Apparatus for determining decomposed spectral image data

Abstract

The invention relates to an apparatus for determining decomposed spectral image data with improved accuracy. The apparatus comprises a spectral image data providing unit (111) for providing spectral image data, a spectral image data decomposition unit (112) for calculating a basis decomposition of the spectral image data, a virtual non-contrast image generating unit for generating a virtual non-contrast image based on the decomposed spectral image data, and a contrast agent residue identifying unit (113) for identifying a residue region in the virtual non-contrast image comprising a contrast agent residue based on expected structural properties of the contrast agent residue, wherein the spectral image data decomposition unit (114) is configured to calculate a new basis decomposition in the residue region. Utilizing the structural properties of the contrast agent residue allows for a very accurate determination of the contrast agent residue and improves the decomposition accuracy in this region.

Description

Technical Field

[0001] The present invention relates to an apparatus, method, and computer program for determining spectral image data of an object's decomposition. Background Technology

[0002] Spectral X-ray imaging is a usable imaging technique, important for applications such as good characterization of human anatomy and function. In spectral imaging, image data of an object is acquired using at least two different radiation spectra, for example, using X-ray radiation including a first wavelength and X-ray radiation including a second wavelength. Therefore, the spectral image data includes multiple attenuation values ​​acquired at at least two different radiation spectra. These multiple attenuation values ​​can be used to calculate the basis decomposition of the spectral image data. The basis decomposition determines the contribution of at least two basis functions to the overall mass attenuation coefficient of the material found in that part of the object, based on the multiple attenuation values ​​at at least two different radiation spectra for each part of the object (e.g., each voxel). Examples of basis functions are functions describing the dependence of the Compton scattering effect and / or photoelectric effect on radiation energy. The contribution of each basis function to the mass attenuation coefficient of the material in a part of the object can be viewed as a weighted average of the basis functions in a mathematical representation of the mass attenuation coefficient. The weights can then be considered as the decomposed spectral image data and can be combined in various ways to generate different spectral images, such as Compton scattering images, photoelectric images, single-energy images, virtual noncontrast images, etc. In particular, images can be derived from decomposed spectral image data with very high contrast between the contrast agent and surrounding tissue present during spectral image data acquisition. However, suboptimal processes, inaccurate image acquisition, and numerical inaccuracies during the decomposition process of spectral image data can all lead to inaccurate decomposition of the spectral image data. Such inaccurate decomposition is particularly pronounced with respect to the contrast agent and results in inaccurate calculations of the distribution of the contrast agent in the region of interest of the object. Since it is important for clinical applications that the distribution of the contrast agent can be accurately determined by the user (e.g., physician), these inaccuracies hinder the clinical use of spectral image data. Summary of the Invention

[0003] The object of this invention is to provide an apparatus, method, and computer program that allows for the determination of decomposed spectral image data with improved accuracy.

[0004] In a first aspect of the invention, an apparatus for determining decomposed spectral image data of an object is provided, wherein the apparatus comprises: a) a spectral image data providing unit configured to provide spectral image data of the object having been injected with a contrast agent, wherein the spectral image data of the object has been acquired using a spectral X-ray imaging device; b) a spectral image data decomposition unit configured to calculate a basis decomposition for the spectral image data, wherein the basis decomposition yields decomposed spectral image data, wherein the decomposed spectral image data indicates the weights of at least two basis functions provided to the spectral image data; and c) a virtual non-contrast image generation unit configured to generate a virtual non-contrast image based on the decomposed spectral image data, wherein the contrast agent contributes... The contrast agent residue is removed from the image data; and d) a contrast agent residue identification unit configured to identify a residual region in the virtual uncontrast image containing contrast agent residue based on the expected structural characteristics of the contrast agent residue, wherein the contrast agent residue is the contribution of the contrast agent to the attenuation represented by the image values ​​of the virtual uncontrast image, wherein the spectral image data decomposition unit is configured to: determine a residual measure for the residual region, wherein the residual measure indicates the intensity of the contrast agent residue in the residual region, and calculate a new basis decomposition in the residual region by recalculating the weights of the basis functions provided to the spectral image data in the residual region based on the residual measure, thereby reducing the contrast agent residue.

[0005] Since the contrast agent residue identification unit is adapted to identify residual regions containing contrast agent residues in the virtual non-contrast image based on the expected structural characteristics of the contrast agent residues, prior knowledge of the expected distribution of the contrast agent in the region of interest of the object can be considered. This allows for very accurate determination of regions containing contrast agent residues, which indicate inaccurate decomposition of the spectral image data in those regions. In other words, the presence of the contrast agent residues identified by the contrast agent residue identification unit indicates inaccurate decomposition of the spectral image data in those regions. Since the spectral image data decomposition unit is then configured to compute a new basis decomposition in the residual regions by recalculating the weights of the basis functions provided to the residual regions, reducing the contrast agent residues, the correction of the decomposition of the spectral image data is based on the very accurately determined residual regions, allowing the correction to be focused on these regions and improving the accuracy of the correction of the decomposition of the spectral image data. Therefore, spectral image data with improved decomposition accuracy can be provided.

[0006] The object from which the spectral image data is acquired can be any object that can be imaged using a spectral X-ray imaging device. Preferably, the object refers to a patient in a medical application, wherein the patient can be a human or an animal. A contrast agent is injected into the object before acquiring the spectral image data using the spectral X-ray imaging device. The contrast agent can be any contrast agent that can be used for the intended medical application of the spectral image data. In a preferred embodiment, the contrast agent refers to an iodine-based contrast agent. However, the contrast agent can also refer to a barium-based contrast agent.

[0007] A spectral image data providing unit, which can also be considered a spectral image data provider, is configured to provide spectral image data of an object, particularly to a spectral image data decomposition unit. The spectral image data providing unit can be a storage unit where the spectral image data of the object is stored and can be retrieved from the storage unit to provide it. The spectral image data providing unit can also be a receiving unit for receiving the spectral image data of the object and for providing the received spectral image data. The spectral image data can be received, for example, directly from a spectral X-ray imaging device or from the storage device of the spectral X-ray imaging device. Furthermore, the spectral image data providing unit can also be considered to include or be part of a spectral X-ray imaging device.

[0008] For example, spectral X-ray imaging devices can be based on dual-source technology, fast kV switching technology, photon counting detector technology, etc. Spectral image data acquired using a spectral X-ray imaging device provides image data of an object acquired using at least two different radiation spectra, and thus indicates the attenuation of the imaged portion of the object with respect to at least two different radiation spectra.

[0009] The spectral image data decomposition unit, which can also be viewed as a spectral image data decomposer, is configured to compute a basis decomposition for the spectral image data, providing decomposed spectral image data as a result of the basis decomposition. During the basis decomposition of the spectral image data, the contributions of at least two different predetermined basis functions to the overall mass decay coefficient of a portion (preferably a voxel) of the object are determined in the form of weights that are weighted in the mathematical representation of the overall mass decay coefficient. The weights depend on the material found in each portion of the region of interest of the object, and therefore vary between different portions of the object, particularly depending on the current material composition in that region. Preferably, the weight of each basis function is determined by the spectral image data decomposition unit for each voxel of the region of interest. The determined weights form the decomposed spectral image data. Based on the decomposed spectral image data, multiple different images can be derived by utilizing different combinations of the decomposed spectral image data. An overview of the decomposition of spectral image data and the generation of different spectral images based on the decomposed spectral image data can be found, for example, in the article “Detector-based spectral CT with a novel dual-layer technology: principles and applications” by N. Rassouli et al. (Insights into Imaging, Vol. 8, pp. 589-598 (2017)).

