DE102013206525A1 - Reconstruction method for generating a tomographic representation of an examination object, with computer and CT system for carrying out this method - Google Patents

Abstract

The invention relates to a method for generating a tomographic representation of an examination volume in an examination object, a computer and a CT system for carrying out this method, the method comprising the following method steps: recording or receiving at least one initial sinogram (P0) from projections a multiplicity of projection directions, - reconstruction of a first tomographic representation (I0) of the examination volume from at least one initial sinogram (P0) - detection of edges in the first representation (I0), - tracing and marking of edge sinogram data in the at least one initial sinogram (P0 ), which originate from the projection of a predetermined area around the edges, - noise reduction of the at least one sinogram (P0), the edge sinogram data experiencing less or no noise reduction than the non-edge sinogram data, - reconstruction of an at least least provisionally final tomographic representation (Ii) from the noise-reduced at least one initial sinogram (Pi).

Description

The invention relates to a reconstruction method for generating a tomographic representation of an examination subject and a computer and a CT system for carrying out this method.

It is well known to use conventional backprojection (FBP) reconstruction techniques in computed tomography to compute three-dimensional image data sets from a plurality of projection data and from a sinogram, respectively. In this case, there is a correlation between image sharpness and image noise in such a way that the image noise deteriorates at the same time as an optimization of the image sharpness, or vice versa, the image sharpness simultaneously deteriorates when the image noise is reduced.

Basically, methods for iterative image reconstruction are known in the prior art. There are two variants, namely raw-data-based and image-data-based iterative reconstruction. Regardless of the type of iterative reconstruction, the same problem exists here of the opposite dependence of the quality with respect to image sharpness and image noise.

It is an object of the invention to find a reconstruction method in which this contrastiveness of the image quality with respect to image sharpness and image noise is broken, so that it is possible to reconstruct tomographic CT images which are simultaneously optimized with respect to their image sharpness as well as their image noise.

This object is solved by the features of the independent claims. Advantageous developments of the invention are the subject of the subordinate claims.

The inventors have recognized that it is possible by skillful combination of iterative steps from both an image-based and raw-data-based iterative reconstruction to cancel out the interaction between image noise and image sharpness and at the same time be able to improve both the image noise and the image sharpness, or by improvement according to the invention The image data with regard to image noise and sharpness with the same image quality to reduce the dose for image production significantly.

und die Vorwärtsprojektion als Funktion

beschrieben werden. Die jeweils anzuwendenden Typen von Vor- und Rückprojektoren können dabei beliebig je nach Applikation gewählt werden. Als Beispiel für einen Rückprojektor sei die gewichtete gefilterte Rückprojektion (WFBP), sowie als Vorwärtsprojektor der Joseph-Projektor erwähnt.In raw-data-based iterative reconstruction, a correction of the reconstruction artefacts is achieved by multiple forward and backward projecting of the image and raw data, mostly using filtered rear-projectors and forward projectors. Occurring artefacts, apart from Strahlaufhärtungs- and scattered beam artifacts, are usually due to excessive Conebeamwinkel and the non-exact FBP (= filtered rear projection). Noise reduction is not performed here. On the contrary, if filters are not used during pre- and post-projection, artifacts may be reduced while at the same time the image noise is increased. The raw-data-based method works exclusively in the raw data space and uses the transition to the image data space only for a faster solution of a linear system of equations. The process of rear projection is intended here as a function

and the forward projection as a function

to be discribed. The applicable types of front and rear projectors can be chosen arbitrarily depending on the application. As an example of a rear projector, the weighted filtered rear projection (WFBP), as well as mentioned as forward projector of the Joseph projector.

Projection data are denoted by P, and reconstructed image data by I. n denotes the iteration step, N the maximum number of iteration steps, started at 0 for the initial volume I ₀ , γ the admixing of the correction term from the current iteration step to the currently reconstructed volume. Whether the correction term is added up or subtracted here depends on the constellation of the equation. The negative sign here only makes it clear that artifacts are "subtracted away".

Equations (1) and (2) describe the raw-data-based iterative reconstruction:

Any intermediate steps have been omitted here for reasons of better readability, however, filter steps are provided to compensate for numerical instabilities.

