Reconstruction of an image in medical imaging

Abstract

The invention relates to a method for reconstructing an image of an object (7) in medical imaging by - Providing (S1) a first truncated projection image (pi) of the object (7), which projects an area of ​​the object to be mapped from a first projection source position (Ci) into a first projection plane, - Providing (S2) a 3D model (M) specific to the object (7) including a modeled area to be mapped which corresponds to the area to be mapped of the object (7), - Generating (S3) a first supplementary forward projection (ei1) of a first supplementary area of ​​the 3D model (M), which extends beyond and borders the modeled area to be mapped, from the first projection source position (Ci) into the first projection plane and - Reconstructing (S6) the image on the basis of the first truncated projection image (pi) and the first supplementary forward projection (ei1) of the first supplementary area.

Description

[0001] The present invention relates to a method for reconstructing an image of an object in medical imaging. Furthermore, the present invention relates to a computed tomography method, an imaging modality, and a computer program.

[0002] For the unambiguous reconstruction of tomographic images from projections, the projection data must completely depict the entire cross-section (in the lateral direction) of the object or patient. If the projections are not complete but truncated, this is called truncation. Truncated projection data occur in medical imaging, for example, when the patient is too large for the detector or when the projections were deliberately only partially acquired (e.g., by collimating the beam path), for example, to minimize radiation dose (volume-of-interest (VO1) imaging).

[0003] To obtain usable reconstruction results, truncated projections are typically extended (extrapolated) using a heuristic extrapolation model with established methods to minimize the resulting image artifacts. Approaches to this are shown in the following two articles: Sourbelle, K., Kachelriess, M., Kalender, WA. Reconstruction from truncated projections in CT using adaptive detruncation. Eur Radiol 15, 1008-1014 (2005). https: / / doi.org / 10.1007 / s00330-004-2621-9 and Hsieh, J., Chao, E., Thibault, J., Grekowicz, B., Horst, A., McOlash, S., Myers, TJ (2004): A novel reconstruction algorithm to extend the CT scan field-of-view. In Medical physics 31 (9), pp. 2385-2391. DOl: 10.1118 / 1.1776673.

[0004] However, a previous volumetric dataset (e.g., a planning CT scan) of the patient often already exists. This can be used for further imaging procedures, for example, to reduce radiation exposure. For this, newly acquired 2D projection data must be registered onto the existing 3D volumetric dataset.

[0005] The goal of registering 2D projection data and 3D volume data of an object with each other is to determine a suitable geometric transformation that establishes a common spatial relationship between the two datasets. An overview of this topic can be found in van de Kraats, EB, Penney, GP, Tomazevic, D., van Walsum, T., Niessen, WJ, “Standardized evaluation methodology for 2-D-3-D registration,” in IEEE Transactions on Medical Imaging, vol. 24, no. 9, pp. 1177-1189, Sept. 2005, doi: 10.1109 / TMI.2005.853240.

[0006] In 2D-3D registration of projection and volume data, two datasets are available: 2D projections of a specific area of ​​the object (ideally the entire object) and a 3D volume dataset of the same object or patient. To transform both datasets into a common spatial reference system, forward projections of the volume dataset (so-called Digitally Reconstructed Radiographs (DRRs)) are typically simulated repeatedly. The parameters of the geometric transformation that result in the smallest error compared to the given projections, in terms of a similarity metric, are determined through optimization. Applying a geometric transformation with these calculated parameters to the given data then performs the determined registration.

[0007] In the case of truncated projection data (i.e., the object was only partially mapped), this approach becomes difficult because, as a rule, only feature-based similarity metrics can be used to compare DRR and existing projection data.

[0008] From the article by D. Kolditz et al., “A Volume-of-interest (VOI) imaging in C-arm flat-detector CT for high image quality at reduced dose”, Medical Physics, 2010, Vol. 37, No. 6, pp. 2719-2730, a method for reconstructing an image of an object in flat-panel detector computed tomography is known, in which a complete low-dose overview image (OV) and a local high-dose VOI image are combined. From the article by J. Wiegert et al. “Projection Extension for Region Of Interest Imaging in Cone-Beam CT”, Academic radiology, 2005, Vol. 12, No. 8, pp. 1010-1023, describes a method for extending truncated projections to avoid truncation artifacts in C-arm-based 3D-ROI imaging, which utilizes prior knowledge by combining forward projections of a previously acquired, non-truncated 3D reference image with the truncated ROI projections.From US patent 2007 / 0195923 A1, a method for reconstructing an image created using a cone beam CT is known, in which truncations are corrected.

