Computer-implemented method for operating an imaging X-ray device, X-ray device, computer program and electronically readable data carrier

Abstract

The invention relates to a computer-implemented method for operating an imaging X-ray device (15), wherein a virtual patient model (7) is provided for an examination procedure of a patient. - the virtual patient model (7) of the patient, which describes at least the surface (6) of the patient, is determined based on patient information describing the patient, - a model position for the patient model (7) is determined in a coordinate system (10) of the X-ray device (15), in which the radiation distribution during an X-ray image of the examination procedure is also known, and - the patient model (7) is positioned according to the model position, wherein, to determine the position of the patient model (7), at least one X-ray image (1) of the examination procedure, which is taken at a current patient position of the patient in the X-ray device (15) and is available in the coordinate system (10) of the X-ray device (15), is supplied as input data to a trained position determination function (3), the output data of which describe an assignment of at least one distinguished point of the patient model (7) to at least one distinguished point of the X-ray image (1), wherein the model position is determined from the output data.

Description

[0001] The invention relates to a computer-implemented method for operating an imaging X-ray device, wherein a virtual patient model is provided for an examination procedure of a patient. - the virtual patient model of the patient, which at least describes the patient's surface, is determined based on patient information describing the patient, - a model position for the patient model is determined in a coordinate system of the X-ray device, in which the radiation distribution during an X-ray image acquisition of the examination procedure is also known, and - the patient model is positioned according to the model position.

[0002] In addition, the invention relates to an imaging X-ray device, a computer program and an electronically readable data carrier.

[0003] Medical X-ray imaging equipment is now commonplace in medical practice. X-rays emitted by an X-ray tube are used to non-invasively acquire images of the patient's internal structures by means of an X-ray detector. Since X-rays can have undesirable effects on humans, it is essential to monitor the radiation dose to which the patient is exposed during an examination, particularly when multiple images are being taken, and to provide the operator with corresponding dose information. Of particular interest is the so-called skin dose, which refers to areas along the patient's surface and is spatially resolved.To enable precise determination of dose information for a patient, accurate knowledge of the patient and their position relative to the imaging setup formed by the X-ray tube and detector—that is, their position within the coordinate system of the X-ray unit—would be necessary. However, this requires complex additional sensors and / or additional X-ray images, which should ideally be avoided.

[0004] Therefore, regarding the determination of dose information, but also for other applications, it has been proposed to use a patient model that can be patient-specific, but preferably is a statistical shape model (SSM) that describes the patient's surface and is sufficiently accurately adapted to the current patient using specific patient information, such as their height, weight, and sex. Such virtual patient models can be used not only for dose analysis but also in many other areas, such as workflow automation, markerless tracking, positioning, and navigation assistance during minimally invasive, image-guided procedures. For these reasons, it may also be possible to map other anatomical features in addition to the patient's surface, for example, at least the point positions of internal organs and the like.Such statistically based patient models are described, for example, in an article by Karthik Shetty et al., “BOSS: Bones, organs and skin-shape model,” Computers in Biology and Medicine 165 (2023) 107383. It is also possible to adapt predefined patient models based on the patient's X-ray images using artificial intelligence; see, for example, the article by Karthik Shetty et al., “Deep Learning Compatible Differentiable X-Ray Projections for Inverse Rendering,” in: Palm, C., Deserno, TM, Handels, H., Maier, A., Maier-Hein, K., Tolxdorff, T. (eds): “Bildverarbeitung für die Medizin 2021. Informatik aktuell.”, Springer Vieweg, Wiesbaden. https: / / doi.org / 10.1007 / 978-3-658-33198-6 70.

[0005] When using patient models, a common challenge is positioning the virtual patient model as accurately as possible within the coordinate system of the X-ray imaging system in order to perform calculations with the model, such as determining dose information. The goal is for the position of the patient model in a virtual computational space to correspond as closely as possible to the patient's actual position as determined for the examination. One known method is to position the patient model within the X-ray system's coordinate system based on user input, for example, from a physician and / or a medical-technical assistant. For instance, it may be known that for a specific procedure, the patient is typically positioned supine with their head 10 cm below the end of the examination table.However, it has been shown that in a large proportion of the actual investigations, the spatially resolved dose is calculated at the wrong location, meaning that the positioning assumption is not correct or at least not accurate enough.

