Material decomposition in dual energy x-ray imaging

Abstract

In a computer-implemented method for material decomposition in dual energy X-ray imaging, a first X-ray image data set (21, 24) corresponding to a first X-ray radiation spectrum and a second X-ray image data set (22, 25) corresponding to a second X-ray radiation spectrum are obtained. At least one material-specific image dataset (13, 14) is generated by applying a decomposition module (10) comprising a first sequence of processing steps or a machine learning function to input data dependent on the first X-ray image dataset (21, 24) and the second X-ray image dataset (22, 25), a filtering module and / or an artifact reduction module is applied to the input data before the decomposition module (10) is applied.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method for material decomposition in dual-energy X-ray imaging, wherein first X-ray image data corresponding to a first X-ray radiation energy spectrum and second X-ray image data corresponding to a second X-ray radiation energy spectrum are obtained. The present invention also relates to a corresponding method for performing dual-energy X-ray imaging, as well as a computer-implemented training method, a data processing device, and a computer program product. Background Art

[0002] In dual-energy X-ray imaging (DE X-ray imaging), different X-ray image data, such as two different X-ray projection images or two different three-dimensional volume reconstructions, are generated by different X-ray radiation energy spectra, especially X-ray radiation energy spectra having different radiation energies or radiation powers. This particularly refers, on the one hand, to high-energy consumption or high-energy scans, and on the other hand, to low-energy consumption or low-energy scans. Different X-ray radiation energy spectra can be achieved by changing the operating parameters of the X-ray source, especially the operating voltage of the X-ray source and / or filtering in the beam path, such as filtering for beam hardening. Dual-energy X-ray imaging can be used to generate X-ray projection images, and can also be used for computed tomography (CT), i.e., the DE-CT method, or cone-beam computed tomography (CBCT), i.e., the DE-CBCT method.

[0003] Since the attenuation coefficients of different materials, such as cerebrospinal fluid (CSF), blood, and X-ray contrast agents, such as iodine- or barium-containing contrast agents, fluctuate to different degrees when the radiation energy changes, images or reconstructions can be generated from different X-ray image data according to different X-ray radiation energy spectra, in which materials, such as contrast agents, are particularly highlighted, and other images can also be generated, in which the materials are suppressed. This is the so-called material decomposition. For example, it can be checked whether obvious bleeding occurs during an intervention. In some clinical images, the contrast agent image is crucial for classifying the examination results.

[0004] Reference values for the attenuation of materials such as CSF or contrast agents in different X-ray radiation energy spectra can be preset for material decomposition. In combination with the actually measured attenuation values, material decomposition can basically be performed by using methods of linear algebra. Therefore, material decomposition provides material-specific image data. According to different objectives, applications, and materials to be separated, the material-specific image data can, for example, include a contrast image or a corresponding contrast reconstruction in which the contrast agent is highlighted. This similarly applies to other materials such as water and fat. Additionally or alternatively, the material-specific image data can also include a virtual non-contrast image (VNC image, English: virtual non-contrast image) or a corresponding VNC reconstruction in which the contrast agent is suppressed.

[0005] In principle, it is desirable to achieve as precise a material decomposition as possible so that not only can materials be qualitatively distinguished, but also quantitative analysis can be performed, for example, to determine the specific dosage of a contrast agent. The factors limiting image quality in CBCT include, in particular, a relatively low signal-to-noise ratio and / or contrast-to-noise ratio, scatter effects, cone-beam artifacts, truncation, etc.

[0006] Currently known heuristic or empirical methods can improve image quality, especially in CBCT methods, which also improves image quality in the results of material decomposition. However, this improved image quality mainly relates to the visual perception of human observers and cannot improve or only insufficiently improves the applicability for quantitative analysis.

[0007] Machine-trained algorithms can also be used to improve quality. However, due to the strong dependence of these algorithms on the training data used, so-called overfitting problems may occur in these algorithms, and thus these algorithms also cannot improve or can only limitedly improve the applicability for quantitative analysis. Summary of the Invention

[0008] The technical problem to be solved by the present invention is to improve the quality of material-specific image data in material decomposition in dual-energy X-ray imaging, thereby especially enabling quantitative material analysis or improved quantitative material analysis.

[0009] The above technical problem is solved by a computer-implemented method for material decomposition in dual-energy X-ray imaging, a method for dual-energy X-ray imaging, a computer-implemented training method, a data processing device having at least one computing unit, an X-ray imaging device, and a computer program product.

[0010] The concept underlying the present invention is to perform material decomposition by differentiable variables based essentially on linear algebra methods and by applying trained functions, especially functions trained by machine learning, the input data of which are X-ray image data sets corresponding to different X-ray radiation energy spectra.

[0011] According to one aspect of the present invention, a computer-implemented method for material decomposition in dual-energy X-ray imaging is disclosed. Here, a first X-ray image data set corresponding to a first X-ray radiation energy spectrum and a second X-ray image data set corresponding to a second X-ray radiation energy spectrum are obtained, the second X-ray radiation energy spectrum being different from the first X-ray radiation energy spectrum. By applying a decomposition module containing a first function to input data depending on the first X-ray image data set and the second X-ray image data set, at least one material-specific image data set is generated. Before applying the decomposition module, a filtering module or an artifact reduction module or a filtering module and an artifact reduction module are applied to the input data.

[0012] The present invention is based on the recognition that if the X-ray image data set is preprocessed by a filtering module and / or an artifact reduction module, the quality of material decomposition that can be achieved by the material decomposition module can be significantly improved.

[0013] The actual material decomposition is based on applying linear algebra methods to differentiable variables. For example, the differentiable variable can be represented by recording the attenuation value obtained according to the second X-ray radiation energy spectrum relative to the attenuation value obtained according to the first X-ray radiation energy spectrum, which is elaborated in detail, for example, in Figure 2 In this diagram, the differentiability of the variable is reflected in that the slope of the line connecting the plotted points reflects the attenuation value for a determined material.

[0014] The actual material decomposition algorithm is generated by the first function and can be integrated as an adaptable or adapted differentiable sequence of processing steps, or as a machine learning function, into any differentiable sequence of machine learning processing steps. The differentiable material decomposition can be embedded in a machine learning algorithm to achieve task-based machine learning. This improves the robustness of the learned algorithm for the task of material decomposition according to X-ray image data sets taken at different energy spectra. Thereby, it becomes possible to calculate the loss terms for particularly important materials. For example, the loss for neurological applications can be calculated based on an iodine map provided as a reference value.

