Method for image processing of CT images - Patent Application 20070233633

Abstract

The present invention provides a computer-implemented method for image processing of a CT image, the method comprising the steps of: performing one or more pre-processing steps on a CT image to obtain a pre-processed CT image, the one or more pre-processing steps including applying an edge-preserving noise removal algorithm; and performing an adaptive spike suppression algorithm on the pre-processed CT image to obtain a processed CT image. The adaptive spike suppression algorithm is configured such that the processed CT image has a reduced number of spikes compared to the pre-processed CT image.

Description

[Technical field]

[0001] The present invention provides a computer-implemented method for image processing of CT images, a method including the method for image processing, a data processing system, a system including the data processing system, a computer program product, and a computer readable medium. [Background technology]

[0002] Edge-preserving denoising processes can provide superior image texture and appearance by removing noise while preserving the underlying image details, but can also lead to an increased number of spikes, especially when the noise distribution has heavy tails, which occur in many situations such as CT photon counting (PhC) systems and low-dose CT scans, among others.

[0003] More specifically, photon-counting CT systems utilize direct conversion detector technology to acquire data at multiple energy levels, and therefore have the potential to provide details of various x-ray interactions, such as the photoelectric effect, Compton scattering, and k-edge energy components, and allow for more detailed quantification. Such a system may also have an improved signal-to-noise ratio (SNR) with respect to an integrating detector.

[0004] Noise in a CT image is inversely proportional to the square root of the average number of photons. When the signal amplitude from electronic noise is comparable in magnitude to that from the detected photons, the signal contribution from electronic noise becomes significant. One advantage of photon-counting detectors is that they significantly reduce electronic noise, which allows for reduced doses.

[0005] The improved SNR also enables tissue differentiation and / or material labeling and quantitative imaging with CT systems, as well as enabling novel imaging techniques such as k-edge imaging and reduction of beam hardening artifacts. However, as briefly mentioned above, a challenge posed to PhC is its heavy tailed noise characteristics.

[0006] By way of background, the concept of heavy-tailed noise is briefly described below. Noise will often have only an approximately Gaussian distribution. In particular, even if the center of the distribution is approximately Gaussian, the tails may not be. The tails of a distribution are regions of density that correspond to large distribution arguments. Heavy tails mean that for large values ​​of the density argument, the density approaches zero more slowly than a Gaussian distribution. The heavy tailed noise characteristic leads to increased occurrence of spikes after noise removal, especially at low doses.

[0007] Addressing this problem by removing spikes often does not produce satisfactory results because there will be too many false spike identifications and real texture and / or contrast will be lost, resulting in an artificially smooth surface that lacks detail. Summary of the Invention [Problem to be solved by the invention]

[0008] It is therefore an object of the present invention to alleviate some of the above mentioned problems and enable improved image quality, especially also in the context of challenging CT imaging scenarios such as PhC systems and low dose. [Means for solving the problem]

[0009] The invention provides according to the independent claims a computer-implemented method for image processing of CT images, a method comprising said method for image processing, a data processing system for performing image processing of CT images, a system comprising said data processing system, a computer program product and a computer readable medium. Preferred embodiments are defined in the dependent claims.

[0010] The present invention provides a computer-implemented method for image processing of a CT image, the method comprising performing one or more pre-processing steps on the CT image to obtain a pre-processed CT image, the one or more pre-processing steps including applying an edge-preserving denoising algorithm, the method further comprising performing an adaptive spike suppression algorithm on the pre-processed CT image to obtain a processed CT image, the adaptive spike suppression algorithm being in particular an algorithm configured such that the processed CT image has a reduced number of spikes compared to the pre-processed CT image.

[0011] The method described in the claims makes it possible to solve at least one of the above problems by performing spike suppression in an adaptive manner and on image data that has been previously denoised by edge-preserving denoising.

