Method and device for predicting X-ray scattered signals and method for correcting scattered beams

Abstract

The present application relates to a method and apparatus for predicting X-ray scatter signals and a method for correcting a scattered beam. The method for predicting X-ray scatter signals for an examination object includes: scanning each of a plurality of phantoms in a scanning manner such that the scattering degree of each phantom is less than a reference scattering degree to obtain first projection data for each phantom; scanning each of a plurality of phantoms in a scanning manner such that the scattering degree of each phantom is equal to the reference scattering degree to obtain second projection data for each phantom; obtaining a true scatter signal for each phantom by subtracting the first projection data from the second projection data; training a learning model based on the second projection data of each phantom and the true scatter signal of each phantom to obtain a trained learning model; and applying the trained learning model to X-ray projection data for the examination object to predict the scatter signal of the X-ray projection data.

Description

Technical Field

[0001] The present application generally relates to the technical field of computed tomography (CT) projection signal correction and image processing, and more specifically, to a method for predicting X-ray scattering signals based on a learning technology. Background Art

[0002] Computed tomography (CT) has been widely used in the field of medical imaging. However, because scattered radiation is an inherent characteristic of X-rays, it is inevitable that scattered radiation will be generated during the CT imaging process, affecting image quality. This scattered radiation degrades CT image quality.

[0003] CT systems typically use anti-scatter grids (ASGs) on detector hardware to reduce the impact of scattered signals. However, with the increasing use of wide-body detectors and the pressure on hardware costs, CT manufacturers are increasingly inclined to use software algorithms to calibrate scattered signals.

[0004] Various scatter correction algorithms and methods have been developed based on analytical models. These algorithms often have varying degrees of computational complexity. For example, some scatter correction algorithms and methods require iterative processes or statistical simulations, such as Monte Carlo simulations. Scatter correction algorithms and methods developed based on analytical models, such as those based on Monte Carlo simulations, are often complex and require iterative processes, which results in high computational effort and, therefore, are not widely used.

[0005] On current CT systems, a convolution kernel algorithm based solely on raw projection data is used. This algorithm is a simple and efficient scatter correction algorithm that does not involve iterative methods. However, this algorithm is highly model-dependent and difficult to scale for a wider range of applications. As the scattered radiation contribution becomes larger and scanning conditions become more complex, current convolution kernel scatter correction algorithms are unable to predict the appropriate scatter signal for all situations.

[0006] Furthermore, convolution kernel design and modeling are key aspects of current scatter correction algorithms, as they estimate the distribution of scatter after it exits the X-ray scanned object and reaches the detector. Each new system type typically requires simulations that rely on system properties as well as measurements of the specified scatter signal from representative phantoms scanned under different scanning conditions. Based on these simulated and measured data, the convolution kernel is then designed and the model parameters adjusted. For systems with larger scatter contributions, more complex models must be applied to compensate for the scatter effects tailored to each situation. Summary of the Invention

[0007] The main purpose of this application is to train a learning model using the real scattering signals of the phantom obtained by scanning the phantom under different scanning conditions and the scattering attenuation data of the entire phantom obtained by scanning the phantom under different scanning conditions with a standard clinical scan width (i.e., the reference scan width mentioned below) as training data, so as to predict the corresponding scattering signals according to different scanning conditions (different object attenuation structures, different object positions, different scanning powers (e.g., kV)), thereby getting rid of the limitation of the current convolution kernel algorithm that can only predict fixed scattering signals with a limited convolution kernel, and obtaining more accurate scattering correction results.

[0008] To achieve the above objectives, according to one aspect of the present application, a method for predicting a scattered signal of X-rays for an examination object is provided, the method comprising: scanning each of a plurality of phantoms in a scanning manner such that a scattering degree of each of the phantoms is less than a reference scattering degree, to acquire first projection data of each of the phantoms; scanning each of the plurality of phantoms in a scanning manner such that a scattering degree of each of the phantoms is equal to the reference scattering degree, to acquire second projection data of each of the phantoms; obtaining a true scattering signal of each phantom by subtracting the first projection data of each phantom from the second projection data of each phantom; training a learning model based on the second projection data of each phantom and the true scattering signal of each phantom to obtain a trained learning model; and applying the trained learning model to the X-ray projection data of the examination object to predict the scattering signal in the X-ray projection data.

[0009] In this article, the baseline scattering degree refers to the scattering degree generated by scanning the phantom under the standard clinical mode. In this way, the phantom is scanned in a scanning mode that makes the contribution of the scattering signal small enough to be negligible to obtain the phantom's non-scattered projection data, and the phantom is scanned in the standard clinical mode to obtain the phantom's scattered projection data. The true scattering signal of the phantom is obtained by subtracting the scattered projection data from the non-scattering projection data. The learning model is trained based on the true scattering signal of the phantom and the projection data of the phantom under the standard scanning. Therefore, since the data used for training are all data measured from the phantom, that is, actually measured data, compared with the method of training the learning model based on simulated data in the prior art, more accurate scattering signals of X-ray projection data can be obtained.