[0015] Unless otherwise stated, all steps of the computer-implemented method can be performed by a data processing device having at least one computing unit. The at least one computing unit is in particular configured or adapted to perform the steps of the computer-implemented method. For this purpose, the at least one computing unit can, for example, store a computer program containing instructions that, when executed by the at least one computing unit, cause the at least one computing unit to perform the computer-implemented method.

[0016] However, by including the respective steps for generating a first X-ray image dataset and a second X-ray image dataset, in particular by means of an X-ray source and an X-ray detector, the non-purely computer-implemented embodiments of the method can be directly derived from each embodiment of the computer-implemented method.

[0017] The first X-ray image dataset and the second X-ray image dataset each represent the same region of the object to be imaged from the same angle. In the case of a C-arm X-ray device or a device for performing CT or CBCT, the scanning positions of the X-ray source and the X-ray detector relative to each other and relative to the object are in particular the same when generating the first X-ray image dataset and the second X-ray image dataset.

[0018] That the input data depends on the first X-ray image dataset and the second X-ray image dataset can in particular be understood as meaning that the input data includes the first X-ray image dataset and the second X-ray image dataset, or includes corresponding preprocessed variants of the first X-ray image dataset and the second X-ray image dataset. The preprocessing can for example include noise reduction, other artifact reduction, image filtering, etc.

[0019] In the context of dual-energy X-ray imaging, a material-specific image dataset can for example be understood as meaning that in the material-specific image dataset, in particular compared to the first X-ray image dataset and the second X-ray image dataset, the material to be determined is particularly highlighted or suppressed. Which material is involved depends on the first function, in particular the parameterization of the first function, which itself can also be trained due to the differentiability of the first function, or, if a trained function is involved, on the training data used to train the first function. Thus, a material-specific image dataset can highlight or suppress for example fat, water, blood, CSF, contrast agent, etc. It can also relate to VNC images.

[0020] The first function can for example be a function trained by using machine learning. The first function can for example be an artificial neural network (KNN), in particular a convolutional neural network (CNN, English: convolutional neural network), or a vision transformer network, or a combination of multiple artificial neural networks.

[0021] The first function can also include an optimized or optimizable function, for example a function optimized by means of a gradient-based method or capable of being optimized by means of a gradient-based method. For example, bilateral filtering can be derived and optimized according to the corresponding parameters.

[0022] The decomposition module can also be part of a more complex algorithm that can include other tasks in addition to the actual material decomposition, such as preprocessing, filtering, and / or three-dimensional reconstruction from multiple X-ray projection images. The first X-ray image dataset and the second X-ray image dataset can in particular be X-ray projection images or three-dimensional reconstructions respectively based on multiple X-ray projection images.

[0023] The first X-ray radiation energy spectrum and the second X-ray radiation energy spectrum are in particular different with respect to their radiation energy, i.e., photon energy or spectral radiation energy distribution. The average radiation energy of the first X-ray radiation energy spectrum is in particular higher than the average radiation energy of the second X-ray radiation energy spectrum. Different X-ray radiation energy spectra can in particular be achieved by different voltages, in particular peak voltages, and the X-ray source is operated at different voltages to generate the corresponding X-rays. Optionally, filtering can additionally be used in the beam path, for example, to perform beam intensification in an X-ray radiation energy spectrum with a higher average radiation energy.

[0024] With the method according to the invention, high-quality automatic material decomposition can be achieved in dual-energy X-ray imaging, so that in particular in the case where the image quality of the X-ray image dataset is not ideal, for example, in the case of CBCT, quantitative analysis of the material-specific image dataset can still be performed.

[0025] According to at least one embodiment, the first X-ray image dataset corresponds to a first X-ray projection image, and the second X-ray image dataset corresponds to a second X-ray projection image.

[0026] According to at least one embodiment, the first X-ray image dataset corresponds to a first reconstructed volume, and the second X-ray image dataset corresponds to a second reconstructed volume.

[0027] The reconstructed volume can here be a three-dimensional reconstruction of the entire object or a local region of the object, in particular a three-dimensional layer of the object, based on X-ray projection images taken from different directions, for example, in a CT method, a CBCT method, or a tomography method.

[0028] According to at least one embodiment, at least one material-specific image dataset includes a contrast agent image dataset.

[0029] In other words, the first X-ray image dataset and the second set of X-ray image datasets are generated after the use of a contrast agent, so that the contrast agent is present in the object. The contrast agent can be an iodine-containing contrast agent. The contrast image dataset is sometimes also referred to as an iodine image. The contrast agent can also be a barium-containing contrast agent or other X-ray positive or X-ray negative contrast agents. This is particularly advantageous if the accumulation of the contrast agent in a region of the object is of particular interest.

[0030] According to at least one embodiment, at least one material-specific image data set includes a non-contrast image data set, in particular a VNC image data set, i.e., for example, a VNC image or a three-dimensional VNC reconstruction.

[0031] In other words, the first X-ray image data set and the second X-ray image data set are also generated here after the use of a contrast agent. In the non-contrast image data set, the contrast agent is suppressed, i.e., an image without a contrast agent is simulated. This is particularly advantageous if the contrast agent is present in the form of an actually unwanted or no longer desired residue.

[0032] According to at least one embodiment, an artifact reduction module is used before the decomposition module. By applying the artifact reduction module that includes at least one trained second function to other input data, an artifact-reduced first X-ray image data set and an artifact-reduced second X-ray image data set are generated, and the other input data depends on, for example, includes the first X-ray image data set and the second X-ray image data set. The input data depends on, for example, includes the artifact-reduced first X-ray image data set and the artifact-reduced second X-ray image data set.

[0033] At least one second function can be, for example, at least one function trained using machine learning. At least one second function can include, for example, KNN, in particular CNN or a vision transformer network, or a combination of multiple KNNs.

[0034] According to different embodiments, one trained second function of at least one trained second function, for example, a trained KNN, can be applied to the first X-ray image data set and the second X-ray image data set to generate an artifact-reduced first X-ray image data set and an artifact-reduced second X-ray image data set.