[0012] The claimed method allows for significant noise reduction, e.g. even up to 90% dose reduction, while effectively suppressing spike noise and preserving texture. The preservation of texture can provide the texture and appearance of conventional CT images, e.g. avoiding images with a plastic or artificial appearance. Thus, the method provides results that are compatible with current procedures that rely on the texture of conventional CT images. Denoising the data in a pre-processing step increases the signal-to-noise ratio of the spikes, which makes them easier to identify and remove. The present disclosure provides an adaptive noise reduction algorithm, in other words, the algorithm implements an adaptive spike noise suppression method (aSNS).

[0013] The method according to the present disclosure allows and is suitable for removing noise in PhC systems and low-dose CT scans, in particular for achieving significant noise reduction and suppression of spike noise while preserving image texture.

[0014] An adaptive algorithm means that the model used to process an image can be adapted for each image processed depending on whether the model meets certain criteria.

[0015] Edge-preserving denoising algorithms known in the art, such as the Philips iDose algorithm, can be applied in the pre-processing step. Applying the denoising algorithm may include fitting the data to a linear or quadratic model using a bilateral filter. This allows reliable removal of spikes and noise while preserving edges and textures.

[0016] As mentioned above, the term "spike suppression" refers to a reduced number of spikes in a processed image compared to a pre-processed image. Image pre-processing increases the signal-to-noise ratio (SNR) of spikes in the input CT image. The spike reduction specifically helps to remove these spikes. The processed image may have a reduced number of spikes compared to the image before pre-processing, but does not necessarily have a reduced number of spikes.

[0017] The CT images in this disclosure may be photon counting CT images, in particular multi-energy level CT images.

[0018] The adaptive spike suppression algorithm may comprise fitting a model, in particular a linear model, to a 3D neighborhood of voxels of the preprocessed image. This allows reliable spike removal while preserving sharp edges.

[0019] Executing the adaptive spike suppression algorithm may include qualifying the model, e.g. a linear model. Qualifying the model may include checking whether the model satisfies one or more predetermined criteria. Executing the adaptive spike suppression algorithm may further comprise adapting one or more model parameters of the model to obtain a modified model, in response to determining that the model does not satisfy the one or more predetermined criteria.

[0020] This allows the method to be applied under a wide range of different conditions, such as a wide range of input images, while still ensuring the quality of the results. If the initial model is not suitable for the given conditions, the features of the present disclosure allow the model to be adjusted.

[0021] The criteria may include, for example, that the weight of a voxel does not exceed a predetermined threshold and / or does not exceed the weight of its nearest neighboring voxels by more than a predetermined amount, and / or that the average weight of a central voxel does not exceed a predetermined threshold.

[0022] Executing the spike suppression algorithm may include repeating the steps of validating a modified model and adapting one or more model parameters until the modified model meets a predetermined criterion.

[0023] In other words, the model can be iteratively improved until an acceptable result is obtained. Such an iterative procedure avoids overcompensation when adapting the model, which could potentially lead to a model that meets certain criteria, but at the same time result in loss of texture and / or contrast.

[0024] The step of executing the spike suppression algorithm may further include applying the model or the modified model to the voxel, particularly in response to determining that the model or the modified model satisfies the predetermined criteria, i.e., the model may be applied to the voxel in response to determining that the model satisfies the predetermined criteria and / or the modified model may be applied to the voxel in response to determining that the modified model satisfies the predetermined criteria.

[0025] In other words, the voxel value is corrected if necessary, i.e. if the voxel in question is determined to be part of a spike based on the model. This step can be performed for multiple voxels, in particular all voxels, of a sub-region of the image or of the entire image. This reduces the number of spikes in the image.

[0026] In the present disclosure, fitting the model may include determining, for each voxel, a weighting factor for each of a number of nearest neighbors, in particular a weighting factor incorporating at least one of a weighting based on a spatial distance between the voxel and the nearest neighbor and a weighting based on a value distance between the voxel and the nearest neighbor, thereby allowing for reliable identification and removal of spikes.

