Method for determining a three-dimensional target image data set

Abstract

Method for determining a three-dimensional target image data set (20) showing at least one area of ​​interest (10, 23) of an image area (9), wherein the image data of the three-dimensional target image data set (20) are reconstructed from two-dimensional projection images (2, 8) recorded under different projection directions, characterized by the following steps: - Acquisition of first projection images (2) without collimation of the radiation source (33) used from first projection directions, and reconstruction of a three-dimensional overview image dataset (4) of the recording area (9) from the first projection images (2), - Selection of the relevant sub-area (10, 23) in the overview image dataset (4), - Recording of second projection images (8) from second projection directions, collimated onto the sub-area (10, 23), wherein the second projection directions differ from the first projection directions, - Reconstruction of the target image data set (20) showing the recording area (9) and the sub-area of ​​interest (10, 23) from all first and second projection images (2, 8), - wherein, as second projection directions, projection directions lying at least partially between two first projection directions along the same recording trajectory (11) are chosen.

Description

[0001] The invention relates to a method for determining a three-dimensional target image data set showing at least one part of interest of a recording area, wherein the image data of the three-dimensional target image data set are reconstructed from two-dimensional projection images recorded under different projection directions.

[0002] Methods for reconstructing a three-dimensional image dataset from two-dimensional projection images acquired from different projection directions are already well-established in the art. For example, it is possible to process the two-dimensional projection images using iterative methods or filtered backprojection techniques to obtain a three-dimensional reconstruction image dataset. Such approaches are typically used in areas where spatial information is also required. In addition to computed tomography (CT), computed tomography-like methods for C-arm X-ray systems and tomosynthesis are also known.This means that computed tomography methods have also been used for some time on other X-ray systems, especially flat detector X-ray systems in interventional and diagnostic settings, to enable high-quality volumetric imaging.

[0003] The achievable three-dimensional image quality within the field of view (FOV) depends on the reconstruction method used - however, the image quality is theoretically limited in principle by the choice of the recording parameters, in particular the recording trajectory with the recording positions and the resolution of the X-ray detector used, i.e. by the number of projection images and the pixel size, as well as by the X-ray dose used.

[0004] Therefore, in the state of the art, it is common practice to select these acquisition parameters accordingly when designing a measurement protocol in order to obtain sufficient 3D image information for meaningful evaluation. For example, it is known to acquire approximately 400 projection images using 2x2 binning for applications with C-arm X-ray systems with the aim of achieving good low-contrast resolution. For screening in breast tomosynthesis, a very high pixel resolution, and thus no binning, is typically chosen during acquisition. The images provide the desired high image quality across the entire acquisition area. However, a disadvantage of such computed tomography-like methods is the high radiation exposure to which the patient is subjected by acquiring the many projection images in the acquisition area.

[0005] The problem here is that it is often unnecessary to obtain the entire three-dimensional field of view with the same image quality. Often, only specific areas of interest need to be evaluated in detail, such as tumors within tissue, cartilage between joint bones, or fractures. These areas should be reproduced with the highest possible image quality, whereas a lower quality would suffice for the surrounding area. However, the precise location of these areas of interest within the image volume is frequently unknown before the scan begins.

[0006] To enable a more detailed examination of a specific area of ​​interest, methods addressing this problem have already been proposed in the prior art. If a three-dimensional image dataset is calculated from an acquired sequence of projection images, maintaining a consistently high quality across the entire recording area, then, if only a sub-area is of interest, a (digital) zoom function can be used in the visualization to magnify that sub-area. In this way, details can be visually highlighted, although the (digital) zoom does not provide any additional image information.

[0007] In an article by D. Kolditz et al., “Volume-of-interest (VOI) imaging in C-arm flat-detector CT for high image quality at reduced dose”, Med. Phys. 37 (6), June 2010, pages 2719–2730, it is proposed to first perform a low-dose scan from which a three-dimensional reconstruction image dataset is reconstructed, in which the region of interest is localized. Subsequently, a second scan is performed along the same scanning path (acquisition trajectory), but with a higher X-ray dose and with collimation onto the region of interest, so that the radiation is focused only on the localized region of interest (often also referred to as the region of interest – ROI).The final image result, i.e. the target image dataset, is then reconstructed from the projection images of the second scan, whereby forward-projected images from the reconstruction dataset can be used to adjust the projection images of the second scan in order to avoid truncation effects.

