Motion artifact correction for phase contrast and dark field imaging

Abstract

A system (IPS) and related method for image processing, in particular dark-field or phase-contrast imaging, to reduce motion artifacts. The system comprises an input interface (IN) for receiving a series of projection images (π) of an object (OB) acquired by an X-ray imaging apparatus (XI) for a given projection direction, the imaging apparatus (XI) being configured for phase-contrast imaging and / or dark-field imaging. A phase-contrast and / or dark-field image generator (IGEN) applies an image generation algorithm to compute a first image based on the series of projection images (π). A motion artifact detector (MD) detects motion artifacts in the first image. If a motion artifact is detected, a combiner (∑) combines parts of the first image with parts of at least one auxiliary image to obtain a combined image. The auxiliary image is computed previously by applying the image generation algorithm gated with respect to a subset of the series of projection images (π). The combined image can be output at an output interface (OUT) as a motion artifact reduced image.

Description

TECHNICAL FIELD

[0001] The present invention relates to an image processing system, an image processing method, a computer readable medium and a computer program element. BACKGROUND

[0002] Dark-field imaging has attracted a lot of interest, in particular in the medical field. Dark-field imaging ("DAX") is a type of X-ray imaging. The contrast in dark-field imaging is related to the amount of small-angle scattering experienced by the X-rays.

[0003] A. Yaroshenko et al. in "Pulmonary Emphysema Diagnosis with a Preclinical Small-Animal X-ray Dark-Field Scatter-Contrast Scanner" (Radiology, vol. 269, No 2, November 2013) report on experimental dark-field imaging with mice.

[0004] It has also been found that phase-contrast ("Φ") images add additional useful insight, in particular when imaging soft tissue.

[0005] It has been noted that sometimes DAX images or phase-contrast Φ images are spoiled by motion artifacts. A. Velroyen in "Dissertation X-ray Phase-Contrast and Dark-Field Imaging of Small Animals: Contrast Enhancement and In Vivo Imaging" (Technische Universitat Munchen, March 2015) reports on a graphical illustration of the extent of artifacts due to respiratory motion. According to A. Velroyen, to minimize motion blur, forced breathing combined with a shorter exposure time would be mandatory to allow for a gated phase-step acquisition.

[0006] US2017 / 0209108A1 discloses a grating device for phase-contrast and / or dark-field imaging of a movable object, comprising a control unit configured to control an actuation unit to position a grating unit in different sampling positions based on a detected motion of the movable object. SUMMARY

[0007] Hence, there can be a need for alternative systems or methods for improved dark-field or phase-contrast imaging.

[0008] ​The object of the present application is solved by one aspect, wherein further embodiments are incorporated in another aspect. It should be noted that the following described aspects of the present application equally apply to the image processing method, the image processing system, the computer program element and the computer readable medium.

[0009] According to a first aspect of the present application, a system for image processing is provided, comprising:

[0010] an input interface for receiving a series of projection images of an object acquired by an X-ray imaging apparatus for a given projection direction, the imaging apparatus being configured for phase-contrast imaging and / or dark-field imaging;

[0011] a phase-contrast and / or dark-field image generator configured to apply an image generation algorithm to compute a first image based on the series of projection images (p);

[0012] a motion artifact detector configured to detect a motion artifact in the first image;

[0013] a combiner configured to combine at least part of the first image with at least part of at least one auxiliary image to obtain a combined image if a motion artifact is detected, the auxiliary image being computed previously by applying the image generation algorithm with respect to a subset of the series of projection images (p); and

[0014] an output interface for outputting the combined image.

[0015] The proposed system allows to solve one reason why it has been observed that dark-field or phase-contrast images suffer from motion artifacts. The reason is as follows. Some image generation algorithms, such as phase-recovery algorithms, use a signal model. The signal model assumes a stationary object. If the object does indeed move, the stationary assumption is violated and this creates image artifacts. The proposed system takes this motion into account by implementing an adaptive gating in the image generation algorithm. In particular, some image generation algorithms comprise a fitting of a signal model to the measured projection images. The fitting operation is formulated as an optimization of an objective function, such as a cost function. In particular, the proposed system can be considered as a two-step approach. Since the motion creates data that is not consistent with the signal modeling that is the basis of some phase-recovery algorithms, it has been found that the objective function after optimization usually returns values that are much larger than statistically expected. In embodiments, this observed fact is exploited in the proposed system to reduce motion artifacts. The combined image comprises information from both the first image and a particular one of the auxiliary images (gated image) to reduce motion artifacts.

[0016] The motion detection by the motion detector is based on information in the first image and / or based on analysis of additional information. This additional information can include analysis of whether the "mismatch" measured by the objective function is within a statistically expected range. In embodiments, the model can be reapplied to compute the mismatch from the first image and the projection images.

[0017] According to one embodiment, the system comprises a selector configured to select, in a selection operation, the auxiliary image from a pool of such auxiliary images computed based on different such subsets.

[0018] According to one embodiment, the image generation algorithm is based on an objective function. The image generation algorithm adjusts the fitting variable(s) of the signal model to fit the measured data (projection images). The algorithm improves the objective function in adjusting the fitting variable(s). Some or the fitting variable(s) represent the DAX or Φ image to be obtained.

[0019] According to one embodiment, wherein the selection operation is based on a score.

[0020] According to one embodiment, the score comprises a value returned by the objective function.

[0021] According to one embodiment, the motion detector is configured to detect the motion artifact based on a value of the objective function for the first image.

[0022] According to one embodiment, the system comprises a sanity checker configured to compute a series of transmission images based on a second subset and pairwise differences with the combined image, and to preserve image information from the first image at image locations in the combined image if the difference values propagate through the series of transmission images. In this embodiment, the sanity checker operates downstream of the combiner.

[0023] However, alternative embodiments are also contemplated in which the sanity checker operates upstream of the combiner. In one such upstream embodiment, the series of projection images received at the input interface are provided by the sanity checker as a subset from the original set of projection images, the sanity checker having previously computed transmission images from the original set of projection images, and the sanity checker SC is configured to exclude one or more images from the original set of projection images based on the transmission images, wherein the excluded one or more projection images represent linear motion at at least one image location.

[0024] The sanity checker is advantageous to help avoid artifacts that can be caused in the presence of temporary aperiodic motion, such as linear motion. If the motion is periodic, then the sanity checker can not be needed.

[0025] According to another aspect, there is provided an image processing method comprising the steps of:

[0026] receiving a series of projection images of an object acquired by an X-ray imaging apparatus for a given projection direction, the imaging apparatus being configured for phase-contrast imaging and / or dark-field imaging;

[0027] applying an image generation algorithm to compute a first image based on the series of projection images;

[0028] detecting motion artifacts in the first image;

[0029] if motion artifacts are detected, combining at least part of the first image with at least part of at least one auxiliary image to obtain a combined image, the auxiliary image being computed previously by applying the image generation algorithm with respect to a subset of the series of projection images; and

[0030] outputting the combined image.