[0006] It was also suggested to register the patient model using acquired X-ray images. However, this requires the patient model to fully represent the patient's internal anatomy so that the same structures are present in both the X-ray images and the patient model. Such complex patient models, which generate large data volumes and are difficult to adapt to specific patients, are rarely used in current technology. Known patient models typically only include, for example, a polygon mesh describing the patient's surface and, if necessary, individual point-like landmarks, such as points representing organs in space, like the center of gravity, the distal end of an organ, the proximal end of an organ, and so on. Classical registration methods are not applicable in these cases.

[0007] Publication US 2019 / 0 380 806 A1 discloses a computer-implemented method in which operating parameters are automatically determined by means of a determination algorithm from input data describing the patient in the form of a patient model and a registration of the patient with a coordinate system of the image acquisition device and are used to control the image acquisition device.

[0008] The invention is therefore based on the objective of enabling a more precise positioning of a virtual patient model for an examination procedure in a coordinate system of the imaging X-ray device.

[0009] This problem is solved according to the invention by a computer-implemented method, an X-ray imaging device, a computer program, and an electronically readable data carrier according to the dependent claims. Advantageous embodiments are described in the sub-claims.

[0010] In a method of the type mentioned at the outset, it is provided according to the invention that, in order to determine the position of the patient model, at least one X-ray image of the examination procedure, which is taken at a current patient position in the X-ray device and is available in the coordinate system of the X-ray device, is supplied as input data to a trained position determination function, the output data of which describe an assignment of at least one distinguished point of the patient model to at least one distinguished point of the X-ray image, wherein the model position is determined from the output data.

[0011] The specific determination of the patient model can utilize various methods that are generally known in the prior art. In preferred embodiments, a statistical shape model (SSM) is provided, which is adapted based on patient information. This patient information can then include, for example, the patient's height, weight, and / or gender. Other patient information that can be used to adapt the statistical shape model is also conceivable, such as age, physique, and the like. However, it is also within the scope of the present invention to determine the virtual patient model in other ways, for example, from image data of a previous examination of the patient, by capturing, for instance, at least partially three-dimensional images and / or surface data.The trained position determination function is, in any case, expediently determined in such a way that it can be used for all conceivable concrete instances of the patient model, in particular by using training data sets for different instances of the patient model.

[0012] Regarding the problematic positioning of the patient model within the coordinate system of the X-ray unit, it is specifically proposed to use a trained positioning function, i.e., methods of artificial intelligence or machine learning. This trained positioning function is provided through machine learning in such a way that it can determine, based on X-ray images of the patient, at least approximately where it is located on the patient, allowing the virtual patient model to be moved accordingly. Since X-ray images show an area of ​​the patient, they also contain information about which area is being scanned and where this area is located abstractly within the patient model.Furthermore, considering that the imaging geometry is known in the coordinate system of the X-ray device, the current patient position can be determined at least partially, which the model position should correspond to as closely as possible.

[0013] If, for example, the coordinate system of the X-ray equipment refers to a patient bed provided as a means of patient positioning, a coordinate on the patient bed can be determined from the assignment of a point in the X-ray image to a point in the virtual patient model, to which the virtual patient model (virtually lying on the virtual patient bed) is moved.This involves directly assigning a point on the virtual patient model, specifically a point on the surface described by the virtual patient model, to a point in the X-ray image, for example, its center or at least one of its corner points. This enables automatic positioning of the virtual patient model (here, the surface model) within the coordinate system of the X-ray unit, taking into account the known imaging geometry within that coordinate system and at least one additional assumption, such as that the patient is positioned on the patient positioning center. This saves time and avoids user errors that could lead to incorrect results from the functions utilizing the virtual patient model.No additional hardware, such as additional sensors like cameras, is needed for this, as it allows the processing of X-ray images that would be taken anyway.

[0014] As already mentioned, the application of the procedure described here is particularly advantageous when direct registration approaches fail because the patient model does not provide the necessary information. For example, a patient model can be used that only describes the patient's surface, i.e., a pure surface model. It is also conceivable that, in addition to the skin surface, the patient model describes at least one distinguished point of at least one organ, particularly as its only additional information. In such a case, the patient model does not describe any bone and / or tissue contours that would allow direct registration. Nevertheless, precise positioning within the coordinate system of the X-ray device can be achieved using the procedure according to the invention.

