Method and system for x-ray image quality assurance

Abstract

A computer implemented method for x-ray image quality assurance is provided. The method comprises: acquiring X-ray image data of an object using a mobile X-ray imaging system, the X-ray image data comprising at least one X-ray image; performing an image-based automatic detection of an imaging error indicative of a misalignment of the X-ray imaging system with respect to the object based on the X-ray image data; and in response to detecting an imaging error indicative of misalignment, causing the display device to display a user interface including an indicator indicative of the presence of misalignment and / or an indicator indicative of a type of misalignment and / or an indicator indicative of a degree of misalignment.

Description

Technical Field

[0001] The present invention provides a method and system for X-ray image quality assurance, a computer program product and a computer readable medium. Background Art

[0002] X-ray imaging is the modality of choice when it comes to verification of wire and tube positioning in the intensive care unit (ICU) and emergency department (ED). This assessment is often performed based on the location of the device (e.g., in terms of the device tip) and its relationship to specific anatomical landmarks or regions. In addition, it is often one of the first modalities to address and follow chest pathology.

[0003] Due to the challenging imaging conditions in the ICU and ED, mobile chest radiography is susceptible to a wide range of imaging problems. In this context, especially projection errors and distortions due to suboptimal positioning of the patient and / or configuration of the X-ray system may lead to erroneous measurements and potential misdiagnosis.

[0004] Mobile chest radiography is assumed to contribute to a large share of all chest X-ray images. In ICU and ED settings, mobile imaging is often employed to verify the correct positioning of lines and tubes such as central venous catheters (CVC) or endotracheal tubes (ETT). Mispositioning of these devices can lead to serious adverse events with potentially life-threatening consequences, such as pneumothorax.

[0005] Therefore, routine confirmation via X-ray imaging is often employed. As an example, for the assessment of ETT position, the distance between the tip of the ETT and the carina bifurcation is often evaluated, a measurement that can be severely affected by projection errors and distortions. Furthermore, as mentioned above, such imaging is often one of the first modalities to resolve and follow chest pathologies. If the radiographs do not conform to standardized projections, these pathologies may be obscured or misinterpreted.

[0006] Mobile X-ray imaging presents challenges due to adverse patient conditions, ICU / ED room conditions, and / or the equipment involved. Therefore, there may be quality issues due to over- or under-exposure, and there may be quality issues due to positioning or rotation of the patient relative to the imaging system, which is common in X-ray imaging, especially mobile imaging.

[0007] Ideally, the median coronal plane of the detector should be parallel to the chest and perpendicular to the central ray of the X-ray system. In the above conditions, such a setup is usually difficult to achieve.

[0008] Deviations can result in head or tail angulation, which not only affects the appearance of anatomical structures in the image, but can also affect the reliability of measurements, AI results, and the assessment of pathology. Angulation is the most frequent quality problem in X-ray imaging.

[0009] As can be seen from the above, sometimes it is difficult or impossible to obtain ideal X-ray imaging conditions, especially for mobile X-ray imaging such as in an ICU or ED room. As a result, image quality may be lacking and patients may be at risk. Summary of the invention

[0010] It is an object of the present invention to allow reducing the negative impact of poor image quality and potential risks to the patient.

[0011] The present invention provides a computer-implemented method for X-ray image quality assurance. The method includes: using a mobile X-ray imaging system to acquire X-ray image data of an object, the X-ray image data including at least one X-ray image; performing image-based automatic detection of imaging errors indicating misalignment of the X-ray imaging system relative to the object based on the X-ray image data; and in response to detecting the imaging errors indicating misalignment, causing a display device to display a user interface, the user interface including an indicator indicating the presence of misalignment and / or an indicator indicating the type of misalignment and / or an indicator indicating the degree of misalignment.

[0012] Thus, the negative impact of poor X-ray image quality may be mitigated by detecting such quality issues, which are particularly relevant, for example, for mobile chest X-ray imaging, and alerting medical personnel, ensuring appropriate use of the X-ray images and ensuring that appropriate actions can be initiated.

[0013] A mobile X-ray imaging system may be understood as an X-ray imaging system, which is mobile in its entirety, in particular without disassembly. In other words, a mobile X-ray imaging system may be understood as an X-ray imaging system, which is configured such that it can be repositioned in its entirety, in particular without disassembly.

[0014] As can be seen from the above, the method involves image-based detection of projection errors, which arise, for example, from patient positioning or distortions, which arise, for example, from an increased distance between the source and the patient in the X-ray acquisition. To this end, different techniques described in more detail below can be employed, including, for example, AI methods for image classification and / or regression and based on segmentation and landmark detection. The method can also provide a user interface that alerts users (e.g., radiologists and ICU / ED staff) to quality issues (such as the presence and / or degree of angulation and / or misalignment between the patient and the X-ray system), thereby allowing appropriate action and / or interpretation of the X-ray images.