[0031] According to one embodiment, the method further comprises:

[0032] i) computing a series of transmission images based on a second subset and a pairwise difference to the combined image, and if the difference values are found to propagate through the series of transmission images, retaining image information from the first image at image locations in the combined image; or

[0033] ii) wherein the received series of projection images is provided as a subset from an original set of projection images, obtained by:

[0034] computing a transmission image from the original set of projection images, and excluding one or more images from the original set of projection images based on the transmission image, wherein the excluded one or more projection images represent linear motion at at least one image location.

[0035] According to another aspect, there is provided an imaging apparatus comprising:

[0036] a system according to any of the preceding embodiments;

[0037] an X-ray imaging apparatus; and

[0038] a display device.

[0039] In one embodiment, the X-ray imaging device is of the slit-scan type or of the full- field type.

[0040] In one embodiment, the X-ray imaging apparatus comprises an interferometer.

[0041] According to another aspect, there is provided a computer program element, which, when being executed by at least one processing unit, is adapted to cause the processing unit to perform the method of any one of the embodiments described above.

[0042] According to another aspect, there is provided at least one computer readable medium having stored the program element.

[0043] The "gated application" of the image generation algorithm to a subset of the projection images includes restricting the image generation algorithm to only projection images from the subset. However, the gated application of the image generation algorithm also includes application to (one or more) projection images outside the subset (so all projection images can be processed), but the projection images in the subset are given more weight relative to the projection images outside the subset.

[0044] The "image generation algorithm" includes in particular a phase retrieval algorithm or other image generation scheme configured to extract or isolate a dark-field image and / or a phase-contrast image from a series of projection images. The acquisition of the series of projection images can be facilitated by including additional imaging hardware structures into the X-ray imaging apparatus, such as in grating-based imaging (e.g., interferometry), coded apertures, or other at least partially radiation-blocking structures with periodic or non-periodic sub-structures, etc. The image generation algorithm can include optimization of a fit of the measurement data (projection images) to a signal model. The optimization is to improve a cost function that measures a mismatch between values according to the signal model and the measurement data.

[0045] The "object" of the imaging is living and includes a human or animal patient or a part thereof, or the object is non-living, such as an item or luggage in a security screening system or a sample object in a non-destructive material testing.

[0046] The "user" includes a person (e.g., a healthcare professional) operating the imaging apparatus and / or system to image the object. BRIEF DESCRIPTION OF DRAWINGS

[0047] Exemplary embodiments of the present application will now be described with reference to the following, not-to-scale, drawings in which:

[0048] Figure 1 is a schematic block diagram of an X-ray imaging apparatus configured for phase-contrast or dark-field imaging;

[0049] Figure 2 is a schematic diagram of a motion artifact correction system;

[0050] Figure 3 is an illustration of the effect of a combining step that combines image information; and

[0051] Figure 4is a flowchart of a method for image processing. DETAILED DESCRIPTION

[0052] Reference Figure 1 Fig. 1 shows a schematic block diagram of an image processing apparatus IA comprising a computerized image processing system IPS and an X-ray imaging device XI (“imager”). The X-ray imaging device is configured for dark-field X-ray (“DAX”) imaging and / or phase-contrast (“Φ”) imaging.

[0053] The image processing system IPS can run one or more software modules or routines on one or more processing units PU, such as one or more computers, servers, etc. The IPS can be arranged external and remote to the imager XI, or the IPS is integrated into or associated with the imager XI, e.g. into one or more processing units PU of the imager XI, such as a workstation. The image processing system IPS can be implemented in a distributed architecture to serve a group of imagers over an appropriate communication network. Some or all components of the IPS can be arranged in hardware, such as an appropriately programmed FPGA (field-programmable gate array), or as a hardwired IC chip.

[0054] Broadly, the image processing system IPS comprises an image generator IGEN to process the projection images π acquired by the imager XI into dark-field and / or phase-contrast images. The projection images π are sometimes referred to as “frames” herein. The image processing system IPS comprises a motion correction system MCS (“motion corrector”) to reduce motion artifacts in the images, in particular due to patient OB motion during acquisition of the projection images π. The motion-corrected images provided by the motion correction system MCS can then be displayed on a display unit MT, or can be stored in a memory for later review, or can be further processed in other ways.

[0055] Although in the Figure 1 embodiments it is envisaged that the imaging device XI directly supplies the projection images π to the image processing system IPS via a wireless or wired connection, this can not be the case in all embodiments. For example, the projection images π can first be stored in a memory, such as a picture archiving system (PACS) of a hospital information system (HIS), etc., and the images to be processed are retrieved by the IPS at a later stage, e.g. upon user request, and then processed.

[0056] Generally, the imaging apparatus XI comprises an X-ray source XR, an X-radiation sensitive detector D and an imaging facilitator structure IFS, such as an interferometer, arranged between the source XR and the detector D. The X-ray source XR generates an X-ray beam XB which is detectable by the detector D. The imaging facilitator structure IFS is a device or set of devices which allows for refraction of the X-ray beam XB and / or conversion of small angle scattering of the beam XB into intensity modulation at the detector D facilitating resolving said modulation into dark-field and / or phase-contrast image signals and, if desired, into attenuation image signals.

[0057] In the following, reference will mainly be made to an interferometric imaging apparatus XI comprising an interferometer as the imaging facilitator structure IFS, but this does not exclude embodiments using other (in particular non-interferometric imaging facilitator structures IFS). Such non-interferometric imaging facilitator structures IFS comprise for example coded aperture systems. Generally, dark-field or phase-contrast is obtained by the imaging facilitator structure IFS imparting a periodic wavefront modulation to the incoming imaging X-ray beam and by the X-ray detector D measuring the resulting wavefront changes induced by the object OB to be imaged.

[0058] Turning now in more detail to the imaging apparatus XI, this can be configured for 2D imaging, such as a radiography apparatus, or for 3D imaging, such as a CT scanner. Between the X-ray source XR and the detector D, an imaging region is defined in which the object OB to be imaged (e.g. the chest of a subject) is located during imaging. In the imaging region, an interferometer is arranged as one embodiment of the imaging facilitator structure IFS. The interferometer comprises a single, two or three (or more) grating structures. As mentioned above, the interferometer is only one embodiment of the image facilitator structure IFS and we will mainly refer to this embodiment in the following, wherein it should be understood that the principles of the present disclosure are not limited to interferometric measurements, but can readily be extended to other grating-based or non-grating-based structures as other embodiments of the image facilitator structure IFS as mentioned above.

[0059] Continuing with the (non-limiting) interferometric measurement embodiment of the image facilitator structure IFS, the periodicity, aspect ratio, etc. of the gratings are such that they induce diffraction of the X-ray beam and / or achieve just enough coherence so that small angle scattering can be detected or derived. Absorbing gratings and phase gratings can be used. In one embodiment, the gratings are formed by lithography or cutting in a silicon wafer to define a periodic pattern of grooves. The gaps between the grooves can be filled with lead or gold for absorbing gratings. Instead of such gratings, crystal structures can be used.