[0015] The patient model can be used in a variety of ways, including for various functions to calculate results, such as workflow automation, but preferably for determining dose information. Specifically, the dose information can be determined as a spatially resolved calculation of the radiation exposure on the positioned patient model, particularly as skin dose information. Due to the automated and, in particular, robust positioning of the virtual patient model within the coordinate system of the X-ray unit, the dose information can be determined with sufficient accuracy and correct positioning.

[0016] In general, a trained function replicates cognitive functions that people associate with other human brains. Through training based on training data (machine learning), the trained function is able to adapt to new circumstances and detect and extrapolate patterns. Another term for "trained function" is "trained machine learning model."

[0017] Generally speaking, the parameters of a trained function can be adjusted through training. Specifically, supervised learning, semi-supervised learning, unsupervised learning, reinforcement learning, and / or active learning can be used. Furthermore, representational learning (also known as feature learning) can be employed. The parameters of the trained function can be adjusted iteratively through multiple training steps. In particular, a specific cost function can be minimized during training. For example, the backpropagation algorithm can be used when training a neural network.

[0018] A trained function can, for example, comprise a neural network, a support vector machine (SVM), a decision tree, and / or a Bayesian network, and / or the trained function can be based on k-means clustering, Q-learning, genetic algorithms, and / or assignment rules. Specifically, a neural network can be a deep neural network, a convolutional neural network (CNN), or a deep CNN. Furthermore, the neural network can be an adversarial network, a deep adversarial network, and / or a generative adversarial network (GAN).

[0019] A convolutional neural network (CNN) is a neural network that uses a convolutional operation instead of general matrix multiplication in at least one of its layers, known as the convolution layer. Specifically, a convolution layer can perform a scalar product of one or more convolution kernels on the incoming data / images to the convolution layer, where the entries of the one or more convolution kernels are the parameters or weights that are adjusted through training. In particular, the inner Frobenius product and the ReLu activation function can be used. A CNN can include additional layers, such as pooling layers, fully connected layers, and normalization layers.

[0020] Convolutional neural networks (CNNs) enable highly efficient processing of input images. A convolutional operation based on different cores can extract a wide variety of image features, allowing relevant image features to be identified during training by adjusting the weights of the convolutional cores. Furthermore, by sharing weights across the convolutional layer cores, fewer parameters need to be trained, thus avoiding overfitting during the training phase. This allows for faster training or a greater number of layers in the CNN, thereby increasing the network's performance.

[0021] It is particularly advantageous in this case that the trained position determination function includes a convolutional neural network and / or an encoder. An encoder analyzes the X-ray image and outputs the relevant features—in this case, the depicted area with respect to the patient model—as a low-dimensional feature vector, which can optionally be decoded from a latent space by a decoder. It has been shown that even simple, fundamentally well-known architectures are sufficient to achieve excellent mapping accuracy.

[0022] In a convenient embodiment of the present invention, at least one of the at least one distinguished point of the X-ray image can be its center point. The center point of the X-ray image describes the attenuation experienced along the central ray of the X-ray field used. Since the acquisition geometry of the X-ray image is known in the coordinate system of the X-ray device, and thus also the position of the central ray, and since the patient can be assumed to be located on the patient positioning device of the X-ray device, in particular the patient table, simple positioning is therefore possible by assigning a point of the virtual patient model to the center point of the X-ray image. Here, the virtual patient model is assumed to be positioned on a virtual image of the patient positioning device, in particular the patient table.It can therefore be provided that the patient model is positioned in such a way that the central beam of the X-ray field, when recording the X-ray image used as input data, the position of which is known in the coordinate system of the X-ray device, passes through the distinguished point of the patient model when the patient model is positioned on a patient support device, in particular a patient bed, of the X-ray device.

[0023] When positioning the patient model at the model position within the coordinate system of the X-ray unit, additional orientation information can be used, either pre-defined by the examination procedure, derived from user input, or determined from the X-ray image. Particularly when positioning is based solely on the central beam, additional information regarding the patient's orientation on the patient support is necessary and useful, at least when different orientations are conceivable. Such orientation information is often already available, for example, assigned to a class of examination procedure, since the same orientations are typically used for certain classes of examinations. It can also be derived from user input and / or from the X-ray image itself, especially automatically.The orientation information can be kept simple, for example, on a patient table, indicating the side on which the head is positioned and whether the patient is lying supine or prone. While determining the orientation information using sensors within the X-ray unit is theoretically possible, it is less preferred, as the procedure described here specifically aims to avoid the need for such additional sensors.