[0009] The object of the present invention is to improve the quality of image reconstruction in medical imaging.

[0010] This problem is solved according to the invention by the subject matter of the independent claims. Advantageous embodiments of the invention are set forth in the dependent claims.

[0011] According to the invention, a method for reconstructing an image (e.g., a tomographic image or 3D image) of an object in medical imaging is provided. The object can be a patient, a part of a patient, or an object within a patient. Such an object would be, for example, a catheter or an implant. Medical imaging can be performed using computed tomography, angiography, and similar techniques. The basis of the respective medical imaging is projections or fluoroscopy of the object. Through medical imaging, at least one image is obtained or reconstructed. It is determined from the raw imaging data. For example, the image is a tomographic image.

[0012] In the first step, a truncated projection image of the object is provided. This image projects a portion of the object to be mapped from a first projection source position onto a first projection plane. The truncated projection image does not represent the object completely, but only partially. Therefore, it is cut off, or truncated. The truncated projection image only depicts the area of ​​the object to be mapped (e.g., Volume of Interest, VOI). The projection rays run from the first projection source position through the object to the first projection plane.

[0013] The provision of the truncated projection image can be accomplished in many ways. For example, the truncated projection image is acquired and stored in a storage device. The truncated projection image is thus captured and made available digitally, for instance. Alternatively, the truncated image can also be received and provided via a data network (e.g., the internet). Another alternative is that the provision of the truncated projection image can also be achieved by reading it from a storage device. Such a storage device could be a hard drive or, if necessary, a portable storage unit.

[0014] In a further step, a 3D model specific to the object is provided, including a modeled area to be mapped, which corresponds to the area of ​​the object to be mapped. Providing the 3D model can, in principle, be done in the same way as providing the truncated projection image just described, for example, by capturing, receiving, reading, and so on. The object-specific 3D model is a volumetric dataset that spatially represents the object. The volumetric dataset, and thus the 3D model, is derived from images previously acquired of the object. The 3D model therefore shows essentially the same details as the object itself. The 3D model also includes an area that corresponds to the real-world area of ​​the object, which is represented by the truncated projection image.The corresponding area in the 3D model is referred to as the modeled volume to be represented. In particular, this can be the modeled volume of interest (VOI). Naturally, there can be (minor) deviations between the real object and the model. For example, a patient's body may undergo deformations due to breathing, etc. Depending on the time elapsed between a current projection image and the previously acquired volume data set, minor changes are also to be expected (e.g., a tumor may have formed, an organ may have changed, etc.).

[0015] According to a further step of the inventive method, a first supplementary forward projection of a first supplementary area, which extends beyond and borders the modeled area to be mapped, is generated from a first projection source position into the first projection plane. Based on the 3D model, a first supplementary forward projection, i.e., a modeled projection image, is calculated starting from the first projection source position. This means that a projection is simulated using the 3D model specific to the object. This simulation, or supplementary forward projection, does not refer, or does not primarily refer, to the area of ​​the object to be mapped, but rather to a first supplementary area. This supplementary area may include a portion of the area to be mapped, but extends beyond it.In any case, the first supplementary area borders the area to be mapped without any gap in between. The aim is to avoid artifacts that arise at the boundaries of the areas to be mapped.

[0016] In a further step, the image is reconstructed based on the truncated projection image and the first supplementary forward projection of the first supplementary region. This reconstruction is performed, for example, using a reconstruction algorithm into which the truncated projection image and the first supplementary forward projection of the first supplementary region are input. Naturally, the reconstruction algorithm can also use further (truncated) projection images and further supplementary forward projections for the reconstruction. Typically, a set of numerous such projection images or supplementary forward projections is used for the reconstruction.

[0017] Advantageously, the first supplementary forward projection, or subsequent supplementary forward projections, provide additional data for image reconstruction, thus improving the reconstruction quality. Furthermore, the additional data from the supplementary forward projection(s) originates from an object-specific model of the object being imaged and is not based solely on heuristics. This also helps to reduce artifacts. Since a previously acquired, object-specific volumetric dataset is used as the model, the method can be considered both an extrapolation and a registration method.