[0027] As mentioned above, in the present disclosure, the step of executing an adaptive spike suppression algorithm may comprise a step of qualifying the model, which includes a step of checking whether the model satisfies one or more predefined criteria. The step of checking whether the model satisfies one or more predefined criteria may comprise a step of checking whether the weight coefficients of one or more voxels are within a predefined range, in particular whether they remain below a predefined threshold. Alternatively or additionally, the step of checking whether the model satisfies one or more predefined criteria may comprise a step of checking whether the weights of one or more voxels exceed the weights of their nearest neighbours by more than a predefined amount and / or whether the weights of some central voxels are within a predefined range, in particular whether they remain below a predefined threshold.

[0028] This makes it possible to determine with high confidence whether the model requires further adaptation.

[0029] In the present disclosure, applying the model may include performing a weighted average for each voxel using the weighting factor or factors determined for each of a number of nearest neighbors, thereby allowing for reliable removal of spikes.

[0030] In this disclosure, fitting a model to a 3D neighborhood of voxels in the preprocessed image may be performed using a filter, and adapting one or more model parameters may include adapting filter parameters, which allows reliable removal of spikes and noise while preserving edges and textures.

[0031] The step of adapting one or more model parameters may comprise adapting a parameter determining the aggressiveness of the weights, in particular adapting a parameter determining the aggressiveness of the weights such that the aggressiveness of the weights increases.

[0032] The aggressiveness of the weights determines how sensitive the model is in identifying spikes: the more aggressive the weight, the more voxels will be identified as spikes and the smoother the overall image will be after spike removal. Adjusting the aggressiveness allows us to modify the model to reliably remove spikes while still preserving texture.

[0033] Alternatively or additionally, adapting one or more model parameters may comprise adapting a parameter controlling a spatial weighting, in particular adapting a parameter controlling a spatial weighting such that the spatial weighting is increased.

[0034] Alternatively or additionally, adapting one or more model parameters may include adapting a parameter determining the number of nearest neighboring voxels, in particular adapting the parameter determining the number of nearest neighboring voxels such that the number of nearest neighboring voxels is increased.

[0035] Optionally, adapting one or more model parameters may include iteratively increasing the aggressiveness of the weights, iteratively increasing the spatial weights, and / or iteratively increasing the number of nearest neighbors, thereby efficiently approaching model parameters that allow reliable removal of spikes while preserving texture.

[0036] In this disclosure, a voxel V i,j,k Weights based on the spatial distance between voxels and their nearest neighbors, e.g.

number

[0037] Alternatively or additionally, in the present disclosure, voxel V i,j,k The weight based on the value distance between voxel and its nearest neighbor is

number

[0038] In the present disclosure, the step of applying the model comprises:

number

[0039] The present disclosure also provides methods, including methods for image processing according to the present disclosure.

[0040] The method may further include capturing the CT image by a photon counting CT imaging system, in particular by a low-dose CT scan, which may be a CT imaging system that detects radiation at a number of different energy levels.

[0041] Alternatively or in addition, the method may further comprise storing the processed CT image in a storage device and / or outputting the processed CT image, inter alia, on a display device. In this way, the processed image can be used by the computing element and / or by a user. For example, the computing element can access the processed CT image on the storage device for use in image-guided navigation and / or for computer image recognition methods. Alternatively or in addition, the user can view the processed CT image on a display device for use in image-guided navigation and / or visual analysis.

[0042] Alternatively or additionally, the method may include using the processed CT images for image-guided navigation and / or for computer image recognition methods.

[0043] CT images processed according to the present disclosure will perform better in all of these use cases since spikes are reliably removed and texture is preserved.

[0044] The present disclosure also provides a data processing system configured to perform one or more pre-processing steps on a CT image to obtain a pre-processed CT image, where the one or more pre-processing steps include applying an edge-preserving denoising algorithm. The data processing system is further configured to perform an adaptive spike suppression algorithm on the pre-processed CT image to obtain a processed CT image. The adaptive spike suppression algorithm can be configured in particular such that the processed CT image has a reduced number of spikes compared to the pre-processed CT image. In particular, the data processing system can be configured to perform any of the steps of the method for image processing of CT images of the present disclosure. In the context of the method steps, reference is made to the above disclosure and to the method claims.