[0035] In other embodiments, one trained second function of at least one trained second function can be applied to the first X-ray image data set to generate an artifact-reduced first X-ray image data set, and another trained second function of at least one trained second function can be applied to the second X-ray image data set to generate an artifact-reduced second X-ray image data set.

[0036] At least one trained second function is particularly trained to reduce artifacts in the X-ray image data set. Artifacts may include noise or other artifacts, such as artifacts due to scattered radiation or truncation, cone beam artifacts, etc.

[0037] Thereby, the quality of the input data of the decomposition module can be improved, and correspondingly, the quality of at least one material-specific image data set can also be improved.

[0038] The second function may also include an optimized or optimizable function, such as a function optimized by means of a gradient-based method or capable of being optimized by means of a gradient-based method. For example, bilateral filtering can be derived and optimized therefrom according to the corresponding parameters.

[0039] The first function and the at least one second function can be trained jointly or separately from each other.

[0040] According to at least one embodiment, a filtering module is applied before the decomposition module. By applying a filtering module comprising at least one filtering function to other input data, a filtered first X-ray image data set and a filtered second X-ray image data set are produced, the other input data depending on the first X-ray image data set and the second X-ray image data set, for example comprising the first X-ray image data set and the second X-ray image data set. The input data depends on the filtered first X-ray image data set and the filtered second X-ray image data set, for example comprising the filtered first X-ray image data set and the filtered second X-ray image data set.

[0041] The at least one filtering function can, for example, be a function trained by using machine learning. The at least one second function can, for example, include KNN, in particular CNN or a vision transformer network, or a combination of multiple KNNs. However, the at least one filtering function can also be at least one conventional filtering function, i.e., a function not trained by using machine learning. In this case, the parameters of the at least one filtering function can be preset and / or changed manually or automatically according to the application.

[0042] The at least one filtering function can, for example, be used for contrast improvement, noise reduction, edge enhancement, etc.

[0043] According to different embodiments, one of the at least one filtering functions can be applied to the first X-ray image data set and the second X-ray image data set to produce a filtered first X-ray image data set and a filtered second X-ray image data set.

[0044] In other embodiments, one of the at least one filtering functions can be applied to the first X-ray image data set to produce a filtered first X-ray image data set, and another one of the at least one filtering functions can be applied to the second X-ray image data set to produce a filtered second X-ray image data set.

[0045] Thereby, the quality of the input data of the decomposition module can be improved, and correspondingly also the quality of at least one material-specific image data set.

[0046] The embodiments that specify the use of the filtering module can also be combined with the embodiments that specify the use of the artifact reduction module. In this case, the artifact reduction module and the filtering module in particular implement different noise reduction algorithms, or not both are used for noise reduction. Here, for example, the artifact reduction module can be applied to other first input data to generate an artifact-reduced first X-ray image dataset and an artifact-reduced second X-ray image dataset, where the other first input data depends on, in particular includes, the first X-ray image dataset and the second X-ray image dataset. Subsequently, the filtering module can be applied to other second input data to generate a filtered first X-ray image dataset and a filtered second X-ray image dataset, where the other second input data depends on, in particular includes, the artifact-reduced first X-ray image dataset and the artifact-reduced second X-ray image dataset. It can also be done vice versa, i.e., first apply the filtering module and then the artifact reduction module.

[0047] According to at least one embodiment, at least one filtering function includes at least one filtering function for bilateral filtering and / or at least one filtering function for differentiable guided filtering.

[0048] According to another aspect of the invention, a method for dual-energy X-ray imaging is provided. Here, a first X-ray image dataset representing an object to be imaged is generated, in particular by generating a first X-ray corresponding to a first X-ray radiation energy spectrum by means of an X-ray source and detecting, in particular by means of an X-ray detector, the fraction of the first X-ray passing through the object. A second X-ray image dataset representing the object is generated, in particular by generating a second X-ray corresponding to a second X-ray radiation energy spectrum by means of an X-ray source and detecting, in particular by means of an X-ray detector, the fraction of the first X-ray passing through the object. A computer-implemented method for material decomposition according to the invention is performed by using the first X-ray image dataset and the second X-ray image dataset.

[0049] According to at least one embodiment of the method for dual-energy X-ray imaging, the method is carried out as a CT method.

[0050] According to at least one embodiment of the method for dual-energy X-ray imaging, the method is carried out as a CBCT method.

[0051] Due to the generally low image quality of the CBCT method, the present invention has particularly advantageous effects here.

[0052] According to at least one embodiment, the first X-ray image dataset corresponds to a first X-ray projection image, and the second X-ray image dataset corresponds to a second X-ray projection image. At least one reconstructed volume is generated based on at least one material-specific image dataset.

[0053] Known reconstruction methods can be used to generate a reconstructed volume based on at least one material-specific image dataset. In particular, multiple material-specific image datasets from different viewing directions are generated, and a reconstructed volume is generated based on the multiple material-specific image datasets.

[0054] According to at least one embodiment, the first X-ray image dataset corresponds to a first reconstructed volume, and the second X-ray image dataset corresponds to a second reconstructed volume.

[0055] The first reconstructed volume and the second reconstructed volume can be generated by using known reconstruction methods. In particular, multiple X-ray projection images from different viewing directions are generated for the first and second X-ray radiation spectra respectively, and corresponding reconstructed volumes are generated based on the respective multiple X-ray projection images.

[0056] According to another aspect of the present invention, a computer-implemented training method is provided. The training method is used to provide a filtering module and / or an artifact reduction module and a trained first function for a computer-implemented method according to the present invention for material decomposition in dual-energy X-ray imaging. A first X-ray training image dataset corresponding to a first X-ray radiation spectrum and a second X-ray training image dataset corresponding to a second X-ray radiation spectrum are obtained. At least one material-specific ground-truth image dataset for the first X-ray training image dataset and the second X-ray training image dataset is obtained. By applying the artifact reduction module or the filtering-and-artifact reduction module to the input training data depending on the first X-ray training image dataset and the second X-ray training image dataset and then applying a decomposition module including an untrained or partially trained first function, at least one predicted material-specific image dataset is generated, especially predicted.