[0018] According to the invention, the first truncated projection image and the first supplementary forward projection of the supplementary area are registered relative to each other. The first truncated projection image and the first supplementary forward projection originate from two different, independent processes. The truncated projection image is generated by a real projection through the real object. In contrast, the supplementary forward projection is simulated by a corresponding projection algorithm. These different image generation processes generally require the resulting images to be registered relative to each other. This means that the two images, namely the truncated projection image and the supplementary forward projection, must be aligned or adjusted relative to each other. This registration can involve translation, rotation, magnification, contrast adjustment, and the like.In a concrete implementation, the translation and rotation of the object can be optimized. Scaling, contrast, etc., are also possible. However, the scaling is usually known from the given metadata of the volume dataset and therefore does not necessarily need to be selected via optimization. After such registration, the two images are generally optimally aligned and can complement each other perfectly.

[0019] Furthermore, the inventive method additionally comprises generating a second supplementary forward projection of a second supplementary area of ​​the 3D model, which extends beyond and borders the modeled area to be imaged, from the first projection source position into the first projection plane. The image is then reconstructed based on this second supplementary forward projection of the second supplementary area. Thus, a second supplementary forward projection is obtained for which, for example, there is no corresponding projection image. Similarly, there is no projection image for the first supplementary forward projection. Preferably, the second supplementary area borders a side of the modeled area to be imaged opposite the first supplementary area.This means that the modeled area to be mapped lies precisely between the first and second supplementary areas of the 3D model. These areas are directly adjacent to each other in pairs. Accordingly, a corresponding series of directly adjacent projections results: the first supplementary forward projection, the first truncated projection image, and the second supplementary forward projection. Thus, the reconstruction of the desired image also occurs on a pair of supplementary forward projections, located, for example, to the left and right of the actual truncated projection image. This helps to avoid artifacts that arise due to the two side edges of the truncated projection image.

[0020] The simulation of the model's forward projections can be performed simultaneously on both sides. For this, the model is positioned with a certain translation and rotation, and then the two complementary forward projections (left and right) are "captured" simultaneously. The simulation is essentially performed with a very large virtual detector large enough to capture the left, center, and right areas. Since the projection in the center is already given, it is not simulated due to computational time constraints. Theoretically, one could also simulate both complementary forward projections (left and right) sequentially. However, since both complementary forward projections (or "complementary projections") should represent the model with the same translation / rotation (to ensure consistent projections), this is generally not practical.

[0021] According to the invention, the two supplementary forward projections correspond to a first pair generated with respect to the first projection source position, and analogously, adjacent to a second truncated projection image in a second projection plane different from the first, a second pair of supplementary forward projections is obtained using the 3D model, generated with respect to a second projection source position different from the first, and the image is reconstructed based on both pairs of supplementary forward projections and both truncated projection images. This means that the image can be reconstructed not only based on one pair of forward projection images, but on at least one further pair of supplementary forward projections or forward projection images.Each pair of supplementary forward projections is generated from, or rather, derived for, a different projection source position. This means that the two truncated projection images are derived from two different projection source positions; that is, the respective sources for the projection are located at different positions. For each pair of supplementary forward projections, the same source position is used during simulation as that of the corresponding truncated projection image. In this way, further triples of forward projection pairs, including their corresponding projection images, can also be provided for reconstruction. This allows for further improvement of the reconstruction quality.

[0022] According to the invention, a first intermediate function is created for the first truncated projection image and the first supplementary forward projection (which together form a first duplicate), and a second intermediate function is created for the second truncated projection image and a corresponding supplementary forward projection (which together form a second duplicate). An error (i.e., an arbitrary error measure) between the two intermediate functions is then calculated and optimized. The first truncated projection image and the first supplementary forward projection are thus aligned. This is achieved using a first intermediate function, which results from a corresponding combination between the projection image and the supplementary forward projection. The intermediate function can be formed based on the projection parameters underlying the first truncated projection image and the first supplementary forward projection.Upon successful registration, truncation is completely eliminated. The size of the supplementary forward projections can be chosen so that the correctly registered volume can be fully mapped.

[0023] According to an analogous method of the invention, a first intermediate function is created for the first truncated projection image and the first pair of supplementary forward projections, and a second intermediate function is created for the second truncated projection image and the second pair of supplementary forward projections. An error (any error measure) between the two intermediate functions is then calculated and optimized. Thus, an intermediate function is created for each of the first triples (first projection image and first pair of supplementary forward projections) and for the second triples (second projection image and second pair of supplementary forward projections), and the error between them is minimized. This registers the supplementary forward projections with their corresponding projection images. The reconstruction quality is further improved by these internally registered triples.