[0008] The invention is therefore based on the objective of providing a way to achieve a further increased image quality in selective sub-areas of a recording area identified as important, without exposing the object being recorded to an excessively high dose of radiation.

[0009] To solve this problem, the following steps are provided according to the invention in a method of the type mentioned at the outset: - Acquisition of initial projection images without collimation of the radiation source used from initial projection directions, and reconstruction of a three-dimensional overview image dataset of the recording area from the initial projection images, - Selection of the relevant sub-area in the overview image dataset, - Recording of two projection images from second projection directions, collimated onto the sub-area, where the second projection directions differ from the first projection directions, - Reconstruction of the target image dataset showing the recording area and the sub-area of ​​interest from all first and second projection images, - where, as second projection directions, projection directions lying at least partially between two first projection directions along the same recording trajectory are chosen.

[0010] The invention proposes a method for improving image quality in a target image dataset, which depicts the entire acquisition area, by ultimately achieving an incremental gain in image information within the selected area of ​​interest in the current acquisition area. This gain is achieved through further scans collimated only on the area of ​​interest, i.e., the second projection images. A fundamental idea is to acquire the new, second projection images from projection directions from which no projection image, i.e., no first projection image, has previously been acquired, for example, from intermediate angular positions. This ensures that the second projection images actually provide new information about the object of investigation, specifically the area of ​​interest.

[0011] The calculation of the current target image dataset always incorporates all projection data, i.e., all first and second projection images. During the imaging process, this results in a dose-saving image in which the selected area of ​​interest is presented in very high quality, particularly with high resolution and / or high contrast, low noise, and few artifacts, while the surrounding area, less relevant for evaluation, is presented in a lower image quality.

[0012] The method according to the invention can therefore be understood as a kind of “magnifying glass”, which, in contrast to conventional magnifying glass or zoom functions, not only produces a digital enlargement / improvement of the image content in the area of ​​interest, but actually delivers a higher image quality by capturing new information - the second projection images - and taking it into account in the calculation.

[0013] Compared to the prior art, the described method therefore offers the following advantages. Firstly, it results in a very low overall dose in the acquisition area compared to a conventional method in which the image quality is set to a uniformly high level across the entire acquisition area. Furthermore, all acquired projection data are relevant for the reconstruction of the final image dataset; this means that projection data from projection directions that have already been considered are never acquired, thus avoiding the acquisition of redundant data. Finally, the method according to the invention generates a target image dataset in full size but with variable quality. This allows for both a rough orientation and very high image quality in selected sub-areas.

[0014] According to the invention, the selection of the sub-area can be at least partially automated, particularly based on segmentation of the overview image dataset. It is therefore conceivable to apply evaluation algorithms, especially segmentation algorithms, optionally using further information such as an anatomical atlas, to the overview image dataset in order to determine the sub-area of ​​interest, preferably fully automatically, for example, based on predefined examination information. A semi-automatic approach is also conceivable, in which various sub-areas are segmented from the overview image dataset and offered to a user for selection.

[0015] However, it is also conceivable that the selection of the area of ​​interest is done entirely manually, for example by a user using an input device, such as a mouse, to directly mark the target area of ​​interest in a representation of the overview image data set.

[0016] In all these cases, the initial projection images serve to build an overview image dataset that provides the necessary orientation to enable the selection of the area of ​​interest.

[0017] In a further embodiment of the present invention, it can be provided that, in addition to collimation, at least one further recording parameter is adjusted when capturing the second projection images. It is therefore conceivable not only to perform collimation on the area of ​​interest, as described, but also to adjust further recording parameters in such a way as to achieve the highest possible quality of the additional information obtained from the second projection images. In particular, it can be provided that the adjustment of the further recording parameter is aimed at increasing the resolution and / or contrast.