Equation (3) is intended to illustrate the additional optional filtering steps, the filters being designated A through F throughout. The respective implementation is application-specific.

In all other equations, the possible, optional filter steps are not specified separately. The equation (3) is used here only to represent the complete equation.

In the image data-based iterative reconstruction, noise reduction is achieved by multiple or iterative application of an image-based adaptive filter, for example a diffusion filter or a bilateral filter. In this case, either data directly from the filtered backprojection I ₀ or from a result of the raw-data-based iterative reconstruction I _{1... N} can be used.

A basic problem here, however, is that the usual image texture is not preserved and high frequencies in the image space are destroyed by multiple convolution, ie by low-pass filtering. This gives the impression of a blocky unnatural noise pattern structure.

Another problem stems from the fact that only a finite resolution during reconstruction, typically a 512x512 image matrix with a field of view of 500 mm, is used. However, the frequencies which are present in the image receiver system and thus in the projections are not transmitted to the image in the process of rear projection due to the low image matrix or spatial resolution in the image. This has the consequence that working on image data whose noise power spectrum has already undergone a low-pass filtering on which the subsequently iterative image processing algorithm can no longer function optimally and optionally per iteration further deterioration of the noise power spectrum is effected.

Another problem is that edges in the image disappear through the iteration. Current algorithms also degrade the edge information by parameterizing the adaptive average filter for noise suppression only by the amount of noise, but taking no account of edge information and thereby spoiling edges. Ground edges are then restored by an additional term, and the edges can only be restored if they could be detected beforehand. Since the image-based iterative reconstruction has a multitude of different processing steps, only an exemplary example of such an image-based iterative reconstruction will be presented in a formula.

gibt einen adaptiven Glättungsfilter an, der über eine pixelweise Rauschinformation

gesteuert wird, sowie

die Kanteninformation. α, β und γ stellen die Möglichkeit dar, zu steuern, welche Beiträge von den Einzelkomponenten zum Gesamtresultat I_n zusammenfließen.I _n denotes the result of an image-based iterative reconstruction. The input is a result of the filtered back projection or a result of an iteration step from the raw data-based iterative reconstruction. Again, analogous to raw-data-based iterative reconstruction, n specifies the iteration step, where n = 0 ... N, but the number of iterative steps N of the image-based iterative reconstruction need not match that of the raw-data-based iterative reconstruction.

indicates an adaptive smoothing filter that has pixel-wise noise information

is controlled, as well

the edge information. α, β and γ represent the possibility of controlling which contributions from the individual components converge to the overall result I _n .

According to the invention, a raw data and an image-based iterative reconstruction are combined. Here, analogous to the conventional raw-data-based iterative reconstruction and for image-data-based iterative reconstruction, an initial filtered backprojection-preferably with a sharp core to obtain the high frequencies-is first performed. In order for all frequencies to actually be transmitted to the image space, the matrix size should be at least 512 × 512, but better is a higher matrix size of 1024 × 1024 in order to transmit the spatial resolution of the projection. Optimal is a matrix size with the same maximum spatial frequency as the image generator system or the projection. However, since the field of view has an additional influence on the spatial frequencies available in the image, it makes sense to provide a variable matrix size that adapts itself to the particular conditions of the reconstruction itself.

sowie eine Rauschschätzung

erfolgen. Anschließend muss eine Vorwärtsprojektion des Volumens I₀ erfolgen, um mittels Gleichung (5) artefaktreduzierte Projektionsdaten P_n zu erzeugen. Siehe hierzu auch Gleichung (2), jedoch wird hier noch eine zusätzlich Vorwärtsprojektion angewandt.This gives an initial volume I ₀ , as already known from the conventional reconstruction. With this volume as a basis can be an edge detection

as well as a noise estimate

respectively. Subsequently, a forward projection of the volume I _{0 must} be made in order to generate artifact-reduced projection data P _n by means of equation (5). See also Equation (2), but here an additional forward projection is used.