[0015] The method of the present disclosure can be used in an X-ray imaging system, for example, for early detection of misaligned devices, as well as parts of a PACS viewer or diagnostic workstation. Alternatively or additionally, the method can be used to notify radiologists and ICU / ED staff of quality issues of X-ray images that may affect diagnostic evaluation, i.e., as a result of distortion and / or angulation. This allows the radiologist or staff to take action, such as ignoring the image and / or triggering or performing image correction and / or acquiring a new image.

[0016] A subject may be a human and is also referred to as a patient in this disclosure.

[0017] Imaging errors manifest themselves in different aspects of the X-ray image. As examples, signs of angulation may include the relationship of the manubrium to the fourth thoracic vertebra (T4), the position of the clavicle relative to the spine, and the shape of the ribs.

[0018] According to the present disclosure, performing image-based automatic detection of imaging errors may include utilizing deep learning methods, in particular utilizing a first deep learning method for image classification-based detection of misalignment and / or utilizing a second deep learning method for landmark detection-based detection of misalignment.

[0019] This allows reliable detection while also being non-intrusive and without the need for the simultaneous use of additional detection modules for the monitored object. In particular, the use of radio tags can be avoided.

[0020] Given the wide availability of data on anatomical structures and the relatively well-defined problem, deep learning methods are well suited and capable of image-based automatic detection of imaging errors. In the case of image classification-based detection, as an example, deep learning methods may involve learning based on real and / or artificially created X-ray images labeled with respect to the presence and / or type and / or degree of misalignment. The X-ray images may then be classified or identified as including misalignment and optionally including a specific type or degree of misalignment.

[0021] In the case of landmark detection based detection of misalignment, a deep learning approach can be used to identify predetermined landmarks in the X-ray image, and based on the available anatomical data, it can be automatically determined that the landmark has a relative position in the X-ray image that deviates from the expected position and this may indicate a misalignment. More details on this will be provided below.

[0022] According to the present disclosure, performing image-based automatic detection of imaging errors may include performing image classification-based detection of misalignments using a first neural network, the first neural network being trained using real and / or artificially created X-ray images marked with respect to said presence and / or said type and / or said extent of misalignment. The first neural network may for example be a convolutional neural network, in particular a DenseNet or a ResNet. However, other methods of machine learning are conceivable. The convolutional neural network may be a deep convolutional neural network.

[0023] The real X-ray images included in the training data used to train the first neural network can be X-ray images obtained at a plurality of different and known alignments or misalignments of the object relative to one or more components of the X-ray system. The artificial X-ray images that can be used and how they can be used will be described in more detail below.

[0024] An advantage of classification-based detection is that it is robust and fast, providing the required amount of accuracy for many applications. In particular, where precise measurements in X-ray images are not required, the accuracy requirements are not particularly high.

[0025] An example of a classification would be a statement good / average / bad on the image quality, which could therefore be an indication for a retake of the image, which could be output, for example, to a technician.

[0026] According to the present disclosure, image classification based detection may include: a first neural network performing image classification and / or regression on X-ray image data of the object to detect imaging errors, and in particular identify the presence and / or type and / or degree of misalignment.

[0027] Image classification and / or regression may be performed in order to determine the misalignment, and in particular the type and / or degree of misalignment, leading to one or more imaging errors that are most similar to one or more imaging errors in the X-ray image.

[0028] Image classification and regression can be used individually or in combination.

[0029] As mentioned above, an example for classification would be good / average / bad statements on image quality. For example, a regression could return a continuous score for the amount (degree) of rotation.

[0030] Methods known in the art can be used for image classification and / or regression.

[0031] According to the present disclosure, performing automatic image-based detection of imaging errors may include: using a second neural network to detect imaging errors based on landmark detection, the second neural network is configured to detect predetermined landmarks in the X-ray image, the landmarks being suitable for deriving imaging errors from the arrangement of the landmarks in the X-ray image, in particular, wherein the second neural network is a convolutional neural network for image segmentation, such as UNet. However, other methods of machine learning are conceivable.

[0032] In the present disclosure, a landmark may be any anatomical structure detectable by X-rays. For example, a landmark may be part of the surface of an anatomical structure. They may be determined, for example, based on the surface contour and / or shape of the depicted anatomical structure. Anatomical maps and / or meshes modeling the surface of anatomical features may be used in this context.

[0033] Based on the landmark positions in the X-ray image and the expected landmark positions, the imaging error can be derived. Landmark-based detection allows the type and / or degree of misalignment to be determined very accurately, and may also allow correction of quantitative measurements made in the X-ray image, such as distances and / or angles.

[0034] According to the present disclosure, landmark detection-based detection of imaging errors may include detecting predetermined landmarks in an X-ray image by a second neural network, determining an arrangement, in particular a distance, between the detected landmarks in the X-ray image, and based thereon detecting imaging errors, and in particular the presence and / or type and / or degree of misalignment.

