METHOD AND SYSTEM FOR PREPROCESSING CT IMAGES

Description

[0001] The invention relates to a method and a system for preprocessing a plurality of simultaneously acquired CT scans at different acquisition energies (CT: computed tomography). In particular, the invention relates to a beam hardening correction for spectral CT scans with noise limitation.

[0002] CT systems capable of simultaneously acquiring multiple spectra allow for the reconstruction of image datasets that themselves represent a material property (e.g., the spatial distribution of base materials) or that depict attenuation coefficients at a given energy, which do not necessarily correspond to the energies used in the measurement. Ideally, these representations are free of artifacts caused by beam hardening, but this is often not the case in reality.

[0003] The problem is particularly relevant for CT systems with quantum-counting detectors, but also for any other multi-spectrum CT principle, e.g., dual source CT, sandwich detector, fast kVp switching, or others.

[0004] It is known in the prior art that material components can be reconstructed from N spectra, provided that they are distinguishable due to their different spectral dependence of the attenuation coefficient (see Figure 2 While precise decomposition of the base material can be achieved, this results in increased noise, particularly when large proportions of non-water-like materials are present. In clinical CT, this applies to bone, metals, or high contrast agent concentrations.

[0005] US 2021 / 0012463 A1 discloses a method according to the preamble of claim 1.

[0006] It is an object of the present invention to provide an alternative, more convenient method and a corresponding system for preprocessing a plurality of simultaneously acquired CT scans at different acquisition energies, with which the disadvantages described above can be avoided and in particular a reduction of noise can be achieved.

[0007] This problem is solved by a method according to claim 1, a system according to claim 7, a computer program product according to claim 10 and a storage medium according to claim 11.

[0008] The method according to the invention serves for preprocessing, in particular for noise reduction, a plurality of simultaneously acquired CT scans with raw data acquired at different acquisition energies. These CT scans are acquired at different energies, whereby, for example, an energy-selective detector, which allows differentiation into several energy channels, and a broadband X-ray source could be used, or a non-energy-selective detector in combination with several, possibly narrower-band, X-ray sources that differ with respect to their emitted spectrum. The scans thus comprise multi-energy CT raw data Pi(k) of an object, e.g., a patient, acquired in N spectral channels, where i = 1, ... N. Here, k, in simplified terms, describes all existing degrees of freedom of the raw data, e.g., projection angle, channel number, detector row.Raw data from at least two CT scans acquired at different energies are therefore available. The majority of CT scans acquired at different (acquisition) energies can also be referred to as a spectral CT dataset, whereby the respective CT scans within the dataset can be assigned to different X-ray energy distributions. As previously explained, this can correspond to an acquisition using N different X-ray spectra or to an acquisition using an energy-selective detector, which allows for spectral differentiation of the raw data into N channels, i.e., spectra.

[0009] The method comprises the steps of claim 1.

[0010] The components Pi of the vector P are the raw data from the CT scans, each at a specific energy level for each vector component. Essentially, this vector corresponds to the raw data, assigning them a sequence or rigid assignment that specifies which raw data (i.e., acquired at which energy, i.e., in which spectral channel) is used for which subsequent calculation. For example, a specific arrangement of the raw data, eg The vector P can already be represented by increasing or decreasing recording energy. Essentially, one could continue to refer to the raw data at this point, but the vector makes the subsequent steps easier to understand. Preferably, the recorded spectra (i.e., the raw data at the different energies) are indexed such that P1 denotes the spectrum with the lowest noise. This means that P1 represents the raw data in the spectral channel that exhibits the least noise. This can eg In the case of a quantum-counting detector, the raw data will be the one assigned to the lowest energy threshold.

[0011] The dimension of the vector is N, which corresponds to the number of recorded energies. This means that it is not necessary to use all raw data from all energies. While it is preferred to create a vector with N components for N recorded energies, it can also be advantageous to record N+1 energies, create a vector with N components, and use a set of raw data for one energy that was not included in the vector as a reference value.

