Method of providing result image data and providing unit and method of providing trained denoising function

Abstract

The invention relates to a computer-implemented method for providing result image data, comprising receiving pre-acquired control data having blood vessel information of an examination region of an examination object, acquiring image data of the examination region by a medical imaging device, providing the result image data by applying a denoising function to input data using the control data, wherein the input data is based on the image data, wherein the control data sets a plausibility condition for structures depicted in the result image data, wherein the result image data is denoised with respect to the image data. The invention further relates to a computer-implemented method for providing a trained denoising function, a providing unit, a system and a computer program product.

Description

Technical Field

[0001] The present invention relates to a computer-implemented method for providing resulting image data, a computer-implemented method for providing a trained denoising function, and providing units, systems, and computer program products. Background Technology

[0002] Image processing algorithms can be used for various purposes in medical imaging, particularly for image data segmentation and / or denoising. During segmentation, for example, objects within blood vessels, especially instruments and / or implants, can be identified in two-dimensional fluorescence fluoroscopy images. Denoising improves image quality by reducing interfering signals.

[0003] As the image quality of the input data deteriorates, for example due to the low radiation dose in X-ray images, the information content and reliability of the processed image data decrease. Structures hidden in noise can only be correctly detected and represented with a limited probability. This can lead to incorrect detection and representation.

[0004] When identifying medical objects, this could mean detecting an object in a location that is not actually present or may not be present. In the case of denoising algorithms, artificial structures may be incorrectly reconstructed as part of anatomical structures or medical objects.

[0005] To date, attempts have been made to optimize the balance between sensitive detection and avoiding false detection of relevant structures by adjusting algorithm parameters or training. Summary of the Invention

[0006] Therefore, the object of the present invention is to improve the imaging of the examination area of ​​the object being examined.

[0007] According to the present invention, this objective is achieved by a computer-implemented method for providing resulting image data, a computer-implemented method for providing a trained denoising function, and the provision of units, systems, and computer program products. Advantageous embodiments with suitable extensions are the subject of this application.

[0008] The solution according to the invention is described below in conjunction with the claimed system and the claimed method. Features, advantages, or alternative embodiments described herein may be attributed to other claimed subjects, and vice versa. In other words, the system claims and embodiments may be improved using features described or claimed in conjunction with the corresponding method. In this case, the functional features of the method are implemented by the physical units of the system.

[0009] Furthermore, the solutions according to the invention are described below with respect to methods and systems for providing resulting image data and methods and systems for providing trained denoising functions. Features, advantages, or alternative embodiments described herein may be attributed to other claimed subjects, and vice versa. In other words, embodiments can be improved to provide trained denoising functions having the features described or claimed in the context of providing resulting image data. Specifically, the data and / or datasets used in the methods and systems for providing resulting image data may have the same properties and characteristics as the corresponding data and / or datasets used in the methods and systems for providing trained denoising functions, and the trained denoising functions provided by the respective methods and systems can be used in the methods and systems for providing resulting image data.

[0010] In a first aspect, the present invention relates to a computer-implemented method for providing result image data. In a first step, pre-acquired control data containing vascular information of an examination area of ​​an object is received. In a further step, image data of the examination area is acquired using a medical imaging device. In a further step, result image data is provided by applying a denoising function to the input data using the control data. The input data is based on the image data. Furthermore, the control data sets credibility conditions for structures depicted in the result image data. Additionally, the result image data is denoised relative to the image data.

[0011] A computer-implemented method can be understood as a sequence of instructions executed by a computer to complete a specific task. This method can be written in a programming language and implemented on a computer system. The aforementioned method can be implemented at least partially, and especially entirely, by a computer.

[0012] The steps of the above method can be performed at least partially simultaneously or sequentially, especially in the order described.

[0013] Receiving pre-acquired control data may include acquiring and / or reading from a computer-readable data storage device and / or receiving it from a data storage unit (e.g., a database). Furthermore, the pre-acquired control data may be provided by one or more medical imaging devices, particularly providing units of the same or different imaging modalities.

[0014] Pre-acquired control data can be acquired at an earlier time point, particularly before the start of the method. Specifically, the pre-acquired control data can be used as input data in the method. This pre-acquired control data can come from, for example, a previous examination or another imaging procedure. The control data may include vascular information of the examination area of ​​the subject.

[0015] The subjects of examination may be, for example, human and / or animal patients and / or human and / or animal patients and / or examination phantoms.

[0016] The examination area may include the spatial region of the object to be examined, particularly the volume, especially the imaged object. The examination area of ​​the object may include, for example, the tissue and / or organ of interest of the object and / or medical object.

[0017] Vascular information may include spatially resolved information, particularly 2D or 3D spatially resolved information, regarding the location of at least one vascular segment (particularly arteries and / or veins) and / or vascular malformations within the examination area of ​​the subject. Advantageously, the vascular information may have a spatially resolved representation, particularly 2D or 3D, of at least one vascular segment and / or vascular malformation within the examination area of ​​the subject, particularly a model and / or image. Advantageously, the vascular information may characterize the spatial location (particularly position and / or orientation), size (particularly length and / or diameter), and / or spatial route of at least one vascular segment. Pre-acquired control data may, for example, include a model of at least one vascular segment, particularly a skeletal model, such as a centerline model and / or volumetric network model. Alternatively or additionally, the control data may include images of at least one vascular segment, particularly medical image data.

[0018] Acquiring image data may include using medical imaging equipment to acquire and / or receive image data. The image data may depict the area of ​​the object being examined in a spatially resolved manner, particularly in two dimensions (2D) or three dimensions (3D). For example, the image data may include 2D projected images of the examination area from one or more projection directions. These projected images may include, for example, X-ray or fluoroscopic images. Alternatively or additionally, the image data may include a three-dimensional volumetric dataset of the examination area. Furthermore, the image data may be resolved in a timely manner, with multiple images of the examination area acquired at different times.

[0019] Image data may include raw data directly provided by the detectors of a medical imaging device (e.g., an X-ray detector). Alternatively or additionally, the image data may have been preprocessed, for example, by reconstructing the raw data. Furthermore, image data may include metadata such as recording parameters, patient information, procedural information, and / or other relevant data.

[0020] Medical imaging equipment can be designed as, for example, magnetic resonance imaging (MRI) systems and / or computed tomography (CT) systems and / or medical X-ray equipment and / or positron emission tomography (PET) systems and / or ultrasound equipment.

[0021] The input data for the denoising function can advantageously be based on image data, particularly including image data. Furthermore, the resulting image data can be provided as the output data of the denoising function. Providing the resulting image data may include storing it on a computer-readable storage medium and / or displaying it on a display unit and / or sending it to a providing unit. In particular, a display unit can be used to display a graphical representation of the resulting image data.

[0022] Denoising functions can be designed to remove or reduce noise in image data. Denoising functions can be designed as classical functions, particularly for digital signal processing, or as trained functions.

[0023] If the denoising function is designed as a classical function, it can be based on filtering methods, such as Gaussian filters, median filters, and / or bilateral filters.

[0024] Noise in image data can be caused, for example, by technical components of the medical imaging equipment and / or physical effects in the imaging background. Alternatively or additionally, noise in image data can be caused by the interaction of imaging (e.g., X-rays and / or magnetic fields) with the object being examined and / or other objects and / or electromagnetic fields. Alternatively or additionally, noise in image data can be caused by the motion of the object being examined, particularly physiological motion.

[0025] Noise and / or denoising in image data can be quantified using metrics such as signal-to-noise ratio (SNR) and / or peak signal-to-noise ratio (PSNR). Furthermore, structural similarity indices (such as structural similarity index measurement (SSIM)) can be used to better capture perceived image quality than purely pixel-based metrics.