[0057] A predetermined loss function is analyzed, which depends on at least one predicted material-specific image dataset and at least one material-specific ground-truth image dataset, especially on the deviation between at least one predicted material-specific image dataset and at least one material-specific ground-truth image dataset. According to the result of the analysis of the loss function, especially the value of the loss function, the parameters of the first function are updated.

[0058] The first function is in particular a function trained by using machine learning, such as a KNN or a combination of multiple KNNs. The parameters of the first function include, for example, a weighting factor and / or a bias factor of one KNN or multiple KNNs.

[0059] In particular, the steps of the training method can be repeated for multiple first and second X-ray training image datasets and the correspondingly assigned ground truth image datasets until a preset interruption criterion or convergence criterion of the loss function is met.

[0060] If at least one predicted material-specific image dataset includes multiple predicted material-specific image datasets, then at least one material-specific ground truth image dataset can include a material-specific ground truth image dataset for each predicted material-specific image dataset. The loss function can include, for example, a respective loss term for each pair of a predicted material-specific image dataset and the correspondingly assigned material-specific ground truth image dataset.

[0061] According to at least one embodiment of the computer-implemented training method, the training method is designed to provide a training method for a trained first function and at least one trained second function. By applying an artifact reduction module comprising at least one untrained or partially trained second function to other input training data, a predicted artifact-reduced first X-ray image dataset and a predicted artifact-reduced second X-ray image dataset are generated, the other input training data depending on, in particular comprising, the first X-ray training image dataset and the second X-ray training image dataset. The input training data depends on, in particular comprises, the predicted artifact-reduced first X-ray image dataset and the predicted artifact-reduced second X-ray image dataset.

[0062] According to at least one embodiment, the parameters of at least one second function are updated based on the result of the analysis of the loss function.

[0063] In these embodiments, the first function and at least one second function are end-to-end trained, which requires less training data and enables the two tasks of artifact reduction and material decomposition to be trained in an optimally coordinated manner.

[0064] At least one second function is in particular at least one function trainable by using machine learning, such as a KNN or a combination of multiple KNNs. The parameters of at least one second function include, for example, a weighting factor and / or a bias factor of one KNN or multiple KNNs.

[0065] According to at least one embodiment, a first ground truth X-ray image dataset with reduced artifacts is obtained for a first X-ray training image dataset, and a second ground truth X-ray image dataset with reduced artifacts is obtained for a second X-ray training image dataset. A preset other loss function is analyzed, the other loss function including a loss term and other loss terms, the loss term depending on the predicted first X-ray image dataset with reduced artifacts and the first ground truth X-ray image dataset with reduced artifacts, in particular depending on the corresponding deviation, and the other loss terms depending on the predicted second X-ray image dataset with reduced artifacts and the second ground truth X-ray image dataset with reduced artifacts, in particular depending on the corresponding deviation. The parameters of at least one second function are updated according to the result of the analysis of the other loss function.

[0066] That is, at least one second function is trained independently of the first function. This is advantageous if, for example, at least one second function is to be provided for different applications.

[0067] According to at least one embodiment of a computer-implemented training method, the training method is designed to provide a training first function and a training method for at least one filtering function. By applying a filtering module including at least one filtering function to other input training data, a predicted filtered first X-ray image dataset and a predicted filtered second X-ray image dataset are generated, the other input training data depending on, in particular including, the first X-ray training image dataset and the second X-ray training image dataset. The input training data depends on, in particular includes, the predicted filtered first X-ray image dataset and the predicted filtered second X-ray image dataset.

[0068] According to at least one embodiment, at least one parameter of at least one filtering function is updated according to the result of the analysis of the loss function.

[0069] In these embodiments, the first function and at least one filtering function are trained end-to-end, which requires less training data and enables the two tasks of filtering and material decomposition to be trained in an optimally coordinated manner.

[0070] At least one filtering function can in particular be at least one function trainable using machine learning, such as a KNN or a combination of multiple KNNs. The parameters of at least one filtering function include, for example, a weighting factor and / or a bias factor of one KNN or multiple KNNs.

[0071] However, at least one filtering function can also include traditional filtering algorithms, such as bilateral filtering. The parameters of at least one filtering function include, for example, the kernel size of a first filter kernel and / or the kernel size of a second filter kernel, such as the standard deviation in the case of a Gaussian filter kernel, and / or the window size of the filter.

[0072] According to at least one embodiment, a filtered first ground truth X-ray image dataset is obtained for a first X-ray training image dataset, and a filtered second ground truth X-ray image dataset is obtained for a second X-ray training image dataset. A preset other loss function is analyzed, the other loss function including a loss term and other loss terms, the loss term depending on a predicted filtered first X-ray image dataset and the filtered first ground truth X-ray image dataset, in particular depending on the corresponding deviation, and the other loss terms depending on a predicted filtered second X-ray image dataset and the filtered second ground truth X-ray image dataset, in particular depending on the corresponding deviation. The parameters of at least one filtering function are updated according to the result of the analysis of the other loss function.

[0073] That is, at least one filtering function is trained independently of the first function. This is advantageous if, for example, at least one filtering function is to be provided for different applications.

[0074] According to another aspect of the invention, a data processing device is provided. The data processing device has at least one computing unit which is adapted to execute a computer-implemented method for material decomposition in dual-energy X-ray imaging according to the invention and / or a computer-implemented training method according to the invention.

[0075] The computing unit can in particular be understood as a data processing device comprising a processing circuit. The computing unit can in particular process data for performing computational operations. The computational operations can also include, if necessary, operations for performing index access to data structures such as look-up tables LUT (English: “look-up table”).

[0076] The computing unit can in particular include one or more computers, one or more microcontrollers and / or one or more integrated circuits, such as one or more application-specific integrated circuits, i.e., ASIC (English: “application-specific integrated circuit”), and / or one or more field-programmable gate arrays, i.e., FPGA, and / or one or more systems on a chip, i.e., SOC (English: “system on a chip”). The computing unit can also include one or more processors, such as one or more microprocessors, one or more central processing units CPU (English: “central processing unit”), one or more graphics processing units GPU (English: “graphics processing unit”), and / or one or more signal processors, in particular one or more digital signal processors DSP. The computing unit can also include a physical or virtual complex of a computer or other units mentioned.