[0010] Furthermore, the data processing process of this application does not directly operate on images, and therefore does not require scanning of the subject's image. Therefore, the above method has no impact on the structure of the existing data processing pipeline, and all other calibration or correction steps (e.g., for correcting for system and detector effects) remain unchanged. Furthermore, existing methods for scatter estimation based on images often have little or no physical evidence, making it difficult to control image quality (such as artifacts and image noise). In contrast, in this application, scatter estimation based on projection data can avoid these issues.

[0011] Further, according to one embodiment of the present application, the learning model is trained based on the second projection data of each phantom and the real scattering signal of each phantom to obtain the trained learning model, including: pre-calculating the second projection data of each phantom based on the physical model to obtain the calculated scattering signal of each phantom; and training the learning model using the calculated scattering signal of each phantom and the real scattering signal of each phantom as training data.

[0012] In this way, since the second projection data of each phantom is preprocessed based on the physical model, more physical evidence can be obtained in the process of scatter signal prediction, thereby making it easier to control image quality.

[0013] Furthermore, according to an embodiment of the present application, the calculated scattering signal of each phantom is used as input training data of the learning model and the real scattering signal of each phantom is used as output training data of the learning model to train the learning model.

[0014] In this way, the calculated scattered signal is still calculated from the physical model with a calibration pre-processor, so that the learning model is only used to learn the real distribution of the generated scattered radiation on the specific system, resulting in reduced learning complexity.

[0015] Furthermore, according to an embodiment of the present application, scanning each phantom in a scanning manner such that the scattering degree of each phantom in the plurality of phantoms is less than a reference scattering degree to acquire first projection data of each phantom includes: performing segmented scanning on each of the plurality of phantoms with a first width less than a reference scanning width under a plurality of different scanning conditions to acquire projection data of each first segment having the first width of each phantom under each scanning condition, wherein the reference scanning width is a width adopted in a standard clinical scan; and acquiring the first projection data of each phantom under each scanning condition by connecting the projection data of each first segment of each phantom acquired under each scanning condition.

[0016] In this way, scatter-free projection data of the phantom can be obtained by performing segmented scanning on the phantom. For example, for a particular phantom, a signal is first acquired using a small collimator width, which can be considered a true scatter-free signal. The small collimator is then moved to acquire the entire signal within the detector range. The signals acquired with each small collimator width are then connected to obtain scatter-free projection data for the entire phantom. In this way, the accuracy of the scatter-free projection data can be improved by controlling the acquisition width, thereby obtaining appropriate scatter-free projection data. Furthermore, in this way, specific scatter signals under different scanning conditions (such as different object attenuation structures, different object positions, and different scanning powers (kV)) can be predicted, allowing the subsequently trained model to compensate for scatter effects specific to different scanning conditions and also being applicable to existing CT systems.

[0017] Furthermore, according to one embodiment of the present application, scanning each phantom in a scanning manner such that the scattering degree of each of the multiple phantoms is equal to the reference scattering degree to acquire the second projection data of each phantom includes: scanning each of the multiple phantoms with a reference scanning width under multiple different scanning conditions to acquire the second projection data of each phantom under each scanning condition.

[0018] In this way, for example, by acquiring signals with a large collimator width for a certain phantom (a typical mode in clinical practice), the acquisition width can be controlled according to the type and application of the CT imaging system to obtain the required scattered projection data, simplifying the computational complexity.

[0019] Further, according to an embodiment of the present application, the first projection data of each phantom and the second projection data of each phantom are both distributions of absorption coefficients collected from each phantom.

[0020] In this way it is possible to make it unnecessary to scan an image of the object during data processing, so it has no impact on the structure of the current data processing pipeline and all other calibration or correction steps remain intact.

[0021] Further, according to one embodiment of the present application, the physical model is δ = γ·T0·exp(-κ·T), where δ is the calculated scattering signal, T is the total absorption coefficient on the entire X-ray scanning path, T0 is the absorption coefficient of each phantom, and κ and γ are model fitting parameters.

[0022] In this way, calculated scattering signals with more physical evidence can be obtained, which can improve the accuracy of the prediction.

[0023] Furthermore, according to one embodiment of the present application, the learning model includes machine learning and deep learning with different network types.

[0024] In this way, the method of the present application can use various learning methods to perform scattering prediction, regardless of which learning technology can be used in the method proposed in the present application, such as machine learning with different network types and deep learning such as convolutional neural networks.

[0025] Further, according to an embodiment of the present application, the plurality of different scanning conditions include at least one of the following: different phantom attenuation structures, different phantom positions, and different scanning powers.

[0026] In this way, even if the scanning conditions are complex and varied, an appropriate scattered signal can still be predicted for all situations.