[0026] Control data specifies the credibility conditions for the structures depicted in the resulting image data. Credibility conditions can be understood as rules or criteria used to evaluate the probability and / or credibility of the structures depicted in the resulting image data.

[0027] The structures depicted in the image data and / or resulting image data may include a variety of anatomical and / or geometric features. Anatomical features may include, for example, tissues, particularly tissue boundaries, organs, hollow organs, particularly vascular portions, and / or anatomical landmarks, particularly orifices. Additionally, geometric features may include, for example, contours, contrast, edges, lines, and / or marked objects.

[0028] The vascular information included in the control data can be used to model the probability of certain structures appearing in the image data, especially in certain regions.

[0029] In particular, reliability conditions can be introduced as additional conditions into the application of denoising functions, especially during the optimization process. Denoising functions can be designed not only to reduce noise in the image data, but also to maximize the consistency between the structures depicted in the resulting image data and the anatomical structures (especially at least one vascular segment) specified in the control data.

[0030] Denoising functions can be designed to reduce noise in image data while preserving relevant structure. These functions can use vascular information contained in pre-acquired control data to assess the plausibility of structures in the resulting image data. Structures classified as unlikely based on confidence criteria can be more strongly denoised or removed from the resulting image data.

[0031] The resulting image data with noise reduction can in particular retain all the characteristics and attributes of image data, where the noise in the resulting image data has been reduced compared to the original image data.

[0032] Using control data during image denoising can advantageously improve the quality of the resulting image data. In particular, it allows for greater noise reduction while preserving relevant structures. This contributes to improved diagnostic and treatment planning.

[0033] In another advantageous embodiment of the proposed computer-implemented method, the denoising function can be designed as a training-only function. At least one parameter of the training function can be adjusted based on a comparison between the training result image data and the comparison image data.

[0034] A trained denoising function maps input data to output data. Specifically, the output data can also depend on one or more parameters of the trained denoising function. One or more parameters of the trained denoising function can be determined and / or adjusted through training. Determining and / or adjusting one or more parameters of the trained denoising function can be based, in particular, on a pair of training input data and associated training output data, especially comparing the output data, where the trained denoising function is applied to the training input data to generate training image data. Specifically, determination and / or adjustment can be based on a comparison of the training image data and the training output data, especially comparing the output data. Generally, a trainable denoising function, i.e., a denoising function whose one or more parameters have not yet been adjusted, can also be called a trained denoising function. By adjusting one or more parameters of the trained denoising function, especially by training the trained denoising function, the trained denoising function can be designed to adapt to new environments and recognize and extrapolate patterns.

[0035] Supervised learning, partially supervised learning, unsupervised learning, reinforcement learning, representation learning, and / or active learning can be used to adjust the trained denoising function. At least one parameter of the trained denoising function can be adjusted iteratively over several training steps.

[0036] Other terms for a trained denoising function are trained mapping rules, mapping rules with trained parameters, denoising functions with trained parameters, and machine learning algorithms. An example of a trained denoising function is an artificial neural network, where the edge weights of the artificial neural network correspond to the parameters of the trained denoising function. Specifically, a trained denoising function can include neural networks such as deep artificial neural networks and / or convolutional neural networks and / or deep convolutional networks and / or adversarial networks and / or deep adversarial networks and / or generative adversarial networks, "support vector machines," decision trees, and / or Bayesian networks. Alternatively or additionally, the trained denoising function can be based on k-means algorithms (k-means clustering), Q-learning, genetic algorithms, and / or association rules.

[0037] Trained denoising functions, particularly neural networks, advantageously have input and output layers. The input layer can be designed to receive input data. Furthermore, the output layer can be designed to provide image data, specifically output data. Both the input and / or output layers can include multiple channels, particularly neurons.

[0038] At least one parameter of the trained denoising function can be adjusted based on a comparison between the training result image data and the comparison image data. Specifically, the trained denoising function can be provided by embodiments of the proposed method for providing the trained denoising function, which will be described below.

[0039] By adjusting the parameters of this function based on comparisons between the training image data and the comparison image data, accurate and reliable noise reduction of the resulting image data can be achieved. This method ensures improved image quality, which is crucial for accurate analysis and diagnosis of medical examination areas.

[0040] In particular, trained denoising functions, especially neural networks, can provide the possibility of establishing a link between the location of a correctly interpreted medical object in image data and the location of blood vessels in control data during the training phase, without relying on model assumptions, such as the influence range of ships in control data.

[0041] In another advantageous embodiment of the proposed computer-implemented method, the image data may include at least one projected image of the examination area from a corresponding projection direction. For each of the at least one projected image, the control data may include a corresponding two-dimensional blood vessel representation of the examination area from the projection direction of the corresponding projected image.

[0042] At least one projected image may include a two-dimensional image of the three-dimensional examination area, which is generated by projecting the three-dimensional examination area onto a two-dimensional plane along a corresponding projection direction. In particular, at least one projected image of the examination area may include a 2D X-ray image or fluorescence fluoroscopic image of the examination area of ​​the object being examined. Medical imaging equipment can be used to acquire at least one projected image. Medical imaging equipment may be designed, for example, as X-ray equipment, particularly a C-arm X-ray device.

[0043] The projection direction, particularly the angle, can be defined as the spatial direction in which the projected image is recorded. The projection direction can be defined, for example, by the angle between the source (particularly an X-ray source) and the area of ​​the object being examined. Specifically, the image data can include multiple projected images that image the area of ​​the object being examined from at least partially, and particularly completely, different projection directions.

[0044] The 2D vascular representation can be designed as a planar representation of at least one vascular portion within the examination area. Advantageously, the control data for each of at least one projection image, and in particular each of multiple projection images, may include a 2D vascular representation of the examination area, the representation, and in particular modeling and / or imaging of the examination area, originating from the same, and in particular, the same projection orientation.

[0045] A 2D vascular representation can be designed as a two-dimensional image of at least one vascular portion of an examination area from the same projection direction as the corresponding projected image. The 2D vascular representation can be designed, for example, as digital subtraction angiography (DSA). Alternatively or additionally, the 2D vascular representation can have a 2D model of the examination area. Alternatively or additionally, the 2D vascular representation can be provided by a virtual projection of a 3D vascular representation along the corresponding projection direction of at least one projected image.

[0046] Using two-dimensional blood vessel representations as control data for each projected image offers the following advantages: the credibility conditions of the structures depicted in the resulting image data can be precisely customized according to the corresponding projection orientation. This improves noise reduction and allows for more accurate imaging of structures in the resulting image data.

[0047] In another advantageous embodiment of the proposed computer-implemented method, the control data may include a three-dimensional representation of blood vessels in the examination area.

[0048] 3D vascular representation can include a volumetric dataset representing, in particular, the spatial arrangement and structure of at least one vascular segment within a modeled and / or imaging examination area. 3D vascular representation can be recorded, for example, using computed tomography angiography (CTA), magnetic resonance angiography (MRA), cone-beam computed tomography angiography, or three-dimensional rotational angiography. In rotational angiography, a C-arm X-ray machine can rotate around the subject to acquire a large number of projected images from different angles, from which a three-dimensional dataset can then be reconstructed.

[0049] 3D vascular representation can advantageously provide detailed spatial information about the route, branches, and / or diameter of at least one vascular segment in the examination area.

[0050] Furthermore, 3D vascular representation can be used to generate corresponding 2D vascular representations for image data, particularly for each projection direction of at least one projected image, for example, through virtual projection along the corresponding projection direction. In this way, control data can be adjusted according to the geometry of the image data acquisition, particularly the geometry of the image data acquisition.

[0051] Using three-dimensional vascular representations as control data offers the technical advantage of enabling more accurate and robust denoising and recognition of structures in the resulting image data, as spatial information about vascular structures can be used more extensively than when using two-dimensional vascular representations.