[0060] In more detail and in one embodiment, an absorbing grating structure G2 is arranged between the detector D and the object OB, while another grating G1 (phase grating) is arranged between the object OB and the X-ray detector D. In some embodiments, there is also an additional grating G0 arranged at the X-ray source XR in case the X-ray source cannot generate native coherent radiation. If the X-ray source produces incoherent radiation (which is usually the case), the (absorbing) grating G0 at the X-ray source (also referred to as source grating) will at least partially transform the X-radiation coming out of the X-ray source into a coherent radiation beam XB. The reverse geometric configuration is also envisaged, where G1 is placed upstream of the object OB, i.e. between XR and OB.

[0061] The at least partially coherent radiation beam XB propagates through the imaging region and interacts with the interferometer IFS and the patient OB. After said interaction, the radiation is then detected in the form of electrical signals at the radiation sensitive pixel elements of the detector D. A data acquisition circuit DAS digitizes the electrical signals into projection (raw) image data π, which are then processed by the IPS in a manner explained in more detail below.

[0062] The imaging apparatus XI can be of the full field of view (FoV) type, where the detector is of the flat panel type. In full FoV imaging systems, the size of the detector D and the size of the IFS correspond to the desired FoV. Alternatively, the detector D and the imaging facilitator structure IFS can be smaller than the intended FoV, such as in slit scan systems as shown in Figure 1 In some of these systems, the detector comprises a series of discrete detector lines. The detector lines are mounted on a scan arm to scan across the intended FoV in different slot positions.

[0063] As Figure 1The slit scanning systems shown are more cost effective than full FoV systems because they require smaller detectors and smaller grating IFSs. The grating IFS is mounted on a scan arm above the detector and scans equally across the FoV. In an alternative slit scanning system, although the detector D has the same size as the desired FoV, the grating is smaller and collimation SC is used to scan only a portion of the FoV (in a "slit") at any one time according to collimation. In both full FoV systems and slit scanning systems with non-moving flat panel detectors, there is a simple one-to-one relationship between pixel positions and hypothetical geometric rays passing through the imaging region to define the imaging geometry configuration. The rays extend from the focal spot of the X-ray source XR and intersect the detector plane at respective pixel positions. Each of the geometric rays corresponds to a respective different single pixel in the pixel. This simple relationship does not exist in some slit scanning systems with smaller detectors, where each geometric ray is seen by many different pixels in different "slits" during scanning. The signals from the different pixels are then processed together by suitable logic for any single geometric ray.

[0064] The image generator IG outputs the dark field signal and / or the phase contrast signal as respective arrays of image values forming dark field images and phase contrast images respectively. These image values or pixel values represent the contrast of the dark field signal and the phase change experienced by the X-radiation in travelling through the object OB for the respective geometric rays respectively.

[0065] In general, when X-radiation interacts with matter, it experiences both attenuation and refraction and thus a phase change. On the other hand, attenuation can be broken down into attenuation from photoelectric absorption and attenuation from scattering. The scattering contribution can in turn be broken down into Compton scattering and Rayleigh scattering. For the current purposes of dark field imaging, of interest is small angle scattering, where "small angle" means that the scattering angle is so small that the scattered photon still reaches the same pixel that it would have reached had it not been scattered.

[0066] The dark field contribution can be modeled as visibility V = V0* e -∫s(z)dz where ε is the spatial distribution of the diffuse nature of the patient OB and the integral is performed along the X-ray beam path and V0is the reference visibility in the case of object interaction (recorded in a calibration measurement). The dark field signal recorded in the dark field image is then D = V / V0.

[0067] Conventional radiographic systems typically cannot resolve the detected signal into dark field contributions. But by using an interferometer IS as shown in Figure 1 or by using other imaging facilitator structures IFS, these contributions can be converted into intensity patterns of fringes that can be analyzed by the image generator IGEN to obtain phase contrast and / or DAX images.

[0068] Now we turn more specifically to the image generator IGEN, which operates on a series of projected images obtained during the phase-stepping operation. Based on this recorded series of projected images, the image generator IGEN computationally resolves the detected fringe pattern in the series of projected data into three contributions or signals: the refraction contribution (also known as the phase contrast signal), the dark field signal component, and the residual attenuation component.

[0069] Because these three contrast mechanisms work together, IGEN processes the detected signal intensity across three signal channels (phase contrast, dark field, and attenuation).

[0070] In imaging systems of the type described above, the ability to achieve dark-field / phase-contrast imaging is achieved as follows: during phase-stepping operations, projection data is acquired at detector D as a series for a given fixed projection direction. The phase of the fringes is typically stepped more than 360°. Phase-stepping operations are typically achieved by inducing movement between the X-ray beam and the image promoter structure IFS or its components. For example, in one embodiment, the analyzer grating G1 (i.e., the grating arranged between the object and the detector) is moved laterally (“scanned”) relative to the optical axis of the X-ray beam. Alternatively, it can also be achieved through methods such as... Figure 1 Phase stepping is achieved by moving the patient during OB or by moving the X-ray source. This phase stepping motion causes a change in the fringe pattern, which can then be recorded in a corresponding series for each step of the motion. For each geometric ray, this series of measurements forms an associated phase profile. The phase profiles are typically sinusoidal, and it has been found that each phase profile encodes the quantity of interest (especially the dark field signal) as well as attenuation and phase changes.

[0071] More specifically, the phase curve of each pixel / geometric ray can be analyzed separately, for example, by fitting it to a sinusoidal signal model as described in Pfeiffer et al., “Hard-X-ray dark-field imaging using a grating interferometer” (Nature Materials 7, pp. 134–137 (2008)), to achieve image generation. Preferably, the three-channel sinusoidal model includes at least three fitting parameters. The three fitting parameters represent three contributions: phase contrast, dark-field signal, and attenuation. The sinusoidal model is fitted to the phase curve by the image generator IGEN to specifically calculate the DAX and / or Φ images as well as the attenuation (also known as “transmission”) image, but this is less of an interest in this paper. It may be necessary to calculate a significantly redundant transmission image to properly account for the three contrast effects, as this would otherwise lead to incorrect contributions in the DAX and / or Φ channels.

[0072] An optimization procedure is used to fit the measured series of projections to the model. The procedure can be understood in terms of a cost function, and the fitting operation can be formulated as an optimization problem. Any suitable optimization scheme is also contemplated, e.g., gradient descent, conjugate gradient, Newton-Raphson, stochastic gradient, maximum likelihood methods, other statistical techniques, etc. Non-analytic methods can also be used, e.g., neural networks or other machine learning techniques.