[0027] According to another aspect of the present application, an apparatus for predicting a scattered signal of an X-ray for an examination object is provided, the apparatus comprising: a first acquisition module configured to scan each of a plurality of phantoms in a scanning manner such that the scattering degree of each of the phantoms is less than a reference scattering degree, thereby acquiring first projection data of each phantom; a second acquisition module configured to scan each of the plurality of phantoms in a scanning manner such that the scattering degree of each of the phantoms is equal to the reference scattering degree, thereby acquiring second projection data of each phantom; a subtraction module configured to obtain a true scattering signal of each phantom by subtracting the first projection data of each phantom from the second projection data of each phantom; a training module configured to train a learning model based on the second projection data of each phantom and the true scattering signal of each phantom to obtain a trained learning model; and a prediction module configured to apply the trained learning model to projection data of X-rays for the examination object to predict the scattering signal in the projection data of the X-rays.

[0028] In this way, the learning model is trained based on the real scattering signal of the phantom and the projection data of the phantom under the standard clinical mode. Therefore, since the data used for training are all data measured from the phantom, that is, actual measured data, compared with the method of training the learning model based on simulated data in the prior art, more accurate scattering signals of X-ray projection data can be obtained.

[0029] According to another aspect of the present application, a method for correcting a scattered beam of X-rays for an examination object is provided, the method comprising: scanning each of a plurality of phantoms in a scanning manner such that a scattering degree of each of the phantoms is less than a reference scattering degree, to acquire first projection data of each of the phantoms; scanning each of the phantoms in a scanning manner such that a scattering degree of each of the plurality of phantoms is equal to the reference scattering degree, to acquire second projection data of each of the phantoms; obtaining a true scatter signal of each of the phantoms by subtracting the first projection data of each of the phantoms from the second projection data of each of the phantoms; training a learning model based on the second projection data of each of the phantoms and the true scatter signal of each of the phantoms to obtain a trained learning model; applying the trained learning model to X-ray projection data for the examination object to predict a scatter signal in the X-ray projection data; and correcting the X-ray projection data for the examination object using the predicted scatter signal to generate scatter-corrected projection data.

[0030] In this way, the learning model is trained based on the real scattering signal of the phantom and the projection data of the phantom under the clinical standard mode. Therefore, since the data used for training are all data measured from the phantom, that is, actual measured data, compared with the method of training the learning model based on simulated data in the prior art, more accurate scattering signals of X-ray projection data can be obtained, thereby more accurately correcting the X-ray scattering signals of the inspection object.

[0031] According to another aspect of the present application, a method for reconstructing an X-ray image of an examination object is provided, the method comprising: scanning each of a plurality of phantoms in a scanning manner such that a scattering degree of each of the phantoms is less than a baseline scattering degree to acquire first projection data of each of the phantoms; scanning each of the phantoms in a scanning manner such that a scattering degree of each of the phantoms is equal to the baseline scattering degree to acquire second projection data of each of the phantoms; obtaining a true scattering signal of each of the phantoms by subtracting the first projection data of each of the phantoms from the second projection data of each of the phantoms; training a learning model based on the second projection data of each of the phantoms and the true scattering signal of each of the phantoms to obtain a trained learning model; applying the trained learning model to X-ray projection data of the examination object to predict a scattering signal in the X-ray projection data; correcting the X-ray projection data of the examination object using the predicted scattering signal to generate scatter-corrected projection data; and reconstructing the scatter-corrected projection data to generate a scatter-corrected reconstructed computed tomography image.

[0032] In this way, the learning model is trained based on the real scattering signal of the phantom and the projection data of the phantom under the standard clinical mode. Therefore, since the data used for training are all data measured from the phantom, that is, actual measured data, compared with the method of training the learning model based on simulated data in the prior art, more accurate scattering signals in the X-ray projection data can be obtained, thereby improving the quality of the reconstructed X-ray image.

[0033] According to another aspect of the present application, a computer storage medium is provided, storing a program, which, when executed, causes a computer to perform the above method.

[0034] In an embodiment of the present application, the actual scatter signals of the phantom acquired by scanning the phantom under different scanning conditions, as well as the scatter attenuation data of the entire phantom acquired by scanning the phantom under different scanning conditions with a standard clinical scan width, are used as training data to train a learning model. This model predicts specific scatter signals based on different scanning conditions (such as different object attenuation structures, different object positions, and different scan powers (e.g., kV)), thereby achieving more accurate scatter correction results. Since the image data of the scanned object is not required during the data processing of the present application, the above method does not require additional iterative processing and has no impact on the structure of the current data processing pipeline. All other calibration or correction steps (e.g., for correcting system and detector effects) remain unchanged. Furthermore, in the present application, scatter prediction based on projection data can avoid image quality issues such as artifacts. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings that constitute part of this application are used to provide a further understanding of this application. The illustrative embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation on this application. In the drawings:

[0036] Figure 1 is a flowchart illustrating a method for predicting a scattered signal of an X-ray for an inspection object according to an embodiment of the present application;

[0037] Figure 2 is a block diagram showing an apparatus for predicting a scattered signal of X-rays for an inspection object according to an embodiment of the present application;

[0038] Figure 3 is a flowchart illustrating a method for correcting a scattered beam of X-rays for an inspection object according to an embodiment of the present application; and

[0039] Figure 4 FIG. 1 is a flowchart illustrating a method for reconstructing an X-ray image of an inspection object according to an embodiment of the present application.