[0024] Furthermore, according to the invention, epipolar lines in the supplementary forward projections are used for calculating the respective intermediate function, wherein the epipolar lines lie in the same plane as the two projection source positions. This means that the epipolar lines of the first doublet (first projection image and first supplementary forward projection) and the second doublet (second projection image and associated supplementary forward projection) or the epipolar lines of the first triplet (first projection image and first pair of supplementary forward projections) and the second triplet (second projection image and second pair of supplementary forward projections) together with the associated projection source positions span a plane.Once the error of two intermediate functions related to corresponding epipolar lines is minimal, all projection images and supplementary forward projections are optimally registered relative to each other.

[0025] The epipolar lines are defined purely by the geometry of the two projections and two source positions. They thus represent the intersection lines of the plane fan around the baseline connecting the source positions with the respective (virtual) detectors.

[0026] For each of these epipolar lines, a value for the respective intermediate function can be calculated. Starting with two intermediate functions, an error function can then be formulated, indicating how much the two intermediate functions differ. If the error between the intermediate function corresponding to projection source position Ci and the intermediate function corresponding to projection source position Cj becomes minimal, it can be assumed that the projections from both positions are consistent with each other.

[0027] The error function only indirectly reveals whether the respective supplementary forward projection matches the projection image. Primarily, it reveals whether one projection with respect to Ci matches, or is consistent with, the other projection with respect to Cj. If this is the case, it can be concluded that the projection image and the supplementary forward projection are also consistent with each other.

[0028] The optimization ultimately takes place in the volume domain (3D). If only a single projection triple or double (2D) "fits," this does not necessarily mean that the volume is correctly oriented in 3D space. For this, another projection triple or double is required. Whether both pairs of projection triples or doubles ultimately "fit" each other can be determined by the error function (as the difference between the two intermediate functions).

[0029] In another embodiment, the values ​​of the respective intermediate functions for the respective supplementary forward projections are calculated before the respective projection images of the object are provided. In particular, the values ​​of the intermediate functions can be calculated in advance. Calculating the error or error function is only meaningful when the truncated projection data is available. The supplementary forward projections are always consistent with each other anyway, since they are generated in this way. For example, an object is positioned in some way in 3D volume space, and the forward projections are simulated in the supplementary area. By design, these outer areas should therefore be consistent with each other. However, the goal is to select the supplementary area to be consistent with the (given) truncated projection image (inner area), and the projections of the supplementary area are optimized accordingly.For this purpose, the object is repositioned in the volume space and projected forward again, etc.

[0030] In a further embodiment, the 3D model is based on images, particularly planning scans, of the object or another object of the same type. Planning scans are often of higher quality than subsequent scans, as they are used for diagnostics and typically originate from a CT scanner. The truncated data, which the invention aims to enhance, are often interventional scans and are acquired, for example, with a C-arm CT (CBCT). The image quality of C-arm CTs is generally lower than that of conventional CTs, at least with regard to contrast resolution. However, the planning scans can also be derived from a previous C-arm CT scan. And, in principle, truncated CT projections could also benefit from the present invention.The corresponding planning data set or prior is used here to increase the scope of the total data obtained, so that ultimately a high image quality of the reconstructed images can be achieved with comparatively low radiation exposure.

[0031] In a preferred embodiment, the image to be reconstructed is a tomographic image or a 3D image. This means that a cross-sectional image or a 3D model of the object is obtained from the available data. The quality of the 3D model itself or the cross-sectional image is improved based on supplementary forward projections, which can be provided using the 3D model of the object.

[0032] The aforementioned task is also solved, in particular, by a computed tomography procedure in which images are reconstructed according to the method described above. The computed tomography procedure can be a standard CT scan, but also a CBCT (cone beam computed tomography) scan, an angiography scan, or similar. In any case, the reconstruction quality can be improved by the above method.

[0033] Furthermore, the above problem is also solved by an imaging modality comprising a data processing unit and a control unit configured to execute the method described above. The data processing unit can include one or more processors, one or more storage units, and optionally, interfaces. The control unit, in turn, can include control elements for imaging components, as well as one or more processors, one or more storage units, and one or more interfaces. The advantages and further developments of the method described above apply analogously to the imaging modality according to the invention. In this case, the aforementioned method steps represent corresponding functional features of sub-elements of the imaging modality.

[0034] Furthermore, a computer program is provided which includes instructions that, when executed by the aforementioned imaging modality, cause it to perform the procedure described above. Similarly, a computer program product or storage medium containing such instructions can also be provided.

[0035] The present invention will now be explained in more detail with reference to the accompanying drawings, which show: Fig. 1. A schematic view of a CT system, Fig. 2 a schematic view of two projections, Fig. 3 supplementary simulated forward projections of areas of an extrapolation model not included in the projection data, Fig. 4 the calculation of intermediate function values ​​along straight lines on a virtual detector of the composite projection data for pairs of projections, and Fig. 5 a schematic flowchart of an embodiment of a method according to the invention.