[0052] In another advantageous embodiment of the proposed computer-implemented method, the control data may include a vascular representation of the examination region. The control data may include a subtraction dataset of the examination region. The subtraction dataset may be provided based on a subtraction of a mask dataset and a filling dataset. The mask dataset may be used to image the examination region during the masking stage. The filling data record may depict the examination region during the filling stage.

[0053] Vascular representation may include a spatially resolved representation, particularly 2D or 3D, of at least one vascular portion of the examination area of ​​the object being examined, especially an image and / or a model.

[0054] Advantageously, the same medical imaging equipment can be used to record both the mask dataset and the padding dataset.

[0055] A mask dataset can be used to image the examination area of ​​the subject during the masking phase. Specifically, a mask dataset can be recorded during the masking phase; during this phase, virtually no contrast agent can be placed in the examination area, particularly in at least one vascular segment. Furthermore, a fill data recording can depict the examination area of ​​the subject during the fill phase. Specifically, the fill data recording can be acquired during the fill phase. The masking phase and the fill phase can be specified as corresponding, particularly different, time periods. During the fill phase, contrast agent, particularly a radiopaque contrast agent, can be placed in the examination area, particularly in at least one vascular segment.

[0056] The subtraction dataset can be advantageously, particularly pixel-by-pixel, provided as the difference between the fill dataset and the mask dataset. Subtracting the mask dataset from the fill dataset can advantageously highlight at least one vascular portion filled with contrast agent within the examination area of ​​the subject while simultaneously suppressing background structures. The mask and fill datasets can be recorded, in particular, as part of digital subtraction angiography (DSA).

[0057] Using a subtractive dataset as part of the control data can be advantageous because it can provide a clear and accurate representation of vascular structures, particularly at least one vascular segment. This can improve the accuracy and reliability of the denoising function by providing a better foundation for the credibility conditions.

[0058] In another advantageous embodiment of the proposed computer-implemented method, the confidence condition can specify a higher probability of structures appearing near blood vessels in the examination area based on vascular information.

[0059] Confidence conditions can be designed as rules or standards that define the probability of certain structures appearing in the resulting image data based on the spatial relationship (e.g., spatial distance) between certain structures and vascular information included in the control data. Confidence conditions can be specified based on vascular information, particularly providing a spatial probability distribution and / or a spatial weighting function for the occurrence of structures near blood vessels in the examination area.

[0060] Advantageously, the credibility condition based on vascular information can specify a higher probability for structures in the image data that are spatially close to blood vessels, particularly at least one vascular segment. For example, a stent-like structure may be significantly less likely to be far from a vascular segment than it is close to one. Proximity to a blood vessel, particularly at least one vascular segment, can be defined as a spatial region located within a predetermined spatial distance from the control data, particularly the blood vessel represented in the vascular information, particularly at least one vascular segment. This spatial distance can be designed, for example, as a threshold. The denoising function can be advantageously designed to classify structures in the image data that are spatially close to blood vessels represented in the control data as true structures with a higher probability, particularly a higher probability and / or weight value, and retain them in the resulting image data. Conversely, structures far from blood vessels represented in the control data may be classified as true structures with a lower probability, particularly a lower probability and / or weight value, and may be interpreted as noise, particularly artifacts.

[0061] The interaction between the confidence condition and the denoising function can be designed such that, when applied to input data, particularly image data, the denoising function represents the spatial relationships between image points of a structure, specifically the structure imaged by the corresponding image points, taking into account at least one nearest vascular segment. The confidence condition can specify the probability of the structure's occurrence discretely or continuously based on the corresponding spatial distance to the blood vessel in the examination area.

[0062] By using this reliability condition, the denoising function can be designed to utilize vascular information from the control data, particularly as anatomical background information, to improve the quality and reliability of the resulting image data, especially in distinguishing relevant structures from noise and / or artificial structures. Based on additional information about vascular location, the spatial probability distribution of the presence of medical objects (e.g., guidewires, catheters, and / or stents) in the image data can be derived, thereby improving the results of the denoising function. In this way, artifacts caused by extravascular misinterpretation can be reduced, and the accurate interpretation and representation of existing medical objects can be improved.

[0063] In another advantageous embodiment of the proposed computer-implemented method, the denoising function may also be designed to identify images of medical objects based on the structure and credibility conditions depicted in the resulting image data, and to provide the resulting image data including identification information.

[0064] In addition to noise reduction, denoising functions can also be designed to identify structures in the resulting image data. Specifically, denoising functions can be designed to identify images of medical objects in the resulting image data.

[0065] Medical objects can be designed as objects or devices that can be used for medical purposes. Medical objects can include surgical and / or diagnostic instruments, such as catheters, guidewires and / or endoscopes, and / or implants, such as stents.

[0066] The resulting image data may include multiple image points, specifically pixels and / or voxels, with corresponding image values ​​(e.g., intensity and / or attenuation values). An image of a medical object can be designed as a representation of the medical object in the resulting image data, specifically the image values ​​of the image points in the resulting image data.

[0067] The image of a medical object in the identified result image data may specifically include the corresponding structure in the identified result image data. The structure in the identified result image data may include the identification, particularly segmentation, of image points in the resulting image data that image the structure. Structure identification may be based, for example, on comparison of pixel image values ​​with thresholds and / or pattern recognition and / or structure recognition. The image of a medical object in the identified result image data may include the identification, particularly the segmentation, of image points in the result image data based on the structure depicted in the result image data, which depicts that structure, particularly the medical object. The structure depicted in the result image data may, for example, have at least one geometric feature that is a characteristic of the medical object to be identified.

[0068] Advantageously, the denoising function can be designed to perform image recognition of medical objects based on additional confidence conditions. The confidence conditions, specified by the control data, can specify the probability that the image of the medical object appears at different locations in the resulting image data. For example, the confidence condition can specify a higher probability that the medical object appears near blood vessels in the examination area based on vascular information.

[0069] The denoising function can be designed to provide result image data with identification information. This identification information can, for example, include annotations, particularly markings and / or highlighting, of the identified medical object in the result image data. Alternatively or additionally, the identification information can indicate coordinates and / or regions, particularly relative to the result image data and / or the area to be examined, where the medical object has been identified, specifically an image of the medical object. The identification information can also include information about the attributes (e.g., operational parameters and / or material parameters) and / or identifiers (e.g., type and / or name and / or category) of the medical object. The identification information can include, for example, the type of the medical object, its location, size, orientation, and / or other relevant attributes.

[0070] This can advantageously enable improved imaging of the examination area containing medical objects.

[0071] In another advantageous embodiment of the proposed computer-implemented method, the timeliness nature of the control data can be determined. The availability of the control data can be checked based on the timeliness nature. If the answer is yes, the acquired control data can be provided to provide the resulting image data. If not, synthetic control data with a predefined, particularly spatially constant, value distribution can be provided to provide the resulting image data, or updated control data can be recorded and provided to provide the resulting image data.

[0072] Timeliness characteristics may include parameters that indicate the timeliness and / or validity of control data. Timeliness characteristics may include, for example, timestamps, version numbers, and / or validity status.

[0073] Determining the timeliness of control data can be done in several ways and involves several technical aspects. Control data may include a timestamp indicating the time it was collected or last updated. Comparing this timestamp to the current system time can be used to determine the age of the control data. Control data may also include a version number, which increments with each update. Comparing this version number to a version defined as the current version can provide information about the up-to-dateness of the control data.

[0074] Alternatively or additionally, checksums for the control data can be calculated and stored. Each time the control data is used, the checksum can be recalculated and compared with the stored checksum to identify changes in the control data. The control data may also include metadata containing information about its validity or timeliness. Advantageously, events requiring updates to the control data can be monitored, such as changes in the positioning of the object under examination and / or changes in the positioning of the imaging device, particularly movement and / or changes in angle, and / or changes in the object under examination, such as medical procedures.