[0040] Figure 5A It shows the two-dimensional distribution of the absorption coefficient of the scanned phantom in a certain projection according to an embodiment of the present application, the scattering signal that needs to be corrected given by the trained learning model applied to the two-dimensional distribution of the absorption coefficient of the scanned phantom in a certain projection, and the two-dimensional distribution of the actual scattering signal of the scanned phantom.

[0041] Figure 5B The figure shows the calculated scattering signal obtained after pre-calculation processing of the physical model according to the two-dimensional distribution of the absorption coefficient of the scanned phantom in a certain projection according to the embodiment of the present application, the scattering signal that needs to be corrected given by applying the trained learning model to the two-dimensional distribution of the absorption coefficient of the scanned phantom in a certain projection, and the two-dimensional distribution of the actual scattering signal of the scanned phantom.

[0042] Figures 6A to 6E The present invention is based on the simple phantom (water phantom), the complex phantom (body phantom) and the real clinical image (head scan). Figure 4 The images constructed by the proposed method are compared with those obtained by applying clinically validated algorithms used in current CT systems. DETAILED DESCRIPTION

[0043] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0044] It should be noted that, unless otherwise specified, all technical and scientific terms used in this application have the same meaning as commonly understood by ordinary technicians in the technical field to which this application belongs.

[0045] In this application, unless otherwise specified, directional words such as "up, down, top, bottom" are usually used with reference to the directions shown in the drawings, or with reference to the components themselves in the vertical, perpendicular or gravity direction; similarly, for ease of understanding and description, "inside and outside" refer to the inside and outside relative to the outline of each component itself, but the above directional words are not used to limit this application.

[0046] Figure 1 FIG. 1 is a flow chart showing a method for predicting scattered signals of X-rays of an inspection object according to an embodiment of the present application. Figure 1 As shown, the method for predicting the scattered signal of X-rays for the inspection object includes:

[0047] In step S100 , each phantom is scanned in a scanning manner such that the scattering degree of each phantom among the plurality of phantoms is smaller than a reference scattering degree, so as to obtain first projection data of each phantom.

[0048] In this step, the baseline scattering level refers to the scattering level generated by scanning the phantom under a standard clinical mode. Scanning each phantom in a scanning mode such that the scattering level of each of the multiple phantoms is less than the baseline scattering level refers to scanning the phantom so that the contribution of the scattering signal from the phantom is so small as to be negligible relative to the phantom's absorption. For example, the scattering level of the phantom is 0.01 times the absorption level of the phantom. In other words, the scattering included in the projection data obtained when the scattering level of the phantom is less than the baseline scattering level has no impact on the image quality and need not be considered. Accordingly, the first projection data may refer to projection data of the entire phantom that can be considered scatter-free.

[0049] Specifically, by scanning a phantom in a scanning manner that makes the scattering degree of the phantom negligible relative to the absorption degree of the phantom, projection data of the phantom that can be considered as scatter-free is obtained. As an implementation method, the first projection data of the phantom can be obtained by: scanning each of the multiple phantoms in segments under multiple different scanning conditions with a first width less than a reference scan width, thereby acquiring projection data of each first segment of each phantom having the first width under each scanning condition, wherein the reference scan width is the width used in standard clinical scanning; and connecting the projection data of each first segment of each phantom acquired under each scanning condition to obtain the first projection data of each phantom under each scanning condition. The different scanning conditions may include different phantom attenuation structures, different phantom positions, and different scanning powers (energies), such as scanning kV.

[0050] As a specific example, for a certain phantom, a small collimator width (i.e., the first width) is first used to collect signals, which can be regarded as real signals without scattering, and the small collimator is moved to collect all signals within the detector range. Then, the signals collected with each small collimator width are connected to obtain the scattering-free projection data of the entire phantom (i.e., the first projection data).

[0051] In step S101 , each phantom is scanned in a scanning manner such that the scattering degree of each phantom among a plurality of phantoms is equal to a reference scattering degree, so as to obtain second projection data of each phantom.

[0052] In this step, the level of scatter generated during the scanning of the phantom is a baseline level of scatter generated during a standard clinical scan. The scatter in the projection data of the phantom obtained from this scan affects the quality of the final image and is non-negligible, thus requiring removal. Accordingly, the second projection data may refer to the acquired projection data including scattered radiation that affects image quality.