[0024] According to the invention, the patient model describes the patient's surface polygonally with vertices of the polygon faces, wherein the distinguished point of the X-ray image is assigned as the distinguished point of the patient model to a vertex of the patient model, in particular to the side of the skin surface facing the X-ray tube of the X-ray device, or to a superior vertex position determined from several adjacent vertices. The patient's surface is thus described in the patient model as a polygon mesh in which corresponding vertices exist. It is now proposed, for example, to determine the nearest vertex to the corresponding point of the X-ray image using the trained position determination function, since the vertices represent clearly defined points in the patient model.For example, the vertices in the polygon mesh describing the surface can be numbered or otherwise labeled. However, it is also conceivable to use a superior vertex position, derived from several vertices, as a distinguished point of the patient model. In this case, the vertex position can be determined, for instance, as the midpoint of several vertices, particularly those associated with an anatomical feature and / or anatomically defined section of the patient's surface, such as the back. The midpoint can be restricted to lying on the surface. The use of vertex positions is particularly advantageous when, for the purposes of using the virtual patient model, a rough positioning relative to the recorded body region is sufficient.For example, multiple vertices can be grouped together, such as those of the left and right hands, as well as other anatomical features or limb segments. This reduces training and computational effort for the trained position determination function and allows for a more robust implementation. In both cases, the orientation information can be taken into account when selecting the patient's side, if necessary.

[0025] For example, to describe the surface of the patient in the patient model, a polygon mesh with one to three thousand vertices, for example two thousand vertices, which may have distances of 1.5 cm and less, can be used.

[0026] A further advantageous embodiment of the present invention provides that the X-ray image used as input data is divided into partial images, with the trained position determination function being applied separately to each partial image to determine partial results for each partial image. To increase accuracy, a current X-ray image can thus be divided into several evaluation areas, namely the partial images, and output data, for example a reference vertex, can be determined for each area. The partial images are therefore evaluated individually, each being fed as a separate input data set to the trained position determination function, so that independent output data sets are obtained as partial results.In a particularly effective advanced training, it can be stipulated that the partial results, especially by comparing the points output in the patient model of adjacent partial images, are checked for consistency to detect outliers, with detected outliers being excluded from determining the model position. Specifically, the proximity in the patient model between all defined distinguished points, especially vertices, can be assessed to check consistency. In this way, outliers can be robustly detected and ignored for positioning the virtual patient model.

[0027] Specifically, to divide the X-ray image into partial images, a regular grid or a grid adapted to the perspective of the X-ray image's acquisition geometry can be superimposed on the image. For example, if the grid is adapted to the perspective distortion caused by the acquisition geometry, the outer partial images may be larger or smaller than the inner partial images.

[0028] Within the scope of the present invention, it is also conceivable that the trained position determination function can be applied to several different divisions of the X-ray image, in particular including an application to the entire X-ray image. For example, it can be provided that the partial results of different divisions are validated against each other. A division here preferably refers to a single partial image, i.e., the use of the entire X-ray image as input data for the trained position determination function.

[0029] The variant involving the determination of initial data for different divisions can be understood as an iterative hierarchical adjustment of the partial images / evaluation areas. For example, the X-ray image can initially be divided into small partial images, which are then iteratively combined and re-evaluated in subsequent iteration steps. Regardless of the specific approach, even the simplest implementation—dividing into several partial images and applying it to the entire X-ray image—advantageously obtains and can incorporate both global and local information.

[0030] Furthermore, the accuracy of the inventive procedure can be increased by using several X-ray images as (separate) input data sets to perform multiple determinations, the results of which can be combined by statistical processing and / or checked for plausibility. Advantageously, at least one initial X-ray image from the examination procedure can be used as the X-ray image. For example, a specific number of the initial X-ray images taken during the examination procedure can be used.

[0031] A further advantageous embodiment of the present invention may provide that a determination of the model position performed at a first time point is checked and / or updated at least at a later time point during the examination procedure by reapplying the trained position determination function to at least one current X-ray image. For example, it may be provided that a determination of the model position is performed every 30 seconds to 5 minutes. This is particularly advantageous when the acquisition of X-ray images serves for image monitoring of a medical procedure, especially a minimally invasive procedure, which may extend over a longer period of time. Moreover, in such cases, it is also extremely advantageous to obtain reliable spatially resolved dose information in order to adjust the acquisition geometry for the X-ray images, which may then be fluoroscopic images, if necessary.