[0012] Matrix A can be predefined (e.g., as a standard matrix) or determined based on data related to the examination (individual matrix A for each different type of examination) or to the patient (matrix A is adapted based on the patient data). If the first vector component P1 has the least noise (or was recorded at the lowest energy), A preferably has the form: A = 1 0 ⋯ 0 # # ⋯ # ⋮ ⋮ ⋱ ⋮ # # ⋯ # with freely selectable entries "#", where a matrix other than the identity is preferred. An example for N = 2, i.e., two energies, would be A = 1 0 1 − 1 .

[0013] Matrix A has dimension N x M, meaning it has N columns and M rows, and multiplication by the N-dimensional vector P results in an M-dimensional vector Q. While P and Q should preferably have the same dimension N, there are also useful applications for an M-dimensional transformed vector Q (and even a different dimension O for the back-transformed vector P'), particularly when the system of recorded spectra is overdetermined with respect to the subsequent number of basis materials. In such cases, an inverse matrix A-1 is not required. It is also preferred in this regard to perform the reduction of spectra before transforming vector P to Q and then working again with a square matrix A. It is therefore preferred that the transformed vector Q also has dimension N (i.e., M = N) and that matrix A is an N x N matrix.

[0014] The matrix A is preferably invertible, i.e., its determinant is non-zero, and especially preferably has row sums of zero in N-1 rows.

[0015] The transformed vector Q is calculated by simply multiplying the vector P by the matrix A. If the matrix A has more than one non-zero entry in a row, then components Qi of Q are mixed terms of two or more Pi. This calculation is known in the art and follows for Qi from the matrix indices A and the vector components (raw data) Pi with i = [0, ..., M] according to the formula Q i = ∑ n = 1 N A in P n .

[0016] Thus, M (preferably with M = N) new projection datasets Q i are generated by a transformation using the N x M matrix A. This is followed by processing of the projection data.

[0017] Even in state-of-the-art technology, data preprocessing can take place, e.g., denoising. In this process, the raw data is preprocessed (e.g., denoised), which would correspond to the P i in this context.

[0018] In contrast, the inventive method processes the components (the data) of the transformed vector Q (i.e., the Qi). This is done using the processing function B, which can, for example, be a noise reduction function. Although noise reduction is a preferred processing method in practice, a number of additional or alternative preprocessing steps can also be performed, such as artifact correction or other objective improvements to the data quality. However, no "arbitrary" changes, such as contrast enhancement, should be made at this stage. "Arbitrary" in this context refers to changes that impair the physical significance of the data in terms of the subsequent material decomposition. Contrast enhancement, for example, would be a visual improvement, but it qualitatively alters the data.

[0019] It is particularly advantageous here that a value in the raw data (at a given energy) remains unchanged, for example, if the matrix multiplication at this point resulted in the component P i being identical to Q j, or if this value is used additionally. This value can be used as a reference value, especially for noise reduction.

[0020] A preferred processing method is structure-preserving noise reduction (often denoted by "Λ") in the projection data Q, where geometric synchronization takes place. For example, assuming that, according to an advantageous choice for the vector P (e.g., the first component with the lowest noise) and the matrix A (e.g., such that Q₁ = P₁), the data of the first spectrum exhibit the best statistics, these could be used as a guide or prior. Such methods are known from the literature.

[0021] The processed vector Q is then referred to as "Q'" and results from the processing function B with Q' = B(Q). In structure-preserving noise reduction, B can be A and Q' = Λ(Q).