[0035] For example, when the arrangement of landmarks in the X-ray image deviates from the expected arrangement, it can be determined that there is a misalignment. In addition, the arrangement can be used to determine the degree of misalignment, and based on this, a corrected distance can be optionally determined to correct for imaging errors caused by misalignment.

[0036] The predetermined landmarks may be landmarks with fixed relative positions, for example landmarks that will not move significantly relative to each other. Their arrangement, in particular the distance, may therefore be a particularly reliable indicator for imaging errors.

[0037] According to the present disclosure, detecting imaging errors indicating misalignment can be performed using artificially created X-ray images. In particular, the artificial X-ray images can be images obtained by applying a plurality of different parameterizations to digitally reconstructed radiographs reconstructed from 3D medical image data (e.g., CT image data) to obtain a plurality of artificial X-ray images with imaging errors indicating different misalignments.

[0038] For example, digitally reconstructed radiographs (DRRs) can be obtained from CT image data to generate artificial X-rays. Using artificial X-rays, by different parameterizations of the DRRs, artificial X-ray images with precise information about misalignment (e.g., level of angulation and / or distance) can be created.

[0039] An advantage of using artificially created X-ray images in the manner described above is that it allows the creation of very large and comprehensive datasets, for example as training data for machine learning. Higher accuracy can also be achieved taking into account that real X-ray image data may have a wide variety of imaging errors and not all of these imaging errors may have been accurately identified and labeled in the dataset.

[0040] According to the present disclosure, the first neural network and / or the second neural network may be trained based on artificially created X-ray images. In particular, the first neural network may be trained using at least artificially created X-ray images marked with respect to the presence and / or type and / or degree of misalignment.

[0041] The use of artificial X-ray images may allow for a large, comprehensive, well-structured training data set that may include different types of misalignments, either alone or in combination, and include different degrees of misalignment. Thus, reliable detection and identification of misalignments may be provided. This may be particularly suitable for classification algorithms.

[0042] According to the present disclosure, using the artificially created X-ray image to detect imaging errors indicative of misalignment may include using the artificially created X-ray image to automatically determine corrected landmarks.

[0043] The artificially created X-ray image may be an X-ray image as described above, but in addition to or instead of being marked as having misalignment, the artificially created X-ray image in this case may also include marked landmarks. Based on this, landmarks detected in the X-ray image may be corrected, for example, by using a machine learning method trained using the marked artificially created X-ray images as training data. The marked artificially created X-ray image may also be used to identify landmarks in the X-ray image, for example, by using a machine learning method trained using the marked artificially created X-ray images as training data.

[0044] Thus, in addition to detecting misalignment, the data may also be corrected to some extent. As an example, the corrected landmarks may be displayed on a display device overlaid with the x-ray image. Optionally, the detected (uncorrected) landmarks may also be displayed on the display device. The distances to which the corresponding positions of the landmarks have been corrected may also optionally be displayed on the display device.

[0045] An advantage is that correct landmark location may be sufficient in many use cases, making it unnecessary to perform more complex, time-consuming and / or resource-intensive methods.

[0046] According to the present disclosure, using an artificially created X-ray image to detect imaging errors indicative of misalignment may include using the artificially created X-ray image to automatically determine a misalignment-induced error or uncertainty margin for measurements made based on the X-ray image.

[0047] For example, where the distance and / or relative orientation between points on an X-ray image (e.g., landmarks depicted therein) are measured in the X-ray image, the actual distance and / or orientation may be different depending on the degree and type of misalignment. The error may quantify the deviation of the measured value from the expected actual value. The error may be corrected and / or taken into account, and in the case where the exact error is unknown or not known with the desired accuracy, an uncertainty margin may be provided that represents an estimated limit on the accuracy of the measurement result in the X-ray image due to the misalignment. Thus, it may be determined, for example, whether the X-ray image is suitable for quantitative assessment and to what extent.

[0048] According to the present disclosure, using an artificially created X-ray image to detect imaging errors indicative of misalignment may include using the artificially created X-ray image to automatically determine corrected measurements by correcting misalignment-induced errors in measurements made based on the X-ray image.

[0049] The artificially created X-ray images may be the artificially created X-ray images described above, and the advantages and features described in that context apply analogously in this context. Using them to correct alignment-induced errors in the measurement results may be particularly advantageous, because the effects of misalignment can be well identified and isolated from other misalignments and / or other imaging errors, so that the correction will produce highly accurate results.

[0050] The method of the present disclosure may include determining the position of predetermined landmarks in one or more of the 3D medical image data and the manually created X-ray images, and based thereon determining imaging errors and / or determining corrected landmark and / or misalignment induced error or uncertainty margins and / or corrected measurement results.

[0051] The method allows the above-mentioned general advantages of artificially created X-ray images to be exploited. In addition, 3D medical image data are particularly suitable for using landmark-based methods because, unlike X-ray images, landmarks can be determined using 3D data (e.g., surface data of anatomical structures) and their positions can be determined in 3D coordinates. This allows high-quality predictions of expected landmark positions for a given type and / or degree of misalignment. Therefore, high accuracy can be ensured.