[0075] The availability of control data can be checked based on its timeliness. Checking the availability of control data can include comparing its timeliness against one or more predefined criteria. This check can include, for example, comparing the timestamp and / or age and / or version of the control data with corresponding thresholds, or checking the validity status of the control data.

[0076] If the answer is yes, the acquired control data can be provided to deliver the resulting image data, specifically for applying a denoising function to the input data. Here, "if yes" means that if the availability check is positive, i.e., the control data is classified as available, then that control data is used for further processing.

[0077] If not, synthetic control data with a predefined, particularly spatially constant, value distribution can be provided to deliver the resulting image data. The synthetic control data can be artificially generated control data with a specific predefined structure and / or distribution, particularly the value distribution. A spatially constant value distribution can mean that the synthetic control data, particularly the data points of the synthetic control data, such as pixels or voxels, have the same value across the entire relevant space. The synthetic control data can be designed to have a neutral effect on the denoising function. A denoising function using control data can always be designed in such a way that the absence of control data is equivalent to providing synthetic control data and only results in spatially uniform processing of the image data, so that there are no disadvantages compared to implementing such a denoising function without control data. If the control data is an integral and necessary part of the denoising function, for example, in an embodiment where the denoising function is used for training, particularly as a neural network, then the absence of control data can be represented by synthetic control data. Due to the above equivalence, in this step, only the beneficial effects of the control data may be lost, but the denoising function remains usable, and the resulting image data remains usable, without having to implement separate variants or operating modes of the denoising function with or without control data. Specifically, if control data is no longer evaluated as current but updated control data is still unavailable, this method can make it possible, for example, to selectively disable the use of control data based on user input.

[0078] Alternatively, if the answer is no, updated control data can be recorded and provided to provide the resulting image data, specifically by applying a denoising function to the input data. Updated control data can be acquired in a similar manner to acquiring control data. Acquiring updated control data can include acquiring new control data or updating the last acquired control data.

[0079] The choice between these two alternatives can be based on various factors, such as the availability of resources for re-imaging or the expected changes in the area being examined since the last acquisition of control data.

[0080] This process ensures that the denoising function always works with current or at least neutral control data. This improves the quality and reliability of the resulting image data. In particular, it prevents outdated control data from causing misinterpretations or artifacts in the resulting image data.

[0081] Advantageously, this method allows for robust and flexible application of the denoising function even if conditions in the examined area change between the acquisition of control data and image data. This is particularly useful in dynamic medical scenarios, such as during interventions where vascular structures may change.

[0082] In another advantageous embodiment of the proposed computer-implemented method, determining the timeliness of the control data may include receiving process information regarding changes in the inspection area of ​​the object being inspected between acquiring control data and acquiring image data prior to the start of the method.

[0083] Process information may include information about the medical procedures or interventions performed, particularly before the procedure begins. Process information may include, for example, information about the type, extent, timing, and / or location of the procedure performed on the object being examined.

[0084] Receiving process information may include acquiring and / or reading from computer-readable data storage and / or receiving from a data storage unit (e.g., a database). Furthermore, process information may be provided by one or more medical imaging devices, particularly providing units of the same or different imaging modalities. Alternatively or additionally, process information may be received based on user input recorded using an input unit.

[0085] Changes to the examination area can be designed to alter the anatomical or physiological conditions of the subject's examination area. These changes may include, for example, the opening or closing of blood vessels, particularly the opening of embolisms or vascular closures, the introduction or removal of medical objects or implants, and / or tissue alterations.

[0086] The acquisition of control data can occur before the start of the method. Specifically, the acquisition of control data is not part of the method itself. Furthermore, changes in the inspection area may occur between the acquisition of control data and the acquisition of image data, particularly before the start of the method.

[0087] Determining the timeliness of information based on process details can include an analysis of the type and / or extent of procedures performed prior to the start of the method. The degree to which the procedure may have altered the examined area, particularly at least one vascular portion, and / or other aspects of the examined area relevant to the control data can be assessed. Advantageously, the validity of the control data can be identified, and particularly evaluated, based on process information.

[0088] This process advantageously allows the method to dynamically adapt to changes in the inspection area. This improves the accuracy and reliability of the resulting image data by ensuring that the control data used correctly reflects the current situation in the inspection area.

[0089] In another advantageous embodiment of the proposed computer-implemented method, the provision of the resulting image data may include displaying a graphical representation of the resulting image data using a display unit.

[0090] The display unit may include, for example, a monitor, display, screen, projector, data glasses, or other devices for displaying a graphical representation of the resulting image data.

[0091] Graphical representations can be designed as visual representations of the resulting image data.

[0092] The graphical representation of the resulting image data displayed using the display unit can include presenting the graphical representation on the screen or projecting the resulting image data onto the display plane of the display unit.

[0093] Graphical displays enable users, especially medical professionals, to visually acquire and interpret resulting image data. This can support diagnostic and treatment planning.

[0094] Advantageously, the graphical representation can be designed to be interactive. For example, users can zoom, rotate, or highlight certain areas. This can be accomplished through inputs on input units connected to the display unit.

[0095] In a second aspect, the present invention relates to a computer-implemented method for providing a trained denoising function. In a first step, pre-acquired or simulated training control data having vascular information of a training examination region of a training examination object is received. In a further step, pre-acquired or simulated training image data of the training examination region is received. In a further step, comparative image data of the training examination region is provided, which is denoised and / or classified compared to the training image data. In a further step, training result image data is provided by applying the trained denoising function to input data using training control data, the input data being based on the training image data, and the training control data specifying credibility conditional image data of the training structure depicted in the training result. In another step, at least one parameter of the trained denoising function is adjusted based on a comparison of the training result image data and the comparative image data. In a further step, the trained denoising function is provided.

[0096] The advantages of the proposed computer implementation method for providing the trained denoising function substantially correspond to the advantages of the proposed computer implementation method for providing the resulting image data. The features, advantages, or alternative embodiments mentioned herein can also be transferred to other claimed subjects, and vice versa.

[0097] Training control data can, in particular, possess all the characteristics and properties of control data that describe the method used to provide the resulting image data, and vice versa. Training control data can be recorded before the method begins, especially via medical imaging equipment, and is particularly similar to the recording of control data. Alternatively, training control data can be simulated.

[0098] The training check region can be formed as a spatial region of the training check object used to train the denoising function. The training check region and the training check object can, in particular, possess all the features and attributes of the check region and check object described with respect to the method used to provide the resulting image data, and vice versa.

[0099] Receiving training control data may include acquiring and / or reading from a computer-readable data storage device and / or receiving it from a data storage unit (e.g., a database). Furthermore, pre-acquired training control data may be provided by one or more medical imaging devices, particularly providing units of the same or different imaging modalities.

[0100] The simulation of training control data can be based on, for example, biophysical models. Alternatively or additionally, statistical models can be created by analyzing real vascular data and then used to generate synthetic training control data. Alternatively or additionally, artificial neural networks, such as generative adversarial networks or variational autoencoders, can be trained to generate realistic vascular structures. Based on anatomical knowledge, rules and / or boundary conditions considered when generating vascular structures can be defined. Furthermore, combinations of different methods can be used to combine the advantages of different approaches and create simulations that are as realistic as possible. Advantageously, different vascular routes and / or branches can be simulated to depict anatomical variations. Pathological changes, such as stenosis, aneurysms, and / or other vascular abnormalities, can be included in the simulation. The simulation can also be adapted to different demographic groups to account for differences in age and sex.

[0101] Vascular information may include spatially resolved information, particularly 2D or 3D spatially resolved information, regarding the location of at least one vascular segment (particularly arteries and / or veins) and / or vascular malformations in a training examination area of ​​a training examination object. The vascular information may advantageously have a spatially resolved representation, particularly 2D or 3D, of at least one vascular segment and / or vascular malformation in a training examination area of ​​a training examination object, particularly a model and / or image. Advantageously, the vascular information may characterize the spatial location (particularly position and / or orientation), size (particularly length and / or diameter), and / or spatial route of at least one vascular segment.