[0018] Various adjustments to other acquisition parameters are conceivable. For example, the X-ray dose could be increased. This could, for instance, contribute to increased contrast in the area of ​​interest. Furthermore, pixel binning could be reduced. If pixels from the X-ray detector are combined in the previously acquired projection images, for example, to accelerate readout and / or evaluation, this binning can be reduced. This means combining fewer pixels or even eliminating them entirely when the second projection images are acquired, resulting in higher resolution in the area of ​​interest. Another example would be to adjust the X-ray spectrum to be suitable for capturing the area of ​​interest.The X-ray spectrum is therefore also one of the imaging parameters that can be adjusted, with different X-ray spectra being particularly suitable for soft tissues, bones, contrast agents and other target substances of interest.

[0019] Advantageously, fewer than 100, in particular 15–25, first projection images can be acquired. In this embodiment, therefore, only a very small number of first projection images, covering a correspondingly small number of projection directions, are obtained, resulting in a very low overall X-ray dose compared to a conventional procedure. It is deliberately accepted that the reconstructed overview image dataset may contain artifacts and inaccuracies; however, such an overview image dataset nevertheless allows for at least a rough localization of the area of ​​interest, in particular at least partially automatically and / or by a user. In extreme cases, it is conceivable to acquire only ten first projection images.

[0020] Second projection directions are chosen that lie at least partially between two first projection directions along the same recording trajectory. It is therefore possible to use the same recording trajectory, for example, a circular path, for recording both the first and second projection images, with the second projection images ultimately filling "gaps" between the recording positions of the first projection images. For example, a recording position for second projection images can always be positioned midway between two recording positions of the first projection images. Alternatively, the "gaps" can be filled with multiple second projection images using corresponding projection directions.

[0021] In a further advantageous embodiment of the method, a reconstruction image dataset can be determined from the second projection images to generate the target image dataset. The target image dataset is then created by a weighted, pixel-by-pixel addition of the image data from the overview image dataset and the reconstruction image dataset. Such additive superimposition is particularly useful when the overview image dataset and the reconstruction image dataset were generated using filtered backprojection algorithms, which, as is known, process the projection data linearly. In this case, any remaining line artifacts in the overview image dataset and / or the reconstruction image dataset are reduced by undersampling. Ultimately, the overview image dataset is updated as soon as new second projection images are acquired.The selected area of ​​interest will be better represented in the target image dataset than in the overview image dataset, but also better than in the reconstruction image dataset, since the entirety of the projection image information available so far is included in the calculation.

[0022] As previously described, weighting can be considered here. Specifically, pixel-dependent weights can be used to normalize the reconstruction. This means that weighting factors can be applied that allow normalization of the reconstructions for each pixel x. If, as described above, the target image dataset is determined using weighted addition, the following can be written: f=a1(x) f1(x)+a2(x) f2(x).

[0023] Here, f denotes the image data of the target image dataset, f1 the image data of the overview image dataset, and f2 the image data of the reconstruction image dataset; a1 and a2 are the weights.

[0024] In a practical, simple choice of weighting factors, it can be provided that the weighting factors a(x) for each pixel x and for both the overview image dataset and the reconstruction image dataset are defined as a(x)=(N1+N2) / (w1(x)+w2(x)) This results in a weighting where N1 denotes the total number of first projection images, N2 the total number of second projection images, w1(x) the number of first projection images in which the pixel x is located in the radiation area illuminated by X-rays, and w2(x) the number of second projection images in which the pixel x is located in the radiation area illuminated by X-rays. In this case, a1(x) = a2(x) = a(x). This achieves effective normalization. Of course, other definitions of the weighting are also conceivable in principle.

[0025] In an alternative embodiment of the present invention, it is also possible to determine the target image data set by a complete reconstruction taking into account the projection data of all first and second projection images. Instead of adding individually reconstructed three-dimensional image data sets, it is therefore also conceivable to perform a complete reconstruction.