[0045] The present disclosure also provides a system comprising the data processing system of the present disclosure. The system may further comprise a CT imaging system configured to capture a CT image or images, in particular the CT imaging system being a photon counting CT imaging system. As previously mentioned, spike removal is particularly beneficial for photon counting CT images that may otherwise result in poor image quality.

[0046] Alternatively or additionally, the system may include a display device configured to output the processed CT images, as previously described, allowing a user to view the processed CT images on the display device, for example, for use in image-guided navigation and / or visual analysis.

[0047] The present disclosure also provides a computer program product having instructions, when executed by a computer, that cause the computer to perform one or more pre-processing steps on a CT image to obtain a pre-processed CT image, the one or more pre-processing steps including applying an edge-preserving denoising algorithm, and to perform an adaptive spike suppression algorithm on the pre-processed CT image to obtain a processed CT image. The adaptive spike suppression algorithm may in particular be configured such that the processed image has a reduced number of spikes compared to the pre-processed image. In particular, the instructions, when executed by a computer, may cause the computer to perform any of the methods for image processing of CT images of the present disclosure.

[0048] The present disclosure also provides a computer-readable medium having instructions that, when executed by a computer, cause the computer to perform one or more pre-processing steps on a CT image to obtain a pre-processed CT image, the one or more pre-processing steps including applying an edge-preserving noise removal algorithm, and to perform an adaptive spike suppression algorithm on the pre-processed CT image to obtain a processed CT image. The adaptive spike suppression algorithm may be configured, among other things, such that the processed CT image has a reduced number of spikes compared to the pre-processed CT image.

[0049] In particular, the instructions, when executed by a computer, can cause the computer to perform any of the disclosed methods for image processing of CT images.

[0050] The features and advantages outlined in the context of the method for image processing likewise apply to methods including the method for image processing, data processing systems, systems including the data processing systems, computer program products, and computer-readable media of the present disclosure.

[0051] Further features, examples, and advantages will become apparent from the following detailed description, which proceeds with reference to the accompanying drawings. [Brief description of the drawings]

[0052] [Figure 1] FIG. 1 shows a flow chart of a method for image processing of CT images according to the invention. [Diagram 2] FIG. 2 shows a schematic, not to scale, view of a system according to the present disclosure. [Diagram 3] FIG. 3 illustrates a flow chart of an exemplary adaptive spike suppression method according to the present disclosure. [Figure 4] FIG. 4 illustrates a flow chart of an example of an edge-preserving denoising method according to the present disclosure. [Figure 5a]FIG. 5a shows a first exemplary CT image before preprocessing, after preprocessing, and after adaptive spike suppression at a certain magnification. [Figure 5b] FIG. 5b shows a first exemplary CT image before preprocessing, after preprocessing, and after adaptive spike suppression at other magnifications. [Figure 6a] FIG. 6a shows a second exemplary CT image before preprocessing, after preprocessing, and after adaptive spike suppression at one magnification. [Figure 6b] FIG. 6b shows a second exemplary CT image before preprocessing, after preprocessing, and after adaptive spike suppression at another magnification. [Figure 7a] FIG. 7a shows a third exemplary CT image before preprocessing, after preprocessing, and after adaptive spike suppression at a certain magnification. [Figure 7b] FIG. 7b shows a third exemplary CT image before preprocessing, after preprocessing, and after adaptive spike suppression at another magnification. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0053] FIG. 1 shows a flow chart of an exemplary method for image processing of CT images according to the present disclosure.

[0054] The method comprises a step S1 of performing one or more preprocessing steps to obtain a preprocessed image, more specifically applying an edge-preserving denoising algorithm, such as the Philips iDose algorithm, which uses a model fitter to fit local structures to a linear (constant) or non-linear model to remove noise, thereby allowing the noise to be removed while preserving the underlying image details.