[0053] As an implementation, each of the multiple phantoms can be scanned at a reference scan width under multiple different scanning conditions, thereby acquiring second projection data for each phantom under each scanning condition. Here, the different scanning conditions can also include different phantom attenuation structures, different phantom positions, and different scanning powers, such as scanning kV.

[0054] As an example, the second projection data of a phantom is obtained by acquiring signals at a large collimator width (ie, the second width) for a certain phantom (a typical mode in clinical practice).

[0055] In this context, the collimator can be a thin sheet installed after the X-ray generator of a CT imaging system and configured to allow only X-rays from within a specific range of directions to scan and image the phantom. The first and second projection data of the phantom can, for example, be the absorption coefficient distribution of the phantom at a specific projection. The phantom can be a simple phantom, such as a water phantom, or a more complex phantom, such as a human head phantom or a human lung phantom.

[0056] Step S102 , obtaining a true scattering signal of each phantom by subtracting the first projection data of each phantom from the second projection data of each phantom; in this step, the true scattering signal refers to a scattering signal value obtained when scanning the phantom.

[0057] Step S103 : training the learning model based on the second projection data of each phantom and the real scattering signal of each phantom to obtain a trained learning model.

[0058] By scanning a diverse and complex phantom, the actual scattering signal values ​​(e.g., a two-dimensional array) obtained in step S102 for each projection direction can be used as the true values ​​(Signal 1) in training the learning model. Simultaneously, the projection data of the phantom in each projection direction obtained in step S101 can be used as Signal 2. For example, by using signals collected under a large collimator width, a two-dimensional distribution of the phantom's absorption coefficient in each projection direction can be obtained (Signal 2). The learning model is trained based on Signal 1 and Signal 2, resulting in a trained learning model that can predict the scattering signal of X-ray projection data of an examination object (e.g., a specific part or parts of a human body).

[0059] In addition, in step S103, the second projection data of each phantom can be pre-calculated based on the physical model to obtain a calculated scattering signal of each phantom, and the calculated scattering signal of each phantom and the real scattering signal of each phantom are used as training data to train the learning model.

[0060] For example, the calculated scattered signal can be a two-dimensional array, and this two-dimensional data can serve as the input of the learning model, while signal 1, also a two-dimensional array, serves as the output of the learning model, thereby pairing the learning model for training. Furthermore, there are no specific requirements for the learning algorithm of the learning model, and examples include, but are not limited to, deep learning, neural network learning such as convolutional neural networks, and machine learning. There are also no specific requirements for the selection of the network.

[0061] As an example, the first projection data of each phantom and the second projection data of each phantom are the distribution of absorption coefficients collected from each phantom. In other words, in the present application, the data collected in the projection space are transmitted X-rays, and each X-ray can be represented by a certain spatially correlated absorption coefficient distribution.

[0062] The physical model can be δ=γ·T0·exp(-κ·T), where δ is the calculated scattering signal, T is the total absorption coefficient over the entire X-ray scanning path, T0 is the absorption coefficient of each phantom, and κ and γ are model fitting parameters.

[0063] like Figure 5A As shown in FIG. 1 , signal 1 represents the real scattering signal obtained in step S102, and signal 2 represents the two-dimensional distribution of the normalized absorption coefficient (i.e., corresponding to the second projection data) of the scanned phantom under a certain projection (for the sake of simplicity and clarity, only the distribution of the absorption coefficient in a certain one-dimensional direction is shown). Signals 1 and 2 can be used to train the learning model. Figure 5A As shown, signal 3 represents the scattering signal that needs to be corrected given by signal 2 when the trained learning model is applied. It can be seen that the scattering signal that needs to be corrected given by the trained learning model for signal 2 is basically consistent with the real scattering signal.

[0064] In addition, if Figure 5B As shown, the calculated scattered signal generated by processing the signal 2 through the above physical model (ie, as shown in FIG. Figure 5B The signal 2') and the signal 1 shown are used as training data pairs to train the learning model to simplify the computational complexity.

[0065] Step S104: Apply the trained learning model to the X-ray projection data of the examination object to predict the scattered signal in the X-ray projection data. Specifically, the X-ray projection data of the examination object is input into the learning model to generate a prediction of the X-ray scattered signal in the projection space.

[0066] In the above-mentioned method according to the present application, projection data with scattering and projection data without scattering can be collected under multiple different conditions for various complex phantoms of different types (phantoms with different girths, heights, etc.) to provide sufficiently rich training data for training the learning model, so that the method of the present application can be applied to various systems, that is, system-independent, and can be applied to scattering prediction under various conditions.

[0067] The above-mentioned scatter correction method of the present application can also be used in combination with scatter mitigation technology such as anti-scatter grids.