[0073] In general, the optimization problem for a signal model M has the following structure:

[0074]

[0075] where, is an at least three-channel modulator function that describes how the three kinds of contrast mechanisms combine to modulate and transform the incident (unperturbed) radiation X into the measured data π, and ||.|| is a suitable similarity measure, e.g., p-norm, (squared) Euclidean distance, etc. The function F, the objective function (in this case, the cost function), measures how well the signal model S "explains" (or "fits") the measured data π, and the optimization task is to best choose the parameters (T, D, φ) of the model, where the similarity measure ||.|| quantifies the goodness of the fit in terms of cost or error. The task in the optimization is to improve the cost function by adjusting the parameters (T, D, φ). In this case, the parameters will be adjusted in the optimization so that the value ("cost") returned by the cost function F is reduced. More than three channels can be used in the signal model M, depending on the number of contrast mechanisms one wishes to consider. In (1), F can be a function of the residuals (terms). Each residual quantifies the corresponding deviation or cost of a given projection image from its prediction under the model S.

[0076] More specifically, in one embodiment, as a special case of (1), each pixel or geometric ray uses the following analytic signal model F = Δ 2 to optimize:

[0077] Δ 2 (T, D, φ) = ∑ i w i (M i - I i T(1 + V i D cos(φ - α i ))) 2 (2)

[0078] where, M i is the measured data (taken from π), the unperturbed radiation "X" is given by I i , V i , and α idenotes, respectively, the blank scan intensity, the visibility and the phase. T, D and φ are, respectively, the three contrast modulators of S() above, i.e. the transmittance, the dark-field and the differential phase of this image point. w i is an optional statistical weight, typically chosen equal to the inverse variance of the measurement data M i or proportional to the inverse variance of the measurement data M i and i indicates the phase step. Starting from S(), the task in (2) is to minimize the cost TΔ 2 on the measurement data M

[0079] The image generation algorithms (1), (2) of the above type are sometimes referred to as “phase retrieval”, but this is a misnomer for the present purposes, as there is also a dark-field image generated in common in the fitting operation, and in fact also a transmittance image as mentioned above. Other Fourier-based methods, such as the method of Pfeiffer et al. (cited above) and related methods, are also contemplated herein as phase retrieval algorithms in embodiments.

[0080] The above phase stepping operation takes time, and this can lead to adverse consequences when imaging a moving object, such as an animal or a human patient. More specifically, in the phase stepping of DAXand Φimaging, a certain number of frames is required for image generation, and each frame acquisition takes time for exposure, data readout and movement of the gratings to the next position. The acquisition of these frames consumes a time interval of about one second. This is of the same order of magnitude as the duration of a heartbeat, so intensity fluctuations due to cardiac motion disturb the modulation fringe pattern of the DAX / Φphase retrieval algorithm, and this can lead to artifacts. These influences can be observed in time-slot scanning systems as shown in Figure 1 , but are expected to occur also in full FoV systems.

[0081] A particular problem caused by the longer acquisition times is the relevance of cardiac motion, as this leads to motion artifacts being recorded. One of the main reasons for the artifacts to manifest is that the patient OB does not comply with the model assumptions (1), (2) on which the image generation is based. In (1), (2) a stationary imaging object is assumed. In particular, in (1), (2) the transmittance T is assumed to be the same for all measurements π in the signal model. However, in the presence of motion, the transmittance T is not the same for all measurements π, but intensity variations are recorded in addition to those expected due to the phase stepping measurements. Unfortunately, if not taken into account, these additional intensity variations are now erroneously attributed to variations in the other two channels, the dark-field signal D and the differential phase φ.

[0082] The proposed motion corrector system MCS is configured to take into account patient motion, thereby reducing motion artifacts in DAX or Phi imaging. In order to remove or at least reduce motion artifacts caused by motion of the patient's OB, in particular respiratory or cardiac motion, the proposed motion corrector system MCS is broadly configured to adaptively apply cardiac and / or respiratory gating in the phase retrieval step. In particular, and as will be more fully described below, a two-step approach is proposed. Since motion produces data that is inconsistent with the model, the cost function after optimization is typically much larger than statistically expected. It has been observed that image locations affected by motion artifacts typically return larger values of the cost function.

[0083] However, the Applicant has found that such gating operations, i.e. discarding or down-weighting portions of data, are a double-edged sword that cannot always successfully trade-off between the two opposite goals, i.e. low image artifacts and low image noise. Moreover, there is limited freedom in gating the acquisition, since it is necessary to simultaneously ensure that enough data is acquired for image generation, in particular for phase retrieval that relies on multiple steps i to have a well-defined stepping curve.

[0084] In more detail, the Applicant has found that motion artifact reduction by simple gating is not necessarily uniform in the whole image. Motion artifact reduction is rather local and different for different image regions. While motion artifacts are greatly reduced in most locations, the amount of artifact reduction seems to depend on the location in the image. This can be particularly true if cardiac motion produces artifacts in one location, but respiratory motion produces artifacts in another location. Moreover, there are even image regions where the artifact level is increasing, for example because the phase stepping curve is poorly sampled due to gating. Moreover, it has been observed that the reduced artifact damage level comes at the cost of increased noise due to the use of fewer data points in the phase retrieval in some gating window schemes. In order to address both goals, i.e. noise reduction and motion artifact reduction in the whole or most of the image, the proposed motion correction system MCS is configured to proceed in two stages. The components to which the proposed motion corrector MCS refers will now be explained in detail with reference to the block diagram of Figure 2

[0085] Broadly, input projection images π are received at an input interface IN. The input images are processed in a manner that will be more fully described below, to produce motion artifact corrected dark field and / or phase contrast images that are output at an output port OUT.

[0086] ​In one stage, a phase retrieval operation is performed based on a series of projection images p obtained during the described phase stepping operation. This series of projection images p is then processed. In particular, the image generator IGEN implements a phase retrieval algorithm to produce a first image, such as a phase contrast and / or dark field image. This image will be referred to herein as the "primary image".

[0087] In yet another stage, which can occur before or after the above-mentioned previous stages, a retrospective gated phase retrieval is then performed in embodiments by the image generator IGEN either in the previous preparation stage or on demand during operation of the motion correction system MCS.

[0088] This gated phase retrieval operation includes sampling by the sampler SP from the original set of projection images p respective different subsets. Based on these subsets, the image generator IGEN performs respective phase retrieval operations only for those different subsets. In (2) above, the summation over i will thus only run over those subsets of indices associated with the sampled projection images. These operations result in different gated phase contrast and / or dark field images, referred to herein as "auxiliary" or "gated images". Thus, each sample subset results in a respective gated image. These gated images can be stored in the memory storage device P.

[0089] For example, one simple under-sampling scheme can include forming different subsets by selecting different samples of a particular fraction p of all available measurement data p, such as p = 50%. Other fraction values of p are also envisaged. However, if p is too small, the fraction p needs to be balanced against an expected rise in noise contribution. The fraction p can be user-adjustable, or the system adaptively adjusts the fraction p, e.g. based on an entropy evaluator that inspects the primary image or the set of projection images. If the global image information is complex, a higher p value can be required, otherwise p can be reduced. Alternatively, as will be further explained below, P can vary as a function of image location during processing. The sampling can be random, or can be guided by other considerations, such as reasonably covering the full 360° phase of the remaining samples, so that proper phase retrieval is still possible. Reasonably can mean, for example, that the maximum gap between phase samples is less than 90°.