[0102] Receiving training image data may include acquiring and / or reading from a computer-readable data storage device and / or receiving it from a data storage unit (e.g., a database). Furthermore, pre-acquired training image data may be provided by one or more medical imaging devices, particularly providing units of the same or different imaging modalities.

[0103] Pre-acquired or simulated training image data can be designed to be image data acquired or generated through simulation before the method used to provide the denoising function for training begins. The training image data can, in particular, possess all the characteristics and properties of the image data described with respect to the method used to provide the resulting image data, and vice versa. Advantageously, the training image data can depict the training examination area. The training image data can advantageously be recorded similarly to the image data, particularly via medical imaging equipment. Alternatively, the training image data can be simulated.

[0104] Simulation of training image data can be based on various methods and algorithms to generate realistic medical training image data. For example, mathematical models, particularly the Radon transform, can be used to generate synthetic X-ray images. Physical simulations, especially biophysical simulations, can simulate the interaction of imaging methods (e.g., X-rays) with different tissue types. For example, the Monte Carlo method can be used. Based on real, particularly acquired and / or measured medical image data, statistical models can be created and then used to generate synthetic training image data. For example, Markov random fields or Gaussian mixture models can be used. Alternatively or additionally, artificial neural networks, such as generative adversarial networks or variational autoencoders, can be trained to generate realistic simulated training image data. Simulations can advantageously take into account various factors, such as anatomical variations, pathological changes, imaging parameters, and / or noise characteristics. By using these simulation methods, a wide variety of training image data can be generated, which helps to improve the robustness and generality of the trained denoising function.

[0105] Comparative image data can depict the examination area of ​​an object in a spatially resolved manner, particularly in 2D or 3D. For example, comparative image data can include 2D projected images of the training examination area from one or more projection directions. These projected images can include, for example, X-ray or fluorescence fluoroscopic images. Alternatively or additionally, comparative image data can include a three-dimensional volumetric dataset of the training examination area. Furthermore, comparative image data can be resolved in a timely manner, with multiple images of the training examination area acquired at different times.

[0106] Comparative image data may also include raw data directly provided by the detectors of medical imaging equipment (e.g., X-ray detectors). Alternatively or additionally, the comparative image data may have been preprocessed, for example, by reconstructing the raw data. Furthermore, comparative image data may include metadata such as recording parameters, patient information, procedural information, and / or other relevant data.

[0107] The comparative image data is denoised and / or classified compared to the training image data. Denoising may mean that interfering image artifacts or noise in the comparative image data have been reduced or removed compared to the training image data. Classification may mean that certain structures or objects have been identified and / or labeled in the comparative image data. The comparative image data can be provided based on the training image data or on other image data. For example, the comparative image data may include high-quality recordings of the training examination area with significantly higher radiation doses or longer recording times. These recordings may have an improved signal-to-noise ratio compared to the training image data. Alternatively or additionally, providing the comparative image data may include a combination of multiple images of the training examination area, particularly averaging, for example, through iterative reconstruction algorithms and / or algebraic reconstruction methods and / or statistical reconstruction methods. Alternatively or additionally, the comparative image data may be provided by applying image processing algorithms (e.g., filter algorithms, particularly nonlocal mean filters and / or wavelet-based denoising methods) to the training image data.

[0108] The classification of comparative image data can include manual annotation or automatic classification, such as the application of automatic segmentation algorithms. Images of anatomical structures, pathological changes, and / or medical objects in comparative image data can be classified, particularly annotated and / or segmented.

[0109] The trained denoising function is applied to the input data using training control data to provide the training result image data. The input data for the trained denoising function is based on the training image data. Specifically, the input data for the trained denoising function may include the training image data.

[0110] Training control data specifies the credibility conditions for the structures depicted in the training result image data. Credibility conditions can be understood as rules or criteria used to evaluate the probability and / or credibility of the structures depicted in the training result image data.

[0111] The input data for the trained denoising function can also be based on training control data, and in particular, it can include training control data.

[0112] If the trained denoising function is designed as a deep neural network, training control data can be fed into different levels of the network. This allows the trained denoising function to consider plausible information at different levels of abstraction. Integration can also be achieved using a conditional generative model, where the training control data is used as conditions to generate the denoised training image data.

[0113] Furthermore, at least one parameter of the trained denoising function is adjusted based on a comparison between the training result image data and the comparison image data. This comparison may include, for example, calculating a difference and / or similarity measure between the training result image data and the comparison image data. The adjustment of at least one parameter of the trained denoising function can be performed iteratively to improve, and in particular optimize, the correspondence between the training result image data and the comparison image data.

[0114] Providing a trained denoising function may include storing it on a computer-readable storage medium and / or transmitting it to a supply unit. The trained denoising function provided by this method can be advantageously used in the proposed method for providing resulting image data.

[0115] The described method for providing a trained denoising function offers various technical advantages. By using training control data containing vascular information, the denoising function can be specifically optimized for processing medical image data with vascular structures. Specifying confidence conditions through the training control data helps the denoising function provide realistic and anatomically plausible results. High quality and reliability of the trained denoising function can be achieved by comparing it with comparative image data and iteratively adjusting the parameters.

[0116] Generally, a trained denoising function can mimic the cognitive function that links humans to other human ghosts. Specifically, by training on training data, a trained denoising function may be able to adapt to new environments and recognize and infer patterns. Another term for “training a denoising function” is “training a function.” Generally, the parameters of the denoising function can be tuned through training. Specifically, supervised training, semi-supervised training, unsupervised training, reinforcement learning, and / or active learning can be used. Furthermore, representation learning (an alternative term could be “feature learning”) can be used. In particular, the parameters of the denoising function can be iteratively tuned over several training steps. In particular, a specific cost function can be minimized during training. In particular, backpropagation algorithms can be used for training neural networks. Specifically, the denoising function can include neural networks, support vector machines, decision trees, and / or Bayesian networks, and / or the denoising function can be based on k-means clustering, Q-learning, genetic algorithms, and / or association rules. Specifically, the neural network can be a deep neural network, a convolutional neural network, or a convolutional deep neural network. Furthermore, the neural network can be an adversarial network, a deep adversarial network, and / or a generative adversarial network.

[0117] A convolutional neural network (CNN) can be a neural network that uses convolution operations instead of general matrix multiplication in at least one of its layers (so-called "convolutional layers"). Specifically, a convolutional layer can perform a dot product between one or more convolutional kernels and the input data / image of the convolutional layer, where the entries for the one or more convolutional kernels can be parameters or weights tuned through training. In particular, the Frobenius dot product and ReLU activation functions can be used. A CNN may include additional layers such as pooling layers, fully connected layers, and normalization layers.

[0118] By using convolutional neural networks, input images can be processed very efficiently because convolution operations based on different kernels can extract different image features. Therefore, by adjusting the weights of the convolutional kernels, relevant image features can be found during training. Furthermore, because the weights in the convolutional kernels are shared, fewer parameters need to be trained, thus preventing overfitting during training and allowing for faster training of more layers in the network, thereby improving network performance.

[0119] According to one aspect, the denoising function may include one or more residual networks (ResNets). Specifically, a ResNet can be an artificial neural network that includes at least one skip connection for skipping at least one layer of the artificial neural network. Specifically, a ResNet can be a convolutional neural network including one or more skip connections, each skip connection skipping one or more convolutional layers. According to some examples, a ResNet can be represented as an m-layer ResNet, where m is the number of layers in the corresponding architecture, and according to some examples, it can take values ​​of 34, 50, 101, or 152. According to some examples, such an m-layer ResNet may each include (m-2) / 2 skip connections.