[0026] In a further advantageous embodiment of the present invention, it can be provided that a truncation correction is applied to the second projection images and / or a reconstruction image dataset reconstructed therefrom. Since the second projection images generally exhibit strong lateral truncation due to the selection of the smaller area of ​​interest, a truncation correction is proposed for the reconstruction of the target image dataset in general, or of the reconstruction image dataset in the first embodiment described above. Specifically, it can be provided that the truncation correction is performed on an extrapolation of projection image data, taking the overview image dataset into account.Since the overview image dataset shows the entire acquisition area, it is possible to extrapolate the projection image data of the second projection images so that they correspond to the image data of the overview image dataset in order to enable good truncation correction. Less preferred, it is also conceivable to perform model-based data extrapolation or to use another truncation correction method to reduce truncation artifacts, for example, a correction method proposed in the article by F. Dennerlein, “Conebeam ROI reconstruction using the Laplace operator”, Proc. Fully 3D 2011, pages 80–83, 2011.

[0027] In a particularly advantageous embodiment of the present invention, it can be provided that a new sub-area of ​​interest, contained in particular within the previously relevant sub-area, is selected as a new overview image dataset within the image dataset. Following this selection, after acquiring new second projection images and collimating them onto the new sub-area of ​​interest, a new target image dataset is reconstructed. This means that the inventive procedure can be repeatedly applied to different sub-areas of interest, particularly with progressively increasing refinement, so that the image quality can be continuously improved in successively defined sub-areas. This corresponds to a kind of incremental magnifying function, whereby further additional information from projection directions that have not yet been scanned is always obtained for the currently selected sub-area of ​​interest.In this way, the image quality can be improved to the desired degree. It is advantageous if the new area of ​​interest is completely contained within the existing area of ​​interest. Therefore, the method according to the invention can ultimately be repeated until the desired image quality is achieved within the area of ​​interest, particularly for a feature relevant for evaluation within the acquisition area. If, as described, progressively smaller areas of interest are selected, the collimation is enhanced with each new scan of two projection images, thus successively reducing the additional dose.

[0028] It should be noted here that, since several recording processes, at least two, are required, it is advantageous to avoid movement of the object being recorded, especially a patient, or at least of the recording area, as much as possible. This can be achieved, for example, by securely fixing the patient or the recording area.

[0029] However, it is also conceivable to implement motion correction within the scope of the present invention. For example, it can be provided that, to correct a movement, the overview image dataset is registered with the reconstruction image dataset, which is determined from the second projection images. In this way, registration information can be obtained that enables the compensation of patient movements between scan runs. It is also possible to initiate registration directly in the projection image space, which can then be taken into account during the reconstruction of image datasets, particularly in the context of backprojection.For this purpose, it may be provided that one or more projection images of the first and second projection images are recorded under the same projection directions; that is, it may be specifically provided that at least one pair of first and second projection images is recorded under the same projection direction, whereby registration information is determined by comparing first and second position images of the same projection direction.

[0030] An X-ray device comprising a control unit designed to carry out the method according to the invention is also conceivable. Such an X-ray device could, for example, be a C-arm X-ray device, with which computed tomography-like projection images can be acquired at different projection angles and three-dimensional image data sets can be reconstructed. It is also conceivable to design the X-ray device as a tomosynthesis X-ray device, with which, for example, the breast and lungs can be examined. All aspects relating to the method according to the invention can be applied analogously to the X-ray device, with which the same advantages can therefore be achieved.

[0031] Such a control unit, i.e., the X-ray unit, can therefore include a reconstruction unit designed to reconstruct the target image data set. A display device can be used to show image data sets, and a corresponding input device may also be provided, particularly if the area of ​​interest is selected at least partially manually. A collimation device, which may include one or more apertures, is provided at the radiation source. The collimation device is automatically adjustable, controlled by the control unit. The acquisition process is also generally controlled by the control unit, in particular the operation of the radiation source and the X-ray detector, as well as the movement of the acquisition setup consisting of the radiation source and the X-ray detector, or of the radiation source alone, along the acquisition trajectory.