[0090] The motion detector MD then inspects the primary image to locate one or more image regions that are or can be compromised by motion artifacts, and marks those one or more locations accordingly, where the respective marking indicates motion corruption, e.g.: "yes" = 1 or "no" = 0. This binary 0 / 1 marking is an example of a "hard" marking. A "soft" or continuous marking can also be implemented, e.g. by providing a "likelihood" of artifact presence, which ranges, for example, from 0 to 1.

[0091] In embodiments, the motion detector MD can be implemented by thresholding a cost function used in the phase retrieval algorithm. For example, in embodiments, the function F = Δ 2 or any other function in equation (1) is used. Other motion detection criteria can also be used. In particular, if the cost function is suitably set to be a noise-weighted least squares cost function, then its value is expected to follow a χ 2 distribution. The likelihood of a χ 2 likelihood function can be the cumulative probability derived from this distribution. Alternatively, in a discrete version, where all image points whose cost function exceeds the expected value of the χ 2 distribution by more than 2 standard deviations can be labelled "yes" = 1.

[0092] The selector SEL selects from the storage device P for some or each image pixel that has been found by the motion detector MD to be corrupted by motion artefacts the respective gated image. In embodiments, the selector SEL selects for a given motion artefact corrupted pixel or patch the gated image that has the smallest cost function F at the respective pixel identified by the motion detector in the primary image. Alternatively or additionally, the selector SEL selects the gated image that has the smallest standard deviation locally, i.e. in the patch, e.g. the gated image that has the smallest level of artefact corruption, or other. Typically, an agreed measure function or score is used to measure the goodness of the selected gated image for each patch or pixel in the primary image.

[0093] The goodness or "best" gated image is measured against a score threshold. There can be more than one "best" gated image, in which case a random operation can be performed to pinpoint a single gated image, or the average of the best gated images can be used.

[0094] The selection operation thus determines the best gated window or best gated image that is best suited to reduce / eliminate the motion artefacts found by the motion detector in the primary image. The selection operation is performed pixel by pixel or regionally ("patch-wise") and is returned for each different pixel / region in the primary image, typically a different gated image. In other words, the selection operation by the selector establishes a mapping in which for each region or pixel in the primary image a respective gated image from the storage device P is associated that is best suited to eliminate or reduce the motion artefacts at the respective location in the primary image.

[0095] The selector map can be adaptive in that the fraction p is adjusted based on an evaluation of the local information in the primary image. Based on this evaluation, P is determined, and based on this value, the pool is then built by the sampler SP, from which the selector SEL then selects the correct gated image. In other words, in certain embodiments, the pool does not always contain the same gated image for a given primary image (but this is indeed the case in some embodiments), but the pool content varies with the image position in the primary image. The evaluation can be done by the mentioned entropy analyzer to establish the image information complexity, and then the fraction p is determined based on the entropy at the considered image position (pixel or patch).

[0096] The combiner Σ then combines the image information from the primary image with the image information from the best gated image found by the selector SEL to eliminate or at least reduce the image artifacts at the respective positions. The combining operation of the combiner Σ is typically done pixel-wise or at least region-wise.

[0097] From the above, it can be understood that all those pixels or regions in the primary image that are not found to be impaired by motion artifacts are maintained and not changed. However, those image pixels (x, y) that have been found to be impaired by motion by the motion detector will be changed based on the associated gated image.

[0098] The combining operation of the combiner Σ can be done by simply replacing the motion corrupted pixel at (x, y) in the primary image by the image information at (x, y) in the associated (and thus “best”) gated image. Optionally, a linear combination of the two pixel values (i.e., from the primary image and the associated gated image) can be performed. In this linear combination, the weight of each pixel / patch can depend on a quality measure, e.g., the value of the cost function at this pixel / patch.

[0099] More specifically, the weight is “sliding” in that a high weight is given to the pixel in the original value or the gated image, depending on the degree of motion corruption that is present. For example, if a high motion corruption is found at the respective position in the primary image, more weight is given to the image value derived from the associated gated image. However, if less motion corruption is found at the respective image position, the weight is reduced to favor the weight of the image value in the primary image. In other words, the weight is proportional to the level of motion corruption found by the motion detector MD at a given image position in the primary image.

[0100] Once the combining operation by the combiner Σ has been completed, the combined image thus found is then output at the output port OUT as a motion artifact reduced image. This can then be stored, displayed on a display device, or otherwise processed.

[0101] The above embodiments of the motion corrector MCS can be understood as "hard" gating embodiments. In other words, when solving (1), (2) to generate the gated image, the image generator IGEN only processes the selected projection images selected by the selector SEL.

[0102] In alternative embodiments, a "soft" gating scheme is envisaged, in which a penalizing weight is applied by the image generator when solving (1), (2) to relatively down-weight the residuals of the non-selected projection images, thereby giving more weight to the residuals of the selected images. In (2), for each stage step i, the penalizing weight is envisaged to be more than and higher than the optional statistical weight w i .

[0103] In other words, in soft gating, more (in particular all) projection images p are taken into account in the optimization of (1), (2). However, the residual terms involving the non-selected projection images are (at least) relatively down-weighted compared to the residual terms involving the selected projection images. The relative down-weighting comprises applying a higher weight to the residuals of the selected projection images and / or applying a lower weight to the residuals of the non-selected projection images. In summary, in soft gating, the excluded (non-selected) projection images are still taken into account for the phase retrieval (1), (2), but their contribution to the objective function F is reduced. Thus, the mode of operation of the image generator based on the above soft gating or hard gating schemes can be collectively referred to herein as a gated application of the image generation algorithm, in which more (in particular all) projection images are equally considered in the optimization (1), (2), compared to the (ungated) image generation of the primary image.

[0104] Optionally, the system MCS can further comprise a sanity checker SC. This is because it has been found that the sampling operation or the selection operation of the selector SEL can sometimes introduce an undesired aliasing effect, which can lead to the removal of real image structure in the combined image. Due to the selection of the subset by the selector SEL or the sampler SP, the aliasing effect can cause real structure in the primary image to be eliminated. The sanity check is only performed for image positions that can be in dispute, i.e. for image positions that are flagged by the motion detector MD.

[0105] The sanity checker SC interacts with the combiner ∑. The sanity checker SC itself performs a selection operation from the measured dataset p, but this selection is based on a subset that is a complement of the set of those selected by the selector. This complementary selection is performed separately for each flagged image position in the primary image, and it is understood that the following applies to each (or some) such flagged image position in the primary image.

[0106] In more detail and as explained above, each gated image selected for combination by the selector has been obtained by projecting a particular sub-selection of the images. Then, the plausibility checker considers for this selection the subset of the projection images that were not selected, i.e. the (set) complement. From this subset of complementary projection images, a corresponding series of transmission images is computed. This can be done, for example, by using a Fourier-based approach, in which the diffraction pattern imprinted by the image facilitator structure IFS, such as an interferometer (G1, G2), is cancelled from the projection images p in the complementary subset. Alternatively, the fringe pattern is subtracted from the images of the stepping curve based on the computed phase contrast image (phase of the fringe pattern) and dark field image (amplitude of the fringe pattern).