[0036] The exemplary embodiments described in more detail below represent preferred embodiments of the present invention.

[0037] Fig. Figure 1 shows a schematic representation of an advantageous embodiment of a proposed X-ray device as an imaging modality or medical CT device (computed tomography device) 1. The CT device 1 can comprise an X-ray source 2, an X-ray detector 3, and a processing unit 11 as a type of signal processing device. The X-ray source 2 and the X-ray detector 3 are arranged opposite each other. The X-ray source 2 can be configured to irradiate the X-ray detector 3 with X-rays along an X-ray incidence direction.

[0038] The CT scanner 1 can also include a gantry 4 with a rotor 5. The X-ray source 2 and the X-ray detector 3 can be arranged on the rotor 5 in a defined configuration, in particular integrated into or attached to the rotor 5. The rotor 5 can be rotatably mounted about a rotational axis 6. The object 7 to be imaged can lie on a patient positioning device 8 and be moved along the rotational axis 6 through the gantry 4. The processing unit 11 can be used to control the CT scanner 1 and to calculate cross-sectional or volumetric images of the object 7. An input device 9, for example a keyboard, and an output device 10, for example a screen and / or a display, can be connected to the processing unit 11, in particular via signal coupling.The input device 9 can advantageously be integrated into the output device 10, for example in the case of an input display, in particular a resistive and / or capacitive input display. The output device 10 can be configured to display a graphical representation of the X-ray image data set or any modeled images.

[0039] The schematic representations contained in the described figures do not indicate scale or proportions.

[0040] Another example (not shown in the figures) involves a monoplanar X-ray system with a C-arm held by a stand, such as a six-axis industrial or articulated robot. At the ends of the stand are an X-ray source (e.g., an X-ray source with an X-ray tube and collimator) and an X-ray image detector as the image acquisition unit. The implementation of the X-ray diagnostic device is not dependent on the industrial robot; conventional (stationary or movable) C-arm devices can also be used.

[0041] Fig. Figure 2 schematically illustrates the execution of forward projections vi and vj. Forward projection vi is generated by projecting a three-dimensional object P with a volume of interest VOI. The projection is made from a projection position Ci into a first projection plane. Similarly, forward projection vj is obtained by projecting the three-dimensional object P from the projection position Cj into a second projection plane that differs from the first. The projection positions Ci and Cj can, for example, be located on a circular path, a spiral path, or a more complex modulated path. By comparing the forward projections vi and vj with actual projections (2D data), the 3D data of the object P can be mapped onto the 2D projection data.

[0042] The present invention considers the registration problem from a new perspective: Instead of calculating simulated forward projections in the region of object P for which the projections are given, regions for which there is no projection data are, for example, calculated according to Fig. 3 simulated forward projection, resulting in simulated complementary forward projections ei1, ei2, ej1, ej2.

[0043] Often, a prior volumetric dataset (e.g., a planning CT scan) of the patient already exists. By registering this so-called prior onto the projection data to be reconstructed, a significantly more accurate, patient-specific extrapolation and thus a better reduction of image artifacts would be possible. For this, 2D projection data must be registered onto 3D volumetric data.

[0044] For the supplementary forward projection, an extrapolation model M is used. This extrapolation model M can be the same as the object model P derived from a prior dataset, or it can supplement the latter. From the extrapolation model M, for example, a first supplementary area and a second supplementary area can be forward-projected from the projection position Ci into the first projection plane, resulting in modeled projection images (also called supplementary forward projections ei1 and ei2 for short). The two supplementary forward projections ei1 and ei2, based on the supplementary areas, are located, for example, to the left and right of a truncated projection image (in Fig. 3 not shown). In the same way, a first supplementary area and a second supplementary area can also be projected forward from the projection position Cj into the second projection plane to obtain the respective supplementary forward projections ej1 and ej2. These two supplementary forward projections ej1 and ej2 are also located, for example, to the left and right of a truncated projection image taken from the projection position Cj (in Fig. 3 is not shown either).

[0045] How Fig. As shown in Figure 4, an actual or truncated projection image pi, taken from the projection position Ci, can be supplemented or combined with the two supplementary forward projections ei1 and ei2 in the first projection plane. Preferably, the supplementary forward projections ei1 and ei2 are designed or arranged such that they directly adjoin the projection image pi on opposite sides. In principle, supplementing the truncated projection image pi is only possible on one side, especially if there is only one-sided truncation. However, supplementing on both sides is advantageous in the case of two-sided truncation. For an exact reconstruction, all projections of the object should be completely truncation-free. The more truncation is present, the more pronounced the image artifacts resulting from the reconstruction of truncated data become.

