Method and device for improving X-ray images

Abstract

The invention relates to a method for improving X-ray images (F), comprising the steps: - Providing a series of X-ray images (F) representing successive images of the same region of interest, - Segmenting the X-ray images (F) of the series by identifying structures (S) in the region of interest, and determining a confidence level for identified structures, which indicates the quality of the identification, - Selecting at least some of these structures (S) as reference structures (R) based on their security value, - Registering the X-ray images (F) of the series against each other, whereby the reference structures (R) are registered against each other according to a registration function (X), - Creating a result image (B) by linking image information from the registered X-ray images (F, Fr) - Outputting the result image (B). Furthermore, the invention comprises a device, a control unit and a medical technology system.

Description

[0001] The invention relates to a method and a device for improving X-ray images, in particular fluoroscopy images, a control device for a medical technology system and a medical technology system.

[0002] In X-ray imaging, particularly fluoroscopic imaging, the visualization of relevant anatomical or procedural structures, such as pathologies, anatomical landmarks, or medical objects, is typically degraded by noise. This is due to the "as-low-as-reasonably-achievable" (ALARA) principle, as the physician is motivated to keep the radiation dose as low as possible for the patient's benefit. However, this increases image noise. Besides spatial noise reduction, temporal averaging of the image information is a common approach to reducing image noise. However, if the structure of interest moves between images, averaging tends to blur the structure rather than enhance its clarity.

[0003] One way to counteract this is to perform image registration across multiple frames before averaging them over time. However, this requires that the structures be visible across several consecutive frames. An example of this is an algorithm where pairs of balloon markers are tracked across multiple frames and then superimposed to obtain a sharper view of an object such as a stent. This allows for the assessment of whether the object's placement was successful. Balloon markers are highly X-ray-absorbing bulges on a guidewire that is advanced through a catheter. These are located, for example, where a stent balloon is positioned.

[0004] Currently, no method exists that is applicable to larger or multiple structures.

[0005] DE 10 2017 201 162 A1 describes a method for operating a medical imaging X-ray device by capturing a previous single image from a series of images containing a multitude of single images of a body part of a patient, wherein a medical component is located in the body part.

[0006] In US patent 2014 / 0079308A1, a method and system for real-time stent enhancement on a live 2D fluoroscopy scene are disclosed. A motion-compensated stent enhancement image is generated from an initial set of images in a fluoroscopy image sequence.

[0007] In US patent 2016 / 0174902A1, a method and a system for the recognition of anatomical objects using neural networks are disclosed.

[0008] US Patent 2013 / 0072788A1 discloses a method and a system for detecting and tracking multiple catheters in a fluoroscopy image sequence.

[0009] DE 10 2004 061 435 A1 describes techniques for analyzing features of lung images by registering areas, especially pleural areas of the lung, in different images generated at different times.

[0010] From DE 10 2015 208 929 B3 a method for registering a three-dimensional image data set of a target area of ​​a patient with a two-dimensional X-ray image of the target area taken in an imaging geometry is known.

[0011] It is an object of the present invention to provide a method and a device for improving X-ray images, a control device for a medical technology system and a medical technology system with which the disadvantages described above are avoided.

[0012] This problem is solved by a method according to claim 1, a device according to claim 10, a control device according to claim 12 and a medical technology system according to claim 13.

[0013] A method according to the invention serves to improve X-ray images. These X-ray images are preferably fluoroscopy images. It comprises the following steps: - Providing a series of X-ray images representing successive images of the same region of interest, - Segmenting the series of X-ray images by identifying structures in the region of interest and determining a confidence level for the identified structures, indicating the quality of the identification; - Selecting at least some of these structures as reference structures based on their confidence level. - Registering the X-ray images of the series against each other, whereby the reference structures are registered against each other according to a registration function, - Creating a result image by linking image information from the registered X-ray images, - Outputting the result image.

[0014] The method for improving X-ray images can be particularly advantageous for noise reduction or, more generally, for signal amplification. The procedure for improving X-ray images can be described as follows: First, a series of consecutive X-ray images of the same region of interest (ROI) are acquired. This could, for example, be a series of X-ray images of a moving object, such as a moving body organ, e.g., a heart, during an examination. The delivery of the images can involve both capturing them and downloading them from a database.

[0015] Next, structures within the images are identified and delineated (segmented) from other structures. This can be done manually by a specialist, but preferably automatically using specialized software. For this to be effective, the structures should be recognizable in the images across several consecutive frames, at least to the user performing the procedure. Image segmentation itself is well-established in the art. For example, the heart, aorta, coronary arteries, and a stent can be segmented in X-ray images. It is preferred to segment two or more structures in the X-ray images.

[0016] It should be noted that the term "segmentation" also includes landmark detection, which can later be used for registration in conjunction with orientation calculations. Landmark detection involves identifying specific, pre-existing points of the structures in the X-ray images as landmarks, thus segmenting the structures by defining these distinctive points. It should be noted that landmark detection can certainly be used for registration without further segmentation, for example, by simply overlaying the landmarks. However, the landmarks of specific structures or areas can also be used to further refine their segmentation.

[0017] In addition to identifying structures, a confidence score is determined that indicates the quality of the identification. This can be done, for example, by reading the probability values ​​of a trained machine learning system for the segmented image areas during segmentation. The confidence score can also be determined subsequently as part of a segmentation review. The important thing is that for each identified (and thus segmented) structure, there is a value that indicates a measure of the quality of the identification (or segmentation) and / or a measure of the completeness of the segmented structure. For clarity, this value can be assumed to lie between 0 and 1, with its magnitude representing a measure of quality.

[0018] At least one of these identified structures is selected as a reference to register the images against each other. These structures are called reference structures for clarity, but they represent a portion (or all) of the segmented structures. The reference structures should be present in all radiographs, otherwise registration becomes difficult. However, there is also a solution for cases where reference structures are not visible in all radiographs. This involves dividing the series of radiographs into subseries and is described in more detail below.

[0019] The selection of reference structures is based on their security values. The exact procedure can be defined in advance. Preference is given to structures whose security value indicates the highest quality (especially those with the highest security values). A range of values ​​can be specified, and all structures within that range can be selected as reference structures. Alternatively, a number N of structures can be specified, and the N structures with the highest security values ​​can be selected. It is also possible to suggest the structures with the highest security values ​​to a user, who then selects the reference structures.

[0020] The individual X-ray images are then registered against each other using the selected reference structures and a corresponding mathematical function (registration function) to make the X-ray images optimally comparable, at least in the area of ​​the reference structures. At least the area of ​​the reference structures is registered. The area of ​​the X-ray images beyond this can also be registered, preferably using the same registration function, or it can remain unregistered.

[0021] It should be noted that not all areas of an X-ray image are always medically relevant. It may well be that only a single organ or medical object is to be examined. This organ or object can then be one of the reference structures, and it is not necessary, for example, to register the edges of the X-ray images with each other. For instance, the positions of the heart, the surrounding vessels, and an artificial heart valve are aligned using the registration function. The principle of image registration itself, as well as the creation of a registration function, is known in the prior art.

[0022] The remaining areas of the X-ray images (outside the reference structures) can also be registered to each other using this registration function. However, this is not strictly necessary. The procedure could be designed so that this option can be enabled or disabled via a user interface. It may even be the case that the areas outside the segmented structures should remain unchanged, and registration should deliberately only occur in an overlay area, e.g., an ellipse. It may well be preferable to average a structure under investigation with a certain surrounding "halo" over time. In this case, registration only needs to take place in this area.

[0023] It should be noted that the registration function may be the same across reference structures, or it may differ in different areas, especially for different reference structures.

[0024] A final image is then generated based on the registered X-ray images. It should be noted that multiple final images can be generated, but at least one final image must be generated. This final image must be based on a number of registered X-ray images. This means that it is created using the image information from several registered X-ray images. It is particularly advantageous that each pixel of the final image is based on the corresponding pixels of the respective registered X-ray images (each at the same image coordinate). This can be achieved, for example, by adding, subtracting, multiplying, or dividing the relevant image values ​​of the registered X-ray images. Essentially, all possibilities offered by modern image processing are available.

[0025] Finally, the generated result images (one or more) are displayed. By using multiple consecutively registered images for a single result image, this result image, when the procedure is applied correctly, has improved quality and accuracy compared to the original X-ray images. This output of result images can be used to check the number of images displayed on a screen, or they can simply be saved.

[0026] In practice, the following procedure is preferred: First, a general segmentation of the frames to be registered (these are the X-ray images) is performed, and then a selection of the structures to be used for registration (reference structures) is made. If the selected structures are no longer usable, a

[0027] The selection of reference structures is changed. Then the registration takes place, followed by the creation of the resulting image.

[0028] To utilize as many structures represented in the frames as possible, a general segmentation method is employed that segments a large number of structures in each frame according to known automated procedures. Various methods can be used, such as segmentation algorithms that distinguish between foreground and background to select only important structures (i.e., from the foreground). Furthermore, various features of the (segmented) structures can also be segmented or recognized to aid subsequent registration. For example, specific landmarks, contrast features, metrics, or edge information can be recognized and then used (optionally additionally) for registration. The quality of the segmentation does not need to be perfect, especially if the center point of the structure of interest and its orientation are most important for the subsequent steps.

[0029] Only well-segmented structures can be selected for registration. It is preferred that the segmentation method outputs a value for each segmented structure, reflecting the certainty of the segmentation and structure recognition ("confidence value"). For example, the confidence value indicates how certain it is that a structure recognized as a "heart" actually represents the heart. Determining such a value is state of the art and will not be discussed further here. This value can be determined, for example, during segmentation or as part of a subsequent recognition process. Segments can then be selected for further investigation, or the N (integer > 0) structures with the highest confidence values ​​can be selected. Alternatively, a single structure can be selected for investigation, along with N additional structures.This can be achieved, for example, by using Monte Carlo Drop Out with deep learning-based methods.

[0030] Various methods (especially deep learning-based ones) can be used for registration to ensure real-time performance (see Fu, Y. et al. “Deep learning in medical image registration: a review. Physics in Medicine & Biology”, 65(20), 20TR01; 2020). Another method could be to calculate the center of gravity with a clear principal axis on each frame and compare the frames to determine the relationship between them. To improve robustness, it is preferable to focus solely on the position and orientation of the structures of interest. Alternatively, template matching can be used to perform robust registration.

[0031] If a selected structure in the frames is no longer easily segmentable or disappears from the image area (or "field of view"), for example, due to anatomical or C-arm movement, it is possible to segment new structures for registration. One way to do this would be to replace the old structure with the next best suitable one, or to select a new reference structure. The registration itself is maintained by the other structures that were previously used, as well as the one that is no longer in use. If this is not possible, the user could be notified that the reference structures are no longer present between two frames and that manual registration by the user is required. Once this is done, new structures can be searched for to register.

[0032] For medical practice, it is preferred that the user can additionally specify how many structures between two frames should be used for registration, e.g., between 1 and 10. Furthermore, it is preferred that the user can specify from which anatomical region the structures should be used for registration (e.g., only from a specific vessel or vascular region). For this purpose, it is advantageous if the current segmentation of the structures is displayed along with the confidence level. The user can then determine which structures should be used, preferably by clicking on them or drawing a contour. It is also preferred that the user can specify that only structures with a certain confidence level should be used for registration, e.g., >95%.

[0033] Voice control is preferred for user input.

[0034] A major advantage of this method is its flexible, comprehensive solution, allowing for the enhancement of various structures based on the image and the physician's input. This enables flexible application across multiple scenarios, requiring validation and attention to only a single component within the product. Furthermore, multiple structures can be used simultaneously, potentially leading to improved performance. Moreover, the method is not limited to individual structures but can be dynamically adapted.

[0035] A device according to the invention serves to improve X-ray images. It comprises the following components: - a data interface designed to receive or retrieve a series of X-ray images representing successive images of the same region of interest, - a segmentation unit designed to segment the series of X-ray images by identifying structures in the region of interest and to determine a confidence score indicating the quality of the identification, - a selection unit designed to select at least some of these structures as reference structures based on their security value, - a registration unit designed to register the X-ray images of the series against each other, wherein the reference structures are registered against each other according to a registration function, - a result unit designed to create a result image by linking image information from the registered X-ray images, and - a data interface designed for outputting result images.

[0036] The device is preferably designed for carrying out a method according to the invention and in particular allows for an improvement in X-ray images. It comprises several interconnected components which, in combination, enable high-quality imaging.

[0037] The first component is the data interface, which is designed to receive or retrieve a series of sequential X-ray images of the same region of interest. An example of this would be reading in a sequence of X-ray images of a moving object.

[0038] These images are then forwarded to the segmentation unit, which identifies and segments structures within the X-ray images. Additionally, a confidence level is determined for the segmented structures, indicating the quality of the identification.

[0039] The selection unit then chooses at least some of these identified structures as reference structures based on their security value (as described in the procedure description). These reference structures are crucial for the subsequent registration of the recordings against each other.

[0040] The registration unit then registers the selection of X-ray images against each other by registering the reference structures according to a registration function and preferably also adjusting at least some of the other areas of the image accordingly. An example of this would be aligning X-ray images of a moving heart by using the structure of the pericardium and the surrounding vessels as reference structures.

[0041] The output unit is used to create a final image by combining image information from the recorded X-ray images. This has already been described in the context of the procedure. Optionally, a denoising unit can be used as the output unit to create a noise-reduced or signal-optimized image. This is preferably done by (temporal) averaging of the recorded X-ray images. For example, the denoising unit could reduce the noise in a sequence of X-ray images by calculating the average intensity of each pixel across all images.

[0042] Finally, another data interface outputs the registered X-ray images or, if available, the noise-reduced image. This allows the enhanced images to be displayed on a monitor or saved for further analysis.

[0043] A control device according to the invention for a medical technology system, in particular a diagnostic system or an X-ray system, preferably a fluoroscopy system, comprises a device according to the invention and / or is designed to carry out a method according to the invention.

[0044] A medical technology system according to the invention is preferably a diagnostic system or an X-ray system, in particular a fluoroscopy system, and comprises a control device according to the invention.

[0045] The invention can be implemented, in particular, in the form of a computer unit with suitable software. The computer unit can, for example, comprise one or more cooperating microprocessors or the like. In particular, it can be implemented in the form of suitable software program components within the computer unit. A largely software-based implementation has the advantage that even previously used computer units can be easily retrofitted by a software or firmware update to operate according to the invention. In this respect, the problem is also solved by a corresponding computer program product with a computer program that can be directly loaded into a memory device of a computer unit, containing program sections to execute all steps of the method according to the invention when the program is run in the computer unit.In addition to the computer program itself, such a computer program product may include additional components such as documentation and / or additional components, including hardware components such as hardware keys (dongles, etc.) for using the software.

[0046] For transport to the computer unit and / or for storage on or in the computer unit, a computer-readable medium, such as a memory stick, a hard drive or other portable or permanently installed data carrier, can be used, on which the program sections of the computer program that can be read and executed by a computer unit are stored.

[0047] Further, particularly advantageous embodiments and developments of the invention result from the dependent claims and the following description, wherein the claims of one claim category may also be further developed analogously to the claims and description parts of another claim category and, in particular, individual features of different embodiments or variants may be combined to form new embodiments or variants.

[0048] Preferably, the resulting image is a noise-reduced or signal-enhanced image. It is generated based on multiple X-ray images by averaging the recorded X-ray images. It is not necessary to use all recorded X-ray images for this purpose, although this can be advantageous. However, if multiple resulting images are to be generated as a series, it is preferred to use only a portion of the recorded X-ray images for each result image. For example, with N recorded X-ray images, M could be used for each result image. <N aufeinanderfolgende registrierte Röntgenaufnahmen für ein Ergebnisbild verwendet werden, wobei für das erste Ergebnisbild die Aufnahmen 1 bis M, für das zweite die Aufnahmen 2 bis (M+1) und so weiter, verwendet werden.

[0049] The creation of result images is preferably achieved by averaging the recorded X-ray images. This means that the image values ​​of corresponding image coordinates are averaged. Since the X-ray images are taken sequentially, temporal averaging is preferred. This essentially means that corresponding image values ​​from successive images are averaged. Here, averaging preferably means that N image values ​​from N recorded X-ray images are added and normalized (e.g., by division by N).

[0050] Preferably, the procedure includes the following steps: - Creating a noise-reduced image as a result image based on the registered X-ray images, preferably by time-averaging the registered X-ray images, and - Outputting the noise-reduced image as the result image.

[0051] Preferably, the procedure includes the following steps: - Creating a signal-optimized image as a result image based on the recorded X-ray images, preferably by time averaging the recorded X-ray images, and - Outputting the signal-optimized image as the result image.

[0052] Image denoising is a well-known technique. However, the unique aspect here is that registering the X-ray images using this method in the area of ​​reference structures allows for particularly effective image denoising.

[0053] Time averaging can be achieved, for example, simply by adding up the image values ​​of the X-ray images with the same image coordinates and normalizing them all with the same value, e.g., the number of images.

[0054] Preferably, at least two structures are identified and defined as reference structures. To utilize as many structures as possible depicted in the frames (X-ray images), a general segmentation method is used that segments a large number of structures in each frame. It is preferred that segmentation algorithms are used that distinguish between foreground and background to select only important structures (i.e., from the foreground). Preferably, a user can specify how many and / or which structures should be used for image registration (e.g., 1 to 10). This information can also be provided as presets. Alternatively or additionally, a user can specify from which anatomical region structures should be used (e.g., only from a specific vessel or from a specific vascular region). This information can also be provided as presets.

[0055] According to a preferred embodiment of the method, segmentation is performed such that various features of the segmented structures are identified. "Features" refers to derived properties of the segmented areas, such as prominent points of an organ or its size. Registration is then preferably performed based on these features. It is preferred that the features comprise one or more predefined components from the group consisting of landmarks, contrast features, metrics, averages, and edge information.

[0056] It should be noted that the quality of the segmentation does not need to be perfect, as the center of the structure of interest and its orientation are most important for the subsequent steps. The segmentation may therefore only include a few prominent landmarks of an organ that define its orientation in space. During registration, these landmarks can then be registered relative to each other, and all intermediate points can be recorded in the X-ray images based on the registration of the limiting landmarks.

[0057] It is preferred that, in one embodiment of the method, the selection of reference structures is based on a confidence value, and a number of those segmented structures with the highest certainty are selected as reference structures, either a predetermined number or those structures with a confidence value above a predetermined threshold are selected as reference structures. The term "confidence value" has already been explained above. Essentially, the determination of this value is known in the context of image segmentation or subsequent image recognition (where the term is also referred to as "certainty value"). In the simplest case, the confidence value for each pixel in a structure indicates the probability that a pixel belongs to that structure. One could then, for example, calculate the average of all pixel probabilities in a region and use this as the confidence value for that region.So how certain is the network that a pixel belongs to its structure?

[0058] It is preferred that the safety value be determined using Monte Carlo Drop Out with deep learning-based methods.

[0059] The segmented areas can be displayed to a user along with their security value, and the user can then select segments, for example, by clicking on the structure. It can be specified that only structures with a security value above a predefined threshold should be used. It should be noted that while it is preferable to select structures with the highest security values, a user might prefer other structures.

[0060] It is preferred that in one embodiment of the method, a deep learning-based method is used for registration to ensure real-time performance. Alternatively or additionally, it is preferred that a calculation of the center of gravity with a principal axis is used on each X-ray image. Alternatively or additionally, it is preferred that a comparison of the results between the X-ray images is used to obtain the relationship between them.

[0061] It is preferred that in one embodiment of the registration method, the position and orientation of the reference structures are taken into account, preferably exclusively.

[0062] It is preferred to use template matching for registration. This is a technique in image processing and pattern recognition where a given image (the template) is compared with a portion of a larger image to find similarities or matches. The goal of template matching is to find the positions in the image where the template best fits. The principle of template matching is known in the prior art and can improve robustness.

[0063] According to the inventive method, a quality measure is generated that indicates the quality of the registration of the reference structures to one another or indicates the difference between similar reference structures. The determination of the quality measure is known in the prior art within the context of determining the quality of image registration. If the quality measure lies outside a predetermined quality range, the series of X-ray images is divided into a plurality of sub-series of X-ray images, with individual reference structures being selected for each sub-series and the method being continued for each sub-series as an independent series.

[0064] If the selected structures are no longer usable, the selection of reference structures can preferably be changed, in particular by replacing an existing reference structure with a different structure from a segmented X-ray image as the new reference structure.

[0065] If automatic segmentation or registration is not possible, it is preferred that manual segmentation or registration be carried out by a human for a subseries.

[0066] According to a preferred embodiment of the method, a quality measure is derived from how well a selected structure can be segmented or whether a structure disappears from the image area, e.g. due to anatomical movement or C-arm movement.

[0067] It is preferred that a change vector be derived from a change, particularly a movement, of a structure in the series of X-ray images, and that this change vector be used for registration. It should be noted that a structure can indeed move along a clearly definable trajectory. This trajectory can be determined by external parameters, e.g., in the case of the movement of a C-arm, or by parameters that can be derived from the individual X-ray images, e.g., in the case of the movement of an object through the bloodstream. If the trajectory of an object is known, or even the trajectories of several landmarks of the object, then a change in this object can be extrapolated from the trajectories, even if the object is not visible. This extrapolated change can also be used for image registration.

[0068] According to a preferred embodiment of the method, subsets of X-ray images are registered against each other based on identical reference structures in both subsets. If this is not possible, the user can be notified that the registration structures are no longer present between two frames and that manual registration by the user is required.

[0069] A preferred device comprises at least one machine learning-capable model that has been trained accordingly for its task. This is preferred: - a machine learning model that has been trained to segment X-ray images, and / or - a machine learning model trained to register X-ray images, and / or - A machine learning model that has been trained to select reference structures. This model may well be designed for image recognition or to identify suitable organs from a given examination procedure.

[0070] Preferably, components of the invention are provided as a "cloud service." Such a cloud service serves to process data, particularly using artificial intelligence, but can also be a service based on conventional algorithms or a service where human evaluation takes place in the background. Generally, a cloud service (hereinafter also referred to simply as "cloud") is an IT infrastructure in which, for example, storage space or computing power and / or application software is provided via a network. Communication between the user and the cloud takes place via data interfaces and / or data transmission protocols. In the present case, it is particularly preferred that the cloud service provides both computing power and application software.

[0071] In a preferred method, data obtained within the scope of the invention is provided to the cloud service via the network. This cloud service comprises a computing system that typically does not include the user's local computer. The method can be implemented using a command structure within a network. The data processed in the cloud is subsequently sent back to the user's local computer via the network.

[0072] The invention is explained in more detail below with reference to the accompanying figures and exemplary embodiments. The same components are designated with identical reference numerals in the various figures. The figures are generally not to scale. They show: Fig. 1 an example of an X-ray system with a device or control unit according to the invention, Fig. 2 a sequence of an embodiment of a method according to the invention, Fig. 3. One way to create a subseries of X-ray images, Fig. 4 an alternative way to create a subseries of X-ray images and register them.