Furthermore, a forward projection of the noise and the edge information must take place. Then, according to equations (6) and (7), a projection of the noise information P ^N and a projection of the edge information P ^{E are obtained} :

auf P_n durchgeführt, die als Adaptionsgrundlage für die Parametrierung des Filters die aus der Rekonstruktion I_n ermittelte Kanten P E / n und Rauschinformation P N / n verwendet wird. Es ist vorteilhaft die Rausch- und Kantenschätzung mit jeder Iteration neu zu berechnen, da durch die rohdatenbasierte iterative Rekonstruktion eventuelle Artefakte bereits reduziert wurden, die diese beiden Operationen verbessern können. Die neue intermediäre Projektion P'_n wird dann gemäß der nachfolgenden Gleichung (8) so berechnet, dass der adaptive Filter unter zu Hilfenahme der Rausch- und Kanteninformation entsprechend der Rauschamplitude mehr oder weniger stark filtert. α, β und γ können hier als freie Parameter angesehen werden, die analog zu Gleichung (4) eine Möglichkeit bieten, die Beiträge der einzelnen Teilergebnisse einzustellen. α gibt den Beitrag der originalen Projektion, β den Beitrag der geglätteten Projektion, γ den Beitrag einer zusätzlichen Kanteninformation an. Damit ist es möglich mit γ Kanten zu verstärken oder abzuschwächen und mit β zusätzlich den Grad der Glättung beizumischen.Subsequently, an adaptive filtering / smoothing

performed on P _n , which as an adaptation basis for the parameterization of the filter, the determined from the reconstruction I _n edges P E / n and noise information P N / n is used. It is advantageous to recalculate the noise and edge estimate with each iteration, since raw art-based iterative reconstruction has already reduced any artifacts that can improve these two operations. The new intermediate projection P ' _n is then calculated according to the following equation (8) so that the adaptive filter more or less strongly filters with the aid of the noise and edge information corresponding to the noise amplitude. α, β and γ can be considered here as free parameters which, analogous to equation (4), offer a possibility of setting the contributions of the individual partial results. α indicates the contribution of the original projection, β the contribution of the smoothed projection, γ the contribution of additional edge information. This makes it possible to amplify or attenuate edges with γ and to additionally add the degree of smoothing with β.

The adaptive filter D should work as follows. First, each detector channel is considered individually. For each detector channel, the forward projected noise information P N / n parameterizes the smoothing width of the filter for the respective detector channel. The smoothing is performed isotropically in all spatial directions (2D), however, based on the forward-projected edge information P E / n determines where the smoothing should be stopped. Thus, a two-dimensional smoothing of a complete projection without the edges to grind. Subsequently, as described in equation (9), a filtered backprojection of all smoothed projections P ' _{n takes place} , whereby backprojection can be done either in a larger matrix (1024 × 1024 or larger) or in the conventional matrix size.

angepasst werden. Für den Fall, dass nachträglich eine bildbasierte iterative Rekonstruktion erfolgen soll, kann diese direkt auf dem final rückprojizierten Bild angewendet werden, wobei analog die Verarbeitungsschritte aus der bekannten bildbasierten iterativen Rekonstruktion gemäß Gleichung (4) verwendet werden können.The equation (9) can now be applied iteratively n times, wherein the parameters α, β and / or γ can be changed per iteration step. Alternatively or additionally, the adaptive smoothing filter can also be used

be adjusted. In the event that an image-based iterative reconstruction is to take place subsequently, this can be applied directly to the final backprojected image, whereby the processing steps from the known image-based iterative reconstruction according to equation (4) can be used analogously.

und Kantenschätzung

nicht wie bisher auf die Bilddaten angewendet, sondern es werden zunächst durch eine initiale Rekonstruktion diese beiden Parameter ermittelt und anschließend durch eine Vorwärtsprojektion auf die Rohdaten beziehungsweise Projektionsdaten angewendet.This above-described method according to the invention thus represents a combination of raw-data-based and image-data-based iterative reconstruction. In this case, the two components of the noise estimate

and edge estimation

not as previously applied to the image data, but initially these two parameters are determined by an initial reconstruction and then applied by a forward projection on the raw data or projection data.

One advantage is that the backprojection of the smoothed data and the intrinsic smear provide a more natural image impression than previous methods of image-based iterative reconstruction. Intrinsic smearing is understood to mean the smear inherent to the rear projector due to the interpolations made.