[0052] The method of the present disclosure may include training a first neural network / said first neural network using real and / or artificially created X-ray images marked as to the presence and / or type and / or extent of misalignment.

[0053] The features and advantages outlined above in the context of the first neural network and in particular in the context of using labeled X-ray images (real and / or artificial) apply analogously in this context.

[0054] The method of the present disclosure may comprise training the second neural network / the second neural network to detect predetermined landmarks in the X-ray image, in particular by image segmentation.

[0055] The features and advantages outlined above in the context of the second neural network and in particular in the context of detecting and using predetermined landmarks apply analogously in this context.

[0056] The method of the present disclosure may include obtaining multiple artificial X-ray images / the multiple artificial X-ray images, in particular, applying multiple different parameterizations to digitally reconstructed radiographs reconstructed based on 3D medical image data (e.g., CT image data) so as to obtain multiple artificial X-ray images / the multiple artificial X-ray images having imaging errors representing different misalignments.

[0057] The features and advantages outlined above in the context of artificial X-ray images apply analogously in this context.

[0058] According to the present disclosure, misalignment may include angular and / or distance deviations from the target alignment of the X-ray system relative to the object. In other words, there may be deviations from the relative target orientation, such as expressed by deviations from one or more target angles between the object and one or more components of the X-ray. Alternatively or additionally, there may be deviations from the target distance between the object and one or more components of the X-ray system. These types of misalignments lead to imaging errors, which are particularly relevant and can be well addressed by the present disclosure, namely, projection errors and distortions.

[0059] According to the present disclosure, imaging errors may include projection errors and / or distortions. For example, projection errors may be caused by a non-ideal relative orientation (e.g., angulation) of an object and an X-ray system during X-ray acquisition. For example, distortion may be caused by a deviation from an ideal distance between an X-ray source and an object during X-ray acquisition (i.e., distance deviation).

[0060] According to the present disclosure, an X-ray image may depict a chest of a subject. In particular, the X-ray image may depict at least one of the following: a sternum, in particular a manubrium sternum, vertebrae, in particular at least one of the following: a thoracic vertebra, such as the fourth thoracic vertebra (T4), a clavicle, and one or more ribs. Chest X-rays are typically obtained by mobile X-ray applications, for example, for use in an ICU or ED setting. These types of X-ray images require special attention and quality control compared to simple diagnostic imaging, for example, for detecting a broken limb, as low quality may have significant adverse effects. Therefore, the claimed method will be particularly relevant in this context.

[0061] According to the present disclosure, the subject may be an immobile person, in particular a person in a recumbent position. Therefore, it may be necessary to arrange the X-ray system in a non-ideal orientation to accommodate an immobile person who is unable to move into a position that would allow a better orientation of the X-ray system. This increases the risk of quality issues, and therefore the claimed method will be particularly relevant in this context.

[0062] According to the present disclosure, the method is performed on objects in ICU and / or ED settings. Objects in intensive care unit (ICU) or emergency department (ED) settings are often immobile and / or difficult to position, especially when they may have monitoring and / or treatment machines attached to them. In addition, their condition may not be robust. Therefore, it may be necessary to arrange the X-ray system in a non-ideal orientation to accommodate objects in ICU or ED settings. This increases the risk of quality issues, and therefore, the claimed method will have particular relevance in these settings.

[0063] The present invention also provides a system for X-ray quality assurance. The system includes a processing system configured to execute and / or control any method steps described in the present disclosure, in particular, to execute the calculation or data processing steps of the present disclosure and / or control data acquisition, for example, by controlling an X-ray imaging system / the X-ray imaging system to acquire X-ray image data and / or controlling a CT imaging system / the CT imaging system to acquire CT image data, and / or controlling a display device to display a user interface.

[0064] A system according to the present disclosure may include a mobile X-ray imaging system configured to acquire X-ray image data and optionally provide the X-ray image data to a processing system.

[0065] Alternatively or additionally, the system according to the present disclosure may include a display device configured to display a user interface.

[0066] Alternatively or additionally, a system according to the present disclosure may include a CT imaging system for obtaining 3D medical image data.

[0067] Alternatively or additionally, a system according to the present disclosure may include one or more instruments, in particular instruments to be inserted into an object and depicted in an X-ray image. For example, an instrument may include a tube and / or a probe.

[0068] The present invention also provides a computer program product comprising instructions, which, when the program is run by a computer, cause the computer to perform and / or control any method steps described in the present disclosure.

[0069] The present invention also provides a computer-readable medium comprising instructions, which when executed by a computer cause the computer to perform and / or control any method steps described in the present disclosure.

[0070] The features and advantages outlined in the context of the method for quality assurance similarly apply to the system, computer program product, and computer-readable medium of the present disclosure.