[0032] An advantageous design regarding the training of the position determination function can provide that, for training the position determination function, - X-ray images to be used as training input data sets are determined by simulation with a known relative positioning of the patient model, which describes the associated training output data sets, and - the position determination function is trained using the training data sets formed by related training input data sets and training output data sets.

[0033] Preferably, simulated X-ray images created with various adapted or determined instances of the patient model have the advantage that the relative positioning, and thus the initial data, are known in advance. Specifically, the simulation can include a Monte Carlo simulation and / or the use of a trained generation function for simulated X-ray images. In Monte Carlo simulation, for example, attenuation based on simulated organ densities, which can be assumed to be homogeneous, can be used. Machine learning-based virtual X-ray images, which can be generated using a corresponding generation function, can also be used to create training datasets.

[0034] It should be noted that, within the scope of the disclosure described herein, a computer-implemented method for providing a trained position determination function is also conceivable, wherein the trained position determination function determines, from input data comprising an X-ray image of a patient that exists in a coordinate system of an X-ray device with which it was taken, output data describing an assignment of at least one distinguished point of a virtual patient model of the patient, which describes at least the surface of the patient, to at least one distinguished point of the X-ray image, wherein - X-ray images to be used as training input data sets are determined by simulation with a known relative positioning of the patient model, which describes the corresponding training output data sets, - the position determination function is trained using the training data sets formed by related training input data sets and training output data sets, and - the trained position determination function is provided.

[0035] A suitable deployment system can be set up to carry out this deployment procedure. A deployment computer program, which, when executed on a computer, causes it to perform the steps of the deployment procedure, and an electronically readable deployment data carrier on which the deployment computer program is stored are also conceivable.

[0036] In addition to the operating method, the present invention also relates to an imaging X-ray device comprising an X-ray emitter, an X-ray detector, a patient positioning device, in particular a patient bed, for positioning a patient to be examined in an examination procedure and a control device comprising: - a model determination unit for determining a virtual patient model of the patient, which describes at least the surface of the patient, based on patient information describing the patient, - a unit of determination for determining a model position for the patient model in a coordinate system of the X-ray device, in which the radiation distribution during an X-ray image acquisition of the examination procedure is also known, and - a positioning unit for positioning the patient model according to the model position.

[0037] The unit of determination has the following characteristics: - an application subunit for applying a trained position determination function to input data comprising at least one X-ray image of the examination procedure, taken at a current patient position in the X-ray device and available in the coordinate system of the X-ray device, wherein the output data of the position determination function describe an assignment of at least one distinguished point of the patient model to at least one distinguished point of the X-ray image, and - a model position subunit for determining the model position from the output data of the position determination function.

[0038] All descriptions relating to the method according to the invention can be applied analogously to the imaging X-ray device according to the invention and vice versa, so that the advantages already mentioned can also be obtained with the X-ray device.

[0039] The control unit can comprise at least one processor and at least one storage medium. It is specifically configured to execute an operating method according to the invention. Functional units are formed by hardware and / or software to perform steps of the operating method according to the invention. In addition to the functional units already mentioned, further functional units are of course conceivable to execute further steps, for example, preferred embodiments. For instance, a training unit for training the position determination function, which can include a simulation subunit, is conceivable.

[0040] The control unit may also include functional units that provide functions utilizing the positioned virtual patient model. For example, a dose calculation unit may be provided to determine dose information as spatially resolved radiation exposure to the positioned patient model, particularly as skin dose information. A workflow unit may also be included and utilize the positioned patient model. Further functional units of the control unit may, for example, include an acquisition unit that controls the acquisition of at least one X-ray image and / or additional X-ray images of the examination. A display device, such as a monitor, may be used to output X-ray images and information.For example, dose information, especially regarding skin dose, can be output by means of such a display device, for example controlled by a user interaction unit of the control device.

[0041] A computer program according to the invention can be directly loaded into a storage medium of a control unit of an X-ray device and comprises program elements such that, when the computer program is executed on the control unit, the latter is caused to carry out the steps of a method according to the invention. The computer program can be stored on an electronically readable data carrier according to the invention, which therefore includes control information stored thereon, comprising at least one computer program according to the invention and configured such that, when the data carrier is used in a control unit of an X-ray device, the latter is configured to carry out a method according to the invention. The data carrier can be a non-transient data carrier, in particular a CD-ROM.