[0010] In a preferred embodiment, the basis functions used for decomposing the spectral image data at least indicate functions defining the energy dependence of attenuation on the Compton scattering effect and functions defining the energy dependence of attenuation on the photoelectric effect. However, other basis functions may also be used alternatively or additionally for spectral image decomposition. For example, as an alternative to or supplement to the Compton scattering effect function and the photoelectric effect function, iodine basis functions indicating the energy dependence of attenuation on the iodine content of the material and / or water basis functions indicating the energy dependence of attenuation on the water content of the material may also be used as basis functions.

[0011] A virtual noncontrasted image generation unit, which can also be considered a virtual noncontrasted image generator, is configured to generate virtual noncontrasted images based on decomposed spectral image data. Specifically, the virtual noncontrasted image generation unit is adapted to generate virtual noncontrasted images based on a predefined combination of decomposed spectral image data. A virtual noncontrasted image is an image in which the contributions of contrast agents (such as iodine) to the basis functions (such as photoelectric and Compton scattering components) of the portion forming the decomposed spectral image are removed, and generally should not include any contribution of the contrast agent to the attenuation of the portion of the object represented by the image values ​​of the virtual noncontrasted image. Therefore, a virtual noncontrasted image can be considered essentially an X-ray image acquired without the use of contrast agents. A detailed description of the method for determining virtual noncontrasted images based on decomposed spectral image data can be found, for example, in the article "An iodine-calcium separation analysis and virtually noncontrasted image generation obtained with single source dual-energy MDCT" by Goshen et al. (IEEE Nuclear Science Symposium Conference Record, pp. 3868-3870 (2008)). The generated virtual non-contrast image can be a 2D, 3D, or 4D image based on spectral image data acquired by a spectral imaging device. Generally, in the context of an image, the term voxel is used to refer to an image element of an image that includes an image value. Therefore, for example, the term voxel is also used when referring to a pixel in the context of a 2D image. Thus, the term voxel does not limit the corresponding image to a 3D image representation.

[0012] A contrast agent residue identification unit, which can be considered a contrast agent residue identifier, is configured to identify residual regions in a virtual non-contrast image containing contrast agent residue based on the expected structural characteristics of the contrast agent residue, from which the overall contrast agent contribution has been removed. Since, according to the general principles of virtual non-contrast images, the image should not include any contribution of the contrast agent to the attenuation represented by the image values ​​of the virtual non-contrast image, any region (i.e., voxel) containing such a contribution (i.e., contrast agent residue) indicates potential inaccuracies in spectral image data decomposition. Prior knowledge of at least one expected structural characteristic of the contrast agent residue allows for a very efficient and accurate search for regions containing the contrast agent residue. The expected structural characteristics of the contrast agent residue that can be used to determine the residual region can involve any kind of structural characteristics, such as the expected shape of the contrast agent residue in the virtual non-contrast image, the expected surroundings of the contrast agent residue, the expected location of the contrast agent residue, the expected intensity distribution of the contrast agent residue, etc. The expected structural characteristics can be determined, for example, based on knowledge of the expected distribution of the contrast agent in the region of interest of an object imaged by a spectral X-ray imaging device. Furthermore, the expected structural characteristics can be determined additionally or alternatively relative to a previous situation, i.e., previous spectral image data, involving similar situations, such as similar objects, similar regions of interest, similar medical indications, similar contrast agents, etc., regarding the current situation. The expected structural characteristics can be predefined, for example, based on the contrast agent used, or can be selected by the user (e.g., a physician), for example, by providing the user with several possible structural characteristics, whereby the user can then select the structural characteristics he / she expects for the current situation. The contrast agent residue identification unit can use any type of filter, segmentation algorithm, or shape determination algorithm to find structural characteristics in the virtual non-contrast image and thus find residual regions.

[0013] In a preferred embodiment, the expected structural characteristic refers to a tubular shape, and the contrast agent residue identification unit is adapted to identify residue regions by applying a vascular filter to a virtual non-contrast image. The tubular shape of contrast agent residue is particularly expected for iodine-based contrast agents, which are used to emphasize vascular structures in the region of interest. However, other contrast agents used to emphasize vascular structures in the region of interest can also result in the expected tubular shape of contrast agent residue. Applying a vascular filter to the virtual non-contrast image allows for very efficient identification of structures in the virtual non-contrast image that include tubular shapes. Since virtual non-contrast images generally should not show the vascular system in detail (because the overall contrast agent contribution is removed from the combination of decomposed spectral image data), tubular shapes identified by the vascular filter can be considered to involve contrast agent residue. However, other structural characteristics can also be used in other cases. For example, in cases where the contrast agent is specifically found within a cancerous nodule, the expected structural characteristic can refer to the expected shape of such a cancerous nodule, for example, substantially circular or elliptical, or a portion of a circular or elliptical nodule. The contrast agent residue identification unit can then be configured to use a corresponding filter or segmentation algorithm to find the structure corresponding to that shape.

[0014] Preferably, the residual area includes only contrast agent residue. However, the residual area may also include at least a portion of the area surrounding the potential contrast agent residue, for example, it may include the edge surrounding one or two voxels of the potential contrast agent residue.

[0015] In an embodiment, the contrast agent residue identification unit is configured to apply a machine learning-based classification algorithm to virtual non-contrast images for detecting residue regions. Preferably, the machine learning-based classification algorithm refers to a deep neural network architecture. Machine learning-based classification algorithms, particularly deep neural network architectures, are particularly well-suited for finding predefined structural features in images if correctly trained during the training phase. For the training phase, multiple training spectral images, particularly training virtual non-contrast images where contrast agent residue has already been identified, can be provided to the machine learning-based classification algorithm. In the training spectral images, contrast agent residue can be identified, for example, by an experienced user (such as an experienced radiologist), or by any other known structural identification algorithm (such as a vascular filter). The residue region containing the identified contrast agent residue can then be provided to the machine learning-based classification algorithm along with the training virtual non-contrast images. During the training phase, the machine learning-based classification algorithm then learns to identify contrast agent residue in the virtual non-contrast images. After training, the machine learning-based classification algorithm can be applied to new virtual non-contrast images, i.e., virtual non-contrast images that are not part of the training virtual non-contrast images. The machine learning-based classification algorithm then provides one or more residual regions as results. In this embodiment, it can be considered that the structural characteristics used to identify residual regions are incorporated into the machine learning-based classification algorithm by selecting training data.