[0068] Corresponding to Figure 1 The method shown, Figure 2 : The present invention shows an apparatus for predicting an X-ray scattering signal for an examination object according to an embodiment of the present application. The apparatus 2 includes: a first acquisition module 20, configured to scan each phantom in a scanning manner such that the scattering degree of each phantom in a plurality of phantoms is less than a reference scattering degree, thereby acquiring first projection data S1 of each phantom; a second acquisition module 21, configured to scan each phantom in a scanning manner such that the scattering degree of each phantom in a plurality of phantoms is equal to the reference scattering degree, thereby acquiring second projection data S2 of each phantom; a subtraction module 22, configured to obtain a true scattering signal Sr of each phantom by subtracting the first projection data S1 of each phantom from the second projection data S2 of each phantom; a training module 23, configured to train a learning model based on the second projection data S2 of each phantom and the true scattering signal Sr of each phantom, thereby obtaining a trained learning model; and a prediction module 24, configured to apply the trained learning model to the X-ray projection data for the examination object to predict the scattering signal in the X-ray projection data.

[0069] Preferably, before providing the second projection data S2 of each phantom to the training module 23, the second projection data S2 can be preprocessed by a preprocessing module (not shown) using a physical model. The physical model can be δ = γ·T0·exp(-κ·T), where δ is the calculated scattering signal, T is the total absorption coefficient along the entire X-ray scanning path, T0 is the absorption coefficient of each phantom, and κ and γ are model fitting parameters. The preprocessing module then provides the calculated scattering signal obtained through the physical model processing to the training module 23, so that the training module 23 uses the actual scattering signal of the phantom and the calculated scattering signal of the phantom as training data input and output pairs to train the learning model.

[0070] In addition, it can be understood that the first acquisition module 21 can perform the corresponding reference Figure 1 The second acquisition module 22 may perform the operation corresponding to the above reference step S100. Figure 1The operation of step S101 described above, the subtraction module 23 can perform the operation corresponding to the above reference Figure 1 The operation of step S102 described above, the training module 23 can perform the corresponding Figure 1 In the operation of step S103 described above, the prediction module 24 may specify the corresponding Figure 1 The operation of step S104 is described.

[0071] Figure 3 The present invention provides a method for correcting a scattered beam of X-rays for an inspection object according to an embodiment of the present application. In the method, steps S100 to S104 are Figure 1 The same operations are performed in steps S100 to S104, which will not be repeated here. In addition, the method further includes: step S105, using the predicted scattering signal to correct the X-ray projection data of the examination object to generate scattering-corrected projection data. In this step, the predicted scattering signal can be subtracted from the X-ray projection data of the examination object (for example, a part of the human body in clinical practice, such as the head, lungs, etc.) to generate scattering-corrected projection data.

[0072] Figure 4 The present invention provides a method for reconstructing an X-ray image of an inspection object according to an embodiment of the present application, wherein steps S100 to S104 are Figure 1 The same operation is performed in steps S100 to S104, and step S105 is the same as Figure 3 The same operation is performed in step S105, which will not be described in detail herein. In addition, the method further includes: step S106, reconstructing the scatter-corrected projection data to generate a reconstructed X-ray image after scatter-correction.

[0073] In addition, although the scatter prediction and correction involved in this application are applied to the original data in a certain projection direction, by Figure 4 The images constructed by the proposed method were compared with those obtained by applying the clinically validated algorithm used in current CT systems. It was verified at the image level that the attenuation drift caused by scattered radiation can be well predicted by the prediction method proposed in this application and that appropriate image quality can be obtained after removing the scattered signal. Figures 6A to 6E The method of this application was verified using a simple phantom (water phantom), a complex phantom (body phantom) and real clinical images (head scans). Figure 6A A simple water model image is shown, Figure 6B shows an anthropomorphic head phantom image, Figure 6C shows an image of an anthropomorphic lung phantom, Figure 6D shows a brain window scan image of a clinical head (real person), and Figure 6E Bone window scan images of a clinical head (real person) are shown, wherein the left image of each figure is obtained using the deep learning-based convolutional neural network assisted scatter correction algorithm of the present application, and the right image is obtained using the existing scatter correction algorithm.