[0046] The size of the supplementary forward projections ei1 and ei2 can be equal to the size of the truncated projection image pi. Alternatively, the supplementary forward projections ei1 and ei2 can also be larger or smaller than the truncated projection image pi. They should be large enough that the object is completely projected in pi+ei1+ei2. However, since it is usually not known exactly "how much of the object is missing," the supplementary forward projections are generally chosen to be slightly larger.

[0047] Similarly, a projection image pj obtained from the projection position Cj can be supplemented by one or more additional forward projections ej1 and ej2 in the second projection plane. Typically, a projection dataset from which one reconstructs contains approximately 500 projections. However, the described methodology with the intermediate functions, etc., does not need to be performed for all 500 projections. It is sufficient to compare a sufficiently large number of projection pairs. Theoretically, two projections (i.e., one projection pair as outlined) are enough to perform the registration. The registration specifies how the model must be positioned in volume space to obtain forward projections consistent with the truncated data. Once these parameters (e.g., translation and rotation) for the volume dataset are available, the volume can be analyzed from other directions (e.g.,(from all 500 projection directions of the truncated scan) and thus has a complete extrapolated projection dataset with which reconstruction can be performed.

[0048] To align the projection images pi and pj with their complementary forward projections ei1 and ei2, ej1 and ej2, epipolar lines li and Ij can be used. These epipolar lines li and Ij lie in a common plane Γ with a baseline Bij passing through the projection positions Ci and Cj. In particular, the values ​​of an intermediate function for the regions ei1, pi, and ei2 can be calculated independently. Through an optimization as described below, the respective registration parameters for positioning the model can be optimized such that the intermediate function values ​​correspond to each other on corresponding epipolar lines. The same can be done for the regions ej1, pj, and ej2 with respect to the epipolar line Ij. Through these optimizations, the complementary forward projections and their respective projection images are optimally aligned.This means that the data are consistent with each other, or that the virtual complementary forward projections to the respective projection images are optimally registered.

[0049] In detail, simulated supplementary forward projections ei1, ei2, ej1, ej2 are combined with the known projection data pi, pj of the remaining object domain. The desired parameters of the geometric transformation are not calculated in this case by optimizing a similarity metric (the DRRs no longer approximate the given projection data, but extend it). Instead, data consistency conditions, which must be satisfied for "correct" or consistent projection data, are used to form an error measure that is minimized through optimization. These consistency conditions are preferably calculated as an intermediate function from pairs of projections from different source positions (see Fig. 4) The registration of the data sets achieved with the present invention is therefore only indirect, since exactly opposite regions of the object are always given and forward-projected. The registration is achieved indirectly by calculating a consistent extrapolation and can thus be understood as a kind of inverse DRR registration method: Instead of approximating the known projection data by forward-projecting a volume data set in the region of the known projection data, the known projection data are extrapolated by forward-projecting a volume data set in the exactly opposite region.

[0050] The intermediate function can be calculated according to Punzet et al.: Punzet, D., Frysch, R., Rose, G. (2018): Extrapolation of Truncated C-arm CT Data using Grangeat-based Consistency Measures. In Frédéric Noo (Ed.): Proceedings of the CT Meeting 2018. The Fifth International Conference on Image Formation in X-ray Computed Tomography. Salt Lake City, UT, USA, 2018. The following equations explicitly refer to this publication, where they are explained in more detail. The intermediate function is derived from Grangeat's fundamental relation as... Si(n^):=−∫S2dm^δl(m^⋅n^)gi(m^)=∂∂ξΓ(n^,ξ)|ξ=Ci⋅n^.

[0051] It thus establishes a connection between the X-ray transformation (the projection data) X(Ci,m^):=∫0∞f(Ci+m^λ)dλ:=gi(m^) and the 3D Radon transformation (plane integrals in volume space) R3(n^,ξ):=∭ℝ3f(r)δ(r⋅n^−ξ)dr:=Γ(n^,ξ) This allows the intermediate function to be calculated either by calculating the plane integral and subsequent derivative in volume space, or by calculating line integrals in the projection data (so to speak, on the detector) along with the derivative in the perpendicular direction. Calculating the intermediate function via plane integrals in volume space is generally not possible in practice. However, the intermediate function values ​​can be calculated from the projections themselves.