[0042] Further advantages and details of the present invention will become apparent from the exemplary embodiments described below and from the drawings. These show: Fig. 1 a flowchart of an exemplary embodiment of the operating method according to the invention, Fig. 2 a schematic representation of the interference process of a used trained position determination function, Fig. 3. A schematic diagram for determining a model position based on input data from the trained position determination function. Fig. 4 a schematic representation illustrating a variant of the operating method according to the invention, Fig. 5 an X-ray device according to the invention, and Fig. 6 the functional structure of a control unit of the X-ray equipment.

[0043] Fig. Figure 1 shows a flowchart of an embodiment of the operating method according to the invention for an X-ray imaging device. In step S1, an examination procedure begins on a patient, during which, in this case, several X-ray images are to be acquired using the X-ray imaging device. The patient is already positioned on a patient support device of the X-ray device, here a patient table, in a patient position for acquiring the X-ray images. Furthermore, orientation information is already available, describing how the patient is oriented (for example, on which side of the longitudinal dimension of the patient table the head is located and whether the patient is in a supine or prone position). Patient information, in particular including the patient's height, weight, and sex, has also already been recorded and is available.

[0044] The examination process can, for example, involve image monitoring during a medical procedure, particularly a minimally invasive one. In this case, fluoroscopic X-ray images are repeatedly taken to visually document the progress of the medical procedure.

[0045] In step S2, a virtual patient model is created, in this case by adapting a statistical shape model (SSM) based on the patient information. Other methods are also conceivable, for example, using previous images of the patient. The patient model can also be created entirely patient-specifically.

[0046] The patient model describes the patient's surface as a polygon mesh with corresponding polygon faces and vertices. Optionally, the patient model can also describe designated points of organs; however, the internal anatomical structure is not fully represented, so registration with acquired X-ray images is not possible.

[0047] In step S3, the first X-ray image of the examination process is then taken.

[0048] In step S4, it is checked whether a condition for positioning the patient model within a coordinate system of the X-ray unit is met. The coordinate system of the X-ray unit can, for example, be defined relative to the patient table. For acquired X-ray images, the acquisition geometry within this coordinate system of the X-ray unit is naturally known, and thus also the position of the X-ray field. This means that the X-ray image exists within the coordinate system of the X-ray unit in such a way that for each pixel, the path of the corresponding ray along which the attenuation was measured is known.

[0049] In step S4, for the initial positioning of the patient model in the coordinate system of the X-ray unit, it may be sufficient that a first X-ray image of the examination has been acquired; however, it is also possible that a specific number of initial X-ray images of the examination must be available. In the illustrated embodiment, the positioning is checked regularly, meaning that the condition of step S4 is also fulfilled if, for example, a certain time has elapsed since the last positioning, which can be chosen, for example, between 30 seconds and 5 minutes.

[0050] In step S5, if the condition of step S4 is met, a trained position determination function is used. The trained position determination function uses input data, each comprising one of at least one initial X-ray image and, at a later time, at least one of at least one current X-ray image to be used. It provides output data that assigns a distinguished point on the patient's surface in the patient model to a distinguished point on the X-ray image.

[0051] In this case, the trained position determination function assigns a vertex of the polygon mesh modeling the surface to the center point of the X-ray image used as input data, as schematically shown by Fig. Figure 2 is shown. This initially displays the X-ray image 1 with center point 2 schematically as input data, which is fed as input to the trained position determination function 3. The trained position determination function 3, which is implemented as at least one CNN, comprises an encoder 4 that provides results describing the corresponding vertex 5 of the patient's surface 6 (schematically indicated here) in the patient model 7. In doing so, the vertex 5 located on the side of the X-ray entry point, which is closest to the central beam, is selected (particularly taking the orientation information into account).

[0052] How Fig. As shown in Figure 3, together with the assumption that the patient is lying on the patient bed 8, which is also schematically indicated there, and the orientation information (e.g., supine position, head first), a suitable model position for the patient model 7 results, since the imaging geometry and thus the X-ray field 9 used for the image acquisition are known to the coordinate system 10 of the X-ray device. This applies in particular to the central ray 11, along which the image value of the center point 2 of the X-ray image 1 was acquired, and which, according to the initial data, must pass through the vertex 5 described by it, as shown in Figure 3. Fig. Figure 3 is illustrated graphically. Therefore, the model position can be derived in step S5 from the initial data, together with the assumption that the patient is lying on the patient bed 8 and / or the orientation information. If there is still uncertainty, for example due to an unfavorable orientation of the central beam, a central position on the patient bed is assumed.