[0120] Skip connections can be viewed as bypasses, routing the output of the previous layer directly through one or more skipped layers to the layer following those skipped layers. The skipped layers do not need to directly adjust the required mappings; instead, they need to adjust the residual mappings of the directly fed outputs to "compensate" for them.

[0121] Compared to direct mapping, adjusting the residual mapping is computationally easier to optimize. Furthermore, this mitigates the vanishing / exploding gradient problem during optimization when training the denoising function: if skipped layers encounter such problems, their contributions can be skipped by regularizing the output of the directly fed layer. An advantage of using ResNet is the ability to train deeper networks. As part of the expected provision of the resulting image data based on image data, this has the advantage of more accurately predicting the resulting image data, and thus contributes to improved denoising.

[0122] Generative adversarial models (GA models) can include a generator function and a discriminator function, where the generator function generates synthetic data and the discriminator function distinguishes between synthetic and real data. By training the generator function and / or discriminator function, on the one hand, the generator function can be configured to generate synthetic data that is incorrectly classified as real by the discriminator function; on the other hand, the discriminator function can be configured to distinguish between real data and synthetic data generated by the generator function. In game theory, generative adversarial models can be interpreted as zero-sum games. Training the generator function and / or discriminator function can, in particular, be based on minimizing a cost function.

[0123] By using a Generative Analytics (GA) model, synthetic data with the same properties as the training dataset can be generated from a set of training data. Training a GA model can be based on unannotated data (unsupervised learning), therefore the amount of work required to train a GA model can be very low.

[0124] In particular, a recurrent denoising function can be a denoising function whose output depends not only on the input values ​​and the parameters of the denoising function adjusted through the training process, but also on the hidden state vector, which can be based on previous inputs used for the recurrent denoising function. Specifically, the recurrent denoising function may include additional memory states or additional structures including time delays or feedback loops.

[0125] In particular, the underlying structure of a recurrent denoising function can be a neural network, which can be called a recurrent neural network. This recurrent neural network can be described as an artificial neural network where the connections between nodes form a directed graph along a time series. Specifically, a recurrent neural network can be interpreted as a directed acyclic graph. Specifically, the recurrent neural network can be a finite impulse response recurrent neural network or an infinite impulse response recurrent neural network (where a finite impulse response network can be coiled up and replaced with a strictly feedforward neural network, while an infinite impulse response network can be uncoiled and replaced with a strictly feedforward neural network).

[0126] In particular, the training of recurrent neural networks can be based on the BPTT algorithm (short for "backward propagation in time"), the RTRL algorithm (short for "real-time recurrent learning"), and / or genetic algorithms.

[0127] By using a cyclic denoising function, input data including sequences of variable length can be used. Specifically, this could mean that the method can be used not only with a fixed number of input datasets (and must be trained differently on any other number of input datasets used as input), but also with any number of input datasets. This could mean that the entire training dataset, regardless of the number of input datasets contained in different sequences, can be used in training, and the training data does not simplify to training data corresponding to a fixed number of consecutive input datasets.

[0128] Transformer networks can be neural network architectures, typically including an encoder, a decoder, or both. In some cases, the encoder and / or decoder may consist of multiple corresponding encoding and decoding layers. Attention mechanisms can preferably be present within each encoding and decoding layer. Attention mechanisms, sometimes called self-attention, associate data elements (e.g., words or pixels) within a set of data elements with other data elements within that set. For example, a self-attention mechanism could allow a model to examine a set of pixels within a medical image and determine the relative importance of other groups of pixels within that medical image to the examined pixel group.

[0129] The encoder can be configured to transform the input data into a digital representation. The digital representation may include a vector for each input token. The encoder can be configured to implement an attention mechanism such that each vector of tokens is influenced by other tokens in the input. Specifically, the encoder can be configured to represent the desired output of the analytic transformer network.

[0130] In particular, the decoder can be configured to transform the input into a sequence of output tokens. Specifically, the decoder can be configured to implement a masked self-attention mechanism, such that each vector of tokens is only affected by other tokens on one side of the sequence.

[0131] Furthermore, the decoder can be autoregressive, meaning it returns intermediate results, such as previously predicted token sequences.

[0132] According to some examples, the input to the decoder can be based on the output of the encoder or equivalent to the output of the encoder.

[0133] In addition, the transformer network may include a classification module, which can be configured to map the output of the encoder or decoder to a set of learned outputs.

[0134] Based on some examples, training a transformer model can be performed in two phases: a pre-training phase and a fine-tuning phase. In the pre-training phase, the transformer model can be trained on a large corpus to learn the underlying semantics of the problem. This pre-trained transformer model can be used with different languages. For some applications described in this paper, fine-tuning may include further training the transformer network using medical text with expert-annotated meanings and / or medical ontologies (e.g., RADLEX and / or SNOMED). Based on some examples, the latter allows the transformer model to specifically learn canonical relations and synonymous medical expressions.

[0135] The advantages of transformer networks may include their ability to efficiently handle long-range dependencies in the input data due to their attention mechanism. Furthermore, the encoder used in transformer networks can process data in parallel, thus saving computational resources during inference. Additionally, due to autoregression, the decoder in a transformer network can iteratively generate output token sequences with high confidence.

[0136] The described model and training steps can be used, in particular, as part of a computer implementation of a method for providing a denoising function for training. Specifically, the trained denoising function can be implemented as one of the described models, such as a convolutional neural network, ResNet, or Transformer network.

[0137] In a third aspect, the present invention relates to a supply unit. The supply unit is designed to perform the proposed computer-implemented method for providing resulting image data.

[0138] The providing unit may include a computing unit, a storage unit, and / or an interface. The providing unit, and particularly its components, may be designed to perform the various steps of the proposed method for providing the resulting image data.

[0139] The interface can be designed to receive pre-acquired control data and acquire image data. The computing unit and / or storage unit and / or interface can be designed to provide the resulting image data by applying a denoising function to the input data using the control data.

[0140] The computing unit can be designed as a processor, microprocessor, digital signal processor, or application-specific integrated circuit. The storage unit can be designed as volatile or non-volatile memory, such as RAM, ROM, flash memory, or hard disk storage. The interface can be designed as a network interface, serial interface, or parallel interface.

[0141] The providing unit can be designed as a standalone device or as part of a larger system. The providing unit can be connected to other components, such as a display unit for displaying the resulting image data and / or an input unit for acquiring user input.

[0142] The advantages of the providing unit essentially correspond to the advantages of the proposed computer-implemented method for providing the resulting image data. Taking control data into account, the providing unit can achieve efficient and reliable processing of image data to provide denoised resulting image data. The features, advantages, or alternative embodiments mentioned herein can also be transferred to other claimed subjects, and vice versa.

[0143] In a fourth aspect, the present invention relates to a system. The system includes a medical imaging device and a supply unit. The medical imaging device is designed to acquire image data and provide it to the supply unit. The supply unit is designed to execute the proposed computer-implemented method for providing resulting image data and / or for providing a trained denoising function.

[0144] Medical imaging equipment can be designed as, for example, magnetic resonance imaging (MRI) systems and / or computed tomography (CT) systems and / or medical X-ray equipment and / or positron emission tomography (PET) systems and / or ultrasound equipment.

[0145] Medical imaging devices can be designed to acquire, and in particular record, image data. They can also be designed to provide the acquired image data to a providing unit. This can include data transmission via a suitable interface, such as a wired or wireless network connection.

[0146] The advantages of this system essentially correspond to the advantages of the proposed providing unit for providing the resulting image data and / or for providing the denoising function for training and / or the proposed computer-implemented method. The features, advantages, or alternative embodiments mentioned herein may also be transferred to other claimed subjects, and vice versa.