[0022] After processing, the processed transformed vector Q' is multiplied by the matrix Z to calculate the inversely transformed vector P' = Z · Q'. This matrix Z has n rows and 0 columns for the M-dimensional vector QM, where, preferably, if the dimension of P' is to be equal to the dimension of the vector P, O = N. The matrix Z is preferably the inverse matrix of A (Z = A - 1 < ), but this is not mandatory. It can, for example, also be an M x N matrix, at least if Q has a different dimension M than the vector P. Z can also be an N x N matrix that is not the inverse of A. This calculation is known in the prior art and follows for P' i from the matrix indices Z im and the vector components Q' m of the formula . P ′ i = ∑ m = 1 M Z im Q ′ m = ∑ m = 1 M Z im B Q m .

[0023] The inversely transformed vector P' is then output for further calculations. The components of the inversely transformed vector P' again correspond to raw data, which, however, has been preprocessed by the preceding processing steps, e.g., noise reduction. This data can now be used for image reconstruction or the reconstruction of material components, as described in the introductory section.

[0024] The system according to the invention for preprocessing a plurality of simultaneously recorded CT scans with raw data that have been recorded at different recording energies is defined by claim 7.

[0025] A control device according to the invention for controlling a computed tomography system comprises a system according to the invention and is therefore also designed to carry out the method according to the invention.

[0026] A computed tomography system according to the invention comprises a control device according to the invention.

[0027] A large portion of the aforementioned system components can be implemented wholly or partially as software modules within a processor of a suitable computing system, for example, a control unit of a computed tomography system. A largely software-based implementation has the advantage that existing computing systems can be easily retrofitted via a software update to operate according to the invention. In this respect, the problem is also solved by a corresponding computer program product containing a program that can be directly loaded into a computing system, with program sections to execute the steps of the inventive method when the program is run on the computing system. In addition to the computer program, such a computer program product may optionally include additional components such as documentation and / or additional hardware components, such as...Hardware keys (dongles, etc.) for using the software are included. A computer-readable medium, such as a memory stick, a hard drive, or other portable or permanently installed data carrier, can be used for transport to and / or storage on or in the computer system or control unit. This medium must contain the program sections of the computer program that can be read and executed by a computer system. The computer system may, for example, include one or more cooperating microprocessors or similar components.

[0028] Further, particularly advantageous embodiments and developments of the invention result from the dependent claims and the following description, wherein the claims of one claim category may also be further developed analogously to the claims and description parts of another claim category and, in particular, individual features of different embodiments or variants may be combined to form new embodiments or variants.

[0029] According to a preferred method, the back-transformed vector P' is used in a subsequent step to determine results from the raw data, preferably to determine Base material data and / or monochromatic data and / or electron density and / or nuclear charge and / or proportions of absorption effects, in particular Compton effect or photoelectric effect.

[0030] All these calculations are known in the prior art, but after applying the method according to the invention, they can be performed with significantly improved (pre-processed) raw data. This allows the results to be optimized.

[0031] According to a preferred method, processing the components of the transformed vector Q constitutes denoising. The processing function B thus denoises the components of the transformed vector Q. It is preferred that a structure-preserving noise reduction is performed, in which one component of the transformed vector, in particular the first component, is used as a guide instance for geometric synchronization. Assuming that, according to the specific, advantageous choice, the data of the first spectrum exhibit the best statistics, these can be used as a guide or prior.

[0032] According to a preferred method, CT images are reconstructed from the components of the transformed vector Q before processing the components of Q. These CT images are then processed with the processing function B, and the processed CT images are preferably reconstructed back to raw data. Thus, denoising of the projection data sets Q i takes place in the image space. After processing, the denoised images are transferred back into the projection data space by a forward projection that inverts the reconstruction. The advantage of noise reduction in the image space is that the signal-to-noise ratio of relevant structures is generally higher in the image space than in the projection data. Structure-preserving denoising can also be performed in this variant.

[0033] Because of the linearity of CT reconstruction, forward projection and transformation A, the order can be swapped in many variations, and also arranged asymmetrically, in the sense that, for example, A is applied in the image space and Z (e.g., A -1< ) ​​in the raw data space.