[0107] Then a pairwise difference image is formed between each transmission image and the proposed combined image obtained by the combiner. The pairwise difference operation can result in the formation of particular peaks in the difference image. A tracking operation is then performed to track those peaks. If a peak is found to propagate through the series of difference images, this is then used to indicate that the corresponding image value in the combined image at the considered location is spurious, and in this case the original image value from the primary image is then retained at the considered marked image location. In other words, if the proposed combined image is rejected, the replacement value is returned by the combiner.

[0108] However, if no such propagation is found, this indicates that the image value produced by the combiner is realistic and represents the true image structure and allows the proposed image value in the combined image to be retained. In other words, the plausibility checker based on the propagation evaluation rejects or retains the image value in the combined image, and if rejected, maintains the original image information from the primary image.

[0109] Additionally or alternatively, a significant indicator value is included at the respective image location to indicate to the viewer that this location cannot be trusted as it can or can not be corrupted by motion artefacts. The significant indicator value can be chosen to be a significantly different indicator value from the values in its neighbourhood, such as "1" or "0" in the case of the grey value rendering palette commonly used on radiographs.

[0110] Figure 3 is a diagrammatic illustration as to how such aliasing effects can arise. Assume in the toy example that the field of view consists of five pixels as shown in Figure 3 and that the stepping curve obtained in the phase stepping operation consists of 5 frames from top to bottom as diagrammatically shown in Figure 3 In the first and second frames, there is a small object ST captured from the left in the second pixel. In the last three frames, the object ST moves one pixel to the right with linear aperiodic motion. If the system as in Figure 2 (or the followingFigure 4 If the processing is performed by locally (each pixel individually) minimizing the residual of the cost function with sampling, as in the method described above, then it can be seen that the resulting image RI will consist of 5 blank pixels.

[0111] This "aliasing" effect may be undesirable if the object ST is considered to be a small structure that is potentially relevant to diagnostic imaging. In particular, if linear motion is captured in the image, local minimization of the cost function residuals may be problematic for diagnostic accuracy.

[0112] The sanity checker SC implements a method based on analyzing the similarity of the received projected images using transmission information. In this way, it can be determined whether a true absorption feature ST exists in the phase-stepping sequence, and whether linear motion (propagation) exists for the frame at the start of motion and for that pixel. The above-described embodiment of the sanity checker SC acts as a post-processor relative to the output of the combiner ∑.

[0113] Alternatively, the soundness checker SC can be used as a preprocessor upstream of the combiner Σ. In this embodiment, transmittance information can be used to constrain the operation of the sampler SP when forming a subset of the projected image from which a DAX / Φ image can be generated for use by the combiner Σ. Specifically, in Figure 3 In the example, for the second pixel starting from the left, only the first two frames (first f, second f) will be allowed to be considered for motion artifact reduction. This embodiment will be explained in more detail below.

[0114] While the proposed system MCS has been found to produce satisfactory results when oscillatory or periodic motion is involved, it may fail and produce erroneous images in cases where non-periodic or non-oscillatory motion exists in the imaged object (OB). Specifically, in the latter case, the sanity checker (SC) is recommended, while it may not be necessary if the system MCS is used where only periodic motion (such as cardiac or respiratory motion) is possible. The sanity checker (SC) can be turned on and off by the user.

[0115] The operation of the sanity checker SC, specifically as a preprocessor, will now be explained in more detail, including additional embodiments.

[0116] This can be achieved by Fourier decomposition as described by e.g. Han Wen et al. in "Fourier X-ray scattering radiography yields bone structural information" (Radiology 251.3 (2009), pages 910-918) or by subtracting the modulation superimposed on the image based on e.g. the phase contrast (modulated phase) and the dark field (modulated amplitude) as computed by the combiner. A third possibility would be to use all subsets of the consecutive projection images p (e.g. if p contains 10 images, then use the images with frame numbers 1, 2, 3) to compute the transmission image.

[0117] In case a method similar to the one described by Han Wen et al. is used, it can be advantageous to introduce a relatively high frequency Moire pattern onto the detector. Thus, the transmission image should be computed from each image of the stepping curve.

[0118] Other embodiments are also envisaged, in which the sanity checker does not act as a post-processor, i.e. on the combined image, but as a filter in a pre-stage to ensure that only certain projection subsets are allowed to be processed into a gating image and admitted into the pool. In some such pre-stage embodiments of the sanity checker SC, the transmission images can be compared to each other - e.g. by subtracting the images from each other - and based on certain thresholds, it is determined whether motion is present and where it is present. Then, those frame numbers in p that are not affected by motion are determined. In addition, a comparison of neighboring pixels of some or all image series p will reveal additional information about whether the motion is linear or oscillatory.

[0119] The pre-selection of the images of the image series p that are allowed to be processed into a gating image for selection by the selector SEL can be performed based on the following logic of the sanity checker: if the absorption level changes significantly between frames, i.e. if an absorption feature SF is present in one or several frames in a pixel and then disappears or appears in a different pixel, then images from the stepping curve that depict this feature should be selected. In other words, it is proposed in embodiments herein to allow only those frames to be processed into a gating image that represents an absorption structure (and then admitted into the pool) and not merely to locally minimize a cost function, as this does not necessarily guarantee that the feature will be preserved in the resulting image Rl.

[0120] The definition of an "absorption feature" will vary from case to case and it is suggested in embodiments to use thresholding on neighboring pixels. In more detail, and in one embodiment, if the deviation of the absorption from neighboring pixels exceeds a certain threshold th, e.g. th = 10%, then this can be used to indicate that an absorption feature ST is present at the considered pixel location.

[0121] If the above filtering against suitable frames not recording linear movement results in an insufficient number (e.g. 1 or 2) of remaining “normal” processing flow in the MSC, it is suggested to directly use the signal from the single transmission image of the subtraction derived pi from the superimposed fringe modulation and place a marker at the corresponding position in the main image (dark field or phase contrast) to clearly indicate to the user that a motion artifact is present. In case a gray scale value palette is used, the marker can be set by e.g. replacing the image value with a non-real value such as “1” or “0”.

[0122] Reference is now made to Figure 4 which shows a flowchart of the proposed image processing method that can serve as a basis for the operation of the motion corrector MCS. Figure 2 It is understood, however, that the method steps described below constitute the teaching per se and are not necessarily bound to the specific architecture shown in Figure 1 or Figure 2 .

[0123] Broadly, the proposed new image processing method for motion artifact correction (in particular for DAX or Phi images) comprises a step S410 in which projection data pi is received as input data. The input data can be motion corrupted. Retrospective gating is used to form gated images and some of these can then be used to “repair” the locally appearing motion corruption in the computed DAX or Phi images.