[0053] It is not necessary to use vertices 5 as distinguished points on the surface 6 of the patient model 7; rather, several vertices 5 can be grouped together, for example, according to anatomical features such as hands, feet, left abdomen, and the like, to form distinguished points, namely vertex positions. For instance, the vertex position could be the midpoint of the vertices 5 to be grouped together. In this way, the accuracy of the positioning is somewhat reduced, but the robustness is increased.

[0054] It should be noted that other or multiple prominent points in the X-ray image, such as corner points, can also be used additionally or alternatively. This can provide at least partial indications regarding the alignment.

[0055] Fig. Section 4 describes a variant that can be used additionally or alternatively in step S4. In this step, the X-ray image 1, in this case using a superimposed grid 12, is divided into sub-images 13, to each of which the trained position determination function 3 is applied separately. The result is a group 14 of vertices 5, whose relative positions on the surface 6, i.e., their proximity, can be used for mutual plausibility checks and / or outlier detection. It is conceivable to determine input data for multiple divisions in step S4, in particular for application to the entire X-ray image 1 and at least one division into several sub-images 13. Mutual plausibility checks can also be performed between the divisions.

[0056] Returning to Fig. In step S6, the patient model 7 is then positioned according to the model position in the coordinate system 10 of the X-ray unit. It can now be used by at least one function of the X-ray unit, for example, to determine spatially resolved dose information during longer examination procedures, such as a spatially resolved skin dose.

[0057] The position determination function 3 was trained using machine learning before its deployment. The training datasets were obtained at least partially through simulation by generating X-ray images for use as training input data at known model positions of different instances of the patient model. This was done using Monte Carlo simulation and / or a trained generation function that implicitly performs at least part of the simulation.

[0058] Fig. Figure 5 shows a schematic diagram of an X-ray device 15 according to the invention. This is an X-ray device 15 with a C-arm 16, such as can be used, for example, in interventional settings. The X-ray device 15 comprises an X-ray source 17 and an X-ray detector 18 as an imaging arrangement, which are mounted opposite each other on the C-arm 16. The C-arm 16 is designed to be movable and can be positioned in different ways relative to the patient table 8 in order to achieve different imaging geometries.

[0059] The operation of the X-ray device 15 is controlled by a control unit 19, the functional structure of which is based on Fig. 6 will be explained in more detail. Accordingly, the control unit 19 initially comprises a storage device 20 for storing various information, such as the X-ray images 1, the alignment information, the patient information, the patient model 7 and the like.

[0060] A control unit (not shown in detail) controls the general operation, for example according to steps S1 and S4. The control unit 19 also includes a recording unit 21, which controls the recording operation of the X-ray unit 15. The acquisition of X-ray images 1 in step S3 is also controlled by means of this unit. A model determination unit 22 is configured to determine the virtual patient model 7 of the patient according to step S2. In a determination unit 23, the model position for the patient model 7 is determined according to step S5. For this purpose, the determination unit 23 includes an application subunit 24 for applying the trained position determination function 3 and a model position subunit 25 for determining the model position from the output data of the position determination function 3, for example according to Fig.3. Finally, the control unit 19 also includes a positioning unit 26 for positioning the patient model 7 according to step S6.

[0061] Regarding the use of the positioned patient model 7, a dose unit 27 is also shown as an example, in which the dose information, which displays the spatially resolved radiation dose acting on the patient, is determined based on the positioned patient model 7 and the acquisition parameters of the X-ray images 1. For example, the dose information can be used for general monitoring (issuing warnings when limit values ​​are exceeded) and / or displayed on a display unit of the X-ray device 15.

[0062] Of course, in addition to the functional units described, other functional units may also be provided, such as a user interaction unit and the like.

[0063] Regardless of the grammatical gender of a particular term, persons with male, female or other gender identities are included.