In addition, in the raw data space one works with the higher system resolution and not in the lower resolution of the reconstructed image matrix and the image resolution iterative reconstruction associated Field Of Views. Any problems that cause resolution degradation will only be present in raw data space if the noise would require too much smoothing.

A further advantage of the method according to the invention is that the previously used and programmed standard reconstruction methods can continue to be used since the data smoothing is performed before the back projection on the projection data. Furthermore, model and system parameters can be incorporated during the reconstruction and thus control the behavior of the reconstruction.

In the image-based iterative reconstruction subsequently no system parameters such. B. finite focus or detector size, more models, these would be able to be performed in the prior art only in the raw data-based iterative reconstruction. With the method according to the invention these advantages are combined.

Furthermore, it can be problematic that in the case of image-based iterative reconstruction, a three-dimensional filtering with an excessive expansion of the filter in the axial direction generates an automatic layer thickness broadening, which can have a negative effect on the image impression.

Another advantage of the method of the invention is that additional algorithms, such as metal artifact correction, beam hardening correction, can be easily inserted into the iterative loop. In the case of image processing or raw data processing algorithms, these can be combined in the inventive method and processed simultaneously or consecutively. This brings a huge time advantage.

A not insignificant core of the invention thus lies in the fact that now, not as previously, only in one or the other domain, ie on the raw data or image data, the processing is performed, but the raw data processing and image data processing are interlinked. Thus, according to the invention, the noise estimation is made on the image data, but the noise reduction is performed on the raw data, additionally based on information (edge detection) from the image data. As a result, unfavorable effects of image-based processing, such as blocky structures or Kantenverschmierungen be avoided.

• – Aufnehmen oder Empfangen mindestens eines initialen Sinogramms aus Projektionen aus einer Vielzahl von Projektionsrichtungen,
• – Rekonstruktion einer ersten tomographischen Darstellung des Untersuchungsvolumens aus mindestens einem initialen Sinogramm,
• – Detektion von Kanten in der ersten tomographischen Darstellung,
• – Rückverfolgung und Markierung von Kantensinogrammdaten in dem mindestens einen initialen Sinogramm, welche aus der Projektion eines vorgegebenen Bereiches um die Kanten stammen,
• – Rauschreduktion des mindestens einen Sinogramms, wobei die Kantensinogrammdaten eine geringere bis keine Rauschreduktion erfahren als die Nicht-Kantensinogrammdaten,
• – Rekonstruktion einer zumindest vorläufig endgültigen tomographischen Darstellung aus dem so rauschreduzierten mindestens einen initialen Sinogramm.

According to these explanations, the inventors propose a method for generating a tomographic representation of an examination volume in an examination subject, which comprises the following method steps:

• Picking up or receiving at least one initial sinogram from projections from a plurality of projection directions,
• Reconstruction of a first tomographic representation of the examination volume from at least one initial sinogram,
• Detection of edges in the first tomographic representation,
• Tracing and marking of edge sinogram data in the at least one initial sinogram resulting from the projection of a predetermined area around the edges,
• Noise reduction of the at least one sinogram, the edge sinogram data having less to no noise reduction than the non-edge sinogram data,
• Reconstruction of an at least provisionally final tomographic representation from the noise-reduced at least one initial sinogram.

Thus, in the tomographic image data, edges are detected and intersected therefrom, for example by a geometric observation, which determines sinogram data representing these edges or their immediate surroundings, and because of this knowledge edge-related sinogram data (edge sinogram data) less to not noise-reduced, while the other sinogram data (non-edge sinogram data), which have no edges or their surroundings, undergo a noise reduction or are more noise-reduced. At the same time, it is also possible to proceed across domains in the noise reduction by estimating the noise intensity in the tomographic representation on a pixel-by-pixel basis and executing the noise reduction at the pixel-corresponding points in the sinogram.

• – eine nur die Kanten enthaltende tomographische Kantendarstellung erzeugt und die Kantendarstellung auf ein Kantensinogramm vorwärtsprojiziert wird,
• – eine nur Rauschinformation enthaltende tomographische Rauschdarstellung erzeugt und die Rauschdarstellung auf ein Rauschsinogramm vorwärtsprojiziert wird, und
• – die Rauschreduktion des initialen Sinogramms unter Verwendung des Rauschsinogramms und des Kantensinogramms ausgeführt wird.