[0071] Additional features, examples and advantages will become apparent from the detailed description made with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 shows a schematic diagram of a system according to the present disclosure;

[0073] Figure 2 A flow chart representing a method according to the present disclosure is shown; and

[0074] Figures 3a to 3c Diagram showing detection techniques.

[0075] List of reference numerals:

[0076] System 1

[0077] Data processing system 2

[0078] X-ray imaging system 3

[0079] X-ray source 3a

[0080] X-ray detector 3b

[0081] Display Device 4

[0082] User Interface 5

[0083] Indicator 5a

[0084] Image 5b

[0085] CT imaging system6

[0086] Data connection 7

[0087] Patient 8

[0088] Listing 8a

[0089] Landmark 9

[0090] Instrument 10 DETAILED DESCRIPTION

[0091] Figure 1 A system 1 according to the present disclosure is illustrated. The system comprises a data processing system 2 and an X-ray imaging system 3 having an X-ray source 3a and an X-ray detector 3b.

[0092] The system further comprises a display device 4. The display device is configured to display a user interface 5. The user interface 5 in the present case outputs an indicator 5a and an X-ray image 5b.

[0093] The system may optionally include a CT imaging system 6 and a data connection 7 connecting one or more of the other elements of the system.

[0094] For illustration purposes, Figure 1 Also shown is a subject, for example a patient 8, whose chest 8a is shown in the X-ray image. The subject is shown in a recumbent position and may be, for example, in an ICU setting.

[0095] In this example, the X-ray source is arranged above the object and the X-ray detector is arranged below the object.

[0096] Figure 1 An example is shown in which a landmark 9 and an optional instrument 10 are depicted, which need not be part of the system.

[0097] In the following, reference Figure 2 , describes a computer-implemented method according to the present disclosure, which can be performed, for example, by a system according to the present disclosure, specifically, by the computer-implemented method described above and in Figure 1 As shown in or performed by the system claimed.

[0098] In step S1 , X-ray image data of an object is acquired using a mobile X-ray imaging system, where the X-ray image data includes at least one X-ray image.

[0099] In step S2, based on the X-ray image data, an image-based automatic detection of imaging errors indicative of a misalignment of the X-ray imaging system relative to the object is performed.

[0100] In step S3 , in response to detecting an imaging error indicating misalignment, causing the display device to display a user interface including an indicator indicating the presence of misalignment and / or an indicator indicating the type of misalignment and / or an indicator indicating the extent of misalignment.

[0101] Optional steps S11 and S12 may be performed before, during or after step S1.

[0102] In optional step S11 , a plurality of artificial X-ray images with imaging errors representing different misalignments are obtained, which in this example are obtained by applying a plurality of different parameterizations to digitally reconstructed radiographs reconstructed from the CT image data.

[0103] In an optional step S12 following step S11 , positions of predetermined landmarks are determined in one or more of the CT image data and the manually created X-ray image.

[0104] Based on this, in an optional step S13 , imaging errors and / or corrected landmarks and / or misalignment induced errors or uncertainty margins and / or corrected measurement results are detected in the X-ray image data.

[0105] In optional step S13, a display device is facilitated to display a user interface, the user interface comprising a visual representation of at least one of the following: detected imaging errors detected in the X-ray image data and / or corrected landmarks and / or misalignment-induced errors or uncertainty margins and / or corrected measurement results. The user interface in this step may be the user interface of step S3.

[0106] In an optional step S21, a first neural network is trained using real and / or artificially created X-ray images marked with respect to the presence and / or type and / or extent of misalignment. The artificial image data may be those obtained in step S11. In an optional step S22, instead of or in addition to step S21, a second neural network is trained to detect predetermined landmarks in the X-ray images, in particular by image segmentation. The landmarks obtained in step S12 may be used to train the second neural network. Steps S21 and S22 may be performed prior to the image-based automatic detection of imaging errors. Their respective outputs may be used as input for the image-based automatic detection of imaging errors.

[0107] According to the present disclosure, an image classification / regression-based method can be used to detect imaging errors. That is, the identification of imaging errors can be formulated as an image classification or regression problem. For example, deep learning methods and specifically deep convolutional neural networks can be used to solve this problem.

[0108] For example, given a dataset of X-ray images labeled, for example, with respect to the presence of angulation, a network (e.g., DenseNet, ResNet, etc.) can be trained to distinguish between classes, for example, normal alignment and different classes of misalignment, in particular angulation. Additionally or alternatively, a label or annotation indicating the severity of the misalignment (in particular angulation) can be provided, for example, as a continuous score. The network can then be trained to predict the presence, type, and optionally the extent of an imaging error (e.g., a projection error). The classification / regression approach can be viewed as providing an end-to-end solution for detecting imaging errors, and it can be considered a black-box approach.