[0034] According to a preferred method, the reconstruction of CT images takes place before the formation of the transformed vector Q and the reverse reconstruction of the CT images takes place after the formation of the reverse-transformed vector P'.

[0035] According to a preferred method, the reconstruction of CT images is carried out after the formation of the transformed vector Q and the reverse reconstruction of the CT images is carried out after the formation of the reverse-transformed vector P'.

[0036] According to a preferred method, the reconstruction of CT images is carried out after the formation of the transformed vector Q and the reverse reconstruction of the CT images is carried out before the formation of the reverse-transformed vector P'.

[0037] According to a preferred method, the reconstruction of CT images takes place before the formation of the transformed vector Q and the reverse reconstruction of the CT images takes place before the formation of the reverse-transformed vector P'.

[0038] According to a preferred method, a component of the vector P is selected as a reference value for processing, in particular the first component, or a component that was recorded at a given energy (and thus, according to preliminary considerations, should exhibit the lowest noise component). The matrix A is determined such that in the row acting on this component during the transformation, the matrix coefficient on the main diagonal has the value 1, and all other matrix coefficients in the row have the value 0. This reference value Pi is then mapped onto a projection data set Qi = Pi. Preferably, the spectra are indexed such that the spectrum with the lowest noise, i.e., in the case of a quantum-counting detector, the raw data associated with the lowest energy threshold, forms the first vector component of P (i.e., P1).

[0039] According to a preferred method, an image is acquired at N+1 energies, and one image acquired at one of these energies is selected as a reference image. From the remaining images, the vector P is generated at N energies, and the reference image, after forming the vector Q = A · P, serves as the reference value for processing. This corresponds in principle to the previously described procedure, in which Qi = Pi forms the reference value, with the formal difference that no matrix multiplication was performed for the reference value; it is therefore present as a kind of "additional" value.

[0040] According to a preferred method, the matrix A is invertible and the inversely transformed vector P' = A -1< · Q' is formed from the inverse matrix of A as matrix Z (i.e., Z = A -1< ) ​​and the processed transformed vector Q' = B(Q).

[0041] According to the procedure, matrix A has row sums of 0 in N-1 rows. For matrix Z, it is preferred that Z has column sums of 0 in N-1 columns. However, as already indicated above, it is preferred that Z is uniquely derived from A and that the form of Z therefore depends on A. The advantage of this requirement for the row sums is that there is a "master spectrum" (e.g., Q1), namely the one generated by the single row with a non-zero row sum. The other intermediate spectra (e.g., Q2 to QM) represent differences when the row sum is zero, particularly with respect to Q1. In a case where A is an N x N matrix and all Pi are equal, then Q2 to QN would be 0 with this property. For pure water objects, this situation would be present in CT (except for noise). Thus, Q2 to QM would represent the deviations from water and could be interpreted quite intuitively in terms of material decomposition.

[0042] According to a preferred method, the components of the back-transformed vector P' are reconstructed into images within the framework of an image reconstruction, and preferably CT images CT(E) for different energies E are generated after or within the framework of this image reconstruction using the formula CT ( E ) = ∑ µ i ( E ) L in ( rP' in ) with the energy-dependent coefficient µ, a reconstructed image rP' i from the i-th component of the vector P' and the line integrals L i.

[0043] It should be noted that in practice two approaches are preferred: Material densities can be reconstructed and then combined in image space to form local monochromatic absorption values. Alternatively, monochromatic line intervals can be generated first, which, after reconstruction, directly represent the local monochromatic attenuations. Except for non-linear optimization during reconstruction, this is mathematically identical, since the inverse Radon transform (as the fundamental CT reconstruction) is a linear operator. Which approach is more practical depends on whether the final image space data should primarily represent base material images or images of different energies.