It is advantageous to design the method in such a way that:

• Generates a tomographic edge representation containing only the edges and the edge representation is projected forward onto an edge sinogram,
• Generating a tomographic noise representation containing only noise information and projecting the noise representation forward to a noise ingram, and
• The noise reduction of the initial sinogram is carried out using the squared ingram and the squared ingram.

Furthermore, the method with respect to the edge detection, the noise reduction and the reconstruction of the at least provisionally final tomographic representation on the noise-reduced at least one initial sinogram can be performed iteratively multiple times.

In this case, in a simplest variant, the number of iterations can be used as the termination criterion for the iteration. Alternatively, the iteration can be completed even after reaching a predetermined calculated image quality or a subjective image impression in the provisionally final tomographic display. However, it is also possible that the iteration is aborted if predetermined image quality criteria no longer improve from iteration to iteration or the improvement falls below a predetermined value.

However, it is also part of the scope of the invention for a final image processing to be carried out on the preliminary final tomographic display following the previously described method. Here, too, information from the previously determined edge display can be used to avoid edge smearing. According to the invention, such a final image processing can also be designed as a method operating in itself iteratively.

In principle, any reconstruction method can be used as the reconstruction method, but it is particularly advantageous and efficient if back-projection filtering and preferably at least once a weighted filtered back-projection (WFBP) is used for the reconstruction.

Furthermore, it is advantageous to use in the reconstruction of an edge-enhancing filter, since the subsequent edge detection is facilitated. The edge detection in the tomographic display can take place either layer by layer in the form of a two-dimensional edge detection or as three-dimensional edge detection via the tomographic 3D representation. One of the following operators can be used for edge detection: Sobel operator, Kirsch operator, Laplace operator.

Furthermore, a Joseph forward projector can be used particularly advantageously for forward projection, a bilateral filter or an anisotropic diffusion filter for noise reduction.

In addition, it is also proposed, following the noise reduction in the sinogram, to add the edge sinogram weighted to this noise reduced sinogram in order to emphasize or enhance the edge information.

Furthermore, a spatial resolution should be used for the tomographic representation, which is equal to or greater than the spatial resolution of the projections or the resulting sinograms.

In addition to the method according to the invention, the inventors also propose a computer with a memory for program code, which is executed during operation, wherein the memory is a program code, which realizes the method described above.

Likewise, a CT system with a computer with a memory for program code, which is executed during operation, is proposed, wherein the memory is a program code, which executes the inventive method.

In the following the invention with reference to the preferred embodiments with reference to the figures will be described in more detail, with only the features necessary for understanding the invention features are shown. The following reference symbols are used: E: edge representation; I ₀ : first tomographic representation; I _i : i tomographic representation; N: noise representation; P ₀ : projections; P _i : Noise-reduced projections; P ^E : edge projections; P ^N : noise projections; S1-S11: process steps; Prg ₁ prg _n : computer programs; 1 : CT system; 2 : first X-ray source; 3 : first detector; 4 : second X-ray source; 5 : second detector; 6 : Gantry housing; 7 : Patient; 8th : Patient couch; 9 : System axis; 10 : Control and computing unit.

They show in detail:

1 : Overview of the process sequence of the invention;

2 : Detailed representation of the method according to the invention in an iterative variant;

3 : CT system for carrying out the method according to the invention.

The 1 first shows the inventive method in a rough overview. Accordingly, in the first method step, first the projections Po or the at least one sinogram are received from a present data pool or generated by scanning an examination object with a CT system.

In the reconstruction step, an initial tomographic representation I _{0 is} reconstructed, which is then processed in two separate processes. These processes can be carried out in parallel or serially, it is essential that the information obtained from this is available in parallel for the further process. In this case, an edge estimation is carried out with the calculation of a tomographic edge representation E and a noise estimation with the calculation of a noise representation N, in each case in the same image space of the initial tomographic representation.