[0109] Examples are Figure 3a to Figure 3c FIG3 illustrates a regression algorithm for automatically determining the amount of rotation. Figure 3b An automatic landmark detection algorithm for rule-based angulation assessment is illustrated. Figure 3c Prediction of landmark detection for adjusted position based on angulation correction is illustrated.

[0110] Figure 3a Determining an angulation score and outputting a warning stating that strong head angulation exists is illustrated only as an example. However, instead, only the presence of angulation may be detected, and the user interface may not indicate the type and degree of angulation. Various other combinations are contemplated. Although Figure 3a Refers to a CNN for distortion regression, but this does not necessarily have to be used.

[0111] In other examples, landmark-based detection may be employed in accordance with the present disclosure. For example, to reflect a radiologist's assessment of an image, a landmark-based approach is applied. Using a deep learning approach for landmark detection / segmentation (e.g., UNet as exemplarily shown in FIG. 3 ), structures like the clavicle, manubrium, and T4 may be located in the image. The (particularly signed) distances between structures in the image give a direct indication of the extent of the projection error. Thus, based on landmark evaluations, such as Figure 3b As shown, a warning that an imaging error exists, and optionally its type and extent, in this case strong head angulation, may be displayed as part of the user interface.

[0112] Since image data and corresponding annotations are often difficult to obtain, data simulation strategies can be employed according to the present disclosure. For example, using computed tomography (CT) images of a patient, digitally reconstructed radiographs (DRRs) can be used to generate artificial X-rays. Using artificial X-rays, with the aid of different parameterizations of DRRs, any number of artificial X-ray images with accurate information about angulation / distance levels can be created. Based on this, a network (e.g., the above-described network) can be trained to detect the presence, type, and / or degree of imaging errors.

[0113] In addition to the locations of landmarks in the artificial X-ray image, the locations of landmarks in the CT may be available, and thus accurate measurements may be possible. Figure 3c An example is shown in . As an example, a network (eg, UNet) is used for landmark detection and can predict corrected landmarks and optionally can also provide absolute values ​​of distances in the image, particularly based on landmarks in the CT image data.

[0114] According to the present disclosure, regression techniques may be used to directly predict accurate measurements from images and / or predict uncertainty margins (regarding image-based measurements) and / or segmentation techniques may be used to predict landmarks that allow for projection error correction for accurate measurements.

[0115] Based on the results of the previously described techniques / algorithms, visual feedback may be given to the user of the system, as described above. For example, depending on the type of result, this may take the form of a binary indicator (e.g., "red light" / "green light"), which indicates that the image is suitable for evaluation. Alternatively or additionally, an indicator having more than two values ​​(e.g., a continuous indicator) may be displayed to the user.

[0116] Although the present invention has been illustrated and described in detail in the drawings and the foregoing description, such illustration and description should be considered as illustrative rather than restrictive. The present invention is not limited to the disclosed embodiments. In view of the foregoing description and the drawings, it will be apparent to those skilled in the art that various modifications may be made within the scope of the present invention as defined by the claims.

[0117] Examples of the Present Disclosure

[0118] 1. A computer-implemented method for X-ray quality assurance, the method comprising:

[0119] Using a mobile X-ray imaging system (3) to acquire (S1) X-ray image data of an object (8), the X-ray image data comprising at least one X-ray image;

[0120] performing (S2) image-based automatic detection of imaging errors indicative of misalignment of the X-ray imaging system (3) relative to the object (8) based on the X-ray image data; and

[0121] In response to detecting an imaging error indicating misalignment, causing (S3) a display device (4) to display a user interface (5), the user interface comprising an indicator indicating the presence of the misalignment and / or an indicator indicating the type of the misalignment and / or an indicator indicating the extent of the misalignment.

[0122] 2. A method according to Example 1, wherein performing the image-based automatic detection of imaging errors includes utilizing a deep learning method, in particular utilizing a first deep learning method for performing image classification-based detection of misalignment and / or utilizing a second deep learning method for performing landmark detection-based detection of misalignment.

[0123] 3. The method of example 1 or claim 2, wherein performing the image-based automatic detection of imaging errors comprises performing image classification-based detection of misalignment using a first neural network trained using real and / or artificially created X-ray images marked with respect to the presence and / or the type and / or the extent of misalignment, in particular, wherein the first neural network is a convolutional neural network.

[0124] 4. A method according to Example 3, wherein the image classification-based detection includes: the first neural network performs image classification and / or regression on the X-ray image data of the object to detect the imaging error, and in particular to identify the existence and / or the type and / or the degree of misalignment.

[0125] 5. A method according to any of the preceding examples, wherein performing the image-based automatic detection of imaging errors comprises: landmark-based detection of imaging errors using a second neural network, the second neural network being configured to detect predetermined landmarks (9) in the X-ray image, the landmarks being suitable for deriving imaging errors from the arrangement of the landmarks in the X-ray image, in particular, wherein the second neural network is a convolutional neural network for image segmentation.