[0032] It should be noted at this point that, as already indicated, the method according to the invention can be used in a wide variety of applications. In particular, the described method is suitable for a variety of acquisition geometries, especially for circular acquisition trajectories with angular ranges of approximately 200° to 360° in C-arm computed tomography, for tomosynthesis-like acquisition geometries, for example in mammography, bone imaging or lung cancer screening, and for special, compact cone beam geometries, such as those used, for example, in breast computed tomography or in dedicated extremity scanners.

[0033] Further advantages and details of the present invention will become apparent from the exemplary embodiments described below and from the drawing. The drawings show: Fig. 1 a flow chart of the inventive method, Fig. 2. A first illustration of the selection of shooting positions along a shooting trajectory, Fig. 3 a schematic sketch of collimation during selection according to Fig. 2, Fig. 4. A second illustration of the selection of shooting positions along a shooting trajectory, Fig. 5 a sketch of collimation for a selection according to Fig. 4, Fig. 6. A graph to illustrate the increase in image quality, Fig. 7 a graph to illustrate the increase in dose in the absorption area, and Fig. 8 an X-ray machine.

[0034] Fig. Figure 1 shows a flowchart of an embodiment of the method according to the invention. In step 1, initial projection images 2 are acquired along a recording trajectory without collimating the radiation source. Specific recording positions are used along the recording trajectory, corresponding to the first projection directions of the initial projection images.

[0035] In step 3, an overview image dataset 4 is reconstructed from the first projection images 2 using an algorithm of filtered back projection, which shows the entire recording area after no collimation has been carried out.

[0036] In step 5, this overview image dataset is optionally evaluated to automatically segment potentially relevant sub-areas. In this example, a representation of the overview image dataset 4 is then displayed in step 6, allowing a user to select a sub-area of ​​interest in which they desire higher-quality image data in a three-dimensional reconstruction. It should be noted that the selection of a sub-area of ​​interest in step 6 can also be performed fully automatically.

[0037] Although in the present embodiment only twenty first projection images were taken, and a low X-ray dose was used, the overview image data set 4 is sufficient to determine with sufficient accuracy the location of the area of ​​interest, which may correspond, for example, to a tumor, cartilage or the like.

[0038] After this first area of ​​interest has been selected in step 6, second projection images 8 are acquired in step 7. This is done, firstly, by collimating onto the area of ​​interest, and secondly, from second projection directions that differ from the first projection directions, as shown in the schematic diagram in Fig. 2 will be explained in more detail.

[0039] The maximum possible field of view (FOV) 9 is shown first, which is depicted in the overview image dataset 4, whose first projection images 2 were acquired without collimation. The area of ​​interest 10 lies within field of view 9. A circular path around the object to be recorded, i.e., field of view 9, is chosen as the acquisition trajectory 11 for the first and second projection images 2 and 8. The circles and stars branch off acquisition positions 12 and 13 along the acquisition trajectory 11, which thus correspond to different projection directions. The acquisition positions 12 (circles) correspond to the acquisition positions of the first projection images 2, while the acquisition positions 13 (stars), which lie between acquisition positions 12, are the acquisition positions for the second projection images 8.Thus, the sampling “gaps” are ultimately filled by the second projection images 8, at least for the area of ​​interest 10.

[0040] Fig. Figure 3 schematically shows the collimation performed. The radiation field incident on the X-ray detector, indicated by box 14, corresponds to the situation without collimation, i.e., for the first projection images 2, while box 15 represents the radiation field incident on the X-ray detector for collimation onto the area of ​​interest 10 for the second projection images 8.

[0041] It should be noted that when acquiring the second projection images 8 in step 7, further acquisition parameters can also be adjusted to increase the quality of the image data, for example, an X-ray dose can be increased, binning can be reduced and / or a suitable X-ray spectrum can be selected.

[0042] In step 16, a truncation correction is then performed on the second projection images 8 after they have been laterally truncated. Data from the second projection images 8 are extrapolated, taking the overview image data set 4 into account; that is, the extrapolation is carried out in such a way that the added projection image data corresponds to the image data contained in the overview image data set 4.