[0044] Preferably, components of the invention are provided as a "cloud service." Such a cloud service serves to process data, particularly using artificial intelligence, but can also be a service based on conventional algorithms or a service where human evaluation takes place in the background. Generally, a cloud service (hereinafter also referred to simply as "cloud") is an IT infrastructure in which, for example, storage space or computing power and / or application software is provided via a network. Communication between the user and the cloud takes place via data interfaces and / or data transmission protocols. In the present case, it is particularly preferred that the cloud service provides both computing power and application software.

[0045] In a preferred method, data is provided to the cloud service via the network. This cloud service comprises a computing system, such as a computer cluster, which typically does not include the user's local computer. This cloud service can be provided, in particular, by the medical facility that also provides the medical technology systems. For example, image acquisition data is sent via a RIS (Radiology Information System) or PACS (Patient Information Communication System) to a (remote) computing system (the cloud). Preferably, the cloud computing system, the network, and the medical technology system form a data-related network. The process can be implemented using a command structure within the network. The data processed in the cloud ("result data") is subsequently sent back to the user's local computer via the network.

[0046] The invention enables beam hardening correction based on a base material model without an increase in noise or the introduction of statistical artifacts. This is a significant advantage over prior art methods. Preventing an increase in noise plays a crucial role in clinical applications. Effective noise suppression is essential, particularly for future applications with more than two spectra, such as those involving exotic contrast agents.

[0047] The exact beam hardening correction improves the quality of base material decompositions and monoenergetic data compared to state-of-the-art methods based on a linearized base material model.

[0048] The invention is explained in more detail below with reference to the accompanying figures and exemplary embodiments. The same components are designated with identical reference numerals in the various figures. The figures are generally not to scale. They show: Figure 1 a roughly schematic representation of a computed tomography system with an embodiment of a control unit with a device according to the invention for carrying out the method, Figure 2 a description of a method for base material decomposition and correction of beam hardening artifacts according to the state of the art, Figure 3 an example of a system according to the invention, Figure 4 a schematic representation of the method according to the invention, Figure 5 an example of a possible sequence of a preferred variant of a method according to the invention, Figure 6an example of a possible sequence of a preferred variant of a method according to the invention, Figure 7 An example of a possible sequence of a preferred variant of a method according to the invention.

[0049] Figure 1Figure 1 shows a schematic representation of a computed tomography system 1 with a control unit 10 for carrying out the method according to the invention. The computed tomography system 1 has, in the usual manner, a scanner 2 with a gantry in which an X-ray source 3 rotates, irradiating a patient who is moved into a measurement chamber of the gantry by means of a table 5, so that the radiation strikes a detector 4 opposite the X-ray source 3. It is expressly pointed out that the embodiment shown in this figure is only one example of a CT scanner and that the invention can also be used with any CT design, for example, with a ring-shaped stationary X-ray detector and / or multiple X-ray sources.

[0050] Similarly, only those components of the control unit 10 are shown that are essential for integrating the invention into the CT system (see also Figure 3In principle, CT systems and associated control devices are familiar to experts and therefore do not need to be explained in detail.

[0051] A core component of the control unit 10 is a processor 11, on which various components are implemented as software modules. The control unit 10 also has a terminal interface 14, to which a terminal 20 is connected, allowing an operator to control the control unit 10 and thus the computed tomography system 1. Another interface 15 is a network interface for connecting to a data bus 21, thereby establishing a connection to a RIS (Radiology Information System) or PACS (Picture Archiving and Communication System).

[0052] The scanner 2 can be controlled by the control unit 10 via a control interface 13; that is, for example, the rotation speed of the gantry, the movement of the patient table 5, and the X-ray source 3 itself are controlled. The raw data RD from the detector 4 is read out via an acquisition interface 12. Furthermore, the control unit 10 has a storage unit 16 in which, among other things, various measurement protocols are stored.

[0053] As a software component, an image data reconstruction unit 18 is implemented on the processor 11, with which the desired image data are reconstructed from the raw data RD obtained via the data acquisition interface 12 and pre-processed by the system 6.