By cross-linking the information from the edge representation E and the noise representation N, a noise reduction on the initial projection data P ₀ or the resulting sinogram is then performed, the sinogram data correlated with edges being less noise-reduced than the sinogram data are not correlated with edges, so that edgewise noise-reduced projection data P _i or a sinogram formed therefrom arises.

With the projection data treated in this way, a preliminary final tomographic representation I _{i is} reconstructed in the provisionally last step, whereby due to the pretreatment of the projection data on the basis of the knowledge gained in the image space with regard to the present edge positioning, a noise reduction could be achieved without simultaneously impairing the edge presentation ,

Optionally, in addition, the representation thus obtained can be improved with image processing methods known per se.

An improved and more detailed illustration of this method according to the invention is described in the 2 shown.

There, the method begins with step S1, where first the raw data collected in the form of projection data P ₀ and in step S2 in a conventional manner, for. B. by means of a filtered FBP or WFBP, be reconstructed. This results in an initial tomographic representation of a reconstructed volume stack I ₀ , consisting of n layers.

The reconstruction kernel to be used here should be chosen so that the subsequent steps for edge detection and noise estimation are optimally executed. The reconstruction kernel should be chosen so that edges are preserved, but the noise is not too pronounced. In general, however, it makes sense to focus on preserving the edges, as the subsequent steps reduce the noise anyway. In addition, algorithms can be used that reinforce the edges.

The edge estimation and noise estimation can be done in parallel.

In step S3, an edge detection is performed on the first tomographic display I ₀ either in 2D, that is, in layers, or in 3D. This can be done by conventional edge detection methods, eg. B. with a Sobel, Kirsch or Laplace operator or more complex documented in the literature methods. The result of this operation is an edge representation E in the same dimension as I ₀ .

The noise estimation in step S4 is based on the evaluation of the noise pixel by pixel over the entire representation I ₀ . In this case, a statistical evaluation of the noise is performed in an environment around the respective pixel. This produces a noise representation N with the same dimension as the first tomographic representation I ₀ . This can be z. This can be done, for example, by analyzing the noise in an environment in two or three spatial directions. The size of the environment should be chosen so that only homogeneous areas are included in the calculation of the noise. In this case, the environment can be chosen so small that only rarely homogeneous areas are found, and the problem of defective noise is negligible.

• a) Es werden parallel die Kantendetektion und die Rauschschätzung durchgeführt. Nachteilig ist, dass die Rauschschätzung keine Information über die Struktur des Datensatzes hat und daher das Rauschen unter Umständen nicht optimal abschätzt. Der Vorteil liegt darin, dass die Performance steigt, da eine parallele Verarbeitung möglich ist.
• a) Es werden die Schritte Kantendetektion und Rauschschätzung nacheinander durchgeführt. Dabei ist es günstig, dass basierend auf der Kanteninformation homogene Bereiche leichter identifiziert werden können und sich die Rauschabschätzung verbessert. Der Nachteil liegt in einer schlechteren Performance, da auf die Kantenschätzung gewartet werden muss.

Thus, after the reconstruction of the initial tomographic display I ₀ two possibilities arise:

• a) Edge detection and noise estimation are performed in parallel. The disadvantage is that the noise estimate has no information about the structure of the data set and therefore may not optimally estimate the noise. The advantage is that the performance increases because parallel processing is possible.
• a) The steps edge detection and noise estimation are performed successively. It is favorable here that, based on the edge information, homogeneous regions can be identified more easily and the noise estimation improved. The disadvantage lies in a poorer performance, since one has to wait for the edge estimate.

Subsequently, in steps S5 and S6, a forward projection of the tomographic edge representation E and the tomographic noise representation N is carried out. Since the denoudation in the raw data domain, ie on the sinogram, is to be carried out, the two tomographic representations N and E from the image data domain must be entered into the raw data domain be transformed. This is done via a forward projection, z. Using a Joseph forward projector. This results in two sinograms: P ^{N of} the forward projected volume N and P ^{E of} the forward projected volume E.