[0124] More specifically, the input data received at step S410 comprises a series of projection images for a given projection direction. Radiography is primarily envisaged herein. A series of projection images of an object PAT are acquired by an x-ray imaging apparatus in a given projection direction. The x-ray imaging apparatus is understood to be configured for phase and / or dark field imaging. In embodiments, this comprises having an image facilitator structure IFS arranged between an x-ray source and a detector D of the imaging apparatus. The image facilitator structure IFS, such as a grating(s), interacts with the radiation passing therethrough and passing through the patient and modulates the radiation into an intensity pattern that can be analyzed by an image processing operation. This image processing operation can comprise a phase retrieval operation to extract, in particular, a phase contrast image signal and / or a dark field image signal.

[0125] In step S420, a phase retrieval operation is thus applied to compute a first image from the series of projection images for a given projection direction, referred to herein as a main image. For this operation, preferably the entire series of projection images is used. The phase retrieval operation can be based on an improvement objective function. In embodiments, the objective function is a cost function and the phase retrieval involves a computation of minimizing the cost function. The minimization can comprise adjusting fitting variables in order to make the cost function return a reduced value. Alternatively, the phase retrieval can also be formulated in terms of maximizing a utility function.

[0126] The main image thus obtained is a phase-contrast or dark-field image, however, since the patient can have moved during the acquisition of the projection images, this phase-contrast or dark-field image is expected to be impaired by motion artifacts due to motion corruption in the projection images. The projection images can comprise motion artifacts caused or imparted by cardiac activity or respiratory activity, etc. The motion can be periodic or aperiodic.

[0127] In step S430, motion artifacts are detected in the first image based on a motion detection algorithm. The motion artifact detection S430 is preferably performed locally, i.e. per image location. In the following, “image location” means to refer to a pixel location or a patch, neighborhood or region around the pixel, such as.

[0128] As mentioned above, in embodiments, the phase retrieval algorithm in step S420 is based on an optimization of an objective function, and the value of the objective function for a given image location, such as a pixel or voxel, is thresholded to decide whether there is image motion or not.

[0129] Alternatively, the motion detection algorithm analyzes the information within the image or considers a short sequence of projection images for the respective image location to establish motion corruption.

[0130] In step S450, if a motion artifact is detected in step S430, then for the considered image location, a part of the first image is combined with a part of the auxiliary image to obtain a combined image.

[0131] The auxiliary image has been previously pre-computed by applying an image generation algorithm to a subset of the series of projection images p received at step S410.

[0132] In more detail and in embodiments, there is a gating operation in the form of a selection step S440 that performs a selection operation. In the selection operation S440, an auxiliary image is selected from a pool of auxiliary images for the considered image position. The selection S440 is based on a score or a measure. The score can represent an amount of motion disruption present at the image position (in the auxiliary image) that corresponds to the considered position in the primary image. According to one embodiment, the standard deviation in the image position (as a tile) is the score, and the selection S440 selects the auxiliary image that has the smallest standard deviation at the corresponding image position or a standard deviation that is less than a threshold. In other embodiments, the cost function value itself is used as the score, and the selection S440 returns the auxiliary image that has the smallest cost or at least a cost that is less than a threshold. Other embodiments of the selection score are also envisaged. The auxiliary images in the pool are preferably pre-computed from different such subsets of some or all of the original projection images n in the sampling step S425. The auxiliary or gating images are computed by a phase retrieval algorithm, preferably the phase retrieval algorithm used in step S420, but this time limited to each subset of projection images. Moreover, instead of selecting some subset of projection images while excluding other projection images, it is also possible to allow all projections, but then the other projections (the non-selected projections) are at least relatively down-weighted with respect to their contribution to the cost function, as mentioned above in the context of "hard" and "soft" gating. Alternatively, a different phase retrieval algorithm is used to obtain the gating images. The computation of the gating images can be performed by the same computational entity (machine or module) as used for step S420 in the computation of the primary image. Alternatively, a different entity running the same or a different phase retrieval algorithm can be used. The pool of gating images can have been computed elsewhere and stored in the storage device P and later accessed by the selection step S440. Figure 2 The pool of auxiliary images can be sampled S425 and computed S430 on demand, or can be pre-computed and held in storage for other image positions of the same primary image. The auxiliary images can be computed before or after step S420. The pool of gating images can keep a constant throughput in part or all of the operation of a given primary image, so that the selection occurs from the same pool. Alternatively, the composition of the pool changes during the execution of the method due to an adaptive sampling step that resamples from the pool according to, for example, the given image position of the primary image currently processed.

[0133] The pool of auxiliary images can be sampled S425 and computed S430 on demand, or can be pre-computed and held in storage for other image positions of the same primary image. The auxiliary images can be computed before or after step S420. The pool of gating images can keep a constant throughput in part or all of the operation of a given primary image, so that the selection occurs from the same pool. Alternatively, the composition of the pool changes during the execution of the method due to an adaptive sampling step that resamples from the pool according to, for example, the given image position of the primary image currently processed.

[0134] In step S460, the primary and secondary images are combined. This combination is local and is preferably performed only for those image positions of the primary image for which motion corruption was found in step S430. The combination can comprise replacing the image value(s) at a position by the image value(s) at the corresponding position in the associated secondary (gated) image. Alternatively, a linear or non-linear combination is formed and this is used to replace the value at the considered image position in the primary image.

[0135] The proposed method can further comprise an optional step S470 in which a plausibility check is performed.

[0136] The plausibility check can comprise computing a corresponding series or transmission image based on different subsets of the projection images. The plausibility check can be performed for some or each of those image positions of the primary image for which motion corruption was found in step S430.

[0137] The plausibility checker step S470 comprises forming pairwise difference images between the series of transmission images and the combined images obtained at step S460. The series of difference images thus obtained is then analyzed to find whether certain difference values or peaks propagate through the series of difference images. An image structure tracking scheme can be used for this. If such a propagation can be established, the image value at the considered image position in the combined image is replaced by the original value at the respective image position in the original first image. In other words, the proposed image value in the combination step is rejected. If, however, such a propagation cannot be established, the image value at the considered image position as computed in the combination step is confirmed and maintained.

[0138] The plausibility check operation at step S470 allows to avoid that a real image structure is eliminated by aliasing effects caused by the selection or sampling operation which forms the basis of the respective secondary image already used in the combination step S470. The plausibility check step is preferably used in the presence of aperiodic or non-oscillatory motion which is expected to have occurred during the acquisition of the original measurement data π. At a subsequent step, the possibly plausibility-checked combined image is then output in the next step as a motion artifact reduced image.

[0139] Instead of applying a plausibility check step S470 as a post-processing step after the combiner step S460, it is also envisaged herein to apply a pre-processing plausibility check step upstream of the combiner step S460. In this embodiment, the projection images are converted into transmission images in step S400, e.g. by a Fourier method and subtraction to eliminate the stripe pattern. A motion analysis is performed, e.g. tracking. Frames representing linear motion at one or more image positions are excluded at step S405, thereby reducing the original set of projection images π. This reduced set of projection images π' is then processed as described above starting from step S430.