[0077] In different embodiments, the computing unit includes one or more hardware and / or software interfaces and / or one or more storage units.

[0078] The storage unit can be designed as a volatile data memory, such as a dynamic random access memory DRAM (English: “dynamic random access memory”) or a static random access memory SRAM (English: “static random access memory”), or as a non-volatile data memory, such as a read-only memory ROM (English: “read-only memory”), a programmable read-only memory PROM (English: “programmable read-only memory”), an erasable programmable read-only memory EPROM (English: “erasable programmable read-only memory”), an electrically erasable programmable read-only memory EEPROM (English: “electrically erasable programmable read-only memory”), a flash memory or a flash EEPROM, a ferroelectric random access memory FRAM (English: “ferroelectric random access memory”), a magnetoresistive random access memory MRAM (English: “magnetoresistive random access memory”) or a phase-change random access memory PCRAM (English: “phase-change random access memory”).

[0079] According to another aspect of the present invention, an X-ray imaging device is provided, which includes a data processing device, an X-ray source, an X-ray detector and at least one control unit according to the present invention. The at least one control unit is configured to control the X-ray source to generate a first X-ray corresponding to a first X-ray radiation energy spectrum. The X-ray detector is configured to generate a first X-ray image data set representing an object to be imaged and, for this purpose, to detect the fraction of the first X-ray passing through the object. The at least one control unit is configured to control the X-ray source to generate a second X-ray corresponding to a second X-ray radiation energy spectrum. The X-ray detector is configured to generate a second X-ray image data set representing the object and, for this purpose, to detect the fraction of the second X-ray passing through the object.

[0080] The generation of the first and second X-ray image datasets by the X-ray detector can be understood in particular as the X-ray detector generating a first X-ray projection image as the first X-ray image dataset and generating a second X-ray projection image as the second X-ray image dataset. If the X-ray image dataset is a reconstructed volume, the X-ray detector generates the corresponding X-ray projection images and at least one computing unit generates the X-ray image dataset.

[0081] At least one control unit can be, for example, part of at least one computing unit of a data processing device.

[0082] Other embodiments of the X-ray imaging device according to the invention are directly derived from the design of the computer-implemented method according to the invention and the design of the computer-implemented training method according to the invention, and vice versa. The individual features and the corresponding explanations and advantages related to the different embodiments of the method according to the invention and the training method according to the invention can be transferred analogously, in particular, to the corresponding embodiments of the X-ray imaging device according to the invention. The X-ray imaging device according to the invention is in particular designed or programmed to execute the computer-implemented method according to the invention. The X-ray imaging device according to the invention in particular executes the computer-implemented method according to the invention.

[0083] According to another aspect of the invention, there is provided a computer program having instructions. When the instructions are executed by a data processing device, the instructions cause the data processing device to execute the computer-implemented method for material decomposition according to the invention and / or the computer-implemented training method according to the invention.

[0084] The instructions can exist, for example, as program code. The program code can be provided, for example, in the form of binary code or assembly program and / or source code of a programming language such as C language and / or program script such as Python.

[0085] According to another aspect of the invention, there is provided another computer program having other instructions. When the other instructions are executed by the X-ray imaging device according to the invention, in particular by the data processing device of the X-ray imaging device, the other instructions cause the X-ray imaging device to execute the method according to the invention for dual-energy X-ray imaging.

[0086] The other instructions can exist, for example, as program code. The program code can be provided, for example, in the form of binary code or assembly program and / or source code of a programming language such as C language and / or program script such as Python.

[0087] According to another aspect of the invention, there is provided a computer-readable storage medium that stores the computer program according to the invention and / or the other computer program according to the invention.

[0088] A computer program, other computer programs, and computer-readable storage media are computer program products having instructions and / or other instructions respectively.

[0089] Other features and combinations of features of the present invention are derived from the drawings, the description of the drawings, and the claims. In particular, other embodiments of the present invention do not necessarily include all features of one of the claims. Other embodiments of the present invention may have features or combinations of features not mentioned in the claims. Description of the Drawings

[0090] The present invention will be explained in more detail below with the aid of specific embodiments and the associated schematic diagrams. In the drawings, identical or functionally identical elements may be provided with the same reference signs. Identical or functionally identical elements are not necessarily repeatedly described for different figures when necessary.

[0091] In the drawings:

[0092] Figure 1 A schematic diagram showing an exemplary embodiment of an X-ray imaging device according to the present invention is shown;

[0093] Figure 2 A schematic principle diagram showing material decomposition in dual-energy X-ray imaging is shown;

[0094] Figure 3 A schematic diagram showing an exemplary embodiment of a computer-implemented method for material decomposition or a corresponding computer-implemented training method according to the present invention is shown;

[0095] Figure 4 A schematic diagram showing another exemplary embodiment of a computer-implemented method for material decomposition or a corresponding computer-implemented training method according to the present invention is shown;

[0096] Figure 5 A schematic diagram showing another exemplary embodiment of a computer-implemented method for material decomposition or a corresponding computer-implemented training method according to the present invention is shown;

[0097] Figure 6 A schematic diagram showing another exemplary embodiment of a computer-implemented method for material decomposition or a corresponding computer-implemented training method according to the present invention is shown;

[0098] Figure 7 A schematic diagram showing another exemplary embodiment of a computer-implemented method for material decomposition or a corresponding computer-implemented training method according to the present invention is shown;

[0099] Figure 8Schematic diagrams showing other exemplary embodiments of a computer-implemented method for material decomposition or a corresponding computer-implemented training method according to the present invention.

[0100] Figure 9 Schematic diagrams showing other exemplary embodiments of a computer-implemented method for material decomposition or a corresponding computer-implemented training method according to the present invention.

[0101] Figure 10 Schematic flowchart showing an exemplary embodiment of a method for dual-energy X-ray imaging according to the present invention;

[0102] Figure 11 Schematic flowchart showing another exemplary embodiment of a method for dual-energy X-ray imaging according to the present invention. Detailed embodiments

[0103] Figure 1 Schematic diagrams showing exemplary embodiments of an X-ray imaging device, for example for CBCT, according to the present invention.