[0147] The invention may also relate to a training unit, which can be designed to perform the proposed method for providing a trained denoising function. The training unit may advantageously include a training interface, a training storage unit, and / or a training computation unit. The training unit can be designed to perform a method for providing a trained denoising function, wherein the training interface, training storage unit, and / or training computation unit can be designed to perform corresponding method steps.

[0148] Advantageously, the training interface can be designed to receive pre-acquired or simulated training control data and pre-acquired or simulated training image data. Furthermore, the training computation unit and / or training storage unit can be designed to provide comparison image data, provide training result image data by applying a trained denoising function to the input data using the training control data, and train the denoising function by adjusting at least one parameter based on a comparison between the training result image data and the comparison image data. Additionally, the training interface can be designed to provide the trained denoising function.

[0149] The advantages of the training course essentially correspond to the advantages of the proposed method for providing a computer implementation of the trained denoising function. The features, advantages, or alternative embodiments mentioned herein can also be transferred to other claimed subjects, and vice versa.

[0150] In a fifth aspect, the present invention relates to a computer program product having a computer program that can be directly loaded into the memory of a computer, having a program portion for performing all steps of the proposed method for providing resulting image data and / or, when the program portion is executed by a computer, providing a trained denoising function.

[0151] Computer program products may include software having source code that has not yet been compiled and bound, or that only needs to be interpreted, or executable software code that only needs to be loaded into a computer's memory, particularly a supply unit, for execution. With the aid of computer program products, computer-implemented methods for providing resulting image data and / or for providing denoising functions for training using a computer, particularly a supply unit, can be performed quickly, identically, reproducibly, and robustly. Computer program products can be configured to perform the method steps according to the invention using a computer, particularly a supply unit.

[0152] Computer program products can be stored, for example, on computer-readable storage media or on a network or server, from where they can be loaded into a computer processor, particularly a supply unit, which can be directly connected to the computer, particularly the supply unit, or as part of the computer, particularly the supply unit. Furthermore, control information of the computer program product can be stored on an electronically readable data carrier. The control information on the electronically readable data carrier can be designed such that when the data carrier is used in a computer, particularly a supply unit, it performs the methods according to the invention. Examples of electronically readable data carriers can be DVDs, magnetic tapes, or USB sticks, on which electronically readable control information, particularly software, can be stored.

[0153] A largely software-based implementation can have the following advantages: previously used computers, particularly supply units, can be easily modified through software updates to operate in accordance with the invention. In addition to the computer program, such a computer program product may optionally include add-ons, such as documentation and / or other add-ons, as well as hardware components, such as hardware keys (dongles, etc.) for using the software.

[0154] Advantageously, the computer program product enables efficient and flexible implementation of the method according to the invention. The computer program product can execute on various computer platforms, thereby enabling the method to be widely applied. Furthermore, the computer program product can be easily updated and extended to integrate new functions or improve the process. Attached Figure Description

[0155] Embodiments of the invention are shown in the accompanying drawings and described in more detail below. In the different drawings, the same reference numerals are used for the same features. This is illustrated in:

[0156] Figure 1 and Figure 2 Schematic diagrams of various advantageous embodiments of the proposed computer-implemented method for providing resulting image data,

[0157] Figure 3 A schematic diagram of an advantageous embodiment of the proposed method for providing a computer-based implementation of a trained denoising function.

[0158] Figure 4 A schematic diagram of an advantageous embodiment of the proposed supply unit,

[0159] Figure 5 A schematic diagram of an advantageous embodiment of the proposed training course,

[0160] Figure 6 A schematic diagram of an advantageous embodiment of the proposed system. Detailed Implementation

[0161] Figure 1 A schematic diagram of an advantageous embodiment of the proposed computer-implemented method for providing result image data EBD is shown. Pre-acquired control data SD REC-SD, containing vascular information of the examination area of ​​the subject, can be received. In a further step, image data BD of the examination area can be acquired using a medical imaging device CAP-BD. In a further step, result image data EBD can be generated by applying a denoising function EF to the input data using the control data SD. The input data can be based on the image data BD. Furthermore, the control data SD can specify the credibility conditions of the structures depicted in the result image data EBD. The result image data EBD can be denoised compared to the image data BD.

[0162] Advantageously, the denoising function EF can be designed as a training function. At least one parameter of the trained denoising function can be adjusted based on a comparison between the training image data and the comparison image data.

[0163] Advantageously, the image data BD may include at least one projected image of the examination area from the corresponding projection direction. The control data SD for each of the at least one projected image may include a corresponding two-dimensional vascular representation of the examination area from the projection direction of the corresponding projected image. Specifically, the control data SD may include a three-dimensional vascular representation of the examination area.

[0164] Alternatively or additionally, the control data SD may include a representation of blood vessels in the examination area. The control data SD may include a subtraction dataset of the examination area, provided based on the subtraction of a mask dataset and a filling dataset. Advantageously, the mask dataset may depict the examination area during the masking stage. Furthermore, the filling data record may depict the examination area during the filling stage.

[0165] Advantageously, the confidence condition can specify a higher probability of the presence of structures near blood vessels in the examination area based on vascular information.

[0166] Advantageously, the denoising function EF can also be designed to identify images of medical objects based on the structure and credibility conditions depicted in the resulting image data EBD, and to provide resulting image data that includes identification information.

[0167] Advantageously, providing the resulting image data EBD may include a graphical representation of the resulting image data EBD displayed using a display unit.

[0168] Figure 2 A schematic diagram of another advantageous embodiment of the proposed computer-implemented method for providing resulting image data EBD is shown. The freshness attribute AE ID-AE of the control data SD can be determined. Additionally, the availability of the CHK control data SD can be checked based on the novelty attribute AE. If it is Y, the acquired control data SD can be provided to provide the resulting image data EBD. If not, synthetic control data SSD with a predefined, particularly spatially constant, value distribution can be provided to provide the resulting image data EBD. Alternatively, updated control data can be acquired and provided to provide the resulting image data EBD (not shown here).

[0169] Advantageously, the determination of the novelty attribute AE of the control data SD ID-AE may include receiving process information about changes in the inspection area of ​​the object to be inspected between acquiring the control data SD and acquiring the CAP-BD before starting the method, such as image data BD.

[0170] Figure 3A schematic diagram of an advantageous embodiment of the proposed method for providing a computer-based denoising function EF for training is shown. Pre-acquired or simulated training control data TSD REC-TSD, containing vascular information of the training examination region of the training examination subject, can be received. In a further step, pre-acquired or simulated training image data TBD of the training examination region REC-TBD can be received. In a further step, comparative image data VBD of the training examination region PROV-VBD can be provided. Compared to the training image data TBD, the comparative image data VBD can be denoised and / or classified. In a further step, training result image data TEBD can be provided by applying the trained denoising function EF to the input data using the training control data TSD. The input data can be based on the training image data TBD. Additionally, the training control data TSD can specify the credibility conditions of the training structure depicted in the training result image data TEBD. In another step, at least one parameter of the trained denoising function EF can be adjusted based on a comparison of the training result image data TEBD and the comparative image data VBD Adj-EF. In a further step, a trained denoising function, PROV-EF, can be provided.

[0171] Figure 4 A schematic diagram of an advantageous embodiment of a supply unit PRVS is shown. The supply unit PRVS may include an interface IF, a computing unit CU, and a storage unit MU. The supply unit PRVS, and particularly its components, may be designed to perform the various steps of the proposed method for providing resulting image data EBD. The supply unit PRVS may, in particular, include or be designed to be a computer.

[0172] The interface IF can be designed to receive pre-acquired control data SD (REC-SD) and acquire image data BD (CAP-BD). The computation unit CU and / or storage unit MU and / or interface IF can be designed to provide the resulting image data EBD by applying a denoising function EF to the input data using the control data SD.