[0074] As can be seen from the various figures, the images obtained by applying the algorithm proposed in this application have very similar image quality to the images obtained by applying the clinically verified algorithm used in current CT systems. The images obtained by the two methods are based on the same original projection data. Figure 6A In the case of a simple water model image, Figure 6A Comparing the left and right images, we can see that there is no difference in the accuracy and uniformity of the water values ​​of the images obtained by the two methods. Figure 6B In the case of the anthropomorphic head phantom image shown, Figure 6B Comparing the left and right images, we can see that there is no visible difference in the contrast, low contrast resolution, tissue morphology, resolution, and noise level of the images obtained by the two methods. Figure 6C In the case of the lung anthropomorphic phantom image shown, Figure 6C Comparison of the left and right images shows that the images obtained by the two methods have high contrast resolution and no visible difference in tissue morphology and structure. There is no visible additional artifact in the image reconstructed by the method of the present application. Figure 6D In the case of a clinical head (real person) brain window scan image, Figure 6D Comparing the left and right images, it can be seen that there is no significant difference in the contrast (gray-white matter contrast), low contrast resolution, noise level, and tissue morphology of the images obtained by the two methods, and there is no visible artifact in the image reconstructed by the method of the present application. Figure 6E In the case of a clinical head (real person) bone window scan image, Figure 6E Comparing the left and right images, it can be seen that there is no difference in the high-contrast resolution of the images obtained by the two methods (at the inner ear), the image morphological details are consistent, and there are no visible artifacts in the images reconstructed by the method of the present application.

[0075] It can be seen from the above-mentioned image-level verification that the X-ray images reconstructed by the method of the present application are close in quality to the images obtained by the clinically verified algorithms used in current CT systems. For example, there is no visible difference in image contrast, low-contrast resolution, tissue morphology, resolution, and noise level, and there are no visible artifacts. However, compared with existing algorithms, the method of the present application can simplify the computational complexity and does not rely on the simulation of system properties and the measurement of specified scattered signals from scanning representative phantoms under different scanning conditions. Therefore, the method of the present application has better robustness.

[0076] In addition, the above method of the present application can be implemented in the form of a program, wherein the program can be stored in a storage medium, and a computer can read the program from the storage medium and execute the program to implement the above method.

[0077] The technology proposed in this application has at least the following technical effects:

[0078] First, this paper proposes a novel data training method, i.e., both the input and output data of the learning model are obtained from actual system measurements. In comparison, most current studies train learning models based on simulated data.

[0079] 2. In this application, the learning method or convolutional neural network CNN may not be fully used as a "black box" for the entire scattering correction. For example, in the technology of this application, the theoretical scattered radiation can be calculated through a physical model during the data preprocessing process, so that the learning model such as CNN is only used to learn the distribution of the generated scattered radiation.

[0080] 3. The technology proposed in this application is not system-specific. The trained model is applicable to all systems with the same model type or configuration, so no adjustments are required during manufacturing.

[0081] 4. The technology of this application is purely based on raw projection data and does not require an image of the scanned object during the data processing process, so it has no impact on the structure of the current data processing pipeline, and all other calibration or correction steps remain the same.

[0082] 5. The technology of this application proposes a framework for using learning methods to perform scatter correction. It does not restrict the learning technology used. Machine learning and deep learning with different types of networks can be used in the framework proposed in this article.

[0083] Obviously, the embodiments described above are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments adopted by ordinary technicians in this field without making creative work should fall within the scope of protection of this application.

[0084] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, tasks, devices, components and / or combinations thereof.

[0085] It should be noted that the terms "first," "second," and the like in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects, and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, such that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0086] The above are merely preferred embodiments of the present application and are not intended to limit the present application. Those skilled in the art will readily appreciate that various modifications and variations are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A method for predicting a scattered signal of X-rays directed to an object under examination, characterized in that The method comprises: scanning each phantom in a scanning manner such that a scattering degree of each phantom among the plurality of phantoms is less than a reference scattering degree, to obtain first projection data of each phantom; scanning each phantom in a scanning manner such that the scattering degree of each phantom in the plurality of phantoms is equal to the reference scattering degree, to obtain second projection data of each phantom; obtaining a true scattering signal of each phantom by subtracting the first projection data of each phantom from the second projection data of each phantom; training a learning model based on the second projection data of each phantom and the true scattering signal of each phantom to obtain a trained learning model; and applying the trained learning model to X-ray projection data of an examination object to predict a scattered signal in the X-ray projection data; The step of training the learning model based on the second projection data of each phantom and the real scattering signal of each phantom to obtain the trained learning model includes: performing pre-calculation processing on the second projection data of each phantom based on the physical model to obtain a calculated scattering signal of each phantom; and The learning model is trained using the calculated scattering signal of each phantom and the true scattering signal of each phantom as training data pairs.

2. The method according to claim 1, characterized in that The learning model is trained by using the calculated scattering signal of each phantom as input training data of the learning model and using the real scattering signal of each phantom as output training data of the learning model.

3. The method according to claim 1, characterized in that Scanning each phantom in a scanning manner such that the scattering degree of each phantom among the plurality of phantoms is less than a reference scattering degree to obtain first projection data of each phantom includes: Segmentally scanning each of the plurality of phantoms with a first width smaller than a reference scan width under a plurality of different scanning conditions, so as to acquire projection data of each first segment of each phantom with the first width under each scanning condition, wherein the reference scan width is a width adopted in a clinical standard scan; and The first projection data of each phantom under each scanning condition is acquired by connecting the projection data of each first segment of each phantom acquired under each scanning condition.

