MACHINE LEARNING IN CONTRAST-ENHANCED RADIOLOGY

Description

[0001] The present invention relates to the technical field of generating artificial contrast-enhanced radiological images using machine learning methods.

[0002] Radiology is a medical specialty that deals with imaging for diagnostic and therapeutic purposes.

[0003] While medical imaging used to primarily use X-rays and X-ray-sensitive films, radiology today includes several different imaging techniques such as computed tomography (CT), magnetic resonance imaging (MRI) and sonography.

[0004] All of these procedures can utilize substances that facilitate the visualization or delineation of specific structures in an object under examination. These substances are called contrast agents.

[0005] In computed tomography, iodine-containing solutions are commonly used as contrast agents. In magnetic resonance imaging (MRI), superparamagnetic substances (e.g., iron oxide nanoparticles, superparamagnetic iron-platinum particles (SIPPs)) or paramagnetic substances (e.g., gadolinium chelates, manganese chelates) are commonly used as contrast agents.

[0006] According to their distribution pattern in the tissue, contrast agents can be roughly divided into the following categories: extracellular, intracellular and intravascular contrast agents.

[0007] Examples of extracellular MRI contrast agents include the gadolinium chelates gadobutrol (Gadovist ®), gadoteridol (Prohance ®), gadoteric acid (Dotarem ®), gadopentetic acid (Magnevist ®), and gadodiamide (Omnican ®). The highly hydrophilic properties of these gadolinium chelates and their low molecular weight lead to rapid diffusion into the interstitial space after intravenous administration. They are excreted via the kidneys after a relatively short period of circulation in the bloodstream.

[0008] Intracellular contrast agents are absorbed to a certain extent into the cells of tissues and then excreted again.

[0009] Intracellular MRI contrast agents based on gadoxetic acid, for example, are characterized by their specific uptake by liver cells, the hepatocytes, their accumulation in functional tissue (parenchyma), and their contrast enhancement in healthy liver tissue before their subsequent excretion into the feces via bile. Examples of such contrast agents based on gadoxetic acid are described in US Pat. No. 6,039,931A; they are commercially available, for example, under the brand names Primovist ® and Eovist ®. Another MRI contrast agent with lower uptake into hepatocytes is gadobenate dimeglumine (Multihance ® ).

[0010] The contrast-enhancing effect of Primovist ® / Eovist ® is mediated by the stable gadolinium complex Gd-EOB-DTPA (gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid). DTPA forms a complex with the paramagnetic gadolinium ion that exhibits extremely high thermodynamic stability. The ethoxybenzyl moiety (EOB) mediates hepatobiliary uptake of the contrast agent.

[0011] Intravascular contrast agents are characterized by a significantly longer residence time in the bloodstream compared to extracellular contrast agents. Gadofosveset, for example, is a gadolinium-based intravascular MRI contrast agent. It was used in the trisodium salt monohydrate form (Ablavar ®< ). It binds to serum albumin, thereby achieving the contrast agent's long residence time in the bloodstream (half-life in blood of approximately 17 hours).

[0012] There are numerous radiological examinations in which a contrast agent is administered to a patient and the dynamic distribution of the contrast agent in the body is monitored using imaging techniques. One example is the detection and differential diagnosis of focal liver lesions using dynamic contrast-enhanced magnetic resonance imaging.

[0013] Primovist ®< can be used to detect tumors in the liver. The blood supply to healthy liver tissue is primarily via the portal vein ( Vena portae ), while the hepatic artery ( Arteria hepatica ) supplies most primary tumors. Following intravenous bolus injection of a contrast agent, a time delay can be observed between the signal enhancement of the healthy liver parenchyma and the tumor.

[0014] In addition to malignant tumors, benign lesions such as cysts, hemangiomas, and focal nodular hyperplasia (FNH) are frequently found in the liver. For appropriate treatment planning, these lesions must be differentiated from malignant tumors. Primovist ®< can be used to detect benign and malignant focal liver lesions. It provides information about the nature of these lesions using T1-weighted MRI. Differentiation is based on the different blood supply to the liver and tumor and the temporal course of contrast enhancement.