[0016] The spectral image data decomposition unit is configured to calculate a new basis decomposition in the residual region by recalculating the weights of the basis functions provided to the spectral image data in the residual region, thereby reducing contrast agent residue. The presence of identified contrast agent residue indicates an inaccurate determination of the weights of the basis functions that form the decomposed spectral image in the residual region. Therefore, the spectral image data decomposition unit can be adjusted to reduce contrast agent residue, preferably by removing contrast agent residue, to correct the weights of the basis functions in the residual region. The spectral image data decomposition unit can be adapted to recalculate the basis decomposition based on a correction value. In particular, the spectral image data decomposition unit can be adapted to correct the weights, for example, by adding or subtracting a correction value from the weights of at least one of the basis functions, and then recalculate the weights of the remaining basis functions according to the decomposition principle, i.e., such that the weighted basis functions produce a portion of the quality attenuation coefficient in the residual region. This correction value can be any value or a predetermined value. In particular, the correction value can be predetermined based on knowledge of the expected contrast agent residue and / or based on knowledge of the decomposition algorithm used. For example, if a decomposition algorithm and a contrast agent are used, and it is known that this combination can lead to an overestimation of the weights of a basis function in some regions, resulting in contrast agent residue in these regions, then the correction value can be selected based on this knowledge and can be subtracted from the weights of the corresponding basis functions to recalculate the basis decomposition.

[0017] In a preferred embodiment, the spectral image data decomposition unit is configured to determine a residual measure of the residual region indicating the intensity of contrast agent residue in the residual region, and to recalculate the decomposed spectral image data based on the residual measure. Preferably, a residual measure is determined for each portion of the residual region (e.g., each voxel), and a correction value for each portion is determined based on the residual measure of that portion. The spectral image data decomposition unit is then preferably configured to determine a correction value based on the residual measure, for example, based on a mathematical function indicating the dependence of the correction value on the residual measure. The correction value can then be added to or subtracted from at least one of the weights of the basis functions to recalculate the basis decomposition. In some embodiments, the correction value can be determined to be equal to the residual measure, such that the residual measure is directly used to correct the weights of the basis functions.

[0018] Preferably, the apparatus is configured to recalculate the basis decomposition, specifically, to iteratively correct the weights of the basis functions in the residual region. Specifically, after the spectral image data decomposition unit has calculated a new basis decomposition, the spectral image data decomposition unit is adapted to provide the newly decomposed spectral image data again to the virtual non-contrast image generation unit. The virtual non-contrast image generation unit is then configured to generate a new virtual non-contrast image based on the newly decomposed spectral image data. The contrast agent residue identification unit is then configured to identify whether the residual region still exists, wherein the spectral image data decomposition unit then recalculates the basis decomposition for the still-existing residual region. For example, the spectral image data decomposition unit may be adapted to add correction values ​​to or subtract correction values ​​from the weights of at least one basis function. The correction values ​​may be the same as in previous iteration steps, or they may be determined for each iteration step, for example, based on a residue measure. Specifically, in this embodiment, the correction values ​​may be arbitrarily small values, such that the weights are corrected sequentially using each iteration step. Furthermore, the calculation of the new basis decomposition and the modification of the basis function weights can be based on a comparison between the image values ​​of the residual region and those of the residual region during previous iterations. For example, correction values ​​can be determined based on this difference.

[0019] If a predetermined stopping criterion is met, preferably, the iteration can stop if the contrast agent residue identification unit is no longer able to identify the residue area (indicating that the contrast agent residue has been completely removed from the corrected decomposed spectral image data). However, other technically reasonable stopping criteria can also be used, for example, based on a determined calibration value. For example, the iteration can also stop if the residue measure is below a predetermined threshold. Furthermore, the iteration can also be used in a semi-automatic manner, wherein at each step, an image generated from the decomposed spectral image data, such as a virtual non-contrast image, is presented to the user, whereby the user can then decide whether to stop the iteration.

[0020] Alternatively, for the iterative method, the apparatus can also be configured to correct the weights of the basis decomposition in the residual region using a direct method to determine a new basis decomposition. In this case, the spectral image data decomposition unit is configured to directly recalculate the basis decomposition without further providing the resulting newly decomposed spectral image data to the virtual non-contrast image generation unit. In the following, preferred embodiments are described for calculating the new decomposition using spectral image data, i.e., for calculating the new basis decomposition, which can be used as part of the iterative method but also as part of the direct method.

[0021] In an embodiment, the spectral image data decomposition unit is configured to compute the new basis decomposition to define a residual region in the virtual non-contrast image containing the residual region, wherein the new decomposed spectral image data is also computed based on the defined residual region. The defined residual region can refer to a region of a predetermined shape that is defined around and completely encompasses the residual region. For example, the residual region can refer to a region of a circular, elliptical, or rectangular shape surrounding the residual region. Therefore, the residual region may be larger than the residual region. In this way, the residual region encompasses the (immediately adjacent) surroundings of the residual region, thus providing additional background information about the residual region. Taking into account the residual region and therefore the information about its surroundings, the spectral image data decomposition unit can use this additional information to compute the new basis decomposition, particularly the corrected spectral image data. Preferably, the spectral image data decomposition unit is adapted to determine a residual measure based on the residual region. For example, the average image value of the residual region can be used as the expected image value of the virtual non-contrast image within the complete residual region, and the residual measure can be determined based on the expected image value compared with the image value in the residual region. A correction value can then be determined based on the residual measure, and the weights in the residual region can then be corrected accordingly. In the example, the residual measure of a portion of the residual region can be determined as an indication of the difference between such a desired image value and the actual image value of that portion of the residual region. Preferably, in this case, the correction value is determined to be equal to the residual measure. However, other mathematical methods can also be used to recalculate the weights so that they substantially relate to the desired image value.

[0022] In an embodiment, the spectral image data decomposition unit is configured to compute the new basis decomposition to determine a histogram defined by image values ​​of each region within the residual region, and to compute spectral image data of the new decomposition based on the histogram. Preferably, the region within the residual region refers to a voxel within the residual region, such that the histogram is defined by the image value of each voxel in the residual region. Specifically, the histogram represents multiple regions, specifically voxels, that include image values ​​(i.e., attenuation values) in predefined image value bins. The spectral image data decomposition unit can then be configured to determine an intensity metric and / or a correction value based on the histogram. For example, the intensity metric of a voxel may refer to the image value of a voxel above a predetermined threshold. In particular, the spectral image data decomposition unit can be adapted to determine the threshold based on the histogram, for example, based on the curvature of the histogram. Preferably, the threshold is determined such that the histogram shows a sharp decrease in the amount of voxels in the region of the threshold. The correction value can then be determined based on the intensity metric, for example, by comparing voxels with image values ​​above the threshold with image values ​​of voxels below the threshold.

[0023] Furthermore, the spectral image data decomposition unit is preferably configured to compute a new basis decomposition, such that all image values ​​above a threshold are removed from the histogram. Preferably, the weights of the decomposed spectral image data are modified for regions containing image values ​​above the threshold, such that the image value is transferred from the weight of one basis function to the weight of another basis function. For example, if the basis functions are functions indicating the Compton scattering effect and functions indicating the photoelectric effect, then the image values ​​of voxels above the threshold in the histogram are transferred from the weight of the function indicating the Compton scattering effect to the weight of the function indicating the photoelectric effect.