[0140] At step S480, the combined image is output (optionally checked for sanity) for storage, visualization, or any other processing.

[0141] The components of the image processing system IPS, in particular the MSC, can be implemented as software modules or routines in a single software suite and run on a general purpose computing unit PU, such as a workstation associated with an imager IM or a server computer associated with a group of imagers. Alternatively, the components of the image processing system IPS can be arranged in a distributed architecture and connected in a suitable communication network.

[0142] Some or all components can be arranged in hardware, such as a suitably programmed FPGA (field programmable gate array) or as a hardwired IC chip.

[0143] One or more features disclosed herein can be configured or implemented as / with circuitry encoded within a computer readable medium and / or combinations thereof. Circuitry can include discrete and / or integrated circuitry, application specific integrated circuits (ASICs), system on a chip (SOC), combinations thereof, machines, computer systems, processors and memories, computer programs.

[0144] In another exemplary embodiment of the present application, a computer program or a computer program element is provided that is characterized by being adapted to execute the method steps of the method according to one of the preceding embodiments, on an appropriate system.

[0145] Accordingly, the computer program element might be stored on a computer unit, which also implements at least a part of the

[0146] This exemplary embodiment of the present application covers both the computer program, as such, and the computer program that is implemented on a computer.

[0147] Further, the computer program element might be able to provide all the steps of a method as described above.

[0148] According to a further exemplary embodiment of the present application, a computer readable medium, such as a CD-ROM, having stored thereon the computer program of a preceding paragraph is presented.

[0149] A computer program can be stored and / or distributed on a suitable medium, such as an optical storage medium or a solid state storage medium supplied together with or as part of other hardware, but can also be distributed in other forms, such as via the internet or other wired or wireless telecommunication systems.

[0150] However, the computer program can also be presented over a network like the World Wide Web and can be downloaded into the working memory of a data processor from such a network. According to a further exemplary embodiment of the present application, a medium for making a computer program element available for downloading is provided, which computer program element is arranged to perform a method according to one of the previously described embodiments of the application.

[0151] It has to be kept in mind that the embodiments of the present application are described with reference to different subjects. In particular, some embodiments are described with reference to method type claims whereas other embodiments are described by way of device type claims. However, a person skilled in the art will gather from the above and the following description that, unless other notified, in addition to any combination of features belonging to one type of subject matter also any combination between features

[0152] While the application has been illustrated and described in detail in the drawings and foregoing description, such illustration and description are to be considered illustrative or exemplary and not restrictive. The application is not limited to the disclosed embodiments. Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practising the claimed application, from a study of the drawings, the disclosure, and the claims.

[0153] In the claims, the word "comprising" does not exclude other elements or steps, and the words "a" or "an" do not exclude a plurality. A single processor or other unit can fulfil the functions of several items recited in the claims. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage. Any reference signs in the claims should not be construed as limiting the scope.

Claims

1. A system (IPS) for image processing, comprising: an input interface (IN) for receiving a series of projection images (π) of an object (OB) acquired by an X-ray imaging apparatus (XI) for a given projection direction, the X-ray imaging apparatus (XI) being configured for phase-contrast imaging and / or dark-field imaging; a phase-contrast and / or dark-field image generator (IGEN) configured to apply an image generation algorithm to compute a first image based on the series of projection images (π); a motion artifact detector (MD) configured to detect a motion artifact in the first image; a combiner (∑) configured to, if a motion artifact is detected at a artifact location in the first image, combine the first image and at least one auxiliary image at a location corresponding to the artifact location in the first image, the auxiliary image being previously computed by a gated application of the image generation algorithm with respect to a subset of the series of projection images (π), to obtain a combined image at the artifact location; and an output interface (OUT) for outputting the combined image.

2. The system of claim 1, wherein, a selector (SEL) configured to select, in a selection operation, the auxiliary image from a pool of such auxiliary images computed based on different subsets.

3. The system of claim 2, wherein, the image generation algorithm is based on an objective function.

4. The system of claim 3, wherein, the selection operation is based on a score.

5. The system of claim 4, wherein, the score comprises a value returned by the objective function.

6. The system of claim 3, wherein, the motion artifact detector (MD) is configured to detect the motion artifact based on a value of the objective function for the first image.

7. The system of any one of claims 1-6, comprising a sanity checker (SC) configured to compute a series of transmission images based on a second subset, form a series of pairwise difference images between the series of transmission images and the combined image, and if a difference value is found to propagate through the series of pairwise difference images, retain image information from the first image at an image location in the combined image.

8. The system of any one of claims 1-6, wherein, the series of projection images received at the input interface (IN) is provided as a subset from an original set of projection images by a sanity checker (SC) that previously computed transmission images from the original set of projection images and that is configured to exclude one or more images from the original set of projection images based on the transmission images, wherein the excluded one or more projection images are indicative of a linear motion at at least one image location.

9. An image processing method, comprising the steps of: receiving (S410) a series of projection images (π) of an object (OB) acquired by an X-ray imaging apparatus (XI) for a given projection direction, the X-ray imaging apparatus (XI) being configured for phase-contrast imaging and / or dark-field imaging; applying (S420) an image generation algorithm to compute a first image based on the series of projection images (π); detecting (S430) a motion artifact in the first image; if a motion artifact is detected at a position of an artifact in the first image, combining (S450) the first image with at least one auxiliary image at a position corresponding to the position of the artifact in the first image, the auxiliary image previously computed by applying the image generation algorithm gated with respect to a subset of the series of projection images (pi); and outputting (S480) the combined image.

10. The method according to claim 9, further comprising: i) computing (S470) a series of transmission images based on a second subset, forming a series of pair-wise difference images between the series of transmission images and the combined image, and if a difference value is found to propagate through the series of pair-wise difference images, retaining image information from the first image at the image position in the combined image; or the method comprising: ii) wherein the received series of projection images is provided as a subset from an original set of projection images, the original set of projection images obtained by: computing (S400) a transmission image from the original set of projection images, and excluding (S405) one or more images from the original set of projection images based on the transmission image, wherein the excluded one or more projection images represent linear motion at at least one image position.

11. An imaging arrangement (IR) comprising: a system (IPS) for image processing according to any of claims 1-8; an X-ray imaging arrangement (XI); and a display device (MT).

12. The apparatus of claim 11, wherein, The X-ray imaging arrangement (XI) is of the slit-scan type or of the full- field type.

13. The apparatus of claim 11 or 12, wherein, The X-ray imaging arrangement (XI) comprises an interferometer (IFS).

14. A computer program product comprising a computer program which, when executed by at least one processing unit (PU), is adapted to cause the processing unit (PU) to perform the method according to claim 9 or 10.

15. A computer readable medium having stored thereon a computer program which, when executed by at least one processing unit (PU), is adapted to cause the processing unit (PU) to perform the method according to claim 9 or 10.

Images