[0055] An example of such an edge-preserving denoising method is given below in the context of Figure 4. However, other edge-preserving denoising algorithms are also contemplated for step S1.

[0056] The method further comprises the step S2 of performing an adaptive spike suppression algorithm on the pre-processed CT image to obtain a processed image, the adaptive spike suppression algorithm may be configured such that the processed CT image has a reduced number of spikes compared to the pre-processed CT image. An example of an adaptive spike suppression method is shown later in the context of FIG.

[0057] 2 shows a schematic, not to scale, view of a system 1 according to the present disclosure. In this example, system 1 includes a data processing system 2. The data processing system is configured to perform one or more pre-processing steps on the CT image to obtain a pre-processed CT image, where the one or more pre-processing steps include applying an edge-preserving denoising algorithm. The data processing system is further configured to perform an adaptive spike suppression algorithm on the pre-processed CT image to obtain a processed CT image. The adaptive spike suppression algorithm may be configured such that the processed CT image has a reduced number of spikes compared to the pre-processed CT image.

[0058] The data processing system may be specifically configured to perform any of the image processing methods of the present disclosure. The data processing system may be a computing device or may be a distributed system of interconnected computing devices, optionally including a cloud device.

[0059] The system 1 shown in Fig. 2 comprises an optional CT imaging system 3 configured to acquire CT images. For example, the CT imaging system may be a photon counting CT imaging system. The CT imaging system may be connected to a data processing system via a first data connection 6.

[0060] The system 1 shown in Fig. 2 also comprises an optional display device 4, which may be configured to display at least the processed image and optionally also the initial image and / or the pre-processed image, which may be connected to a data processing system via a second data connection 5.

[0061] As shown in FIG. 2, if the system comprises a CT imaging system 3 and a display device 4 these can also optionally be connected to each other via a third data connection 7 .

[0062] The data connections 5, 6 and 7 may each be a wired or wireless data connection.

[0063] It should be understood that the system need not include all of the above elements and may have elements not described above.

[0064] An example of an adaptive spike suppression method according to the present disclosure applied to pre-processed (i.e., denoised) image data obtained, for example, after application of iDose, is described below with reference to the flowchart shown in FIG.

[0065] As mentioned above, since the data has been denoised, the signal-to-noise ratio of spikes is higher and spikes are easier to identify and remove. The adaptive spike suppression is hereinafter also referred to as adaptive spike noise suppression (aSNS). It should be understood that the above steps are exemplary steps for illustrative purposes only.

[0066] As can be seen from the flowchart in FIG. 3, an exemplary aSNS “noise remover” includes a Linear Model Fitter component that fits a linear (constant) model to the 3D neighborhood of a voxel, a Model Qualifier component that checks if the model meets the aSNS requirements, a Model Parameter Adapter component that adapts the parameters of the model if the model does not meet the predefined requirements, and a Model Applicator component that applies the model to the voxel. Examples of each of the above elements are provided below: Each of the following elements is an example element for illustrative purposes only.

[0067] [Linear Model Fitter] The linear model fitter element can be constructed as follows: For each voxel V of the volume of interest, i,j,k For voxels, the weighting coefficients for the N nearest neighbors are given by Equation 1: i’,j’,k’ =w spatial i’,j’,k’ w HU i’,j’,k’ is determined based on, where w spatial i’,j’,k’ is V i,j,k represents the weighting for neighboring voxels according to the spatial distance to HU i’,j’,k’ is V in HU (Hounsfield unit) space. i,j,k The value of represents the weight for neighboring voxels according to their distance.