[0104] The X-ray imaging device has a data processing device 2, an X-ray source 4, an X-ray detector 3 and at least one control unit according to the present invention. The at least one control unit can be, for example, part of the data processing device 2. The at least one control unit is arranged to control the X-ray source 4 to generate a first X-ray corresponding to a first X-ray radiation spectrum. The X-ray detector 3 is arranged to generate a first X-ray projection image representing the object 5 to be imaged and to detect the fraction of the first X-ray passing through the object 5. The at least one control unit is arranged to control the X-ray source 4 to generate a second X-ray corresponding to a second X-ray radiation spectrum. The X-ray detector is arranged to generate a second X-ray projection image representing the object 5 and to detect the fraction of the second X-ray passing through the object 5.

[0105] The data processing device 2 has at least one computing unit which is arranged to execute a computer-implemented method for material decomposition according to the present invention and a corresponding computer-implemented training method according to the present invention.

[0106] The data processing device 2 hereby uses the first X-ray projection image as the first X-ray image data set and the second X-ray projection image as the second X-ray image data set, or generates a first volume reconstruction 11, 21 as the first X-ray image data set based on the first X-ray projection image and generates a second volume reconstruction 12, 22 as the second X-ray image data set based on the second X-ray projection image.

[0107] Figure 2Shows a schematic principle diagram of three-material decomposition in dual-energy X-ray imaging by CT method or CBCT method.

[0108] The attenuation values according to the first X-ray radiation energy spectrum are plotted on the horizontal axis, for example corresponding to the tube voltage of the X-ray source 4 at 125 kV, and the attenuation values according to the second X-ray radiation energy spectrum are plotted on the vertical axis, for example corresponding to the tube voltage of the X-ray source 4 at 70 kV. The attenuation values are given in particular in Hounsfield-Einheit (HE). Each voxel of the three-dimensional volume reconstruction can then be assigned two-dimensional coordinates in the Figure 2 view. In Figure 2 , the point P1 corresponds to the measured attenuation of the voxel, the point P2 corresponds to the attenuation of the first reference material, such as CSF, and the point P3 corresponds to the attenuation of the second reference material, such as blood. Figure 2 The attenuation of iodine as the third reference material is not shown in

[0109] because the value of the corresponding point P4 is much higher.

[0110] Figure 3 Shows a schematic diagram of an exemplary embodiment of a computer-implemented method for material decomposition according to the invention or a corresponding computer-implemented training method.

[0111] The decomposition module 10 containing the first function is applied to the filtered or artifact-reduced or filtered and artifact-reduced reconstructed first volume reconstruction 11 and the filtered or artifact-reduced or filtered and artifact-reduced reconstructed second volume reconstruction 12, and on this basis provides at least one material-specific volume reconstruction 13, 14, such as an iodine reconstruction 14 and a VNC reconstruction 13. In the application case, the first function has already been trained, and the computer-implemented method thus ends, for example.

[0112] During the training of the first function, a loss function is analyzed based on at least one material-specific volume reconstruction 13, 14 and at least one material-specific true volume reconstruction 15, 16, and the parameters of the first function are updated by an optimization module 17 according to the result of the analysis of the loss function.

[0113] In particular, a first loss term L1 is analyzed based on the VNC reconstruction 13 and the corresponding assigned true VNC reconstruction 15, and a second loss term L2 is analyzed based on the iodine reconstruction 14 and the corresponding assigned true iodine reconstruction 16. The loss terms L1, L2 can be, for example, known loss terms or based on known loss terms such as mean variance, mean absolute error, mean absolute percentage error, structural similarity index, histogram-based loss, etc. The loss function can be, for example, the sum or weighted sum of the loss terms L1, L2.

[0114] The optimization module 17 can use known methods, such as backpropagation, to update the parameters of the first function.

[0115] The method can be carried out similarly based on the corresponding X-ray projection images instead of the volume reconstructions 11 and 12. At least one material-specific X-ray projection image is obtained instead of at least one material-specific volume reconstruction 13, 14.

[0116] Figure 4 Shows a schematic diagram of other exemplary embodiments of an implementation based on Figure 3 of a computer-implemented method for material decomposition or a corresponding computer-implemented training method according to the present invention.

[0117] The volume reconstructions 11, 12 here are artifact-reduced volume reconstructions 11, 12. An artifact reduction module including a second function 18, such as a second KNN, is applied to the original volume reconstructions 21 and 22 corresponding to the first and second X-ray radiation energy spectra and provides the artifact-reduced volume reconstructions 11, 12. The remaining steps correspond to the steps described with reference to Figure 3 In the application scenario, the second function has been trained, and the computer-implemented method thus ends, for example.

[0118] During the training of the second function, a preset other loss function is analyzed. The other loss function includes a loss term L3 and other loss terms L4. The loss term L3 depends on the artifact-reduced first volume reconstruction 11 and the artifact-reduced first true volume reconstruction 19, and the other loss terms L4 depend on the predicted artifact-reduced second volume reconstruction 12 and the artifact-reduced second true volume reconstruction 20. The parameters of the second function 18 are updated by the optimization module 17 according to the result of the analysis of the other loss function.

[0119] The loss terms L3 and L4 can be, for example, known loss terms or based on known loss terms such as mean variance, mean absolute error, mean absolute percentage error, structural similarity index, histogram-based loss, etc. Other loss functions can be, for example, the sum or weighted sum of the loss terms L3 and L4.

[0120] The optimization module 17 can use known methods, such as backpropagation, to update the parameters of the first function. The optimization module 17 can in particular train the first function and the second function 18 separately from each other.

[0121] In some embodiments, the second function 18 is replaced by a second function 18a and other second functions 18b. The second function 18a is applied to the original volume reconstruction 21 and provides an artifact-reduced volume reconstruction 11. The other second functions 18b are applied to the original volume reconstruction 22 and provide an artifact-reduced volume reconstruction 12. This method is schematically shown in Figure 6 shown.

[0122] In an alternative embodiment, the other loss function is discarded, and the first function and the second function 18 are jointly trained end-to-end based on a loss function having loss terms L1 and L2. Thus, the ground truth volume reconstructions 19 and 20 are not required. As Figure 5 schematically shown, these embodiments can also be combined with embodiments in which the second function 18 is replaced by second functions 18a and 18b.