[0043] After truncation correction, a three-dimensional reconstruction image data set 18 is determined in step 17 using a filtered backprojection algorithm.

[0044] In step 19, the overview image dataset 4 is updated to a target image dataset 20 based on the reconstruction image dataset 18, whereby a weighted pixel-wise addition of the overview image dataset 4 and the reconstruction image dataset 18 is performed. The weights used for normalization are the weights a(x) defined above for both image datasets 4 and 18.

[0045] Alternatively, it is also possible to perform a complete reconstruction based on the first projection images 2 and the second projection images 8. However, it is essential in all embodiments of the method according to the invention that all first and second projection images 2, 8 are taken into account in order to determine the target image data set 20.

[0046] The target image dataset 20 now contains the entire recording area 9 and, in higher quality, the area of ​​interest 10. It can be displayed to a user. The user can then decide, in step 21, whether they want a further quality improvement in another area of ​​interest. If so, the target image dataset 20 is used as a new overview image dataset, and a new area of ​​interest is selected. Steps 6-19 are then repeated for this new area to obtain a further updated and improved target image dataset 20. If the quality is sufficient for the user, the process is terminated in step 22.

[0047] The situation when selecting a second, new sub-area of ​​interest 23, which advantageously lies within the existing sub-area of ​​interest 10, is described in the Fig. 4 and Fig. Figure 5 illustrates this. The acquisition trajectory 11 is used again for all acquisitions of projection images 2, 8. The acquisition positions 12 (circles) for the first projection images and the acquisition positions 13 (stars) for the first selected area of ​​interest 10 are again visible. In this case, the acquisition positions 24 for the second, newly selected region of interest 23 were chosen such that they lie between one of the acquisition positions 12 and one of the acquisition positions 13. Of course, other possibilities are conceivable in which fewer second projection images 8 are collimated onto the area of ​​interest 23.

[0048] Fig. 5 shows analogous to Fig. 3. The collimation performed is indicated by boxes 14, 15, and 25, which represent radiation fields on the X-ray detector. Box 25 symbolizes the collimation for the region of interest 23, which lies within the previously identified region of interest 10. Therefore, box 25 lies within box 15.

[0049] If, as shown, the further sub-areas of interest 23 are always selected within the existing sub-areas of interest 10, the patient's radiation exposure increases only slightly despite the gain in information for the actually relevant characteristic. This is to be illustrated by the graphs in the Fig. 6 and Fig. Figure 7 is symbolized. There, the image quality in the last selected area of ​​interest 23 is plotted against the number of projection images, assuming that each acquisition of second projection images 8 yields the same number of second projection images 8 as the initial acquisition of first projection images 2 for the entire acquisition area. Then there is a linear relationship, as indicated by the line 26. Position 27 refers to the acquisition of first projection images 2 in step 1, at position 28 the second projection images 8 for the first selected area of ​​interest 10 have been added, and at position 29 the second projection images 8 for the second selected area of ​​interest 23 have been added.

[0050] In Fig. Figure 7 shows the total dose across the image area plotted against the amount of projection, again with positions 27, 28, and 29 indicated. Due to collimation, as curve 30 shows, the radiation exposure in the image area increases only extremely slowly.

[0051] Fig. Figure 8 shows a schematic diagram of an X-ray device 31. This is a C-arm X-ray device 21, which has a C-arm 32 on which a radiation source 33 and a flat-panel X-ray detector 34 are arranged opposite each other. Various imaging trajectories, in particular circular paths, can be realized by means of the degrees of freedom of movement of the C-arm 32. A collimation device 35 is provided on the radiation source 33, via which the, for example, with respect to the Fig. 3 and Fig. The collimation of the beam cone described in section 5 can be achieved.

[0052] The operation of the X-ray device 31 is controlled by a control unit 36, which is designed to carry out the method according to the invention. This means that the imaging arrangement on the C-arm 32, as well as the collimation unit 35, can be controlled accordingly to acquire the projection images 2, 8 along the imaging trajectory 11 and to perform the corresponding collimation, as well as any further adjustments to the imaging parameters that may be required. The control unit 36 ​​also includes a reconstruction unit 37 in which the corresponding reconstructions and any corrections described above can be performed.