[0068] Weighting w for adjacent voxels according to spatial distance spatial i’,j’,k’ is expressed as Equation 2:

number

number

[0069] [Model Certification Department] The model validation element is the obtained weight w i’,j’,k’ can be analyzed to determine whether the model meets a predefined criterion. If the weight of the central voxel exceeds a predefined threshold, or the weight of the nearest neighboring voxels is greater than a predefined amount, or the average weight of the central voxel exceeds a certain threshold, this may indicate that the model needs to be adapted. For example, if the weight of the central voxel significantly exceeds the weight of the nearest neighboring voxels, this means that the voxel does not receive enough support from the neighboring voxels. If the model does not meet the criterion, the method proceeds to adapt the model parameters, for example using a model parameter adaptor element. Otherwise, the model is applied, for example by a model applicator element.

[0070] [Model Parameter Adaptation] The Model Parameter Adapter element can adapt the model by modifying (in particular increasing) one or more of the following parameters: mult (this is the algorithm parameter that controls the aggressiveness of the weights); σ spatial (this is an algorithm parameter that controls the spatial weighting); and The number of neighboring voxels, N.

[0071] It should be noted that, according to the present disclosure, after adapting the model parameters, the model validator element can revalidate the adapted model. These steps of adapting and validating the model can be repeated until the model validator element determines that no further adaptation is necessary. The model is then applied, for example, by the model application element.

[0072] [Model application part] The model application element is expressed as Equation 4:

number

[0073] A flow chart of an exemplary denoising algorithm that may be used to preprocess CT images in accordance with the present disclosure is shown in Figure 4. The denoising unit receives image data for preprocessing, selects an appropriate linear or quadratic model, and removes noise from the image data to obtain preprocessed denoised image data to be used as input for the spike suppression unit of the present disclosure.

[0074] In the following, three examples are described, each of which is shown in Figures 5a and 5b, 6a and 6b, and 7a and 7b, to illustrate the effect of the method of the present disclosure. The image data in these examples is acquired using a Spectral Photon Counting CT scanner, more specifically, a Philips Spectral Photon Counting CT (SPCCT) scanner. Very low dose data was used for the evaluation. The highlighted rectangles in the figures are each square regions with a size of 100x100 pixels for illustrative purposes only.

[0075] As an example for identifying spikes and their intensities, a spike threshold can be defined. To do so, for example, the median Q2, the first quartile Q1 and the third quartile Q3 can be determined and the interquartile range IQR=Q3-Q1 can be calculated for the square area. In this case, the spike threshold can be defined as spike threshold=Q3-Q2+1.5*IQR for positive spike values ​​and as spike threshold=Q2-Q1+1.5*IQR for negative spike values.

[0076] Any pixel with abs(pixel value-local median)>spike threshold can be identified as a spike, while the intensity of a spike can be defined as abs(pixel value-local median), where the local median can be calculated as the median value within a 5x5 square of pixels.

[0077] 5a and 5b show a first example. More specifically, from left to right, a first exemplary CT image is shown at different magnifications before preprocessing (labeled FBP, short for Filtered Back Projection), after preprocessing (labeled iDose), and after adaptive spike suppression (labeled aSNS, short for Adaptive Spike Noise Suppression). The selected square region is highlighted in each image, and spikes are marked by circles. FIG. 5b is a close-up of the selected square region. In this example, the initial image is an FBP image, and the preprocessed image is obtained by applying the iDose algorithm (an example of an edge-preserving denoising algorithm according to the present disclosure) for preprocessing. The processed image is obtained by applying the spike suppression method of the present disclosure to the preprocessed image. In the FBP image, there are four spikes. The average spike intensity is 820 HU. In the iDose image, the number of spikes increases to 29. The average spike intensity is 446 HU. In the aSNS image, the number of spikes is reduced to 2. The average spike intensity is 284 HU.