In step S7, the de-noise of the projection P _{0 takes place} . For this purpose, a noise filtering of the projection P _{0 is} carried out on the basis of the determined noise and edge information P ^N and P ^E. In the simplest case, a Gaussian filter can be used whose strength is parameterized via the noise information P ^N , wherein, taking into account the edge information P ^E, an edge-preserving filter, eg. As bilateral filter or anisotropic diffusion filter is used. In this case, the edge information represents a stop function (barrier) over which the edge-preserving filters must not smoothen. This results in a noisy projection P _i . The index i should denote the current iteration.

Subsequently, optionally, the edge information P ^E can be added to P _{i for} weighting the edges. P _i = P _i + γ P ^e

The final rear-projection of P _i then takes place with step S8. This can be achieved by conventional rear projection methods, eg. B. filtered rear projection done. Based on whether the iteration is considered complete, the final kernel can be used as the reconstruction kernel. The result is a new noise-reduced tomographic representation I _i in the form of a volume stack.

The next step S9 determines whether the iteration has ended. If so, in step S11 the output of the final tomographic representation follows. If no, the system jumps back to a new initial reconstruction.

If the iteration has not ended yet, other cores can be used that optimally express the edge and noise information as described in step 1. Here, too, post-processing filters can be used which obtain the edge information or highlight other characteristics of the image information.

Whether an iteration has ended or not can be determined in the simplest way by the number of iterations or by evaluation criteria of the image quality or by an image quality that no longer improves significantly. A rating can z. Example, be that the noise of the original image has been lowered by a defined factor. In an empirical termination of the iteration, the number of iterations was determined by measuring on a collective.

Optionally, prior to the output of the final tomographic image I _i , it may still pass through the image post-processing step S 10 shown in dashed lines. This z. B. subsequent algorithms are applied, the z. For example, add the edge information E from the previous iteration or from the first iteration weighted to the representation I _i to subsequently reinforce the edge information. In addition, an iterative image-based method according to the prior art, for. For example, IRIS (Iterative Reconstruction in Image Space) can be applied to the volume stack in order to carry out subsequent de-icing. Here, too, it is optionally possible to resort to the edge information determined in step S3 and to make the smoothing less pronounced in the region of existing edges than in regions without edges.

The method according to the invention described above can be carried out, for example, separately with previously generated projection data on a separate computer or also directly in a CT system with a computer used as a computing and control unit. Such an exemplary CT system 1 is in the 3 shown schematically. Such a CT system usually has a gantry housing 6 , in which there is a gantry, not shown here, on the at least one X-ray source 2 with an oppositely arranged detector 3 are attached. During scanning and to generate projection data, the rotating part of the gantry rotates with the radiator 2 and the detector 3 while a patient 7 with the help of a movable patient bed 8th along the system axis 9 continuously or sequentially through the measuring field in the gantry housing 6 is pushed. Optionally, also another spotlight detector system 4 . 5 arranged angularly offset on the gantry, which then makes it possible to simultaneously capture further projections.

This CT system is controlled 1 through the control and computing unit 10 which has a memory for corresponding computer programs Prg ₁ -Prg _n . The method according to the invention can also be carried out with such a computer, wherein corresponding program code is stored in the memory of the computer, which executes the method according to the invention during operation.

Overall, the invention proposes a method for generating a tomographic representation by raw data and image data processing, in which noise and edge information from projection data and image data are used across raw data and image data in order to reduce noise while retaining edges in the generated tomographic representation to effect. In the context of the invention, in this process, in particular the method steps described in claims 1 to 18, in particular in the combination shown there, are carried out.

Although the invention has been further illustrated and described in detail by the preferred embodiment, the invention is not limited by the disclosed examples, and other variations can be derived therefrom by those skilled in the art without departing from the scope of the invention.