[0024] In an embodiment, the spectral image data decomposition unit is configured to compute the new basis decomposition to determine similar regions in the virtual non-contrast image for the residual region, wherein a region is considered similar to the residual region if the image characteristics of the virtual non-contrast image in a region are similar to the image characteristics in the residual region; and the new decomposed spectral image data is computed based on a comparison of the image values ​​of the similar regions and the residual region. Preferably, the spectral image data decomposition unit is configured to determine similar regions by computed the new basis decomposition in the following manner: by comparing the image values ​​of the residual region with the image values ​​of multiple candidate regions via applying a predefined distance norm and determining the nearest candidate region as the similar region based on the predefined distance norm. The distance norm may be, for example, L... 2 Norm, but it can also be any other suitable distance norm, such as L. 1Norms, mutual information measures, total variation measures, etc. Based on the distance norm, similar regions are identified as candidate regions with minimum values ​​(i.e., minimum distances) after applying a predefined distance norm. Specifically, the result of applying the distance norm can be considered as a similarity measure between residual regions and candidate regions, and similar regions can be determined based on this similarity measure. The spectral image data decomposition unit can be adapted to use similar regions as a model of how residual regions should behave when correctly decomposed image data is applied. Therefore, the spectral image data decomposition unit can be configured to compute new basis decompositions such that residual regions substantially correspond to similar regions in order to reduce contrast agent residue. For example, the spectral image data decomposition unit can be adapted to determine a correction value for a portion of the residual region based on a comparison of image values ​​in that portion of the similar region and in that portion of the residual region. However, other possibilities can also be considered, such as determining the correction value based on the average image values ​​of the similar regions, etc.

[0025] In a preferred embodiment, the spectral image data decomposition unit is configured to calculate the new basis decomposition to compare the similar region with the residual region by determining the difference between each image value and the spatially corresponding image value of the residual region, and to calculate the new decomposed image data based on the determined difference. The determined difference can be determined as a residual measure, wherein, preferably, a correction value is determined to be equal to the residual measure. In particular, the spectral image data decomposition unit can be configured to add the determined difference as a correction value to the weights of one basis function and recalculate the weights of other basis functions accordingly. If the basis functions refer to functions indicating the Compton scattering effect and functions indicating the photoelectric effect, the spectral image data decomposition unit is preferably adapted to remove the difference from the weights of the function indicating the Compton scattering effect and add the difference to the weights of the function indicating the photoelectric effect.

[0026] In an embodiment, the spectral image data decomposition unit is configured to compute the new basis decomposition to determine more than one similar region for the residual region, and to determine a weighted average of all determined similar regions, wherein the residual region is compared to the weighted average of the similar regions. Specifically, the weighted average of similar regions can be defined as a similar region. More than one similar region can be determined as a candidate region comprising a similarity metric indicating a similarity above a predefined threshold. Alternatively, more than one similar region can be determined by identifying a predefined number of candidate regions as similar regions, wherein the identified candidate regions comprise the highest similarity compared to all other candidate regions. The weights of the weighted average can be predetermined weights, can be automatically selected based on previous similarity cases (e.g., cases with the same region of interest), or can be selected and / or changed by the user. The weights can then be based on, for example, knowledge of the reliability of image values ​​in different imaging regions of the spectral image data.

[0027] In an embodiment, the spectral image data decomposition unit is configured to compute a new basis decomposition to: define the residual region as a missing data region; use an image inpainting algorithm to estimate information in the missing data region; and compute a residual metric based on the estimated information. For example, image inpainting algorithms disclosed in A. Crinisi et al.'s paper "Region filling and object removal by exemplar-based image inpainting" (IEEE Trans. Image Process., (2004)) or deep learning-based image inpainting algorithms disclosed in R. Yeh et al.'s paper "Semantic Image Inpainting with Perceptual and Contextual Losses" (IEEE Conf. on Computer Vision and Pattern Recognition, (2017)) can be used to inpaint residual regions in virtual non-contrast images, wherein the inpainted virtual non-contrast image can be used as the basis for computing a new basis decomposition, specifically to correct the weights of the basis functions in the residual region. For example, weights can be determined such that the virtual non-contrast image generated based on the spectral image data with the new decomposition is substantially similar to the inpainted virtual non-contrast image. To achieve this, the apparatus can be configured to utilize an iterative method that stops when the difference between the new virtual non-contrast image (i.e., the virtual non-contrast image generated based on the decomposed spectral image data of the last iteration step) and the image-restored virtual non-contrast image is below a predetermined threshold. Additionally or alternatively, the residual measure can be determined based on the image-restored virtual non-contrast image, for example, by determining the difference between the image value of the residual region and the image value of the corresponding portion of the image-restored virtual non-contrast image. In this case, it is preferable to determine a correction value equal to the residual measure.

[0028] In another aspect of the invention, a method is provided for determining decomposed spectral image data of an object, wherein the method comprises: a) providing spectral image data of the object having been injected with a contrast agent, wherein the spectral image data of the object has been acquired using a spectral X-ray imaging device; b) calculating a basis decomposition for the spectral image data, wherein the basis decomposition yields decomposed spectral image data, wherein the decomposed spectral image data indicates the weights of at least two basis functions provided to the spectral image data; c) generating a virtual non-contrast image based on a predefined combination of the decomposed spectral image data, wherein the contrast agent contributes... The contrast agent residue is removed from the image data; d) a residual region in the virtual uncontrast image containing the contrast agent residue is identified based on the expected structural characteristics of the contrast agent residue, wherein the contrast agent residue is the contribution of the contrast agent to the attenuation represented by the image values ​​of the virtual uncontrast image; e) a residual measure of the residual region is determined, wherein the residual measure indicates the intensity of the contrast agent residue in the residual region; and f) a new basis decomposition in the residual region is calculated by recalculating the weights of the basis functions provided to the spectral image data in the residual region based on the residual measure, thereby reducing the contrast agent residue.

[0029] In another aspect of the invention, a computer program for correcting the decomposition of spectral image data of an object is provided, wherein the computer program includes a program code module for causing the apparatus to perform the steps of the method described above when the computer program is run by the apparatus described above.

[0030] It should be understood that the apparatus, method, and computer program described in the specification have similar and / or identical preferred embodiments, particularly as defined in the specification.

[0031] It should be understood that the preferred embodiments of the present invention may also be any combination of the dependent claims or the above embodiments with the corresponding independent claims.

[0032] These and other aspects of the invention will become apparent from the embodiments described below, and will be illustrated with reference to the embodiments described below. Attached Figure Description

[0033] In the following figures:

[0034] Figure 1 An embodiment of a spectral imaging system, including means for determining the decomposition of spectral image data of an object, is illustrated schematically and exemplary.