[0015] The contrast enhancement achieved with Primovist ®< can be divided into at least two phases: a dynamic phase (comprising the so-called arterial phase, portal venous phase and late phase) and the hepatobiliary phase, in which a significant uptake of Primovist ®< into the hepatocytes has already taken place.

[0016] With the contrast enhancement achieved by Primovist® during the inflow phase, typical perfusion patterns are observed, providing information for characterizing lesions. The visualization of vascularity helps characterize lesion types and determine the spatial relationship between tumor and blood vessels.

[0017] On T1-weighted MRI images, Primovist ® leads to a significant signal enhancement in healthy liver parenchyma < 10-20 minutes after injection (in the hepatobiliary phase), while lesions containing no or few hepatocytes, e.g. metastases or moderately to poorly differentiated hepatocellular carcinomas (HCCs), appear as darker areas.

[0018] Temporally monitoring the distribution of contrast agent over the dynamic and hepatobiliary phases offers a good opportunity for the detection and differential diagnosis of focal liver lesions. However, the examination takes a comparatively long time. During this period, patient movement should be avoided to minimize motion artifacts in the MRI images. This prolonged restriction of movement can be uncomfortable for the patient.

[0019] US2014 / 0376794A1 discloses a method for accelerating the acquisition of multi-channel and multi-scan data in a clinical MRI protocol by estimating correlation functions from previously acquired or reconstructed images with the same or different contrast and resolution. Using an MRI scanner, the method collects a plurality of previous MRI image scan datasets, obtains current MRI scan data, and reconstructs the current MRI scan dataset using an aggregate of the plurality of previous MRI image datasets.

[0020] The present invention provides means that allow one or more radiological images to be synthetically generated. The synthetically generated radiological images are predicted using a machine learning model based on a temporal sequence of measured radiological images that exhibit temporally varying contrast agent enhancement. This has the advantage that a radiological examination can be accelerated because not all radiological images that are important for a diagnosis have to be measured; one or more radiological images can be predicted (calculated) based on measured radiological images; the examination time can thus be shortened. Furthermore, the prediction of radiological images takes place in the frequency domain (and not, as is usual, in the spatial domain).This makes it possible to separate contrast information from detail information in radiological representations of an examination area, limit training and prediction to the contrast information, and then feed the detail information back in after the prediction. This approach reduces computational effort, for example. Furthermore, working in the frequency domain means greater tolerance for poor image registration.

[0021] A first object of the present invention is a computer-implemented method according to claim 1 comprising the steps Receiving a plurality of first representations of an examination region of an examination object in the frequency domain, wherein at least a portion of the first representations represents the examination region during a first time period after an application of a contrast agent, feeding the plurality of first representations to a prediction model, wherein the prediction model has been trained on the basis of first reference representations and second reference representations of the examination region of a plurality of examination objects, to generate one or more second reference representations from the first reference representations, at least a portion of which represents the examination region during a first time period after an application of a contrast agent in the frequency domain, which second reference representations represent the examination region during a second time period in the frequency domain,wherein the second time period is temporally after the first time period, receiving one or more predicted representations of the examination area in the frequency domain from the prediction model, wherein the one or more predicted representations represent the examination area during a second time period, wherein the second time period is temporally after the first time period, transforming the one or more predicted representations into one or more representations of the examination area in the spatial domain, outputting and / or storing the one or more representations of the examination area in the spatial domain.

[0022] Another object of the present invention is a system according to claim 9 comprising a receiving unit, a control and computing unit and an output unit, wherein the control and computing unit is configured to cause the receiving unit to receive a plurality of first representations of an examination region of an examination object in the frequency domain, wherein at least a portion of the first representations represents the examination region during a first time period after an application of a contrast agent, to supply the plurality of first representations to a prediction model, wherein the prediction model has been trained using first reference representations and second reference representations of the examination region of a plurality of examination objects, to generate one or more second reference representations from the first reference representations, at least a portion of which represents the examination region during a first time period after an application of a contrast agent in the frequency domain, which second reference representations represent the examination region during a second time period in the frequency domain,wherein the second time period is temporally after the first time period, to receive one or more predicted representations of the examination area in the frequency domain from the prediction model, wherein the one or more predicted representations represent the examination area during a second time period, wherein the second time period is temporally after the first time period, to transform the one or more predicted representations into one or more representations of the examination area in the spatial domain, to cause the output unit to output and / or store the one or more representations of the examination area in the spatial domain.