[0078] 6a and 6b show a second example. More specifically, from left to right, a second exemplary CT image is shown at different magnifications before preprocessing, after preprocessing, and after adaptive spike suppression. The selected square region is highlighted in each image, and the spikes are marked by circles. FIG. 6b shows a magnified view of the selected square region. In this example, the initial image is an FBP image, and the preprocessed image is obtained by applying the iDose algorithm. The processed image is obtained by applying the spike suppression method of the present disclosure to the preprocessed image. There are 12 spikes in the FBP image. The average spike intensity is 436 HU. In the iDose image, the number of spikes increases to 31. The average spike intensity is 189 HU. In the aSNS image, the number of spikes decreases to 2. The average spike intensity is 154 HU.

[0079] 7a and 7b show a third example. More specifically, from left to right, a third exemplary CT image is shown at different magnifications before preprocessing, after preprocessing, and after adaptive spike suppression. The selected square region is highlighted in each image, and the spikes are marked by circles. FIG. 7b shows a magnified view of the selected square region. In this example, the initial image is an FBP image, and the preprocessed image is obtained by applying the iDose algorithm. The processed image is obtained by applying the spike suppression method of the present disclosure to the preprocessed image. In the FBP image, there are 9 spikes. The average spike intensity is 426 HU. In the iDose image, the number of spikes increases to 22. The average spike intensity is 176 HU. In the aSNS image, the number of spikes decreases to 2. The average spike intensity is 137 HU.

[0080] The results of the above three examples are summarized in Table 1 below and show that spike suppression reduces the number of spikes by 91-94% and the average spike intensity by 19-36%.

[0081] [Table 1]

[0082] Thus, both the images and the calculations show a significant improvement in the number of spikes and their intensity.

[0083] While the present invention has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered as illustrative and not restrictive, i.e., the invention is not limited to the disclosed embodiments. In view of the above description and drawings, it will be apparent that various modifications can be made by those skilled in the art within the scope of the present invention as defined by the claims. [Explanation of symbols]

[0084] S1 Perform one or more preprocessing steps S2 Implements the adaptive spike suppression algorithm 1 System 2 Data Processing System 2 3 CT imaging system 4 Display device 5 Data Connection 6 Data Connection 7 Data Connection

Claims

1. 1. A computer-implemented method for image processing of CT images, the computer-implemented method comprising: performing one or more pre-processing steps on the CT image to obtain a pre-processed CT image, the one or more pre-processing steps including applying an edge-preserving denoising algorithm; performing an adaptive spike suppression algorithm on the preprocessed CT image to obtain a processed CT image; 10. A computer-implemented method comprising:

2. 2. The computer-implemented method of claim 1, wherein the step of performing the adaptive spike suppression algorithm comprises fitting a model, in particular a linear model, to a three-dimensional neighborhood of voxels of the preprocessed image.

3. Executing the adaptive spike suppression algorithm comprises: Qualifying the model, which includes checking whether the model meets one or more predetermined criteria; responsive to determining that the model does not satisfy the one or more predetermined criteria, adapting one or more model parameters of the model to obtain a modified model; The computer-implemented method of claim 2 , comprising:

4. 4. The computer-implemented method of claim 3, wherein executing the spike suppression algorithm comprises repeating the steps of qualifying the modified model and adapting one or more model parameters until the modified model meets the predetermined criteria.

5. 4. The computer-implemented method of claim 3, wherein executing the spike suppression algorithm further comprises applying the model or the modified model to the voxel in response to determining that the model or the modified model satisfies the predetermined criteria.

6. and / or wherein the step of fitting the model comprises determining, for each voxel, a weighting factor for each of a plurality of nearest neighboring voxels, in particular a weighting factor incorporating at least one of a weight based on the spatial distance between the voxel and the nearest neighboring voxel and a weight based on the value distance between the voxel and the nearest neighboring voxel; and / or wherein the step of executing the adaptive spike suppression algorithm comprises a step of qualifying the model comprising checking whether the model satisfies one or more predetermined criteria, and wherein the step of checking whether the model satisfies one or more predetermined criteria comprises checking whether the weighting coefficients of one or more voxels are within a predetermined range, in particular remain below a predetermined threshold; and / or and wherein applying the model comprises performing, for each voxel, a weighted average using a weighting factor or factors determined for each of a plurality of nearest neighboring voxels. The computer-implemented method of claim 2.