Claims

[1] Computer-implemented method for operating an X-ray imaging device (15), wherein a virtual patient model (7) is provided for an examination procedure of a patient - the virtual patient model (7) of the patient, which describes at least the surface (6) of the patient, is determined based on patient information describing the patient, - a model position for the patient model (7) is determined in a coordinate system (10) of the X-ray device (15), in which the radiation distribution during an X-ray image acquisition of the examination procedure is also known, and - the patient model (7) is positioned according to the model position, characterized by, that to determine the position of the patient model (7), at least one X-ray image (1) of the examination procedure, which is taken at a current patient position in the X-ray device (15) and is available in the coordinate system (10) of the X-ray device (15), is supplied as input data to a trained position determination function (3), the output data of which describe an assignment of at least one distinguished point of the patient model (7) to at least one distinguished point of the X-ray image (1), wherein the model position is determined from the output data, and wherein the patient model (7) describes the surface (6) of the patient polygonally with vertices (5) of the polygonal surfaces, wherein the distinguished point of the X-ray image (1) belongs to a side of the skin surface (6) facing the X-ray tube (17) of the X-ray device (15),The vertex (5) of the patient model (7) or a superior vertex position determined from several adjacent vertices (5) is assigned as a distinguished point of the patient model (7). [2] Method according to claim 1, characterized by , that the patient model, in addition to the skin surface (6), also describes, in particular as the only additional information, at least one distinguished point of at least one organ. [3] Method according to claim 1 or 2, characterized by , that at least one of the at least one distinguished point of the X-ray image (1) is its center (2). [4] Method according to claim 3, characterized by, that the patient model (7) is positioned such that the central ray (11) of the X-ray field (9) passes through the distinguished point of the patient model (7) when the patient model (7) is positioned on a patient support means of the X-ray device (15) is taken. [5] Method according to any of the preceding claims, characterized by , that when positioning the patient model (7) at the model position in the coordinate system (10) of the X-ray device (15), additionally an orientation information is used which is associated with the examination process and / or can be derived from a user input and / or can be determined from the X-ray image (1). [6] Method according to any of the preceding claims, characterized by, that the X-ray image (1) used as input data is divided into partial images (13), whereby the trained position determination function (3) is applied separately to each partial image (13) to determine partial results for each partial image (13). [7] Method according to claim 6, characterized by , that the partial results are checked for consistency, in particular by comparing the points output in the patient model (7) of adjacent partial images (13) to detect outliers, whereby detected outliers are not used in determining the model position. [8] Method according to claim 6 or 7, characterized by , that the application of the trained position determination function (3) is carried out for several different divisions of the X-ray image (1), in particular including an application to the entire X-ray image (1). [9] Method according to any of the preceding claims, characterized by, that the trained position determination function (3) includes a Convolutional Neural Network and / or an encoder (4). [10] Method according to any of the preceding claims, characterized by , that to train the position determination function (3) - X-ray images to be used as training input data sets are determined by simulation with a known relative positioning of the patient model, which describes the associated training output data sets, and - the position determination function is trained using the training data sets formed by related training input data sets and training output data sets. [11] Method according to claim 10, characterized by that the simulation includes a Monte Carlo simulation and / or the use of a trained generation function for simulated X-ray images. [12] X-ray apparatus (15) for carrying out a method according to one of claims 1 to 11, comprising an X-ray emitter (17), an X-ray detector (18), a patient positioning device for positioning a patient to be examined in an examination procedure and a control device (19) comprising: - a model determination unit (22) for determining a virtual patient model (7) of the patient, which describes at least the surface (6) of the patient, based on patient information describing the patient, - a determination unit (23) for determining a model position for the patient model (7) in a coordinate system (10) of the X-ray device (15), in which the radiation distribution during an X-ray image acquisition of the examination procedure is also known, and - a positioning unit (26) for positioning the patient model (7) according to the model position, characterized by, that the unit of determination (23) exhibits: - an application subunit (24) for applying a trained position determination function (3) to input data comprising at least one X-ray image (1) of the examination procedure, which is recorded at a current patient position of the patient in the X-ray device (15) and is available in the coordinate system (10) of the X-ray device (15), wherein the output data of the trained position determination function (3) describe an assignment of at least one distinguished point of the patient model (7) to at least one distinguished point of the X-ray image (1), and - a model position subunit (25) for determining the model position from the output data of the position determination function (3). [13] Computer program which, when executed on a control device (19) of an imaging X-ray device (15), causes the latter to perform the steps of a method according to any one of claims 1 to 11. [14] Electronically readable data carrier on which a computer program according to claim 13 is stored.

Images