[0035] Figure 2 A flowchart illustrating a method for determining the decomposition of an object using spectral image data is shown, and

[0036] Figure 3 and Figure 4 The illustration shows an exemplary result of an image generated from decomposed spectral image data. Detailed Implementation

[0037] Figure 1 An embodiment of a spectral imaging system, including means for determining the decomposition of spectral image data of an object, is illustrated schematically and exemplary. In this embodiment, the spectral imaging system 100 includes a spectral imaging unit 120, such as a dual-energy X-ray CT imaging unit, for acquiring spectral image data of a region of interest of a patient 121 located on a patient stage 122. Prior to acquiring the spectral image data, a contrast agent, such as an iodine contrast agent, is injected into the patient 121. The spectral imaging system 100 includes means 110 for determining the decomposition of spectral image data of the patient.

[0038] The device 110 includes a spectral image data providing unit 111, a spectral image data decomposition unit 112, a virtual non-contrast image generation unit 113, and a contrast agent residue identification unit 114. Furthermore, the device 110 may include an input unit 115 (e.g., a mouse and keyboard) and / or a display unit 116, for example, for displaying a virtual non-contrast image or other spectral images generated based on the decomposed spectral image data.

[0039] In this example, the spectral image data providing unit 111 is configured to receive spectral image data acquired by the spectral imaging unit 120. The spectral image data providing unit 111 can then provide the spectral image data to the spectral image data decomposition unit 112 for further processing.

[0040] The spectral image data decomposition unit 112 is configured to calculate the basis decomposition of the spectral image data, wherein the basis decomposition produces decomposed spectral image data. The decomposed spectral image data indicates the weights of at least two basis functions provided to the spectral image data. Preferably, one of the basis functions refers to a function indicating the Compton scattering effect attenuating the material, while the second basis function indicates the photoelectric effect attenuating the material. In this case, the spectral image data decomposition unit can be adjusted, for example, to follow the following principles for calculating the decomposed spectral image data based on the basis decomposition of the spectral image data.

[0041] Spectral or multi-energy CT imaging yields at least two attenuation values ​​for each voxel acquired at at least two different radiation energies, wherein the attenuation values ​​for each voxel form spectral image data. The spectral image data can be used to determine the contribution to the attenuation of the material or different basis functions. Preferably, the contributions of the photoelectric effect, Compton scattering effect, and other effects (e.g., one or more k-edge effects) to the mass attenuation coefficient of the material are determined in the basis decomposition of the spectral image data. The resulting decomposed spectral image data allows for the reconstruction of virtual monochromatic images, iodine concentration maps, virtual non-contrast images, etc., as well as conventional CT images.

[0042] One possible basis decomposition is based on the contributions of functions indicating the energy dependence of the photoelectric effect and the Compton scattering effect as basis functions. In this case, the energy dependence attenuation of the material in water can be approximated by the following equation for the two functions (i.e., photoelectric effect). Compton scattering effect A linear combination of the contributions of the decay curves:

[0043]

[0044] In weight and The contributions of the photoelectric effect and Compton scattering effect are found, where, for this example, the weights are for each part of the imaged object, and in particular for each voxel i in the image representation of the imaged object. and This generates decomposed spectral image data. The decomposed spectral image data can be reconstructed from the corresponding spectral image data, for example, based on instructions. and line integral l P and l S To reconstruct. If the spectral image data contains data from two different energy spectra F... H and F L The attenuation values ​​(L = low energy and H = high energy) can be obtained from the spectral image data using the following formula:

[0045]

[0046] Here, P H / L This refers to spectral image data for high and low energy spectra respectively, for each part of the object, where E is the energy and the term F is the energy. H / L (E) represents the high and low spectral fluxes, obtained by modeling the flux emitted from the up-beam to the detector pixels and the detector response. The difference between these two spectra can arise, for example, from tube modulation or two detector layers. Item l P / S It is the equivalent path for photoelectric and scattering in water, and These are the energy-related attenuation coefficients of these two mechanisms, representing the basis functions of the photoelectric effect and the Compton scattering effect. P and l S The solution can be found by reversing the above equation, which is equivalent to starting from p. L and p H domain to p s and p p Mapping function of the domain:

[0047] p s ,p p =D(p) L ,p H )

[0048] Here, by p s and p p Multiply by a fixed scalar attenuation coefficient from l P / S Get p s and p p This is to facilitate the later processing of dimensionless prepared values. D(p) can be solved, for example, using a polynomial function or a lookup table. L p H Further details regarding the decomposed spectral image data that can be utilized by the spectral image data decomposition unit 112 can be found in the article "Energy-selective reconstructions in x-ray computerized tomography" by Alvarez et al., Physics in Medicine & Biology 21.5 (1976). The mathematical process is an example of how decomposed spectral image data can be obtained. Similar principles can be utilized for other combinations of basis functions.

[0049] In practice, the decomposition process is usually suboptimal, mainly due to the suboptimal separation of flux at the detector, including both high-energy and low-energy photons, and the decomposition function D(p). L p H This is caused by nonlinearity, etc. This leads to inaccurate estimation of contrast agent concentration, as well as visible image artifacts in virtual monochrome images, which may hinder the clinical application of spectral CT.

[0050] Then, the virtual non-contrast image generation unit 113 is configured to generate a virtual non-contrast image based on the decomposed spectral image data, wherein the overall contrast agent contribution has been removed from the image data. If the basis functions include at least a function indicating the photoelectric effect on attenuation and a function indicating the Compton effect on attenuation, the weights (i.e., contributions) of these two functions form a portion of the decomposed spectral image data and can be represented in the form of a photoelectric image and a Compton scattering image. The virtual non-contrast image generation unit 113 can be configured to generate a virtual non-contrast image based on the photoelectric and Compton scattering images, i.e., based on the decomposed spectral image data, for example, by removing the contrast agent contribution to the photoelectric and Compton scattering images. A detailed description of the method for generating virtual non-contrasted images from the decomposed spectral image data that can be used from the virtual non-contrasted image generation unit 113 can be found, for example, in the article “An iodine-calcium separation analysis and virtually non-contrasted image generation obtained with single source dual energy MDCT” (Goshen et al., IEEE Nuclear Science Symposium Conference Record, pp. 3868-3870 (2008)).

[0051] The contrast agent residue identification unit 114 is then configured to identify residual regions in a virtual non-contrast image that include potential contrast agent residues based on the expected structural characteristics of the contrast agent residues. If a predefined combination of decomposed spectral image data (for each voxel, the contribution of at least two basis functions to the total mass attenuation coefficient of the material found in that part of the object) is performed with a less accurate decomposition of the image spectral image data, the virtual non-contrast image will have residual regions including the remaining contribution of the contrast agent, which should generally have been removed when the virtual non-contrast image was generated. The contrast agent residue identification unit 114 can be configured to offer a user a selection of the applied contrast agent, wherein, based on the selected contrast agent, the expected structural characteristics can then be predetermined. However, the contrast agent residue identification unit 114 can also be adapted to present the user with different expected structural characteristics, wherein the user can then select the most expected structural characteristic of the contrast agent residues in the current situation. In the example below, it is assumed that the contrast agent is an iodine-based contrast agent and is specifically used to enhance the visibility of the vascular structures of patient 121. However, in other examples, the contrast agent can be any other contrast agent and can be used to enhance other anatomical structures of the patient 121.