[0054] System 6 and its operation are described in more detail below.

[0055] Figure 2This paper presents a state-of-the-art method for decomposing the base material and correcting beam hardening artifacts. The raw data RD comprises a plurality of (not yet reconstructed) images at a plurality of energies, which can be equated here with the vector components P i. These images are reconstructed into images using several reconstruction units G1, G2, G3 of image reconstruction unit 18. A reconstruction unit G1, G2, G3 is used for each vector component. Using the line integrals L1, L2, L3 of the base material components, the reconstructed raw data are given as a function of the material components. Due to the strict monotonicity in all components, the mapping from RN< → RN< can be inverted, so that L n = G n (P 1 ,...,PN ) can be determined for n = 1,...,N. The reconstruction of the L n raw data, e.g.Filtered backprojection then yields a CT image that represents the local material fractions of the i-th base material. CT reconstruction of the Ln data provides the spatial distribution of the n-th base material, a so-called base material decomposition.

[0056] Although this approach allows for a precise decomposition of the base material, it leads to noise amplification, particularly when large proportions of non-water-like materials are present. In clinical CT, this applies to bone, metals, or high contrast agent concentrations.

[0057] Figure 3Figure 1 shows an example of a system 6 according to the invention for preprocessing a plurality of simultaneously acquired CT scans at different acquisition energies using a method according to the preceding claims. The system is shown here in a rough schematic representation, with boxes representing the functional units. Raw data RD are input into the system from the left. The symbol "P 1" indicates that the individual data sets of the raw data RD, which were acquired at the different acquisition energies, i.e., in different spectral channels, essentially form the subsequent vector entries of the vector P. The system comprises the following components: A vector unit 61, which is designed to form a vector P with the raw data RD of the CT scans at one energy per vector component, such that the dimension of the vector P corresponds to the number N of acquired energies. These can be all energies, but theoretically also fewer.

[0058] A matrix unit 62, which in this example is designed to determine an N x N matrix A. In a simple case, this matrix can be stored in it, but it can also store several matrices and select a suitable matrix for each specific investigation.

[0059] A transformation unit 63, designed to generate a transformed vector Q = A · P by multiplying the vector P by the matrix A. This unit and subsequent units can, in particular, be implemented by software modules on a computing unit.

[0060] A processing unit 64, designed to process the data of the transformed vector Q with a processing function B, and to form the processed transformed vector Q' = B(Q).

[0061] A reverse transformation unit 65 designed to form the reverse-transformed vector P' = Z · Q' from the (in this example) N x N matrix Z and the processed transformed vector Q' = B(Q).

[0062] An output unit 66 designed for outputting or further processing the back-transformed vector P' with the pre-processed raw data RD'.

[0063] Figure 4 Figure 1 shows a schematic representation of the inventive method for preprocessing a plurality of simultaneously recorded CT images, which were recorded at a plurality of (recording) energies, i.e. in different spectral channels.

[0064] In step I, a vector P is created from the raw data RD of the CT scans, each acquired at one energy level per vector component Pi. In this example, three sets of raw data RD were acquired at three different energies. Vector component P1 is intended to contain the raw data RD from the scan at the lowest energy, vector component P2 the raw data RD at a medium energy, and vector component P3 the raw data RD at the highest energy. The dimension of vector P is therefore N = 3.

[0065] In step II, an N x N matrix A is determined. In this example, this matrix has a one in its first row on the main diagonal and zeros in the rest. The remaining rows can essentially be chosen arbitrarily, but the vector Q should later contain all the information from the raw data RD. Although A could theoretically also be the identity, the procedure works best when A is not equal to the identity.

[0066] In step III, a transformed vector Q = A · P is generated by multiplying the vector P by the matrix A. Due to the special form of the matrix, in this example: Q₁ = P₁, meaning it is the raw data set RD that was recorded at the lowest energy.