[0023] A further subject matter of the present invention is a computer program product according to claim 10 comprising a computer program that can be loaded into a working memory of a computer system and causes the computer system to carry out the following steps: Receiving a plurality of first representations of an examination region of an examination object in the frequency domain, wherein at least a portion of the first representations represents the examination region during a first time period after an application of a contrast agent, feeding the plurality of first representations to a prediction model, wherein the prediction model has been trained on the basis of first reference representations and second reference representations of the examination region of a plurality of examination objects, to generate one or more second reference representations from the first reference representations, at least a portion of which represents the examination region during a first time period after an application of a contrast agent in the frequency domain, which second time period lies after the first time period,Receiving one or more predicted representations of the examination area in frequency space from the prediction model, wherein the one or more predicted representations represent the examination area during a second time period, wherein the second time period is temporally after the first time period, Transforming the one or more predicted representations into one or more representations of the examination area in spatial space, Outputting and / or storing the one or more representations of the examination area in spatial space. ,

[0024] A further object of the present invention is a use of a contrast agent in a method according to claim 11 for predicting at least one radiological image, the method comprising the following steps: Generating a plurality of first representations of an examination region of an examination object in the frequency domain, wherein at least a portion of the first representations represents the examination region during a first time period after an application of the contrast agent, feeding the plurality of first representations to a prediction model, wherein the prediction model has been trained using first reference representations and second reference representations of the examination region of a plurality of examination objects, to generate one or more second reference representations from the first reference representations, at least a portion of which represents the examination region during a first time period after an application of a contrast agent in the frequency domain, which second time period lies after the first time period,Receiving one or more predicted representations of the examination area in frequency space from the prediction model, wherein the one or more predicted representations represent the examination area during a second time period, wherein the second time period is temporally after the first time period, Transforming the one or more predicted representations into one or more representations of the examination area in spatial space, Outputting and / or storing the one or more representations of the examination area in spatial space. ,

[0025] The invention is explained in more detail below, without distinguishing between the subject matter of the invention (method, system, computer program product, use). Rather, the following explanations are intended to apply analogously to all subject matter of the invention, regardless of the context (method, system, computer program product, use) in which they occur.

[0026] If steps are mentioned in a particular order in this description or in the claims, this does not necessarily mean that the invention is limited to that order. Rather, it is conceivable that the steps may be performed in a different order or even in parallel; unless a step builds on another step, which absolutely requires that the subsequent step be performed (which will become clear in individual cases). The specified sequences thus represent preferred embodiments of the invention.

[0027] The present invention allows the shortening of the time span of the radiological examination of an examination object.

[0028] The "object of investigation" is usually a living being, preferably a mammal, most preferably a human.

[0029] The "examination area" is usually a part of the object being examined, for example, an organ or part of an organ. The "examination area," also known as the recording volume (English: field of view , FOV), represents a volume that is depicted in radiological images. The examination area is typically defined by a radiologist, for example, on an overview image (English: localizer ). Of course, the examination area can alternatively or additionally be determined automatically, for example based on a selected protocol.

[0030] The term "radiological examination" refers to all imaging procedures that allow insight into an examination subject using electromagnetic radiation, particle radiation, or mechanical waves for diagnostic, therapeutic, and / or scientific purposes. The term "radiology" within the meaning of the present invention includes, in particular, the following examination methods: computed tomography, magnetic resonance imaging, sonography, positron emission tomography, echocardiography, and scintigraphy.

[0031] In a preferred embodiment of the present invention, the radiological examination is a magnetic resonance imaging examination.