[0052] In a preferred example, iodine-based contrast agent residues used to enhance vascular structures within the region of interest for patient 121 are most likely to result in contrast agent residues with tubular shapes. In this case, the contrast agent residue identification unit 114 can be configured to apply a vascular filter to the virtual non-contrast image, enabling the identification of regions in the virtual non-contrast image that include tubular structures. These identified regions are considered to be residue regions containing contrast agent residues.

[0053] Alternatively, the contrast agent residue identification unit 114 can be configured to use a machine learning-based classification algorithm, preferably a deep learning-based classification algorithm, on the virtual uncontrast image to identify residual regions in the virtual uncontrast image. Preferably, a neural network with a U-net-shaped architecture is used. Examples of deep neural network architectures that can be used by the contrast agent residue identification unit 114 to identify residual regions can be found in the article "Fully convolutional networks for semantic segmentation" (Long et al., Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 3431-3440 (2015)).

[0054] The result of such a network or vascular filter could, for example, be to enhance the mask of voxels belonging to a virtual non-contrast image containing contrast agent residue. Therefore, based on the identification of the residue region, the portion of the virtual non-contrast image including the contrast agent residue is known.

[0055] Based on the residual region, the spectral image data decomposition unit 112 is configured to calculate a new basis decomposition at least for the residual region by recalculating the weights provided to the basis functions at least in the residual region, thereby reducing contrast agent residue. In a preferred embodiment, the spectral image data decomposition unit 112 is configured to define a residual region that includes the residual region in the virtual non-contrast image. For example, the spectral image data decomposition unit 112 may be adapted to define an elliptical, circular, rectangular, or arbitrary-shaped region surrounding the residual region as the residual region. Therefore, the residual region may be larger than the residual region. In this way, the residual region includes the (immediately adjacent) surroundings of the residual region, thereby providing additional background information about the residual region. Based on this defined residual region, the spectral image data decomposition unit 112 is preferably adapted to determine a residual measure indicating the intensity of contrast agent residue in the residual region. The residual measure can also be considered as indicating the level of inaccuracy in the material decomposition of the residual region.

[0056] To determine the residual measure and also to reduce or remove contrast agent residue as much as possible, the spectral image data decomposition unit 112 can be adapted to determine a histogram for the residual region. The histogram is defined as providing the number of voxels in the residual region falling into each bin of multiple image values, i.e., including the image values ​​falling into each bin respectively. The spectral image data decomposition unit 112 can then be adapted to determine the residual measure based on image values ​​above an image value threshold. The threshold can be predetermined or determined based on the histogram such that image values ​​above the threshold belong to the "right tail" of the histogram. The "right tail" refers to image values ​​that follow a sharp decrease in the direction of higher image values ​​along the histogram curvature. A residual measure can then be determined for each voxel that includes image values ​​above the threshold, such that the image values ​​of these voxels are removed. For example, the residual measure can be equal to the image values ​​found above the threshold. In this case, the correction value for each voxel that includes image values ​​above the threshold can be determined to be equal to the residual measure and can be transferred from a weight indicating a function of the Compton scattering effect to a weight indicating a function of the photoelectric effect.

[0057] Alternatively, the spectral image data decomposition unit 112 can be adapted to calculate the new basis decomposition based on similar regions, i.e., regions in the virtual uncontrast image similar to the residual region. In this embodiment, the spectral image data decomposition unit 112 is configured to search for regions similar to the inaccurate residual region. The spectral image data decomposition unit 112 is then configured to determine the residual measure indicating the level of inaccuracy in material decomposition as the difference between the image values ​​of voxels in the residual region and the corresponding voxel image values ​​in the similar region.

[0058] For example, for a given residual region indicating inaccurate material decomposition, the spectral image data decomposition unit 112 can be configured to use L 2 Norm, as a distance measure between regions, finds the most similar regions from a set of candidate regions sampled from a virtual non-contrast image:

[0059]

[0060] Among them, apart from voxels suspected of being inaccurate material decomposition, χ p It is an indicator function for all voxels equal to 1. The following candidate regions are then identified as similar regions: for said candidate region, the distance metric indicates that it is most similar to the residual region compared to all other candidate regions.

[0061] The spectral image data decomposition unit 112 can then be configured to determine the residual measure as the difference between the image value of voxel p(x) in the residual region and the image value of the corresponding voxel p′(x) in the similar region. In this case, the spectral image data decomposition unit 112 can be configured to determine the residual measure as a correction value, and correct the image value of the virtual non-contrast image in the residual region, for example, by subtracting the correction value of the voxel in the residual region from the corresponding voxel in the virtual non-contrast image.

[0062]

[0063] A new basis decomposition can then be calculated based on the corrected virtual uncontrast image, for example, by recalculating the weights of the basis functions such that the weights are combined to the corrected image values ​​in the residual regions of the virtual uncontrast image. Furthermore, the spectral image data decomposition unit 112 can be configured to directly apply the correction values ​​to the weights of the basis functions, i.e., the decomposed spectral data. For example, this can be achieved by applying the correction values ​​according to the above formula to the Compton scattering image p. s To calculate new values ​​for the weights indicating the contribution of the Compton scattering effect to attenuation, i.e., new image values ​​for the Compton scattering image, the photoelectric image p can be updated as follows. p That is, the weight indicating the contribution of the photoelectric effect:

[0064]

[0065] However, the spectral image data decomposition unit 112 can also be adapted to utilize other mathematical possibilities to correct the decomposed spectral image data based on the determined residual measure.

[0066] Additionally or alternatively, the spectral image data decomposition unit 112 may also be adapted to find more than one similar region, rather than only one being the most similar candidate region. For example, the spectral image data decomposition unit 112 may be adapted to determine the three most similar candidate regions as similar regions based on a selected distance norm. The spectral image data decomposition unit 112 may then be configured to estimate a residual measure, i.e., an inaccurate material decomposition, the difference between the image values ​​of voxels representing residual regions and the weighted average of the image values ​​of the corresponding voxels of the most similar regions.

[0067] Additionally or alternatively, to determine the residual region, the spectral image data decomposition unit 112 may also be adapted to define the residual region as a missing data region and use an image inpainting algorithm to calculate new decomposed spectral image data. For example, the spectral image decomposition unit 112 may simply consider all voxels belonging to the residual region including contrast agent residue, i.e., potentially inaccurate spectral CT material decomposition values, as missing voxels. The spectral image data decomposition unit 112 can then use a known image inpainting algorithm to estimate the information in the missing voxels. Based on the estimated information in the missing voxels, the spectral image data decomposition unit 112 can then be configured to determine a new basis decomposition. For example, after applying the image inpainting algorithm, the difference between the image value of a voxel in the residual region and the image value of that voxel can also be calculated as a residual measure, which can also refer to a correction value. The correction value (here, the difference) can be used to correct the decomposed spectral image data, i.e., to calculate the new basis decomposition. One possibility is that, for this case, if the basis decomposition refers to a function indicating the Compton scattering effect and a function indicating the photoelectric effect, the difference is added to the weights of the Compton scattering function and subtracted from the weights of the photoelectric function. However, other mathematical methods can also be used to correct the weights of the basis functions during decomposition based on information provided by the image inpainting algorithm.