[0053] In the embodiment described here, it was assumed that the object to be recorded, in particular a patient, remained motionless. As already explained before the figure description, motion correction algorithms, especially through registration, can also be implemented in the case of patient movement within the recording area.

[0054] Although the invention has been illustrated and described in detail by the preferred embodiment, the invention is not limited by the disclosed examples and other variations can be derived by the person skilled in the art without leaving the scope of protection of the invention.

Claims

[1] Method for determining a three-dimensional target image data set (20) showing at least one area of ​​interest (10, 23) of an image area (9), wherein the image data of the three-dimensional target image data set (20) are reconstructed from two-dimensional projection images (2, 8) taken under different projection directions, characterized by the following steps: - Acquisition of first projection images (2) without collimation of the radiation source (33) used from first projection directions, and reconstruction of a three-dimensional overview image dataset (4) of the recording area (9) from the first projection images (2), - Selection of the relevant sub-area (10, 23) in the overview image dataset (4), - Recording of second projection images (8) from second projection directions, collimated onto the sub-area (10, 23), wherein the second projection directions differ from the first projection directions, - Reconstruction of the target image data set (20) showing the recording area (9) and the sub-area of ​​interest (10, 23) from all first and second projection images (2, 8), - wherein, as second projection directions, projection directions lying at least partially between two first projection directions along the same recording trajectory (11) are chosen. [2] Method according to claim 1, characterized by , that the selection of the sub-area (10, 23) is at least partially automatic, in particular based on a segmentation of the overview image data set (4). [3] Method according to claim 1 or 2, characterized by, that when taking the second projection images (8) at least one other recording parameter is adjusted in addition to the collimation. [4] Method according to claim 3, characterized by , that the adjustment of the further recording parameter results in an increase in resolution and / or contrast. [5] Method according to claim 3 or 4, characterized by , that as a further imaging parameter an X-ray dose is increased and / or binning is reduced and / or an X-ray spectrum suitable for imaging the sub-area is set. [6] Method according to any of the preceding claims, characterized by , that fewer than 100, in particular 15-25, first projection images (2) are taken. [7] Method according to any of the preceding claims, characterized by, that to determine the target image data set (20) a reconstruction image data set (18) is determined from the second projection images (8), whereupon the target image data set (20) is obtained by a particularly weighted, pixel-wise addition of the image data of the overview image data set (4) and the reconstruction image data set (18). [8] Method according to claim 7, characterized by that the pixel-dependent weights are used to normalize the reconstruction. [9] Method according to claim 7 or 8, characterized by , that the weighting factors a(x) for each pixel x and both the overview image dataset (4) and the reconstruction image dataset (18) are as a(x)=(N1+N2) / (w1(x)+w2(x)) result, where N1 is the total number of first projection images (2), N2 is the total number of second projection images (8), w1(x) is the number of first projection images (2) in which the image point x is located in the radiation area illuminated by X-rays, and w2(x) is the number of second projection images (8) in which the image point x is located in the radiation area illuminated by X-rays. [10] Method according to any one of claims 1 to 6, characterized by , that the target image data set (20) is determined by a complete reconstruction taking into account the projection data of all first and second projection images (2, 8). [11] Method according to any of the preceding claims, characterized by , that a truncation correction is applied to the second projection images (8) and / or a reconstruction image data set (18) reconstructed from them. [12] Method according to claim 11, characterized by, that the truncation correction is performed as an extrapolation of projection image data taking into account the overview image data set (4). [13] Method according to any of the preceding claims, characterized by , that in the target image data set (20) a new sub-area of ​​interest (23) is selected as a new overview image data set (4), in particular contained in the previous sub-area of ​​interest (10), whereupon a new target image data set (20) is reconstructed after recording new second projection images (8) under collimation onto the new sub-area of ​​interest (23).

Images