7. 4. The computer-implemented method of claim 3, wherein fitting the model to a three-dimensional neighborhood of voxels in the preprocessed image is performed using a filter, and adapting the one or more model parameters comprises adapting filter parameters.

8. adapting the one or more model parameters - adapting the parameters determining the aggressiveness of the weights, in particular adapting the parameters determining the aggressiveness of the weights so that the aggressiveness of the weights increases; - adapting the parameters controlling the spatial weights, in particular adapting the parameters controlling the spatial weights so that said spatial weights increase; and adapting the parameter determining the number of nearest neighbor voxels, in particular adapting the parameter determining the number of nearest neighbor voxels so that the number of nearest neighbor voxels increases; The computer-implemented method of claim 3 , comprising at least one of:

9. 9. The computer-implemented method of claim 8, wherein adapting the one or more model parameters comprises iteratively increasing the aggressiveness of the weights, iteratively increasing the spatial weights, and / or iteratively increasing the number of nearest neighbor voxels.

10. Voxel V ｉ，ｊ，ｋ and the weight based on the spatial distance between the voxel and its nearest neighbor, where dx is the size of the voxel in the axial plane, dz is the size of the voxel in the Z direction, and σ ｓｐａｔｉａｌ is an algorithm parameter that controls the aggressiveness of the weights, Voxel V ｉ，ｊ，ｋ The weight based on the value distance between voxels and the nearest neighbor voxels is is determined as, where n^ ｉ，ｊ，ｋ is Voxel V ｉ，ｊ，ｋ is the local noise level estimate of , mult is an algorithm parameter that controls the aggressiveness of the weights, and / or applying the model applying a weighted average as １ ｉ，ｊ，ｋ is Voxel V ｉ，ｊ，ｋ represents the noise-free estimate of ｉ’，ｊ’，ｋ’ is Voxel V ｉ，ｊ，ｋ is the weighting factor of the nearest neighbor voxel of ｉ’，ｊ’，ｋ’ =w ｓｐａｔｉａｌ ｉ’，ｊ’，ｋ’ w ＨＵ ｉ’，ｊ’，ｋ’ is; The computer-implemented method of claim 2.

11. 10. A method comprising the computer-implemented method of claim 1 for image processing, comprising: capturing said CT image by a photon counting CT imaging system, in particular by low-dose CT scanning; and / or - storing the processed CT image in a storage device and / or outputting the processed CT image, in particular on a display device; and / or using the processed CT images for image-guided navigation and / or computer image recognition methods; The method further comprises:

12. performing one or more preprocessing steps on the CT image to obtain a preprocessed CT image, wherein the one or more preprocessing steps include applying an edge-preserving denoising algorithm; and performing an adaptive spike suppression algorithm on the preprocessed CT image to obtain a processed CT image; In particular, carrying out the method according to claim 1 Data processing system.

13. 13. A system comprising the data processing system of claim 12, said system comprising: a CT imaging system for capturing said CT images, in particular a photon counting CT imaging system; and / or A display device that outputs the processed CT image The system further comprises:

14. When executed by a computer, the computer: performing one or more pre-processing steps on the CT image to obtain a pre-processed CT image, the one or more pre-processing steps including applying an edge-preserving denoising algorithm; performing an adaptive spike suppression algorithm on the preprocessed CT image to obtain a processed CT image; Execute A computer program comprising instructions that, in particular, cause the computer to carry out the method according to any one of claims 1 to 11.

15. When executed by a computer, the computer: performing one or more pre-processing steps on the CT image to obtain a pre-processed CT image, the one or more pre-processing steps including applying an edge-preserving denoising algorithm; performing an adaptive spike suppression algorithm on the preprocessed CT image to obtain a processed CT image; Execute A computer readable medium comprising instructions that in particular cause said computer to carry out the method of any one of claims 1 to 11.

Images