[0032] Magnetic resonance imaging, abbreviated MRI or MR (English: Magnetic Resonance Imaging ), is an imaging technique that is used primarily in medical diagnostics to depict the structure and function of tissues and organs in the human or animal body.

[0033] In MR imaging, the magnetic moments of protons in a subject are aligned in a basic magnetic field, resulting in macroscopic magnetization along a longitudinal direction. This magnetization is then deflected from its rest position by applying radiofrequency (RF) pulses (excitation). The return of the excited states to their rest position (relaxation), or the magnetization dynamics, is subsequently detected as relaxation signals using one or more RF receiver coils.

[0034] For spatial encoding, rapidly switched magnetic gradient fields are superimposed on the basic magnetic field. The acquired relaxation signals, or the detected and spatially resolved MR data, are initially available as raw data in frequency domain and can be transformed into spatial domain (image space), for example, by subsequent inverse Fourier transformation.

[0035] In native MRI, tissue contrasts are generated by the different relaxation times (T1 and T2) and the proton density. T1 relaxation describes the transition of the longitudinal magnetization to its equilibrium state, where T1 is the time required to reach 63.21% of the equilibrium magnetization before resonance excitation. It is also called the longitudinal relaxation time or spin-lattice relaxation time. Similarly, T2 relaxation describes the transition of the transverse magnetization to its equilibrium state.

[0036] In radiological examinations, contrast agents are often used to enhance contrast.

[0037] "Contrast agents" are substances or mixtures of substances that enhance the visualization of structures and functions of the body in imaging procedures such as X-ray diagnostics, magnetic resonance imaging and sonography.

[0038] Beispiele für Kontrastmittel sind in der Literatur zu finden (siehe z.B. A. S. L. Jascinth et al.: Contrast Agents in computed tomography: A Review, Journal of Applied Dental and Medical Sciences, 2016, Vol. 2, Issue 2, 143 - 149; H. Lusic et al.: X-ray-Computed Tomography Contrast Agents, Chem. Rev. 2013, 113, 3, 1641-1666; https: / / www.radiology.wisc.edu / wp-content / uploads / 2017 / 10 / contrast-agents-tutorial.pdf, M. R. Nough et al.: Radiographie and magnetic resonances contrast agents: Essentials and tips for safe practices, World J Radiol. 2017 Sep 28; 9(9): 339-349; L. C. Abonyi et al.: Intravascular Contrast Media in Radiography: Historical Development & Review of Risk Factors for Adverse Reactions, South American Journal of Clinical Research, 2016, Vol. 3, Issue 1, 1-10; ACR Manual on Contrast Media, 2020, ISBN: 978-1-55903-012-0; A. Ignee et al.: Ultrasound contrast agents, Endosc Ultrasound. 2016 Nov-Dec; 5(6): 355-362).

[0039] MR contrast agents exert their effect by altering the relaxation times of the structures that absorb the contrast agent. Two groups of substances can be distinguished: paramagnetic and superparamagnetic substances. Both groups of substances contain unpaired electrons that induce a magnetic field around the individual atoms or molecules. Superparamagnetic contrast agents lead to a predominant T2 shortening, while paramagnetic contrast agents essentially lead to a T1 shortening. The effect of these contrast agents is indirect, as the contrast agent itself does not emit a signal but only influences the signal intensity in its surroundings. An example of a superparamagnetic contrast agent is iron oxide nanoparticles (SPIO). superparamagnetic iron oxide Examples of paramagnetic contrast agents are gadolinium chelates such as gadopentetate dimeglumine (trade name: Magnevist ®< among others), gadoteric acid (Dotarem ®< , Dotagita ®< , Cyclolux ®< ), gadodiamide (Omniscan ®< ), gadoteridol (ProHance ®< ) and gadobutrol (Gadovist ®< ).

[0040] With the aid of the present invention, one or more synthetic radiological images of an examination region can be predicted. The prediction is carried out using a predictive model.

[0041] The prediction model is a computer-aided model configured to predict one or more second representations of the examination region of the examination object in the frequency domain based on a plurality of first representations of an examination region of an examination object in the frequency domain. At least a portion of the plurality of first representations represents the examination region during a first time period after the application of a contrast agent. The at least one second representation represents the examination region during a second time period.