[0068] The embodiments described above for reducing contrast agent residue and thus correcting the decomposed spectral image data can also all be used in an iterative method. In such an iterative method, the spectral image data decomposition unit 112 is adapted to apply at least one of the methods described above and provide the newly decomposed spectral image data to the virtual non-contrast image generation unit 113 again. Based on the newly generated virtual non-contrast image, the contrast agent residue identification unit 114 can again identify the residue area if at least the contrast agent residue still exists in the virtual non-contrast image. Then, the spectral image data decomposition unit 112 can again apply one of the methods described above to further reduce the contrast agent residue. In particular, the spectral image data decomposition unit 112 can apply the same or different methods as the previous iterative steps. The iteration can be stopped when the contrast agent residue identification unit 114 cannot identify the remaining potential contrast agent residue, or for example, when the remaining potential contrast agent residue includes a residue quantity below a predetermined threshold. Alternatively or additionally, the virtual non-contrast image generated by each iteration step of the iteration can also be presented to a user, such as a physician, using, for example, display unit 116, whereby the user can then use input unit 115 to decide whether the iteration should be performed or whether the iteration can be stopped.

[0069] The obtained corrected and decomposed spectral image data can then be used to generate multiple spectral images, such as Compton scattering images, photoelectric images, virtual noncontrast images, single-energy images, etc., including improved accuracy due to the increased accuracy of the decomposed spectral image data. Furthermore, the spectral image data decomposition unit 112 can be adapted to generate improved virtual noncontrast images directly based on residual measures, for example, by directly applying correction values ​​derived from the residual measures to residual regions in the virtual noncontrast image. The device 110 can then be adapted to present at least one of the generated spectral images to a user (e.g., a physician) on the display unit 116.

[0070] Figure 2 An embodiment of a method for determining the decomposed spectral image data of an object that has been injected with contrast agent before, for example, spectral image data is acquired using spectral imaging unit 120 is illustrated schematically and exemplaryly. In a first step 210, method 200 includes providing spectral image data of an object that has been acquired using, for example, spectral X-ray imaging unit 120. In a second step 220, a basis decomposition for the spectral image data is calculated as described above. Based on the obtained decomposed spectral image data, a virtual non-contrast image is generated in step 230, from which the overall contribution of contrast agent to the decomposed spectral image is removed. In the virtual non-contrast image, residual regions characterized by remaining contrast agent contributions are identified in step 240, wherein the residual regions include contrast agent residues. The residual regions are identified in particular based on the expected structural characteristics of the contrast agent residues. In step 250, new decomposed spectral image data is then calculated by calculating a new basis decomposition in the residual regions to reduce the contrast agent residues. New decomposed spectral image data is calculated by determining a residual measure of the residual region, where the residual measure indicates the intensity of the contrast agent residue in the residual region, and the weights of the basis functions provided to the spectral image data in the residual region are recalculated based on the residual measure. Optionally, the method can be used iteratively, where, in this case, as indicated by arrow 260, after step 250, the method returns to step 230 again by providing new decomposed spectral image data as the basis for generating a new virtual non-contrast image. Iteration can be stopped by using a suitable stopping criterion as described above with respect to apparatus 110.

[0071] Appendix Figure 3 and 4 An exemplary result of an image generated from decomposed spectral image data is illustrated. Specifically, Figure 3A comparison is provided between images generated based on uncorrected decomposed spectral image data and images generated based on corrected decomposed spectral image data. The image in panel 310 refers to a full-slice image, while the image in panel 320 refers to a magnified view of a contrast agent residue as identified in the image of panel 310. Columns 341 and 343 refer to images generated based on decomposed spectral image data without contrast agent residue correction, and columns 342 and 344 refer to images generated from corrected decomposed spectral image data. The first row 331 refers to a Compton scattering image, the second row 332 refers to a photoelectric image, and the last row 333 refers to a virtual non-contrast image. The small rectangles shown in the image of panel 310 indicate portions of the image where residues based on uncorrected decomposed spectral image data can be found. Examples can be shown in these images that contrast agent residues (e.g., visible in the virtual non-contrast image based on uncorrected decomposed spectral image data) are eliminated in the virtual non-contrast image based on corrected decomposed spectral image data. Therefore, it is clear that the corrected spectral image data decomposition according to the above embodiment is more accurate.

[0072] It is possible Figure 4 The same increase in accuracy is shown in the images. Figure 4 In the diagram, the image slices in the left column 410 again refer to virtual non-contrast images based on uncorrected decomposed spectral image data, while the images in the right column 420 refer to virtual non-contrast images generated based on corrected decomposed spectral image data. It can be clearly seen that the contrast agent residue visible in the right-hand images has been eliminated in the images in the left column 420. Therefore, these images also demonstrate that, after the application of the present invention as described above, the obtained decomposed spectral image data is more accurate.

[0073] Spectral CT has been proposed as a technique for better characterizing human anatomy and function with a variety of clinical applications. Spectral CT utilizes multiple attenuation values ​​acquired at various photon energies to determine the contribution of various components to the overall mass attenuation coefficient of a material. However, available suboptimal decomposition processes introduce inaccuracies in material decomposition, leading to inaccurate contrast agent concentration estimates, among other things. These inaccuracies hinder the clinical application of spectral CT.

[0074] For example, this invention proposes a shape constraint method for spectral CT material decomposition, which reduces the amount of inaccuracy in material decomposition by analyzing the shape of potential residues and subsequently applying corrections to the decomposition (e.g., for photoelectric and Compton scattering components) using a personalized statistical model. This method improves the accuracy of spectral CT material decomposition, resulting in better quality spectral results, such as virtual noncontrast images.

[0075] For example, spectral image data with at least two energy levels that allow for spectral analysis can be acquired. Examples include, but are not limited to, CT images of the anatomical structure of interest reconstructed from CT projection data acquired using a dual-layer detector system that separates the X-ray flux at the detector into two energy levels. Spectral CT data acquired by a photon-counting scanner can also be used. Material decomposition inaccuracies can then be detected by analyzing virtual, non-contrast images, which should have no contrast agent contribution. Decomposition accuracy can then be corrected directly in the virtual, non-contrast images and / or photoelectric and / or Compton scattering images, involving recalculating the decomposition using a personalized statistical model.

[0076] While it is preferred to inject an iodine-based contrast agent into the patient before acquiring spectral images, other contrast agents, such as barium-based contrast agents, can also be used.

[0077] Although in the embodiments described above for correcting the spectral image data decomposition, the spectral image data decomposition unit is adapted to determine the residual measure and, based on the residual measure, to determine the correction value, in other embodiments, the spectral image data decomposition unit is also adapted to recalculate, particularly to correct the basis decomposition when the residual measure is uncertain. For example, the spectral image data decomposition unit may be adapted to apply a machine learning algorithm to a residual region that has already been trained to correct the basis decomposition based on the residual region.

[0078] Those skilled in the art should understand that the apparatus for determining the spectral image of the decomposed spectral image of the present invention may further include a spectral X-ray imaging apparatus 120 as the source of the spectral image data to be objected.