[0126] 6. A method according to Example 5, wherein the landmark detection-based detection of imaging errors comprises: detecting the predetermined landmarks (9) in the X-ray image by means of the second neural network, determining the arrangement of the detected landmarks, and in particular the distances between the detected landmarks, and based thereon detecting the imaging errors, and in particular the presence and / or type and / or degree of misalignment.

[0127] 7. The method according to any one of the preceding examples,

[0128] wherein detecting imaging errors indicative of misalignment is performed using artificially created X-ray images,

[0129] In particular, the artificial X-ray image is an image obtained by applying multiple different parameterizations to digitally reconstructed radiographs reconstructed based on 3D medical image data so as to obtain multiple artificial X-ray images with imaging errors representing different misalignments, wherein the 3D medical image data is, for example, CT image data.

[0130] 8. The method according to Example 7,

[0131] wherein the first neural network and / or the second neural network is trained based on said artificially created X-ray images, in particular wherein said first neural network is trained using at least said artificially created X-ray images marked with respect to said presence and / or said type and / or said extent of misalignment; and / or

[0132] Wherein, using the artificially created X-ray image to detect imaging errors indicative of misalignment comprises: using the artificially created X-ray image to automatically determine corrected landmarks; and / or

[0133] Wherein, using the artificially created X-ray image to detect imaging errors indicative of misalignment comprises: using the artificially created X-ray image to automatically determine a misalignment-induced error or uncertainty margin of a measurement result made based on the X-ray image; and / or

[0134] Wherein, using the artificially created X-ray image to detect imaging errors indicative of misalignment includes: using the artificially created X-ray image to automatically determine a corrected measurement result by correcting the misalignment-induced errors of the measurement results made based on the X-ray image.

[0135] 9. A method according to Example 7 or Example 8, wherein the method includes: determining (S12) the position of a predetermined landmark in the 3D medical image data and in one or more of the artificially created X-ray images; and detecting (S13) an imaging error and / or determining the corrected landmark and / or the misalignment-induced error or uncertainty margin and / or the corrected measurement result based on the position.

[0136] 10. The method according to any of the preceding examples, comprising:

[0137] training (S21) a first neural network / said first neural network using real and / or artificially created X-ray images marked with respect to said presence and / or said type and / or said extent of misalignment; and / or

[0138] Training (S22) the second neural network / the second neural network to detect predetermined landmarks in the X-ray image, in particular training (S22) the second neural network / the second neural network to detect predetermined landmarks in the X-ray image by image segmentation; and / or

[0139] Obtain (S11) multiple artificial X-ray images / the multiple artificial X-ray images, in particular, apply multiple different parameters to digitally reconstructed radiographs reconstructed based on 3D medical image data (for example, CT image data) so as to obtain multiple artificial X-ray images / the multiple artificial X-ray images with imaging errors representing different misalignments.

[0140] 11. The method according to any one of the preceding examples,

[0141] wherein the misalignment comprises an angular and / or distance deviation from a target alignment of the X-ray system (3) relative to the object (8), and / or

[0142] Wherein, the imaging error includes projection error and / or distortion; and / or

[0143] wherein the X-ray image depicts a chest (8a) of the subject (8), and / or

[0144] wherein the subject (8) is a stationary person, in particular a person in a recumbent position, and / or

[0145] Therein, the method is performed on a subject (8) in an ICU and / or ED setting.

[0146] 12. A system (1) for X-ray quality assurance, the system comprising:

[0147] A processing system (2) configured to perform and / or control the steps of the method described in any one of Examples 1 to 11,

[0148] The system (1) particularly comprises:

[0149] a mobile X-ray imaging system (3) configured to acquire said X-ray image data and optionally provide said X-ray image data to said processing system; and / or

[0150] a display device (4) configured to display the user interface (5); and / or

[0151] A CT imaging system (6) for obtaining the 3D medical image data; and / or

[0152] One or more instruments (10), in particular instruments to be inserted into the object (8) and depicted in the X-ray image.

[0153] 13. A computer program product comprising instructions which, when said program is run by a computer, cause said computer to perform and / or control the steps of the method according to any one of examples 1 to 11.

[0154] 14. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to perform and / or control the steps of the method according to any one of examples 1 to 11.

Claims

1. A computer-implemented method for X-ray quality assurance, the method comprising: Using a mobile X-ray imaging system (3) to acquire (S1) X-ray image data of an object (8), the X-ray image data comprising at least one X-ray image; performing (S2) image-based automatic detection of imaging errors indicative of misalignment of the X-ray imaging system (3) relative to the object (8) based on the X-ray image data; as well as In response to detecting an imaging error indicating misalignment, causing (S3) a display device (4) to display a user interface (5), the user interface comprising an indicator indicating the presence of the misalignment and / or an indicator indicating the type of the misalignment and / or an indicator indicating the extent of the misalignment.

2. The method according to claim 1, wherein: Performing the image-based automatic detection of imaging errors includes utilizing deep learning methods.