[0123] The method can be similarly carried out for the replacement of the reconstructed volumes 11 and 12 based on the corresponding X-ray projection images. Replacing at least one material-specific volume reconstruction 13 and 14 results in at least one material-specific X-ray projection image.

[0124] Figure 7 shows a schematic diagram of other exemplary embodiments of a computer-implemented method for material decomposition according to the invention or a corresponding computer-implemented training method based on Figure 4 of the embodiment. In Figure 7 of the embodiment, the second function 18 is replaced by a common filtering function 23, such as a common bilateral filter. The filtering function 23 does not have to be designed as a KNN, however, the parameters of the filtering function, such as the filter kernel size and / or window size, can also be trained separately from the first function based on the loss terms L3 and L4 or end-to-end with the first function.

[0125] Figure 8 shows a schematic diagram of other exemplary embodiments of a computer-implemented method for material decomposition according to the invention or a corresponding computer-implemented training method based on Figure 7Schematic diagrams of other exemplary embodiments of the embodiments. The common filtering function 23 is replaced here by two separate filtering functions 23a, 23b, which are applied to the original volume reconstruction 21 or the original volume reconstruction 22. As referred to Figure 5 as described, the training of the filtering functions 23a and 23b is carried out end-to-end with the first function.

[0126] Figure 9 shows a schematic diagram of other exemplary embodiments of a computer-implemented method for material decomposition or a corresponding computer-implemented training method according to the present invention, based on Figure 7 Schematic diagrams of other exemplary embodiments of the embodiments. The common filtering function 23 is replaced here by two separate filtering functions 23a, 23b, which are applied to the original volume reconstruction 21 or the original volume reconstruction 22. As referred to Figure 6 as described, the training of the filtering functions 23a and 23b is carried out separately from the first function.

[0127] Figure 10 shows a schematic flow chart of an exemplary embodiment of a dual-energy X-ray imaging method according to the present invention, which uses a computer-implemented method for material decomposition according to the present invention.

[0128] Here, a plurality of first X-ray projection images 24 are generated for the first X-ray radiation energy spectrum. Based on this, a plurality of artifact-reduced and / or filtered first X-ray projection images 26 are generated. Based on this, a first volume reconstruction 21 is generated and further processed as described above with reference to Figures 4 to 9 described.

[0129] A plurality of second X-ray projection images 25 are generated for the second X-ray radiation energy spectrum. Based on this, a plurality of artifact-reduced and / or filtered second X-ray projection images 27 are generated. Based on this, a second volume reconstruction 22 is generated and further processed as described above with reference to Figures 4 to 9 described.

[0130] Figure 11 shows a schematic flow chart of another exemplary embodiment of a dual-energy X-ray imaging method according to the present invention, which uses a computer-implemented method for material decomposition according to the present invention.

[0131] Here, a plurality of first X-ray projection images 24 are generated for the first X-ray radiation energy spectrum. Based on this, a plurality of artifact-reduced and / or filtered first X-ray projection images 26 are generated. Based on this, a first volume reconstruction 11 is generated and further processed as described above with reference to Figure 3 described.

[0132] Multiple second X-ray projection images 25 are generated for a second X-ray energy spectrum. Based thereon, multiple artifact-reduced and / or filtered second X-ray projection images 27 are generated. Based thereon, a second volume reconstruction 12 is generated and further processed as described above with reference to Figure 3 as set forth.

[0133] As described above, particularly with reference to the drawings, the present invention achieves an improvement in the quality of material-specific image data during material decomposition in dual-energy X-ray imaging, thereby particularly enabling improved quantitative material analysis.

Claims

1. A computer-implemented method for material decomposition in dual energy X-ray imaging, wherein: - obtaining a first X-ray image data set (21, 24) corresponding to a first X-ray radiation energy spectrum and a second X-ray image data set (22, 25) corresponding to a second X-ray radiation energy spectrum; - generating at least one material-specific image data set (13, 14) by applying a decomposition module (10) comprising a first sequence of processing steps or a machine learning function to input data depending on a first X-ray image data set (21, 24) and a second X-ray image data set (22, 25), wherein a filtering module or an artifact reduction module or both a filtering module and an artifact reduction module are applied to the input data before applying the decomposition module (10).

2. The computer-implemented method of claim 1, wherein: - the first X-ray image data set (21, 24) corresponds to the first X-ray projection image and the second X-ray image data set (22, 25) corresponds to the second X-ray projection image; or The first X-ray image data set (21, 24) corresponds to a first reconstruction volume and the second X-ray image data set (22, 25) corresponds to a second reconstruction volume.

3. The computer-implemented method according to claim 1 , wherein: At least one material-specific image data set (13) comprises a contrast medium image data set (14) and / or a virtual non-contrast image data set (13).

4. The computer-implemented method according to any one of the preceding claims, characterized in that - generating an artifact-reduced first X-ray image dataset (11) and an artifact-reduced second X-ray image dataset (12) by applying an artifact reduction module comprising at least one trained second function (18, 18a, 18b) to further input data, the further input data being dependent on the first X-ray image dataset (21, 24) and the second X-ray image dataset (22, 25), and - the input data depend on a first artifact-reduced X-ray image data set (11) and a second artifact-reduced X-ray image data set (12).

5. The computer-implemented method according to any one of claims 1 to 3, characterized in that - generating a filtered first X-ray image data set and a filtered second X-ray image data set (22, 25) by applying a filtering module comprising at least one filtering function (23, 23a, 23b) to further input data, the further input data being dependent on the first X-ray image data set (21, 24) and the second X-ray image data set (22, 25); and - The input data depends on the filtered first X-ray image data set and the filtered second X-ray image data set.

6. The computer-implemented method of claim 5, wherein: The at least one filter function (23, 23a, 23b) comprises at least one filter function (23, 23a, 23b) for bilateral filtering and / or at least one filter function (23, 23a, 23b) for differentiable guided filtering.

7. A method for dual energy X-ray imaging, wherein: - generating a first X-ray image data set (21, 24) representing an object (5) to be imaged by generating first X-rays corresponding to a first X-ray radiation energy spectrum and detecting a portion of the first X-rays that passes through the object (5); - generating a second X-ray image data set (22, 25) representing the object (5) by generating second X-rays corresponding to a second X-ray radiation energy spectrum and detecting the portion of the first X-rays that passes through the object (5); - performing a computer-implemented method according to one of the preceding claims.