[0173] Figure 5 A schematic diagram of an advantageous embodiment of a training unit TRS is shown. The training unit TRS can be designed to perform the proposed computer-implemented method for providing the PROF-EF of the trained denoising function EF. The training unit TRS can advantageously include a training interface TIF, a training storage unit TMU, and / or a training computation unit TCU. The training unit TRS can be designed to perform a method for providing the PROV-EF of the trained denoising function EF, wherein the training interface TIF, the training storage unit TMU, and / or the training computation unit TCU can be designed to perform the corresponding method steps.

[0174] Advantageously, the training interface TIF can be designed to receive pre-acquired or simulated training control data TSD and pre-acquired or simulated training image data TBD REC-TSD, REC-TBD. Furthermore, the training computation unit TCU and / or training storage unit TMU can be designed to provide comparison image data VBD PROV-VBD, to provide training result image data TEBD TSD by applying the trained denoising function EF to the input data using training control, and to adjust at least one parameter of the trained denoising function EF based on a comparison of the training result image data TEBD and the comparison image data VBD ADJ-EF. Additionally, the training interface TIF can be designed to provide the trained denoising function EF PROV-EF.

[0175] Figure 6 A schematic diagram of an advantageous embodiment of a system including medical imaging equipment (particularly a C-arm X-ray device 37) and a supply unit PRVS is shown.

[0176] The C-arm X-ray device 37 can be designed to record image data BD of the object 31 being examined, particularly a projected image. The C-arm X-ray device may include an X-ray source 33 and a detector 34, particularly an X-ray detector, which are arranged opposite each other and on a common C-arm 38, also referred to as the C-arm.

[0177] To record medical image data (BD), the arm 38 of the C-arm X-ray device 37 can be movably mounted about one or more axes. The system may also include a patient positioning device 32 on which an examination subject 31 is disposed. The examination subject 31 may be, for example, a human and / or animal patient and / or an examination phantom, such as a vascular phantom.

[0178] The supply unit PRVS can send signal 24 to the X-ray source 33 to record image data BD. The X-ray source 33 can then emit an X-ray beam. When the X-ray beam strikes the X-ray sensitive layer, particularly the surface, of the detector 34 after interacting with the object under inspection 31, the detector 34 can send signal 21 to the supply unit PRVS. The supply unit PRVS can, for example, receive image data BD REC-BD based on signal 21.

[0179] Furthermore, the C-arm X-ray device 37 may have an input unit 42 (e.g., a keyboard) and a display unit 41 (e.g., a monitor and / or display). The input unit 42 may preferably be integrated into the display unit 41, for example, in a capacitive and / or resistive input display. The medical C-arm X-ray device 37 can be controlled by input made by a medical operator on the input unit 42. For this purpose, the input unit 42 may, for example, send a signal 26 to the providing unit PRVS. The display unit 41 may advantageously be designed to display a graphical representation of the resulting image data EBD. For this purpose, the providing unit PRVS may send a signal 25 to the display unit 41.

[0180] The schematic diagrams included in the accompanying drawings do not represent any scale or size relationship.

[0181] Finally, it should be reiterated that the methods and apparatus described in detail above and illustrated are merely exemplary embodiments, and those skilled in the art can modify them in various ways without departing from the scope of the invention. Furthermore, the use of the indefinite article "a" or "an" does not preclude the fact that the features in question may appear multiple times; similarly, the terms "unit" and "element" do not preclude the fact that the components in question consist of multiple interacting sub-components, which may also be spatially distributed.

[0182] In the context of this application, the expression “based on” can be understood specifically in the sense of “using”. In particular, the expression that the first feature is generated based on the second feature (or alternatively, determined, determined, etc.) does not preclude the first feature from being generated based on the third feature (or alternatively, determined, determined, etc.).

Claims

1. A computer-implemented method for providing Resulting Image Data (EBD), comprising: - Receive (REC-SD) pre-acquired control data (SD), the control data containing vascular information of the examination area of ​​the examination object (31), - Use medical imaging equipment to acquire (CAP-BD) image data (BD) of the examination area. - Providing (PROV-EF) result image data (EBD) by applying a denoising function (EF) to the input data using control data (SD). The input data is based on image data (BD). Among them, the control data (SD) sets the credibility conditions for the structures depicted in the resulting image data (EBD). The resulting image data (EBD) is denoised relative to the image data (BD).

2. The method according to claim 1, in, The denoising function (EF) is designed to be a trained function. In this process, at least one parameter of the trained function is adjusted based on a comparison between the training result image data (TEBD) and the comparison image data (VBD).

3. The method according to claim 1 or 2, in, Image data (BD) includes at least one projected image from the inspection area in the corresponding projection direction. The control data (SD) for each of at least one projection image includes a corresponding two-dimensional representation of the blood vessels in the examination area from the projection direction of the corresponding projection image.

4. The method according to any one of the preceding claims, in, Control data (SD) includes a three-dimensional representation of the blood vessels in the examined area.

5. The method according to any one of the preceding claims, in, Control data (SD) includes a representation of blood vessels in the examined area. The control data (SD) includes a subtraction dataset of the inspection area, which is provided based on the subtraction of the mask dataset and the padding dataset. The mask dataset depicts the inspection areas during the masking stage. Among them, the fill data record (FD) depicts the inspection area in the fill phase.

6. The method according to any one of the preceding claims, in, The credibility condition based on vascular information is set to increase the probability of the structure appearing near the blood vessels in the examination area.

7. The method according to any one of the preceding claims, in, The denoising function (EF) is also designed to identify images of medical objects based on the structure and credibility conditions depicted in the Result Image Data (EBD) and to provide Result Image Data (EBD) with identification information.

8. The method according to any one of the preceding claims, in, Determine the timeliness (AE) of the control data (SD). The availability of the control data (SD) is checked (CHK) according to the timeliness (AE) property. If it is (Y), then the obtained control data (SD) is provided to provide the resulting image data (EBD). If not (N), then synthetic control data (SSD) with a predefined, in particular spatially constant value distribution is provided to provide Result Image Data (EBD), or updated control data is acquired and provided to provide Result Image Data (EBD).

9. The method according to claim 8, in, Determining the timeliness (AE) of the control data (SD) includes receiving process information regarding the changes between the control data (SD) acquired before the start of the method and the image data (BD) acquired (CAP-BD) of the inspection area of ​​the inspection object.

10. The method according to any one of the preceding claims, in, Providing the Result Image Data (EBD) includes displaying a graphical representation of the Result Image Data (EBD) by means of a display unit (41).

11. A computer-implemented method for providing a trained denoising function (EF), comprising: - Receive (REC-TSD) pre-acquired or simulated training control data (TSD), which contains vascular information of the training examination area of ​​the training examination subject. - Receive (REC-TBD) training inspection area pre-acquired or simulated training image data (TBD). - Provides comparative image data (VBD) for the (PROV-VBD) training inspection region. The comparison image data (VBD) is denoised and / or classified relative to the training image data (TBD). -Provide (PROV-EF) training result image data (TEBD) by applying a trained denoising function (EF) to the input data using training control data (TSD). The input data is based on training image data (TBD). Among them, the Training Control Data (TSD) sets credibility conditions for the training structure depicted in the Training Result Image Data (TEBD). - Adjust at least one parameter of the trained denoising function (EF) based on a comparison of the training result image data (TEBD) and the comparison image data (VBD). (ADJ-EF) - Provides a trained denoising function (EF) (PROV-EF).

12. A providing unit (PRVS) configured to perform a computer-implemented method according to any one of claims 1 to 10.

13. A system comprising a medical imaging device and a supply unit (PRVS) according to claim 12, in, Medical imaging equipment is designed to acquire image data (BD) and provide it to a provisioning unit (PRVS).

14. A computer program product having a computer program that can be directly loaded into the memory of a computer, the computer program having program segments that, when executed by the computer, perform all the steps of the method according to any one of claims 1 to 11.

Images