4. The method according to claim 3, characterized in that Scanning each phantom in a scanning manner such that the scattering degree of each phantom in the plurality of phantoms is equal to the reference scattering degree to obtain second projection data of each phantom includes: Each of the plurality of phantoms is scanned with the reference scanning width under the plurality of different scanning conditions, so as to acquire the second projection data of each phantom under each scanning condition.

5. The method according to claim 4, characterized in that The first projection data of each phantom and the second projection data of each phantom are both distributions of absorption coefficients collected from each phantom.

6. The method according to claim 1 or 5, characterized in that The physical model is δ=γ·T0·exp(-κ·T), where δ is the calculated scattering signal, T is the total absorption coefficient over the entire X-ray scanning path, T0 is the absorption coefficient of each phantom, and κ and γ are model fitting parameters.

7. The method according to claim 1, characterized in that The learning models include machine learning and deep learning with different network types.

8. The method according to claim 3 or 4, characterized in that The plurality of different scanning conditions include at least one of the following: different phantom attenuation structures, different phantom positions, and different scanning powers.

9. A device for predicting a scattered signal of X-rays directed to an object under examination, characterized in that The device comprises: A first acquisition module scans each phantom in a scanning manner such that the scattering degree of each phantom among the plurality of phantoms is less than a reference scattering degree, thereby acquiring first projection data of each phantom; a second acquisition module, scanning each phantom in a scanning manner such that the scattering degree of each phantom among the plurality of phantoms is equal to the reference scattering degree, to acquire second projection data of each phantom; a subtraction module, which obtains a real scattering signal of each phantom by subtracting the first projection data of each phantom from the second projection data of each phantom; a training module, which trains a learning model based on the second projection data of each phantom and the real scattering signal of each phantom to obtain a trained learning model; and a prediction module, configured to apply the trained learning model to X-ray projection data of an examination object to predict a scattered signal in the X-ray projection data; The step of training the learning model based on the second projection data of each phantom and the real scattering signal of each phantom to obtain the trained learning model includes: performing pre-calculation processing on the second projection data of each phantom based on the physical model to obtain a calculated scattering signal of each phantom; and The learning model is trained using the calculated scattering signal of each phantom and the true scattering signal of each phantom as training data pairs.

10. A method for correcting a scattered beam of X-rays directed to an object under examination, characterized in that The method comprises: scanning each phantom in a scanning manner such that a scattering degree of each phantom among the plurality of phantoms is less than a reference scattering degree, to obtain first projection data of each phantom; scanning each phantom in a scanning manner such that the scattering degree of each phantom in the plurality of phantoms is equal to the reference scattering degree, to obtain second projection data of each phantom; obtaining a true scattering signal of each phantom by subtracting the first projection data of each phantom from the second projection data of each phantom; training a learning model based on the second projection data of each phantom and the true scattering signal of each phantom to obtain a trained learning model; applying the trained learning model to X-ray projection data of an examination object to predict a scattered signal in the X-ray projection data; and Correcting projection data of the X-rays for the examination object using the predicted scatter signal to generate scatter-corrected projection data; The step of training the learning model based on the second projection data of each phantom and the real scattering signal of each phantom to obtain the trained learning model includes: performing pre-calculation processing on the second projection data of each phantom based on the physical model to obtain a calculated scattering signal of each phantom; and The learning model is trained using the calculated scattering signal of each phantom and the true scattering signal of each phantom as training data pairs.

11. Method for reconstructing an X-ray image of an object under examination, characterized in that The method comprises: scanning each phantom in a scanning manner such that a scattering degree of each phantom among the plurality of phantoms is less than a reference scattering degree, to obtain first projection data of each phantom; scanning each phantom in a scanning manner such that the scattering degree of each phantom in the plurality of phantoms is equal to the reference scattering degree, to obtain second projection data of each phantom; obtaining a true scattering signal of each phantom by subtracting the first projection data of each phantom from the second projection data of each phantom; training a learning model based on the second projection data of each phantom and the true scattering signal of each phantom to obtain a trained learning model; applying the trained learning model to X-ray projection data of an examination object to predict a scattered signal in the X-ray projection data; Correcting projection data of the X-rays of the examination object using the predicted scatter signal to generate scatter-corrected projection data, and reconstructing the scatter-corrected projection data to generate a scatter-corrected reconstructed computed tomography image; The step of training the learning model based on the second projection data of each phantom and the real scattering signal of each phantom to obtain the trained learning model includes: performing pre-calculation processing on the second projection data of each phantom based on the physical model to obtain a calculated scattering signal of each phantom; and The learning model is trained using the calculated scattering signal of each phantom and the true scattering signal of each phantom as training data pairs. 12 . A computer storage medium storing a program which, when executed, causes a computer to perform the method according to claim 1 .

Images