulation software, or the mold body is processed and then the attribute parameters of the mold body are imported into the Monte Carlo simulation software.
8. An apparatus for generating an X-ray image pair for training, comprising:
an interface configured to receive a configuration of the second configuration parameter by a user based on the ideal imaging condition and the real imaging condition of the target X-ray imaging system; and
a processor configured to perform the method of generating an X-ray image pair for training of any of claims 1-7.
9. A non-transitory computer readable medium having instructions stored thereon, wherein the instructions, when executed by a processor, perform the steps of the method of generating an X-ray image pair for training of any of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310459477.5A CN116188619B (en) | 2023-04-26 | 2023-04-26 | Method, apparatus and medium for generating X-ray image pair for training |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310459477.5A CN116188619B (en) | 2023-04-26 | 2023-04-26 | Method, apparatus and medium for generating X-ray image pair for training |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116188619A CN116188619A (en) | 2023-05-30 |
| CN116188619B true CN116188619B (en) | 2023-09-01 |
Family
ID=86444648
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310459477.5A Active CN116188619B (en) | 2023-04-26 | 2023-04-26 | Method, apparatus and medium for generating X-ray image pair for training |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116188619B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104840211A (en) * | 2015-05-18 | 2015-08-19 | 上海联影医疗科技有限公司 | Scattering correction method for projected image and device |
| CN108169256A (en) * | 2016-12-07 | 2018-06-15 | 同方威视技术股份有限公司 | Multi-power spectrum x-ray imaging system and the method for identifying article using multi-power spectrum x-ray imaging system |
| KR20200059712A (en) * | 2018-11-21 | 2020-05-29 | 이희신 | System and method for scatter correction of x-ray image |
| CN111448590A (en) * | 2017-09-28 | 2020-07-24 | 皇家飞利浦有限公司 | Scatter correction based on deep learning |
| CN111597762A (en) * | 2020-05-29 | 2020-08-28 | 南京林业大学 | Method for calculating characteristic peak intensity of X-ray fluorescence spectrum |
| CN111914392A (en) * | 2020-06-23 | 2020-11-10 | 上海联影医疗科技有限公司 | X-ray imaging equipment and modeling method and device of X-ray image |
| CN112434460A (en) * | 2020-11-11 | 2021-03-02 | 北京航星机器制造有限公司 | General security check machine ray shielding design method based on Geant4 |
| CN114219820A (en) * | 2021-12-08 | 2022-03-22 | 苏州工业园区智在天下科技有限公司 | Neural network generation method, denoising method and device |
| CN115828642A (en) * | 2023-02-15 | 2023-03-21 | 北京航空航天大学 | Unity-based GPU (graphics processing Unit) accelerated X-ray digital imaging simulation method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019056309A1 (en) * | 2017-09-22 | 2019-03-28 | Shenzhen United Imaging Healthcare Co., Ltd. | Method and system for generating a phase contrast image |
| EP4020375A1 (en) * | 2020-12-23 | 2022-06-29 | Koninklijke Philips N.V. | System and methods for augmenting x-ray images for training of deep neural networks |
-
2023
- 2023-04-26 CN CN202310459477.5A patent/CN116188619B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104840211A (en) * | 2015-05-18 | 2015-08-19 | 上海联影医疗科技有限公司 | Scattering correction method for projected image and device |
| CN108169256A (en) * | 2016-12-07 | 2018-06-15 | 同方威视技术股份有限公司 | Multi-power spectrum x-ray imaging system and the method for identifying article using multi-power spectrum x-ray imaging system |
| CN111448590A (en) * | 2017-09-28 | 2020-07-24 | 皇家飞利浦有限公司 | Scatter correction based on deep learning |
| KR20200059712A (en) * | 2018-11-21 | 2020-05-29 | 이희신 | System and method for scatter correction of x-ray image |
| CN111597762A (en) * | 2020-05-29 | 2020-08-28 | 南京林业大学 | Method for calculating characteristic peak intensity of X-ray fluorescence spectrum |
| CN111914392A (en) * | 2020-06-23 | 2020-11-10 | 上海联影医疗科技有限公司 | X-ray imaging equipment and modeling method and device of X-ray image |
| CN112434460A (en) * | 2020-11-11 | 2021-03-02 | 北京航星机器制造有限公司 | General security check machine ray shielding design method based on Geant4 |
| CN114219820A (en) * | 2021-12-08 | 2022-03-22 | 苏州工业园区智在天下科技有限公司 | Neural network generation method, denoising method and device |
| CN115828642A (en) * | 2023-02-15 | 2023-03-21 | 北京航空航天大学 | Unity-based GPU (graphics processing Unit) accelerated X-ray digital imaging simulation method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116188619A (en) | 2023-05-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110570492B (en) | CT artifact suppression method, device and medium based on neural network | |
| Sun et al. | Improved scatter correction using adaptive scatter kernel superposition | |
| Van Slambrouck et al. | Metal artifact reduction in computed tomography using local models in an image block‐iterative scheme | |
| CN103900931B (en) | Multi-energy-spectrum CT imaging method and imaging system | |
| Schüller et al. | Segmentation‐free empirical beam hardening correction for CT | |
| CN107802280B (en) | Correction curve generation method, projection image correction system, and storage medium | |
| JP7185236B2 (en) | Apparatus and method for correcting scatter in a radiography system | |
| CN115270075B (en) | Energy spectrum data correction method of dual-energy CT system based on tungsten target X-ray tube | |
| Altunbas et al. | A unified scatter rejection and correction method for cone beam computed tomography | |
| WO2020044060A1 (en) | X-ray imaging system | |
| CN111435542A (en) | Providing a differential image data set and providing a training function | |
| Schmidt et al. | Addressing CT metal artifacts using photon‐counting detectors and one‐step spectral CT image reconstruction | |
| Elhamiasl et al. | Low-dose x-ray CT simulation from an available higher-dose scan | |
| Li et al. | X-ray photon-counting data correction through deep learning | |
| Richtsmeier et al. | Metal artifact correction in photon‐counting detector computed tomography: metal trace replacement using high‐energy data | |
| Kim et al. | Convolutional neural network–based metal and streak artifacts reduction in dental CT images with sparse‐view sampling scheme | |
| Li et al. | MDM-PCCT: Multiple dynamic modulations for high-performance spectral PCCT imaging | |
| CN116188619B (en) | Method, apparatus and medium for generating X-ray image pair for training | |
| CN111435530B (en) | Providing a restricted image data set and/or a differential image data set | |
| CN110246199B (en) | Projection domain data noise removing method for energy spectrum CT | |
| CN111914392A (en) | X-ray imaging equipment and modeling method and device of X-ray image | |
| Danyk et al. | Using clustering analysis for determination of scattering kernels in X-ray imaging | |
| Mettivier et al. | Evaluation of scattering in cone-beam breast computed tomography: A Monte Carlo and experimental phantom study | |
| Zhang et al. | Modularized data‐driven reconstruction framework for nonideal focal spot effect elimination in computed tomography | |
| Yu et al. | Heel effect adaptive flat field correction of digital x‐ray detectors |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |