阅读说明：本技术 用于运行医学成像设备的方法和成像设备 (Method for operating a medical imaging device and imaging device ) 是由 A.格默尔 G.克莱因斯齐格 B.克雷赫 H.孔泽 J.马加拉基亚 S.施耐德 M.韦滕 于 2019-06-26 设计创作，主要内容包括：本发明涉及一种用于运行医学成像设备的方法和成像设备。为了使得能够在这种成像检查的过程中更好地提供图像,用于在进行成像检查时运行医学成像设备(1)的方法提供如下方法步骤：-提供(S1)身体区域的初始图像(2)；-拍摄身体区域的更新图像(3)；以及-借助事先经过训练的人工神经网络(5),由初始图像(2)和更新图像(3)产生三维的后续图像(4)。(The invention relates to a method for operating a medical imaging device and to an imaging device. In order to enable better image provision during such an imaging examination, the method for operating the medical imaging device (1) in the performance of an imaging examination provides the following method steps: -providing (S1) an initial image (2) of a body region; -taking an updated image (3) of the body region; and-generating a three-dimensional subsequent image (4) from the initial image (2) and the updated image (3) by means of a previously trained artificial neural network (5).)

1. Method for operating a medical imaging device (1) in the performance of an imaging examination, having the following method steps:

-providing (S1) an initial image (2) of a body region;

-taking an updated image (3) of the body region; and

-generating a three-dimensional subsequent image (4) from the initial image (2) and the updated image (3) by means of a previously trained artificial neural network (5).

2. Method according to claim 1, characterized in that a three-dimensional initial image (2) of the body region is provided as the initial image (2).

3. The method according to any of the preceding claims, characterized in that the updated image (3) represents a body region at a later point in time than the initial image (2).

4. Method according to any of the preceding claims, characterized in that an initial image (2) is generated during the computer tomography and an updated image (3) is taken by means of a moving X-ray device, in particular by means of a C-arm X-ray device.

5. Method according to any of the preceding claims, characterized in that in the generation of the three-dimensional subsequent image (4), the initial image (2) of the body region is at least partially updated from the updated image (3).

6. The method according to any of the preceding claims, characterized in that the initial image (2) and the updated image (3) characterize a body region during different phases (21, 22, 23) of the same surgical procedure (20).

7. The method according to claim 6, characterized in that the artificial neural network (5) is trained in a manner specific to one of the different phases (21, 22, 23).

8. Method according to any of the preceding claims, characterized in that structural changes on the body region occurring between the initial image (2) and the updated image (3) are determined and taken into account when generating the three-dimensional subsequent image (4).

9. The method according to claim 8, characterized in that a change of position of bone fragments and/or of a medical object on a body region is determined as the structural change.

10. Method according to claim 9, characterized in that for taking into account the structural change a change of position of bone fragments and/or of the medical object is determined and the change of position of bone fragments and/or of the medical object is taken into account by moving the bone fragments or the representation of the medical object in the initial image.

11. The method according to any of the preceding claims, characterized in that a three-dimensional subsequent image (4) is iteratively generated, wherein in a subsequent iteration step (S4) the three-dimensional subsequent image (4) is used as a new initial image (2) and, together with a new updated image (3), a new three-dimensional subsequent image (4) is generated.

12. Method according to any of the preceding claims, characterized in that a plurality of initial images (2) are provided and/or a plurality of updated images (3) are taken, wherein a three-dimensional subsequent image (4) is generated from the plurality of initial images (2) and/or the plurality of updated images (3).

13. The method according to any one of the preceding claims, characterized in that the artificial neural network (5) is trained at least partially by means of a test image (7), wherein the test image (7) contains an updated image (26) of a fracture, in particular of an earlier imaging examination, and/or a simulated updated image (25) generated from a three-dimensional representation of the fracture.

14. The method according to any one of the preceding claims, characterized in that the artificial neural network (5) is trained in a manner specific to the body region to be examined.

15. A medical imaging device (1) for performing imaging examinations, having:

-a providing unit (20) for providing an initial image (2) of a body region;

-a capturing unit (21) for capturing an updated image (3) of the body region; and

-an artificial neural network (5) trained according to a plan for generating a three-dimensional subsequent image (4) from the initial image (2) and the updated image (3).

Technical Field

The invention relates to a method for operating a medical imaging device during an imaging examination. A second aspect of the invention relates to a medical imaging apparatus.

Background

Examples of imaging devices are X-ray devices, computed tomography apparatuses and magnetic resonance tomography apparatuses. Thus, the imaging examination may comprise an X-ray examination, a computed tomography and/or a magnetic resonance tomography. Such an imaging examination is for example performed to generate an image of a body region of a patient. Such images may be used, for example, to perform surgery. The imaging examination is not particularly relevant here for surgery. For example, imaging examinations are performed before, after, and/or during the middle stage of a surgical procedure. The method particularly relates to the technical evaluation of raw or preprocessed data of a medical imaging device to provide an image of the mentioned body region.

Such an imaging examination may be used, for example, to support a surgical procedure for treating a bone fracture. In this case, the images can be generated by means of an imaging method before, after or during a surgical procedure for treating a bone fracture or fracture. By means of an imaging examination, the treating physician performing the surgery can obtain a spatial view of the fracture and/or the anatomy of the introduced medical object or implant. 3D shots taken during surgery are high overhead, associated with high radiation loads, and not possible for many body regions. Furthermore, in particular, the 3D X radiographic examination with a mobile X-ray device, in particular a C-arm device, is associated with a large time expenditure. Continuous 3D photographing cannot be performed.

The surgeon obtains a spatial view by acquiring individual two-dimensional projection images. In this case, the persuasiveness of the spatial view depends, in particular, on the respective projection direction of the individual projection images being selected by the physician in a smart manner. By repeating the production of such projection images for different projection directions and positions of the mobile X-ray device, the spatial view can be expanded or refined. How many projection images are needed for this is largely related to the ability and experience of the physician. Each additional projection image leads to an extension of the operative time of the surgical operation and to an increase of the radiation dose to the patient.

Disclosure of Invention

The object of the invention is to provide a better image representation during an imaging examination of this type.

According to the invention, the above technical problem is solved by the subject of the invention. Advantageous embodiments and suitable embodiments are the subject matter of the following description.

A first aspect of the invention relates to a method for operating a medical imaging system during an imaging examination, comprising the following method steps:

-providing an initial image of a body region,

-taking an updated image of the body region, an

-generating a three-dimensional subsequent image from the initial image and the updated image by means of a previously trained artificial neural network.

The initial image of the body region may be a projection image of the body region taken in advance. Thus, for example, three-dimensional subsequent images can be generated from the initial images, i.e. the projection images and the X-ray images, by means of an artificial neural network. In providing the initial image, a plurality of initial images may be provided, which may be processed in the same way as a single initial image during the course of the method. In particular, the plurality of initial images have different projection directions with respect to the body region.

The update image may be a two-dimensional projection image. The update image is in particular an X-ray image, advantageously a two-dimensional X-ray image. The X-ray images can be recorded by means of a mobile X-ray device, in particular by means of a C-arm device. Alternatively, the update image can be recorded by means of an ultrasound device, a magnetic resonance tomography apparatus, an optical camera, and any other imaging device. Accordingly, the update image may be, for example, an ultrasound image, a magnetic resonance tomography image, or a camera image.

In particular, the update images are not captured during the sequence of images performed to generate the three-dimensional representation. Thus, the plurality of updated images may be any images having any display of a body region.

The initial image may be taken by an imaging method different from the updated image or by an imaging device different from the updated image. In this case, it may be provided that the medical X-ray device receives an initial image from another medical imaging device and subsequently provides the initial image for further processing. The images taken with different imaging methods or different imaging devices, i.e. the initial image and the updated image, can then be further processed into three-dimensional subsequent images. In other words, a three-dimensional subsequent image is generated from the mentioned images. However, it is also possible to take the initial image and the updated image by means of the same imaging method and/or by means of the same imaging device. Accordingly, an initial image can be acquired by means of a mobile X-ray device, in particular a C-arm device. In this case, in particular a plurality of initial images are acquired, so that spatial information about the body region is already provided by the plurality of initial images.

The body region may for example be part of the (human) skeletal system. For example, the body region is a single bone, multiple bones, or a joint. Alternatively, the body region may be a soft tissue of the (human) body, such as the lung, liver, blood vessels, digestive tract. In the case of soft tissue, the diffusion of the contrast agent can be visualized by the method. In summary, with the aid of the method, for example, soft tissue, bones or joints can be examined. However, any other body area is also conceivable. Further, the method is not limited to the human body.

It may be provided that, in the method step of capturing the update image, a plurality of update images are captured. In particular, the plurality of update images are all processed in the same way as for a single update image. In particular the plurality of update images have different projection directions with respect to the body region.

The artificial neural network may be trained such that, by means of the artificial neural network, a three-dimensional subsequent image may be generated from the initial image and the updated image. The more comprehensive the training of the artificial neural network, the more accurate the three-dimensional subsequent images can be generated. In particular, the more complete or comprehensive the training of the artificial neural network, the fewer initial or updated images are required to generate a subsequent image in three dimensions. In training the artificial neural network, the artificial neural network may be provided with a priori knowledge about the generation of the three-dimensional representation of the anatomical structure from the updated image and/or the initial image. The artificial neural network may use this a priori knowledge to generate subsequent images in three dimensions. The three-dimensional subsequent image is in particular a three-dimensional representation of the body region. In particular, the artificial neural network is trained to use one or more updated images in combination with the initial image to generate a three-dimensional subsequent image, wherein the three-dimensional subsequent image and the one or more updated images do not have a predetermined relationship to one another. In the case of a plurality of update images, the artificial neural network can be trained to generate a three-dimensional subsequent image from the initial image and the plurality of update images, wherein the plurality of update images do not have a predetermined relationship with one another. The absence of a predefined relationship with respect to one another means, in particular, that the individual images are not part of a predefined image sequence for generating the three-dimensional representation, as described above. In other words, the training artificial neural network advantageously uses one or more initial images and one or more updated images of the body region in arbitrary relation to each other to generate a three-dimensional subsequent image.

It may be advantageous to use the shooting parameters of the updated image to generate the subsequent image. The update image is, for example, a projection image, in particular an X-ray image taken by means of a moving X-ray device, in particular a C-arm X-ray device. In this example, the shooting parameters may include one or more of the following parameters: shooting direction (angle, orbit and C-arm orientation), C-arm position, acceleration voltage, current intensity (especially tube current), charge amount (especially the product of current intensity and exposure time), and beam geometry. The artificial neural network may use these capture parameters to generate subsequent images in three dimensions. In particular, the artificial neural network can be trained or have corresponding a priori knowledge in order to use the acquisition parameters for better generation of the three-dimensional subsequent images.

According to one embodiment, a three-dimensional initial image of the body region is provided as the initial image. The three-dimensional initial image is in particular a three-dimensional representation of a body region. As mentioned above, the update image may be a two-dimensional update image, advantageously a two-dimensional X-ray image. It is therefore provided that, when a three-dimensional subsequent image is generated, a three-dimensional subsequent image is generated from the three-dimensional initial image and the two-dimensional updated image, in particular the X-ray image. In other words, according to this embodiment, for the generation of the three-dimensional subsequent image, either the three-dimensional initial image on the one hand and the two-dimensional X-ray image on the other hand are used or the three-dimensional initial image on the one hand and the two-dimensional X-ray image on the other hand are jointly processed. In this way, images of different dimensions (two-dimensional and three-dimensional) are enabled to be processed with respect to each other by means of an artificial neural network.

According to one embodiment, it is provided that the X-ray images represent a body region at a later point in time than the initial images. In other words, the updated image may be more up to date than the initial image. In this way, the information content of the initial image and the updated image can be mixed with each other particularly advantageously. For example, the artificial neural network may merge information about the spatial structure of the body region from the initial image with current position information, e.g. of bones or medical objects, from the updated image. In this way, it is possible to generate a three-dimensional subsequent image based on one or several update images with particular advantage.

According to one embodiment, an initial image is generated during the computer tomography and an X-ray image is recorded by means of a mobile X-ray device, in particular by means of a C-arm X-ray device. In other words, in this case, the update image is an X-ray image. In this example, the initial image is a three-dimensional representation or a three-dimensional reconstruction of the body region. Computed tomography imaging, for example, can characterize a body site prior to a surgical procedure. In contrast, the updated image may characterize a body region during a surgical procedure or during an intermediate stage of a surgical procedure. Thus, an initial image representing the body region prior to the surgical procedure may be merged with an updated image representing the body region during the surgical procedure to produce a subsequent image in three dimensions. In this way, on the one hand, comprehensive earlier three-dimensional information can be combined with the current two-dimensional update image, in particular the X-ray image. Thereby, by evaluating the three-dimensional initial image, the need for a current updated image for generating a three-dimensional subsequent image being a current three-dimensional representation of the body-part may be reduced.

According to one embodiment, it is provided that, during the generation of the three-dimensional subsequent image, the, for example, three-dimensional initial image of the body region is at least partially updated on the basis of the X-ray image. For example, the three-dimensional representation provided by the initial image is at least partially updated according to the more current updated image. In this case, the current image information can be combined with the overall three-dimensional information of the initial image on the basis of a priori knowledge of the artificial neural network trained beforehand. The three-dimensional subsequent image of the body region or the three-dimensional representation provided by the three-dimensional subsequent image may be updated from the one updated image or from the plurality of updated images. For example, structural changes relative to the initial image are identified from the updated image, and the initial image used to generate the three-dimensional subsequent image is adjusted accordingly. In this way, a reliable current three-dimensional representation can be provided in the form of a three-dimensional subsequent image, even with several update images.

In particular, it is provided that the initial image and the updated image represent body regions during different phases of the same surgery. For example, it is provided that an initial image is taken during a preceding phase of the surgery for comparison with the updated image. The initial image may be acquired, for example, during a preliminary examination prior to the start of a surgical procedure. Subsequently, updated images may be taken during the surgical procedure. Examples of different stages of surgery for treating bone fractures are: initial examination for the judgment of the injury and selection of the entry point to the fracture, reduction of the fracture, clamping of the reduced bone fragments and finally fixation of the bone fragments by means of screws and/or punches (Locheisen). Thus, in one particular example, an initial image may be generated during a preliminary examination, and an updated image may be taken during the resetting, clamping, and/or securing. For example, during different ones of the exemplarily mentioned phases, a plurality of update images is taken. In this way, images of the first several stages of the surgery may be used to generate or provide subsequent images of the body region for later stages of the surgery. In this case, the artificial neural network is trained to use such updated images to generate subsequent images.

According to one embodiment, it is provided that the artificial neural network is trained in a manner specific to one of the different phases. For example, for multiple ones of the different phases, respective trained artificial neural networks are provided. For example, an artificial neural network trained in a manner specific to the reset phase is provided. For example, another artificial neural network trained in a manner specific to the clamping and/or fixation phase is provided. For example, in an additional method step, an artificial neural network trained for the currently existing phase may be selected from a plurality of artificial neural networks. In this way, the respective a priori knowledge generated by training the artificial neural network can be matched particularly well to the respective phase.

According to one embodiment, it is provided that structural changes in the body region occurring between the initial image and the updated image are determined and taken into account when generating the three-dimensional subsequent image. For example, changes that occur in the updated image compared to the initial image are determined on a targeted basis. This may then be taken into account when updating the initial image from the updated image. In particular, structural changes can be tracked by moving image portions in the initial image. This may be done, for example, in a motion compensated manner for motion occurring over the body region between the initial image and the updated image.

According to one embodiment, a change in the position of the bone fragments and/or of the medical object on the body region is determined as a structural change. Examples of medical objects are clips, punches and screws for fixing bones or bone fragments in body regions. Bone fragments and/or medical objects may move during a surgical procedure and/or be disposed on a body region. The mentioned position changes may occur as a result. The change in position is determined during the course of the method. Moving bone fragments and/or introducing medical objects is explicitly not part of the claimed method here. The different stages of the surgery, its execution and the surgery itself are clearly not part of the claimed method. The artificial neural network may take into account changes in the position of bone fragments and/or medical objects to generate a three-dimensional subsequent image from the initial image. In particular, the change in position is taken into account in order to update the three-dimensional initial image by means of an artificial neural network.

According to one embodiment, it is provided that, in order to take into account structural changes, a change in the position of the bone fragments and/or of the medical object is determined and the change in the position of the bone fragments and/or of the medical object is taken into account by moving the bone fragments or the representation of the medical object in the initial image. In other words, the change in position is taken into account by the movement of bone fragments or a representation of the medical object in the initial image. That is, a representation of a bone fragment or a medical object may be moved in the initial image. In particular, the movement can be carried out in accordance with a predetermined change in position. In other words, the bone fragments or the representation of the medical object in the initial image may be moved to the current position with respect to the body region acquired by the updated image. This is advantageous in particular when the initial image is a three-dimensional initial image. If the initial image is a two-dimensional initial image, the generation of a three-dimensional subsequent image from the initial image and the updated image is only enabled by the updating of the initial image, since the position of the bone fragments and/or the medical object with respect to the body region must be consistent in the initial image and the updated image in order to generate the three-dimensional representation of the body region.

According to one embodiment, it is provided that a three-dimensional subsequent image is iteratively generated, wherein in a subsequent iteration step the three-dimensional subsequent image is used as a new initial image and is generated together with a new updated image. In other words, it can be provided that the update images are repeatedly captured and that a respective three-dimensional subsequent image is generated from the respective update image and the respective initial image. In particular, each three-dimensional subsequent image is used iteratively as a new initial image to generate a corresponding subsequent three-dimensional subsequent image. The accuracy of the three-dimensional information provided by the three-dimensional subsequent image or the three-dimensional subsequent image can thus be iteratively improved by means of the successive update images.

According to one embodiment, provision is made for a plurality of initial images to be provided and/or for a plurality of X-ray images to be recorded, wherein three-dimensional subsequent images are generated from the plurality of initial images and/or the plurality of X-ray images. In other words, as the initial image, a plurality of initial images may be provided. Alternatively or additionally, as the update image, a plurality of update images may be taken. Then, a three-dimensional subsequent image may be generated from the plurality of initial images and the one updated image, from the one initial image and with the plurality of updated images, or from the plurality of initial images and the plurality of updated images. For example, in one iteration step of the iterative method, as update images, a plurality of update images are captured and a three-dimensional subsequent image is generated therefrom in one iteration step. In the following iteration steps, a plurality of updated images can again be recorded and new three-dimensional subsequent images can be generated therefrom. In the case of a plurality of update images, these are advantageously taken during the same phase of the surgery accordingly. In this way, the processing overhead in the case of multiple update images and multiple initial images can be reduced.

According to one embodiment, it is provided that the artificial neural network is trained at least partially by means of the test image. The test image may for example comprise an updated image of the fracture, in particular an earlier imaging examination (in particular an X-ray image of an earlier X-ray examination), and/or a simulated X-ray image generated from a three-dimensional representation of the fracture. Additionally, the test image may comprise an updated image, in particular an X-ray image, of an artificially created fracture on a human body, in particular open for medical examination. For example, an artificial fracture is created by breaking the bone and the process is acquired with a plurality of updated images. The corresponding updated image can show the fractured bone in the unbroken state and with different degrees (e.g. one or more fractures).

Additionally, the test images may be associated with corresponding shooting parameters (see above). An artificial neural network may then be trained based on the test images and associated shooting parameters. The artificial neural network may be trained, with the test images and associated capture parameters, to use the capture parameters of the updated images to generate subsequent images in three dimensions.

The updated images of the earlier imaging examinations, in particular the X-ray images, can each show the respective fracture at different times, i.e. in the untreated state, during intermediate steps of the respective surgery and after completion of the respective surgery.

For example, a simulated updated image, in particular a simulated X-ray image, can be simulated or generated from the three-dimensional representation of the fracture. The three-dimensional representation of the fracture may be a computed tomography image. By means of computed tomography, a two-dimensional projection of the respective simulated update image can be realized particularly simply and comprehensively. In this way, a simulated updated image of a particularly large data volume can be generated with little overhead. The simulated updated image may be generated from a plurality of three-dimensional representations of the fracture, which show fractures of different periods, e.g., unbroken and with different degrees of fracture.

By the test images showing different degrees of fracture and corresponding updated images of unbroken or completely healed bone, the artificial neural network can additionally learn in which way the fracture is treated during the surgical procedure. In this way, bone fragments and/or changes in the position of the medical object can be acquired and/or interpreted particularly well by the artificial neural network. In this way, three-dimensional subsequent images can be generated particularly reliably. In summary, it is shown how comprehensive and convincing test data can be provided to train artificial neural networks by different methods for providing test images.

According to one embodiment, it is provided that the artificial neural network is trained in a manner specific to the body region to be examined. For example, the respective artificial neural networks are trained for different body regions. An artificial neural network trained for the current body region may then be selected based on the body region currently to be examined. For example, the body regions are the knee, tibia, arm joints, arm bones, and shoulder. The artificial neural network can then be trained in a manner that is exactly specific to one of these exemplarily mentioned body regions. In this way, the a priori knowledge for generating the three-dimensional subsequent images can be matched particularly well to the body region to be examined.

A second aspect of the invention relates to a medical imaging apparatus for performing imaging examinations, having

A providing unit for providing an initial image of a body region,

a capturing unit for capturing an updated image of the body region, an

According to the plan

And a trained artificial neural network for generating a three-dimensional subsequent image from the initial image and the updated image. The medical imaging device is advantageously configured for performing a method for performing an imaging examination, which method has the features described within the scope of the present application. The providing unit may have a storage unit for storing the initial image and/or a receiving unit for receiving the initial image. For example, the medical imaging device is designed to take an initial image with the aid of a recording unit or to receive an initial image from another medical imaging device. The medical imaging device is in particular a mobile X-ray device, in an advantageous manner a C-arm device.

The features of the method for operating a medical imaging system disclosed within the scope of the present application thus also extend the medical imaging system here. In particular a medical imaging device, has corresponding components configured for carrying out the method steps and features of the method.

Drawings

The invention will now be explained in detail with the aid of a number of figures. Here:

fig. 1 shows a schematic view of a medical imaging device;

fig. 2 shows a schematic overview of the time course of an exemplary embodiment of the method; and

fig. 3 shows a flow chart of an exemplary method.

Detailed Description

Fig. 1 shows a medical imaging device 1, here a so-called C-arm X-ray device. Here, the imaging device 1 has an X-ray source 13 and a detector 14. Furthermore, the imaging device 1 comprises a capturing unit 12 for capturing the update image 3. In this embodiment the update image 3 is an X-ray image. Additionally, the imaging apparatus 1 comprises a control device 10 with a provision unit 11 and an artificial neural network 5. In the ready-to-operate state of the medical imaging device 1, the artificial neural network 5 is trained according to the plan.

If the medical imaging device 1 is brought into a planned relative position with respect to the patient to be examined, an updated image 3 of the body region of the patient can be taken.

Fig. 2 shows an exemplary time sequence of a method for operating the medical imaging device 1. The respective update images 3 are taken during respective different phases 22, 23, 24 with respect to the surgical procedure 20. The different stages 22, 23, 24 relate to the same surgical procedure 20. The surgery 20 is also shown on the time axis T, however the surgery 20 is not part of the method. Surgery 20 may be provided, for example, to treat a fracture of a patient to be examined. The status of the fracture and/or the treatment of the fracture can be determined or checked by means of an imaging examination. Stages 22, 23, 24 may be intermediate stages of the surgical procedure 20. Here, during a respective intermediate phase of the surgical procedure 20, the update image 3 is captured by the medical imaging device 1. In a first phase 21, the so-called preliminary examination phase, before the actual surgery 20 begins, a three-dimensional initial image 2 or a plurality of two-dimensional initial images 2 is generated or recorded. For the purpose of distinguishing from the following initial image 2, the initial image 2 or initial images 2 of the first stage 21 are referred to as a first initial image 6 or first initial images 6. With the initial image 2 and the updated image 3, the method enables a better overview of the body region. The surgical procedure 20 is specifically not part of the present method.

In the case of a three-dimensional first initial image 6, this can be acquired by means of a computed tomography apparatus. In this case, the imaging apparatus 1 or the provision unit 11 receives the first initial image 6 from the computed tomography apparatus. In this case, the reception can also take place indirectly, for example by means of a data carrier for temporary storage. For this purpose, the provision unit 11 may have an interface, for example a USB connection or a network connection. In the case of a plurality of first initial images 6, these can be captured by means of the imaging device 1. In the case of a plurality of first initial images 6, these first initial images may still be received from another imaging device. The plurality of initial images 6 are in this case corresponding two-dimensional X-ray images recorded by means of the recording unit 12. A plurality of first initial images 6 or one first initial image 6 may be received from the photographing unit 12 by the providing unit 11 and temporarily stored. The first stage 21 may be a preliminary examination performed shortly before the beginning of the surgical procedure 20. In other words, the surgical procedure 20 may be performed directly after the first stage 21. Thus, the plurality of first initial images 6 or one first initial image 6 represents the body region at the beginning of the surgical procedure 20.

During the preliminary examination or first stage 21, for example, an appropriate entry point for the surgical procedure 20 may be determined.

The second stage 22 may interrupt or pause the surgical procedure 20 in time. During the second stage 22, a plurality of update images 3 of the body region, here X-ray images, are taken by means of the camera unit 12. A three-dimensional subsequent image 4 is iteratively generated from each updated image 3 and the previous one of the initial images 12. The subsequent image 4 generated is used accordingly as the initial image 2 for generating the subsequent image 4. This can be seen best by means of the arrows in fig. 2. In particular, a first subsequent image 4 of the subsequent images 4 is generated from a first initial image 6 or a plurality of first initial images 6 and a first updated image 3 of the updated images 3. A second subsequent image 4 of the subsequent images 4 is also generated from the second update image 3 of the update images 3 and the first subsequent image 4 of the subsequent images 4. Thus, the first subsequent image 4 of the subsequent images 4 is used as an initial image for generating the second subsequent image 4 of the subsequent images 4. Here, the expressions "first" and "second" should be understood with respect to the time axis T. Thus, the second subsequent image 4 of the subsequent images 4 represents a body region at a later point in time than the first subsequent image 4 of the subsequent images 4.

This is also shown in fig. 3. In step S1, the artificial neural network 5 is provided with the corresponding initial image 2. The artificial neural network is trained in a previous step S0, which is described in more detail below. In step S2, a corresponding update image 3 is taken and fed to the artificial neural network 5. In step S3, the artificial neural network 5 generates a corresponding subsequent image 4 from the initial image 2 and the updated image 3. In an iteration step S4, the resulting three-dimensional subsequent image 4 is used as a new initial image 2. The new initial image 2 may in turn be fed together with a new updated image 3 to an artificial neural network 5 for generating a new subsequent image 4.

After the second stage 22, the surgical procedure 20 (fig. 2) may continue. After the end of the surgical procedure 20, there is a third stage 23. Accordingly, the updated image 3 of the third stage 23 characterizes the body region at the end of the surgical procedure 20. The updated image 3 of the second stage 22 characterizes the body region during the intermediate stage of the surgical procedure 20. The generation of the three-dimensional subsequent image 4 in the third stage 23 is performed similarly to the generation of the three-dimensional subsequent image 4 in the second stage 22. In the generation of the respective first subsequent image 4 of one of the phases 22, 23, the initial image 2 from the respective previous phase 21, 22 is used. To generate the first subsequent image 4 of the second stage 22, one first initial image 6 or a plurality of first initial images 6 from the first stage 21 is used. To generate the first subsequent image 4 of the subsequent images 4 of the second stage 23, the subsequent image 4 from the second stage 22 is used as the initial image 2. Thus, in the mentioned case, the respective subsequent image 4 is generated from a respective one of the initial images 2 and a respective one of the updated images 3 characterizing the body region during the different phases of the surgical procedure 20.

During the surgical procedure 20, structural changes may occur on the body region. In the case of a fracture, such a structural change may occur, for example, as a result of a change in the position of bone fragments and/or of the medical object on the body region. For example, during surgery 20, bone fragments are moved relative to one another or aligned with one another in order to treat a bone fracture. Alternatively or additionally, during the surgical procedure 20, the medical object may be placed on a body region or moved. Examples of medical objects are clips, screws and punches for fixing bone fragments. With respect to the time axis T in fig. 2, the mutual alignment of the bone fragments can be set during the medical procedure, for example between the first stage 21 and the second stage 22. In this case, the bone fragments can be reduced by the treating physician. Between the second stage 22 and the third stage 23, the introduction of a medical object for fixing bone fragments can be provided. The movement of bone fragments and the fixation of bone fragments is definitely not part of the method. However, the present method enables the treating physician to prepare the movement of bone fragments and/or the progression of the placement of the medical object in a visible manner. In this case, the spatial representation of the body region can be provided to the treating physician in the form of a subsequent image 4. The three-dimensional subsequent image 4 may be a spatial representation of a body region. In particular, for creating the three-dimensional subsequent image 4, the updated images 3 taken in the different stages 22, 23, 24 of the surgery 20 are used. In other words, a three-dimensional subsequent image 4 can be formed from the updated image 3 captured in the different stages 22, 23, 24 of the surgical procedure 20.

In order to be able to generate the three-dimensional subsequent image 4 by means of the artificial neural network 5, the artificial neural network must first be trained in step S0. Here, the a priori knowledge is generated by an artificial neural network, which is able to produce a three-dimensional subsequent image 4 from the a priori knowledge. Without generating this a priori knowledge, the three-dimensional subsequent image 4 cannot be easily produced, since an underdetermined (unsubscribed) system is involved in the case of few updated images 3 or very different updated images 3 (due to structural changes during the surgical procedure 20). Here, "underdetermined" means that too few different projection directions are provided by updating the image 3. This undercharacterization of the system with respect to the updated image and the initial image can be compensated by a priori knowledge of the artificial neural network. In an advantageous manner, the a priori knowledge of the artificial neural network 5 relates to the general structure of the body region. This is based on the following considerations: specific body regions of multiple persons have a large similarity.

The artificial neural network 5 is trained with the aid of training data, which in particular comprise test images 7. The training data or test image 7 may for example comprise a simulated updated image 25, an updated image 26 of an earlier surgical or earlier imaging examination and an updated image 27 of an artificially broken bone of the body released for a medical examination or study. The three-dimensional representation used to generate the simulated update image 25 may be provided, for example, by computed tomography imaging. Such computed tomography images of an artificial break are performed, for example, for the updated image 27 or according to an earlier imaging examination.

For the different update images 25, 26, 27, it is accordingly advantageous to represent the respective fracture by the respective update image 25, 26, 27 in the different periods. The different periods may for example relate to different degrees of misalignment of non-fractured bones, single-fractured bones and multiple-fractured bones or bone fragments. For example, respective sets of updated images 25, 26, 27 are generated for bones that are not fractured or reduced, for bones that are fractured once, for bones that are fractured multiple times, and for different degrees of misalignment of bone fragments. In this manner, the artificial neural network 5 may also be trained with respect to the progression of structural changes on the body region during the course of the surgical procedure 20.

The training data or test image 7 may be specific to a particular body region. Thus, the artificial neural network 5 may be trained in a manner specific to a particular body region. Examples of body regions that the training data or test image 7 may represent are: the knee, tibia, arm joint or antecubital fossa, forearm bone and shoulder, and the artificial neural network 5 can be trained for these body areas. This list should be understood as non-limiting. For the imaging examination, an artificial neural network 5 can be selected from a plurality of artificial neural networks, which is dedicated to the examination of the body region.

It is provided that structural changes on the body region occurring between the initial image 2 and the updated image 3 are determined by the artificial neural network 5. This structural change is taken into account in order to generate a three-dimensional subsequent image 4. This consideration is made in particular by determining a change in position of bone fragments and/or medical objects between the initial image 2 and the updated image 3. If there is such a change in position, the change in position is taken into account by moving the corresponding bone fragments or the corresponding representation of the medical object in the initial image 2. In other words, when in the updated image 3 the position of the corresponding bone fragments and/or medical object changes, the bone fragments or the representation of the medical object in the initial image may be moved by the artificial neural network 5. This may be understood as motion compensation, in which the motion of bone fragments and/or of the medical object is compensated by shifting the corresponding representation in the initial image 2. In this way, the position of the bone fragments or medical objects displayed in the initial image 2 can be matched to the more current updated image 3.

In summary, it is shown by way of example how a better representation of a body region can be provided in the form of a three-dimensional follow-up image.