[0079] Those skilled in the art, through studying the accompanying drawings, the disclosure, and the claims, can understand and implement other variations of the disclosed embodiments when practicing the claimed invention.

[0080] In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude a plurality.

[0081] A single unit or device can perform the functions of several items recited in the claims. Although some measures are recited in different dependent claims, this does not mean that combinations of these measures cannot be used advantageously.

[0082] Processes such as providing spectral image data, calculating decomposed spectral image data, generating virtual non-contrast images, and identifying residual regions, which are performed by one or more units or devices, can be performed by any other number of units or devices. These processes can be implemented as program code modules of a computer program and / or as dedicated hardware.

[0083] Computer programs can be stored / distributed on suitable media, such as optical storage media or solid-state media supplied together with or as part of other hardware, but computer programs can also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.

[0084] No reference numerals in the claims should be construed as limiting the scope.

[0085] This invention relates to an apparatus for determining decomposed spectral image data with improved accuracy. The apparatus includes: a spectral image data providing unit for providing spectral image data; a spectral image data decomposition unit for calculating a basis decomposition of the spectral image data; a virtual non-contrast image generation unit for generating a virtual non-contrast image based on the decomposed spectral image data; and a contrast agent residue identification unit for identifying residual regions in the virtual non-contrast image containing contrast agent residue based on the expected structural characteristics of the contrast agent residue, wherein the spectral image data decomposition unit is configured to calculate a new basis decomposition in the residual regions. Utilizing the structural characteristics of the contrast agent residue allows for very accurate determination of the contrast agent residue and improves the decomposition accuracy in these regions.

Claims

1. An apparatus for determining decomposed spectral image data of an object (121), wherein, The apparatus (110) comprises: spectral image data providing unit (111) configured to provide spectral image data of the object having injected a contrast agent, wherein the spectral image data of the object has been acquired using a spectral X-ray imaging apparatus (120), spectral image data decomposition unit (112) configured to compute a basis decomposition for the spectral image data, wherein the basis decomposition results in decomposed spectral image data, wherein the decomposed spectral image data is indicative of weights of at least two basis functions provided to the spectral image data, virtual non-contrast image generating unit (113) configured to generate a virtual non-contrast image based on a predefined combination of the decomposed spectral image data, and contrast agent residual identifying unit (114) configured to identify a residual region in the virtual non-contrast image comprising a contrast agent residual based on expected structural properties of the contrast agent residual, wherein the contrast agent residual is a contribution of the contrast agent to an attenuation of a portion of the object represented by image values of the virtual non-contrast image, wherein the spectral image data decomposition unit (112) is configured to: determine a residual measure for the residual region, wherein the residual measure is indicative of an intensity of the contrast agent residual of the residual region, and correct the weights of the basis functions provided to the spectral image data in the residual region by adding or subtracting a correction value to the weights of at least one of the basis functions such that the contrast agent residual is reduced, wherein the correction value is determined based on the residual measure, wherein the spectral image data decomposition unit (112) is configured to define a residual zone in the virtual non-contrast image containing the residual region, wherein determining the residual measure is further based on the defined residual zone.

2. The apparatus of claim 1, wherein, The spectral image data decomposition unit (112) is configured for computing the new basis decomposition to determine for the residual zone a histogram defined by image values of each region within the residual zone and to compute new decomposed spectral image data based on the histogram.

3. The apparatus of claim 1, wherein, The spectral image data decomposition unit (112) is configured for computing the new basis decomposition to determine a similar zone in the virtual non-contrast image for the residual zone, wherein a zone is similar to the residual zone if an image property of the virtual non-contrast image in the zone is similar to the image property in the residual zone, and to compute new decomposed spectral image data based on a comparison of the image values of the similar zone and the residual zone.

4. The apparatus of claim 3, wherein, The spectral image data decomposition unit (112) is configured for computing the new basis decomposition to determine the similar zone by comparing the image values of the residual zone with image values of a plurality of candidate zones by applying a predefined distance norm and determining a closest candidate zone as the similar zone based on the predefined distance norm.

5. The apparatus of any one of claims 3 and 4, wherein, The spectral image data decomposition unit (112) is configured for computing the new basis decomposition to compare the similar regions to the residual region by determining a difference between each image value and an image value spatially corresponding to the residual region, and is configured to compute the new decomposed image data based on the determined difference.

6. The apparatus of any one of claims 3 and 4, wherein, The spectral image data decomposition unit (112) is configured for computing the new basis decomposition to determine more than one similar region for the residual region, and is configured to determine a weighted average of all determined similar regions, wherein the residual region is compared to the weighted average of the similar regions.

7. The apparatus of claim 1, wherein, The spectral image data decomposition unit (112) is configured for computing the new basis decomposition to perform the following operations: defining the residual area as a missing data area, estimating information in the missing data area using an image inpainting algorithm, and computing the residual measure based on the estimated information.

8. The apparatus of any of claims 1-4, wherein, The expected structural characteristic refers to a tubular shape, and wherein the contrast agent residual identification unit (114) is adapted to identify the residual area by applying a vascular filter to the virtual non-contrast image.

9. The apparatus of any of claims 1-4, wherein, The contrast agent residual identification unit (114) is configured to apply a machine learning based classification algorithm to the virtual non-contrast image to detect the residual area.

10. The apparatus of claim 9, wherein, The machine learning based classification algorithm refers to a deep neural network architecture.

11. The apparatus of any of claims 1-4, wherein, The contrast agent refers to an iodine contrast agent.

12. The apparatus of any one of claims 1-4, wherein, The basis functions refer at least to a function defining an energy dependence of an attenuation on a Compton scattering effect and a function defining an energy dependence of the attenuation on a photoelectric effect.

13. A method for determining a spectral image data of a decomposition of an object, wherein, The method (200) comprises: providing (210) spectral image data of the object having injected a contrast agent, wherein the spectral image data of the object has been acquired using a spectral X-ray imaging apparatus (120), computing (220) a basis decomposition for the spectral image data, wherein the basis decomposition results in decomposed spectral image data, wherein the decomposed spectral image data indicates weights of at least two basis functions provided to the spectral image data, generating (230) a virtual non-contrast image based on a predefined combination of the decomposed spectral image data, and identifying (240) a residual area in the virtual non-contrast image comprising a contrast agent residual based on an expected structural characteristic of the contrast agent residual, wherein the contrast agent residual is a contribution of the contrast agent to an attenuation of a portion of an object represented by image values of the virtual non-contrast image, determining a residual measure for the residual area, wherein the residual measure indicates an intensity of the contrast agent residual of the residual area; computing (250) a new basis decomposition in the residual area by adding or subtracting a correction value to the weights of at least one of the basis functions provided to the spectral image data in the residual area such that the contrast agent residual is reduced, wherein the correction value is determined based on the residual measure, and wherein the method further comprises defining a residual region in the virtual non- contrast image containing the residual area, wherein determining the residual measure is further based on the defined residual region.

14. A computer program product for correcting decomposed spectral image data of an object, wherein, The computer program comprises program code means for causing a computer to carry out the steps of the method as claimed in claim 13 when said computer program is carried out by the computer.

Images