[0042] The term "plurality" means a number of at least two. Typically, the plurality of initial representations used for prediction is no greater than ten.

[0043] The second period occurs after the first period.

[0044] Fig. 1(a) und Fig. 1(b) serve to clarify. In Fig. 1(a) und Fig. 1(b) Timelines are shown. Defined points in time are marked on the timelines. The point in time t 0 indicates the time at which a contrast agent is administered to a subject. The dashed frame indicates the period of time during which the subject undergoes a radiological examination, for example, the time the subject spends in a magnetic resonance imaging (MRI) or computed tomography (CT) scanner.

[0045] Fig. 1(a) schematically illustrates a typical course of a radiological examination. The object to be examined is placed in a tomograph. At the time t -1 the object under examination is in the tomograph. At the time t -1, a first radiological image is generated, ie, a representation of an examination area of ​​the sub-object is created. At this time ( t-1 ) no contrast agent has been administered to the examination subject, ie it is a contrast-free (native) representation. At the time t 0, a contrast agent is applied to the examination object in the tomograph. At the times t 1 , t 2 and t 3, further representations of the object under investigation are generated. The object under investigation then leaves the tomograph. The object under investigation was in the same room for a comparatively long period of time. T t in the tomograph. During this time, four representations of the examination area were generated by measurement.

[0046] Fig. 1(b) schematically illustrates a course of a radiological examination according to the present invention. The examination subject is placed in a tomograph. At the time t -1 the object under examination is in the tomograph. At the time t-1, a first representation of an examination area of ​​the sub-object is created. At this time ( t -1 ) no contrast agent has been administered to the examination subject, ie it is a contrast-free (native) representation. At the time t 0, a contrast agent is applied to the examination object in the tomograph. At the times t 1 and t 2, further representations of the object under investigation are generated. The object then leaves the tomograph. The representation, which in the case of Fig. 1(a) at the time t 3 has been generated by measurement technology, in the case of the Fig. 1(b) predicted (calculated) of the course shown. The object under investigation was located over a period T a in the tomograph, whereby this time period T a is shorter than the time span T t in Fig. 1(a) . The radiological examination takes in case of Fig. 1(b) not as long as in the case of Fig. 1(a) , it is therefore more pleasant for the object of investigation; however, in both cases representations of the investigation area are created, which represent the investigation area at the times t -1 , t 1 , t 2 and t 3. The representations that were made at the times t -1 , t 1 and t 2 were generated by measurement technology, are first representations of an examination area of ​​an examination object within the meaning of the present invention, in which at least a part represents an examination area during a first period of time after an application of a contrast agent, namely the representations at the times t 1 , t 2 and t 3 ; the representation at the time t-1 represents the examination area in a period before the application of the contrast agent. The representation at the time t 3 is calculated by the prediction model according to the invention on the basis of the representations t -1 , t 1 and t 2. It represents the study area during a second period, where in the case of Fig. 1(b) the second period follows the first period.

[0047] In a preferred embodiment of the present invention, the first time period begins before the application of the contrast agent or with the application of the contrast agent. It is advantageous if one or more representations of the examination area are generated that show the examination area without contrast agent (native images), since a radiologist can already obtain important information about the health status of the examination subject from such images. For example, a radiologist can detect bleeding in native MRI images.

[0048] In order for the prediction model according to the invention to be able to make the predictions described here, it must be configured accordingly in advance.

[0049] The term "prediction" means that at least one representation of an examination region, which represents the examination region during a second time period in the frequency domain, is calculated using a plurality of first representations of the examination region in the frequency domain, wherein at least a part of the plurality of first representations represents the examination region during a first time period after an application of a contrast agent.

[0050] The prediction model is preferably created (configured, trained) using a self-learning algorithm in a supervised machine learning process. Training data is used for learning. This training data comprises a plurality of representations of an examination area for each examination object of a plurality of examination objects. The examination area is usually the same for all examination objects (e.g., a part of a human body or an organ or a part of an organ). The representations of the training data set are also referred to as reference representations in this description. The term "plurality" preferably means more than 10, even more preferably more than 100.