[0052] Epipolar geometry allows the identification of epipolar lines in pairs of projections from the same acquisition (typically a circular trajectory of approximately 200° for CBCT scans) and their use for calculating the intermediate function. Since the intermediate functions, or rather their values ​​calculated from both projections, refer to the same plane integral in volume space, the intermediate functions from both projections should correspond. This yields a consistency condition for internally consistent pairs of projections: the greater the difference between the intermediate functions of two projections, the more inconsistent the two projections are with each other. The deviation of two such intermediate functions from each other implies that certain discrepancies (geometric deviations, artifacts, noise, etc.) must exist between the two projections.

[0053] For the described procedure, this consistency measure is used to find a consistent extrapolation of the limited data for several pairs of projections. The epipolar lines li, Ij can be identified in the projection data composed of given projections pi, pj and forward-projected DRRs ei1, ei2, ej1, ej2, and intermediate functions can be calculated (see Fig. 4).

[0054] In the article by Punzet et al., the described consistency measure is already used to optimize commonly used heuristic extrapolations, for example with an ellipsoid, with respect to consistent projections. The crucial difference of the method described here lies in the use of consistency conditions to utilize an existing volume dataset (prior CT or CBCT) not only for extrapolation, but also to simultaneously register it onto the truncated projection dataset, instead of relying on simple extrapolation models. Crucial for the rapid computation of the optimal extrapolation is the ability to pre-calculate the values ​​of the intermediate functions of the extrapolation DRRs ei1, ei2, ej1, ej2 as soon as the prior dataset is available. In this way, the space of optimization parameters is roughly sampled in advance.At the time of the actual reconstruction, only the intermediate function values ​​of the respective three sub-areas need to be optimally combined. Subsequent local optimization can be used to refine the result.

[0055] Fig. Figure 5 shows an embodiment of a method according to the invention in a flowchart. In a first step S1, a first truncated projection image of the object is provided (in practice, a pair of projection images will usually be necessary for sufficient consistency), which represents a region of the object to be imaged in a projection plane. Providing the projection image can mean that it is acquired, recorded, received, read out, or the like. The projection image is truncated and thus cut off on at least one side.

[0056] In a further step S2, a 3D model specific to the object is provided, including a modeled area to be mapped, which corresponds to the area of ​​the object to be mapped. The 3D model is thus also provided by acquiring, recording, receiving, reading, or similar means. Preferably, it is acquired beforehand using planning scans.

[0057] In a further step S3, a first supplementary forward projection, or a supplementary forward projection pair, of a first supplementary area is generated using the 3D model. This area extends beyond and borders the modeled area to be mapped. Thus, at least one supplementary forward projection is obtained not using a heuristic 3D model, but rather using a 3D model specific to the object.

[0058] Steps S1 and S3 can be repeated for multiple truncated projection images. Likewise, step S3 can be repeated on its own to project multiple supplementary areas forward for a single truncated projection image.

[0059] The calculation of the intermediate function(s) and the minimization of the error via optimization of the volume dataset parameters can follow in an optional step S4. Subsequently, in a step S5, the projection of all supplementary areas with optimal registration parameters (from the optimization) can be performed.

[0060] In a final step S6, the image is reconstructed based on the truncated projection image and the first supplementary forward projection of the first supplementary area. The desired image is thus reconstructed using data that has been supplemented according to an object-specific model.

[0061] The scope of application for the present invention lies in the described scenario where—for reasons of dose reduction or limited detector size—only truncated projection data are available for image reconstruction. Particularly in radiotherapy, but also in interventional procedures, planning scans in the form of a CT prior are typically available. Extrapolation using this prior should generally provide a better extrapolation than simple heuristic methods. For this extrapolation, correct registration of the data sets is necessary or can be considered a byproduct of the described invention. The ultimate benefit is improved image quality of the reconstructed images and a reduced dose for the patient.

[0062] Regardless of the grammatical gender of a particular term, persons with male, female or other gender identities are included.