[0067] In step IV, the components of the transformed vector Q are processed with a processing function B, thereby forming the processed transformed vector Q' = B(Q). This processing can, for example, be noise reduction.

[0068] In step V, the inversely transformed vector P' = Z · Q' is formed from the N x N matrix Z and the processed transformed vector Q' = B(Q). In this example, Z is precisely the inverted matrix A - 1< . Therefore, the components of the inversely transformed vector P' contain the data sets of the (now preprocessed) raw data RD' at the different energies (each component corresponds to a different energy). It is even possible that P' 1 = P 1, at least if this value itself was not processed but used as a reference value for other processing, e.g., in structure-preserving denoising.

[0069] In step VI, the inversely transformed vector P' containing the preprocessed raw data RD' is output, e.g. for further processing. Figure 2 .

[0070] Figure 5Figure 6 shows an example of a possible sequence of a preferred variant of a method according to the invention. From left to right, the raw data RD, in the form of a vector P, are transformed into a transformed vector Q by a transformation unit 63 through matrix multiplication with matrix A. In a processing unit 64, the transformed vector Q is denoised using the denoising function A, whereby a structure-preserving denoising is performed with component Q1 as the comparison component. The results of this denoising (i.e., essentially the processed transformed vector Q') are back-transformed to the back-transformed vector P' in the inverse transformation unit 65. This is done by multiplying Q' by the inverse matrix of A.

[0071] The components of the inversely transformed vector P' are then output on the right, and images and possibly material components are reconstructed using the reconstruction units G1, G2, G3.

[0072] Figure 6 shows an example of a possible sequence of a preferred variant of a method according to the invention. This sequence is based on the one described in Figure 5 The difference is that the processing of the transformed vector Q does not take place in the space of the raw data RD, but rather, using reconstruction routines (marked here with R{}) of an image reconstruction unit 18, the components Q i of the transformed vector Q are first reconstructed into images, these are denoised with the processing unit 64 using the denoising function A, and then the denoised images are transformed back into the space of the raw data RD with the reverse reconstruction unit 19 (marked here with S{}).

[0073] In this example as well, component Q1 serves as a reference value and is not denoised. Therefore, it does not necessarily need to be back-transformed, as it can simply be passed through. Only reconstruction into image space is necessary, since the denoising takes place within that space.

[0074] Figure 7 This shows an example of a possible sequence of a preferred variant of a method according to the invention. This is similar to Figure 6 with the difference that the image reconstruction takes place before the vector P is entered into transformation unit 63. Here, unlike in Figure 6 The component Q1 can also be reconstructed, but this depends entirely on the type of matrix A and Z.

[0075] Finally, it should be noted once again that the methods described in detail above, as well as the illustrated computed tomography system 1, are merely exemplary embodiments which can be modified in various ways by a person skilled in the art without departing from the scope of the invention. Furthermore, the use of the indefinite articles "one" or "an" does not preclude the possibility that the features in question may be present multiple times. Likewise, the terms "unit" and "module" do not preclude the possibility that the components in question consist of several interacting sub-components, which may also be spatially distributed. The expression "a number" is to be understood as "at least one".