[0073] Fig. Figure 1 shows a rough schematic of an X-ray system 1, as an example of a system for taking X-ray images F (see also Fig. 2), with a control unit 2. This is equipped with a device 5 for carrying out the method according to the invention.

[0074] The X-ray system 1 has a radiation source 3 in the usual way, which here represents an X-ray source, and during an X-ray exposure F irradiates a body area of ​​a patient P, so that the radiation hits a detector 4 opposite the radiation source 3.

[0075] Only those components of control unit 2 that are essential for explaining the invention are shown. X-ray systems 1, in general, do not require a detailed explanation.

[0076] The control unit 2 comprises a device 5 for improving X-ray images F, by means of a corresponding method such as is described, for example, in Fig. 2 is shown and includes a data interface 6, a segmentation unit 7, a selection unit 8, a registration unit 9, and a denoising unit 10 as the result unit 10.

[0077] Data interface 6 is used to receive or retrieve a series of X-ray images F, which represent successive images of the same region of interest. Additionally, data interface 6 is used to output the recorded X-ray images F or, preferably, if available, the noise-reduced image B.

[0078] The segmentation unit 7 is used to segment the X-ray images F by identifying structures S in the region of interest.

[0079] The selection unit 8 serves to select at least some of these structures S as reference structures R based on their security value.

[0080] The registration unit 9 serves to register the X-ray images F to each other, whereby the reference structures R are registered to each other according to a registration function X.

[0081] The denoising unit 10 (result unit 10) is used to create a noise-suppressed image B, preferably by time-averaged the registered X-ray images F, Fr.

[0082] Fig. Figure 2 shows, from left to right, a sequence of an embodiment of a method according to the invention for improving X-ray images F.

[0083] First, a series of X-ray images F is provided (left), e.g. after an X-ray examination following Fig. 1. These X-ray images F represent a series of temporally successive images of the same region of interest.

[0084] Then the X-ray images F of the series are segmented by identifying structures S in the region of interest (2nd image from the left).

[0085] After segmentation, a subset of these structures S is selected as reference structures R (3rd image from the left). In the example shown, the heart and aorta were selected as reference structures R.

[0086] The X-ray images F of the series are then registered against each other, with the reference structures R being registered against each other according to a registration function X. It should be noted that usually one of the X-ray images F remains unchanged, and the other X-ray images Fr are registered against it. The "registered X-ray images" F, Fr are then the one unchanged X-ray image F and the other registered X-ray images Fr.

[0087] These registered X-ray images F, Fr can then be output. In the case shown here, a denoised image B is generated from the registered X-ray images F, Fr and then output (right).

[0088] Fig. Figure 3 shows one possibility for creating a subseries T1 of X-ray images F. In this case, the aorta was no longer visible in some X-ray images F (cf. Figure 3). Fig. 2) Therefore, for a subseries T1, different structures S were selected as reference structures R. Now, a registration can be performed with these, which corresponds to the one in Fig. 2 shown corresponds to.

[0089] Fig. Figure 4 shows an alternative method for generating a T2 subseries of X-ray images F and registering them. Here, the structures S are so deformed that an automated process does not produce satisfactory results. Therefore, image registration is performed manually.

[0090] Here, it can be checked whether registration is still possible at the (temporal) boundaries of subseries T1 and T2. If so, subseries T1 and T2 can then be registered together.

[0091] Finally, it should be noted once again that the invention described in detail above merely represents exemplary embodiments, which can be modified in various ways by a person skilled in the art without departing from the scope of the invention. Furthermore, the use of the indefinite articles "a" or "an" does not preclude the possibility that the features in question may be present multiple times. Likewise, terms such as "unit" do not preclude the possibility that the components in question consist of several interacting sub-components, which may also be spatially distributed. The term "a number" should be read as "at least one." Regardless of the grammatical gender of a particular term, persons of male, female, or other gender identities are included.

Images