3. The method according to claim 1, wherein: Performing the image-based automatic detection of imaging errors includes utilizing a first deep learning method for image classification-based detection of misalignment and / or utilizing a second deep learning method for landmark detection-based detection of misalignment.

4. A method according to any one of the preceding claims, wherein: Performing said image-based automatic detection of imaging errors comprises: performing image classification-based detection of misalignment using a first neural network, wherein said first neural network is trained using real and / or artificially created X-ray images marked with respect to said presence and / or said type and / or said extent of misalignment, in particular, wherein the first neural network is a convolutional neural network.

5. The method according to claim 4, wherein: The image classification-based detection includes: the first neural network performing image classification and / or regression on the X-ray image data of the object to detect the imaging error.

6. The method according to claim 4, wherein: The image classification based detection comprises: the first neural network performing image classification and / or regression on the X-ray image data of the object to detect the imaging error, thereby identifying the presence and / or the type and / or the extent of misalignment.

7. A method according to any one of the preceding claims, wherein: Performing the image-based automatic detection of imaging errors comprises: landmark detection-based detection of imaging errors using a second neural network, the second neural network being configured to detect predetermined landmarks (9) in the X-ray image, the landmarks being suitable for deriving imaging errors from an arrangement of the landmarks in the X-ray image, in particular, wherein the second neural network is a convolutional neural network for image segmentation.

8. The method according to claim 7, wherein: The landmark detection-based detection of imaging errors comprises: detecting the predetermined landmarks (9) in the X-ray image by the second neural network, determining the arrangement of the detected landmarks, and detecting the imaging errors based on the arrangement.

9. The method according to claim 8, wherein: Determining the arrangement of the detected landmarks comprises determining distances between the detected landmarks, and / or wherein, The presence and / or the type and / or the extent of misalignment is determined based on the arrangement of the detected landmarks.

10. The method according to any one of the preceding claims, wherein: Detection of imaging errors indicative of misalignment is performed using artificially created X-ray images.

11. The method according to claim 10, wherein: The artificially created X-ray image is an image obtained by applying a plurality of different parameterizations to a digitally reconstructed radiograph reconstructed from 3D medical image data, such as CT image data, so as to obtain a plurality of artificial X-ray images with imaging errors representing different misalignments.

12. The method according to claim 10 or 11, in, The first neural network / the first neural network and / or the second neural network / the second neural network are trained based on the artificially created X-ray images.

13. The method according to claim 12, wherein: Said first neural network is trained using at least said artificially created X-ray images marked with respect to said presence and / or said type and / or said extent of misalignment.

14. The method according to any one of claims 10 to 13, wherein: Using an artificially created X-ray image to detect imaging errors indicating misalignment includes: using the artificially created X-ray image to automatically determine at least one of the following: a corrected landmark; an error or uncertainty margin caused by misalignment of a measurement result made based on the X-ray image; and a corrected measurement result obtained by correcting the error caused by misalignment of the measurement result made based on the X-ray image.

15. The method according to any one of claims 10 to 14, wherein: The method comprises: determining (S12) the position of a predetermined landmark in the 3D medical image data and in one or more of the artificially created X-ray images; and detecting (S13) an imaging error and / or determining the corrected landmark and / or the misalignment-induced error or uncertainty margin and / or the corrected measurement result based on the position.

16. A method according to any one of the preceding claims, comprising: A first neural network / said first neural network is trained (S21) using real and / or artificially created X-ray images marked with respect to said presence and / or said type and / or said extent of misalignment.

17. A method according to any one of the preceding claims, comprising: The second neural network / the second neural network is trained (S22) to detect predetermined landmarks in the X-ray image, in particular by image segmentation.

18. The method according to any one of the preceding claims, in, The misalignment comprises an angular and / or distance deviation from a target alignment of the X-ray system (3) relative to the object (8).

19. A method according to any one of the preceding claims, wherein: The imaging error includes projection error and / or distortion.

20. The method according to any one of the preceding claims, in, The subject (8) is a person who is immobile, in particular a person who is in a recumbent position, and / or Therein, the method is performed on a subject (8) in an intensive care unit (ICU) and / or emergency department (ED) setting.

21. A system (1) for X-ray quality assurance, the system comprising: A processing system (2) configured to perform and / or control the steps of the method according to any one of claims 1 to 20.

22. The system according to claim 21, comprising: a mobile X-ray imaging system (3) configured to acquire said X-ray image data and optionally provide said X-ray image data to said processing system; and / or a display device (4) configured to display the user interface (5); and / or A CT imaging system (6), which is used to obtain the 3D medical image data; and / or One or more instruments (10), in particular instruments to be inserted into the object (8) and depicted in the X-ray image.

23. A computer program product comprising instructions for causing a computer to perform and / or control the steps of the method according to any one of claims 1 to 20 when said program is run by the computer.

24. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to perform and / or control the steps of the method according to any one of claims 1 to 20.

Images