Claims

1. Computer-implemented method for pre-processing a plurality of simultaneously recorded CT images with raw data (RD), which have been recorded with different recording energies, the method comprising the steps: - forming a vector P with the raw data (RD) of the CT images with in each case one energy per vector component so that the dimension of the vector P corresponds to a number N of recorded energies, - determining an N x M matrix A, - generating a transformed vector Q = A · P by multiplication of the vector P with the matrix A, - processing the components of the transformed vector Q with a processing function B, and forming the processed, transformed vector Q' = B(Q), wherein the processing of the components of the transformed vector Q is a denoising and the processing function B causes the denoising of the components of the transformed vector Q, - forming the back-transformed vector P' = Z · Q' from the M x O matrix Z and the processed, transformed vector Q' = B (Q), wherein the matrix A can be inverted and the back-transformed vector P' = A-1 · Q' is formed from the inverse matrix of A as matrix Z and the processed, transformed vector Q'= B(Q), - outputting or further processing the back-transformed vector P', characterised in that - the first component of the vector P is selected as a comparison value for the processing and the matrix A is determined such that in the line which acts on this component during the transformation, the matrix coefficient on the main diagonal has the value 1, and all other matrix coefficients of the lines have the value 0, wherein the matrix A in N-1 lines has the line total of 0, and - during the denoising of the components of the transformed vector Q, a structure-containing noise reduction is carried out, in which the first component Q1 of the transformed vector Q is used as a guide entity for a geometric synchronisation, wherein the other components Q2 to QN of the transformed vector Q represent differences with respect to the first component Q1 of the transformed vector Q with other intermediary spectra from the master spectrum of the first component Q1 of the transformed vector Q.

2. Method according to claim 1, wherein the back-transformed vector P' is used in a subsequent step to determine results from the raw data (RD), preferably to determine - base material data and / or - monochromatic data and / or - electron density and / or - nuclear charge and / or - portions of absorption effects, in particular Compton effect or photoelectric effect.

3. Method according to one of the preceding claims, wherein CT images are reconstructed before processing the components of the transformed vector Q, these CT images are processed with the processing function B and the processed CT images are preferably back-reconstructed again to form raw data (RB').

4. Method according to claim 3, wherein - the reconstruction of CT images takes place before the formation of the transformed vector Q and the back reconstruction of the CT images takes place after the formation of the back-transformed vector P', - the reconstruction of CT images takes place after the formation of the transformed vector Q and the back reconstruction of the CT images takes place after the formation of the back-transformed vector P', - the reconstruction of CT images takes place after the formation of the transformed vector Q and the back reconstruction of the CT images takes place before the formation of the back-transformed vector P', or - the reconstruction of CT images takes place before the formation of the transformed vector Q and the back-reconstruction of the CT images takes place before the formation of the back-transformed vector P'.

5. Method according to one of the preceding claims, wherein the matrix Z in N-1 column has the column sum of 0.

6. Method according to one of the preceding claims, wherein the components of the back-transformed vector P' are reconstructed within the scope of an image reconstruction to form images and CT images CT(E) are preferably generated after or within the scope of this image reconstruction for different energies E using the formula CT(E)=Σµi(E)Li(rP'i) with the energy-dependent coefficient µ, a reconstructed image rP'i from the i'th component of the vector P' and the line integrals Li.

7. System (6) for preprocessing a plurality of simultaneously recorded CT recordings with raw data (RD), which have been recorded with different recording energies, comprising means for executing a method according to the preceding claims, and comprising: - a vector unit (61) designed to form a vector P with the raw data (RD) of the CT recordings with in each case one energy per vector component, so that the dimension of the vector P corresponds to the number N of recorded energies, - a matrix unit (62) designed to determine an N x M matrix A, - a transformation unit (63) designed to generate a transformed vector Q = A · P by multiplication of the vector P with the matrix A, - a processing unit (64) designed to process the components of the transformed vector Q with a processing function B and forming the processed transformed vector Q' = B(Q), - a back-transformation unit (65) designed to form the back-transformed vector P' = Z · Q' from the M x O matrix Z and the processed, transformed vector Q' = B(Q), - an output unit (66) designed to output or further process the back-transformed vector P'.

8. Control facility (10) comprising a system (6) according to claim 7.

9. Computed tomography system (1) comprising a control facility (10) according to claim 8.

10. Computer program product, comprising commands, which, upon execution of the program by a computer, trigger this to execute the steps of the method according to claim 1 to 6.

11. Computer-readable storage medium, comprising commands, which, upon execution by a computer, trigger this to execute the steps of the method according to claim 1 to 6.