Claims

[1] Method for reconstructing an image of an object (7) in medical imaging characterized by - Providing (S1) a first truncated projection image (pi) of the object (7), which projects an area of ​​the object to be mapped from a first projection source position (Ci) into a first projection plane, - Providing (S2) a 3D model (M) specific to the object (7) including a modeled area to be mapped which corresponds to the area to be mapped of the object (7), - Generating (S3) a first supplementary forward projection (ei1) of a first supplementary area of ​​the 3D model (M), which extends beyond and borders the modeled area to be mapped, from the first projection source position (Ci) into the first projection plane, whereby the first truncated projection image (pi) and the first supplementary forward projection (ei1) are registered relative to each other, - Generating a second supplementary forward projection (ei2) of a second supplementary area of ​​the 3D model (M), which extends beyond the modeled area to be mapped and, in particular, borders it on a side of the modeled area to be mapped opposite the first supplementary area, from the first projection source position (Ci) into the first projection plane, - Reconstructing (S6) the image on the basis of the first truncated projection image (pi) and the first supplementary forward projection (ei1) of the first supplementary area and the second supplementary forward projection (ei2) of the second supplementary area, wherein the two complementary forward projections (ei1, ei2) correspond to a first pair generated with respect to the first projection source position (Ci), analogously adjacent to a second truncated projection image (pj) in a second projection plane different from the first, a second pair of complementary forward projections (ej1, ej2) is obtained by means of the 3D model (M) in the second projection plane, generated with respect to a second projection source position (Cj) different from the first, and the reconstruction of the image is carried out on the basis of both pairs of complementary forward projections (ei1, ei2, ej1, ej2) as well as both truncated projection images (pi, pj), wherein a first intermediate function is created for the first truncated projection image (pi) and the first supplementary forward projection (ei1), and a second intermediate function is created for the second truncated projection image (pj) and a corresponding supplementary forward projection (ej1, ej2), and an error between the two intermediate functions is calculated and optimized, and wherein epipolar lines (li, Ij) in the supplementary forward projections (ei1, ei2, ej1, ej2) are used for calculating the respective intermediate function, and the epipolar lines (li, Ij) together with the two projection source positions (Ci, Cj) lie in one plane. [2] Method for reconstructing an image of an object (7) in medical imaging characterized by - Providing (S1) a first truncated projection image (pi) of the object (7), which projects an area of ​​the object to be mapped from a first projection source position (Ci) into a first projection plane, - Providing (S2) a 3D model (M) specific to the object (7) including a modeled area to be mapped which corresponds to the area to be mapped of the object (7), - Generating (S3) a first supplementary forward projection (ei1) of a first supplementary area of ​​the 3D model (M), which extends beyond and borders the modeled area to be mapped, from the first projection source position (Ci) into the first projection plane, whereby the first truncated projection image (pi) and the first supplementary forward projection (ei1) are registered relative to each other, - Generating a second supplementary forward projection (ei2) of a second supplementary area of ​​the 3D model (M), which extends beyond the modeled area to be mapped and, in particular, borders it on a side of the modeled area to be mapped opposite the first supplementary area, from the first projection source position (Ci) into the first projection plane, and - Reconstructing (S6) the image on the basis of the first truncated projection image (pi) and the first supplementary forward projection (ei1) of the first supplementary area and the second supplementary forward projection (ei2) of the second supplementary area, wherein the two complementary forward projections (ei1, ei2) correspond to a first pair generated with respect to the first projection source position (Ci), analogously adjacent to a second truncated projection image (pj) in a second projection plane different from the first, a second pair of complementary forward projections (ej1, ej2) is obtained by means of the 3D model (M) in the second projection plane, generated with respect to a second projection source position (Cj) different from the first, and the reconstruction of the image is carried out on the basis of both pairs of complementary forward projections (ei1, ei2, ej1, ej2) as well as both truncated projection images (pi, pj), wherein a first intermediate function is created for the first truncated projection image (pi) and the first pair of supplementary forward projections (ei1, ei2), and a second intermediate function is created for the second truncated projection image (pj) and the second pair of supplementary forward projections (ej1, ej2), and an error between the two intermediate functions is calculated and optimized, and wherein epipolar lines (li, Ij) in the supplementary forward projections (ei1, ei2, ej1, ej2) are used for calculating the respective intermediate function, and the epipolar lines (li, Ij) together with the two projection source positions (Ci, Cj) lie in one plane. [3] Method according to one of claims 1 or 2, wherein values ​​of the respective intermediate function for the respective supplementary forward projections (ei1, ei2, ej1, ej2) are calculated before the respective projection images (pi, pj) of the object are provided. [4] Method according to any of the preceding claims, wherein the 3D model (M) is based on previously obtained images of the object (7) or of another object of the same type as the object. [5] Method according to any of the preceding claims, wherein the image to be reconstructed is a tomographic image or a 3D image. [6] Imaging modality comprising a data processing unit and a control unit configured to perform a method according to any of the preceding claims. [7] Computer program comprising instructions which, when the program is executed by the imaging modality according to claim 6, cause it to execute the method according to any one of claims 1 to 5.

Images