LIST OF REFERENCE NUMBERS

e

edge representation

I ₀

first tomographic representation

I _i

i-th tomographic representation

N

noise representation

P ₀

projections

P _i

Noise-reduced projections

P ^E

Edge projections

P ^N

Noise projections

Prg ₁ prr _n

computer programs

S1-S11

steps

1

CT system

2

first X-ray source

3

first detector

4

second X-ray source

5

second detector

6

gantry

7

patient

8th

patient support

9

system axis

10

Control and computing unit

Claims (21)

A method for generating a tomographic representation of an examination volume in an examination subject, comprising the following method steps: 1.1. Recording or receiving at least one initial sinogram (P ₀ ) from projections from a plurality of projection directions, 1.2. Reconstruction of a first tomographic representation (I ₀ ) of the examination volume from at least one initial sinogram (P ₀ ), 1.3. Detection of edges in the first representation (I ₀ ), 1.4. Tracing and marking of edge sinogram data in the at least one initial sinogram (P ₀ ) resulting from the projection of a given area around the edges, 1.5. Noise reduction of the at least one sinogram (P ₀ ), where the edge sinogram data is less to no noise reduction than the non-edge sinogram data, 1.6. Reconstruction of an at least provisionally final tomographic representation (I _i ) from the noise-reduced at least one initial sinogram (P _i ) Method according to the preceding claim 1, characterized in that: 2.1. generates a tomographic edge representation (E) containing only the edges and the edge representation (E) is projected forward onto an edge sinogram (P ^E ), 2.2. generates a tomographic noise representation (N) containing only noise information, and projects the noise representation ( ^N ) forward to a squid sinogram (P ^N ), and 2.3. the noise reduction of the initial sinogram (P ₀ ) is performed using the noise sinogram (P ^N ) and the edge sinogram (P ^E ). Method according to one of the preceding claims 1 to 2, characterized in that the method with respect to the edge detection, the noise reduction and the reconstruction of the at least provisionally final tomographic representation from the noise-reduced at least one initial sinogram (P _i ) iteratively performed multiple times. Method according to the preceding claim 3, characterized in that the iteration is completed after a predetermined number of iterations. Method according to the preceding claim 3, characterized in that the iteration is completed after reaching a predetermined calculated image quality or a subjective image impression in the provisionally final tomographic representation. Method according to one of the preceding claims 1 to 5, characterized in that finally takes place an image processing on the provisionally final tomographic representation (I _i ). Method according to the preceding claim 6, characterized in that in the final image processing information from the edge representation (E) are used to avoid Kantenverschmierung. Method according to one of the preceding claims 6 to 7, characterized in that the final image processing is designed as an inherently iterative method. Method according to one of the preceding claims 1 to 8, characterized in that for the reconstruction at least once a weighted filtered rear projection is used. Method according to one of the preceding claims 1 to 9, characterized in that in the reconstruction of an edge-enhancing filter is used. Method according to one of the preceding claims 1 to 10, characterized in that in the tomographic representation, a two-dimensional edge detection is performed in layers. Method according to one of the preceding claims 1 to 10, characterized in that in the tomographic representation, a three-dimensional edge detection is performed. Method according to one of the preceding claims 1 to 12, characterized in that one of the following operators is used for edge detection: Sobel operator, Kirsch operator, Laplace operator. Method according to one of the preceding claims 1 to 13, characterized in that for forward projection, a Joseph Vorprojektor is used. Method according to one of the preceding claims 1 to 14, characterized in that a bilateral filter is used for noise reduction. Method according to one of the preceding claims 1 to 14, characterized in that an anisotropic diffusion filter is used for noise reduction. Method according to one of the preceding claims 1 to 16, characterized in that, following the noise reduction on the noise-reduced sinogram (P _i ), the edge sinogram (P ^E ) is added in a weighted manner. Method according to one of the preceding claims 1 to 17, characterized in that the generated tomographic representation (I _i ) and the projections used (P ^E , P ^N ) each have a spatial resolution, wherein for the tomographic representation (I _i ) uses a spatial resolution is equal to or greater than the spatial resolution of the projections (P ^E , P ^N ) or the resulting sinograms. Method for generating a tomographic image from CT projection data by raw data and image data processing, wherein noise and edge information from projection data and image data are used across raw data and image data to effect noise reduction while retaining the edge in the generated tomographic image. Computer having a memory for program code which is executed during operation, characterized in that a program code (Prg ₁ -Prg _n ) is present in the memory, which executes the method according to one of the claims 1 to 19. CT system ( 1 ) with a computer with a memory for program code, which is executed in operation, characterized in that in the memory a program code (Prg ₁ -Prg _n ) is present, which carries out the method according to one of the claims 1 to 19.

Images