8. The method according to claim 7, wherein: The method is performed as a computed tomography method or a cone-beam computed tomography method, and - the first X-ray image data set (21, 24) corresponds to the first X-ray projection image, the second X-ray image data set (22, 25) corresponds to the second X-ray projection image, at least one reconstructed volume is generated from at least one material-specific image data set (13, 14); or The first X-ray image data set (21, 24) corresponds to a first reconstruction volume and the second X-ray image data set (22, 25) corresponds to a second reconstruction volume.

9. A computer-implemented training method for providing a trained first function for use in a computer-implemented method according to one of claims 1 to 6, wherein - obtaining a first X-ray training image data set corresponding to the first X-ray radiation energy spectrum and a second X-ray training image data set corresponding to the second X-ray radiation energy spectrum; - obtaining at least one material-specific real image data set (15, 16) for the first X-ray training image data set and the second X-ray training image data set; - generating at least one predicted material-specific image data set (13, 14) by applying a decomposition module (10) comprising a first sequence of processing steps or a machine learning function to input training data, said input training data being dependent on a first X-ray training image data set and a second X-ray training image data set; - wherein a filtering module or an artifact reduction module or a filtering module and an artifact reduction module are applied to the input data before applying the decomposition module (10); - analyzing a predefined loss function (L1, L2), which is dependent on at least one material-specific predicted image data set (13, 14) and at least one material-specific real image data set (15, 16); and - Update the parameters of the first function according to the results of the analysis of the loss functions (L1, L2).

10. A computer-implemented training method for providing a trained first function and at least one trained second function (18, 18a, 18b) for use in the computer-implemented method according to claim 4, wherein - performing a computer-implemented training method according to claim 9; - generating a predicted artifact-reduced first X-ray image data set (21, 24) and a predicted artifact-reduced second X-ray image data set (22, 25) by applying an artifact reduction module comprising at least one untrained or partially trained second function (18, 18a, 18b) to further input training data, the further input training data being dependent on the first X-ray training image data set and the second X-ray training image data set, and - the input training data is dependent on a predicted artifact-reduced first X-ray image data set (21, 24) and a predicted artifact-reduced second X-ray image data set (22, 25); and - Updating the parameters of at least one second function (18, 18a, 18b) according to the results of the analysis of the loss function (L1, L2).

11. A computer-implemented training method for providing a trained first function and at least one trained second function (18, 18a, 18b) for use in a computer-implemented method according to claim 4, wherein - performing a computer-implemented training method according to claim 9; - obtaining a first real X-ray image dataset with reduced artifacts (19) for the first X-ray training image dataset, and obtaining a second real X-ray image dataset with reduced artifacts (20) for the second X-ray training image dataset; - generating a predicted artifact-reduced first X-ray image data set and a predicted artifact-reduced second X-ray image data set by applying an artifact reduction module comprising at least one untrained or partially trained second function (18, 18a, 18b) to further input training data, the further input training data being dependent on the first X-ray training image data set and the second X-ray training image data set; and - the input training data depends on the predicted artifact-reduced first X-ray image dataset and the predicted artifact-reduced second X-ray image dataset; - analyzing a preset other loss function (L3, L4), wherein the other loss function comprises a loss term (L3) and other loss terms (L4), wherein the loss term depends on a predicted first artifact-reduced X-ray image data set and a first artifact-reduced real X-ray image data set (19), and wherein the other loss term depends on a predicted second artifact-reduced X-ray image data set and a second artifact-reduced real X-ray image data set (20); and - Updating the parameters of at least one second function (18, 18a, 18b) according to the results of the analysis of the other loss functions (L3, L4).

12. A computer-implemented training method for providing a trained first function and at least one filter function (23, 23a, 23b) for use in a computer-implemented method according to claim 5 or 6, wherein - performing a computer-implemented training method according to claim 9; - obtaining a filtered first real X-ray image dataset for the first X-ray training image dataset, and obtaining a filtered second real X-ray image dataset for the second X-ray training image dataset; - generating a predicted filtered first X-ray image data set (21, 24) and a predicted filtered second X-ray image data set (22, 25) by applying a filtering module comprising at least one filtering function (23, 23a, 23b) to further input training data, the further input training data being dependent on the first X-ray training image data set and the second X-ray training image data set; and - the input training data depends on a predicted filtered first X-ray image data set (21, 24) and a predicted filtered second X-ray image data set (22, 25); - analyzing a preset other loss function (L3, L4), wherein the other loss function comprises a loss term (L3) and other loss terms (L4), wherein the loss term depends on a predicted filtered first X-ray image data set (21, 24) and a filtered first real X-ray image data set, and the other loss term depends on a predicted filtered second X-ray image data set (22, 25) and a filtered second real X-ray image data set; and - Updating the parameters of at least one filter function (23, 23a, 23b) according to the results of the analysis of the other loss functions (L3, L4).

13. A data processing device (2) having at least one computing unit, which is adapted to carry out a computer-implemented method according to one of claims 1 to 6 and / or a computer-implemented training method according to one of claims 8 to 12.

14. An X-ray imaging device (1), comprising a data processing device (2) according to claim 13, an X-ray source (4), an X-ray detector (3) and at least one control unit, wherein - at least one control unit is configured to control the X-ray source (4) to generate first X-rays corresponding to a first X-ray radiation energy spectrum; - the X-ray detector (3) is configured to generate a first X-ray image data set (21, 24) representing the object (5) to be imaged and to detect the portion of the first X-ray that passes through the object (5); - at least one control unit is configured to control the X-ray source (4) to generate second X-rays corresponding to a second X-ray radiation energy spectrum; and The X-ray detector (3) is provided to generate a second X-ray image data set (22, 25) representing the object (5) and to detect a portion of the second X-ray radiation that passes through the object (5).

15. A computer program product having - instructions which, when executed by a data processing device (2), cause the data processing device (2) to perform a computer-implemented method according to one of claims 1 to 6 and / or a computer-implemented training method according to one of claims 8 to 12; and / or - other instructions which, when executed by the X-ray imaging apparatus (1) according to claim 14, cause the X-ray imaging apparatus (1) to perform the method according to claim 7 or 8.

Images