[0051] For each examination object, the training data comprises i) a plurality of first reference representations of the examination region in the frequency domain, at least a part of which represents the examination region during a first time period after application of a contrast agent, and ii) one or more second reference representations of the examination region in the frequency domain, which represent the examination region during a second time period.

[0052] The prediction model is trained to predict (calculate) the one or more second reference representations from the plurality of first reference representations for each object of study.

[0053] In machine learning, the self-learning algorithm creates a statistical model based on the training data. This means that the algorithm doesn't simply memorize the examples; instead, it "recognizes" patterns and regularities in the training data. This allows the prediction model to evaluate even unknown data. Validation data can be used to test the accuracy of the evaluation of unknown data.

[0054] The prediction model can be trained using supervised learning (SLL). supervised learning ), i.e., pairs of data sets (first and second representations) are presented to the algorithm one after the other. The algorithm then learns a relationship between the first representations and the second representations.

[0055] Self-learning systems that are trained using supervised learning are widely described in the state of the art (see, for example, C. Perez: Machine Learning Techniques: Supervised Learning and Classification, Amazon Digital Services LLC - Kdp Print Us, 2019, ISBN 1096996545, 9781096996545).

[0056] Preferably, the prediction model is or includes an artificial neural network.

[0057] An artificial neural network comprises at least three layers of processing elements: a first layer with input neurons (nodes), an Nth layer with at least one output neuron (node), and N-2 inner layers, where N is a natural number and greater than 2.

[0058] The input neurons are used to receive first representations. The output neurons are used to output one or more second representations for a plurality of first representations.

[0059] The processing elements of the layers between the input neurons and the output neurons are connected in a predetermined pattern with predetermined connection weights.

[0060] The artificial neural network is preferably a so-called convolutional neural network (CNN for short).

[0061] A convolutional neural network is capable of processing input data in the form of a matrix. A CNN essentially consists of filters (convolutional layer) and aggregation layers (pooling layer) that alternately repeat, and finally, one or more layers of "normal" fully connected neurons (dense / fully connected layer).

[0062] The neural network can be trained, for example, using a backpropagation method. The goal is to achieve the most reliable mapping possible from given input data to given output data. The quality of the mapping is determined by an error function (English: loss function ). The goal is to minimize the error function. The backpropagation method trains an artificial neural network by changing the connection weights.

[0063] In the trained state, the connection weights between the processing elements contain information regarding the relationship between first representations and one or more second representations, which can be used to predict one or more second representations showing an examination area during a second time period for a new plurality of first representations (e.g., of a new examination object), at least a part of which shows the examination area during a first time period after application of a contrast agent.

[0064] A cross-validation method can be used to split the data into training and validation sets. The training set is used for backpropagation training of the network weights. The validation set is used to test the predictive accuracy of the trained network when applied to unknown data.

[0065] Further details on setting up and training artificial neural networks can be found in the state of the art (see, for example: S. Khan et al.: A Guide to Convolutional Neural Networks for Computer Vision, Morgan & Claypool Publishers 2018, ISBN 1681730227, 9781681730226, WO2018 / 183044A1, WO2018 / 200493, WO2019 / 074938A1, WO2019 / 204406A1, WO2019 / 241659A1).

[0066] Preferably, the prediction model is a Generative Adversarial Network (GAN) (see e.g.: http: / / 3dgan.csail.mit.edu / ).

[0067] In addition to the representations, further information about the object under investigation, the examination area, the examination conditions and / or the radiological examinations can be used for training, validation and prediction.

[0068] Examples of information about the subject being examined include: gender, age, weight, height, medical history, type, duration, and amount of medications already taken, blood pressure, central venous pressure, respiratory rate, serum albumin, total bilirubin, blood sugar, iron levels, respiratory capacity, and the like. This information can be retrieved from a database or an electronic patient record, for example.

[0069] Examples of information on the area of ​​investigation include: previous illnesses, operations, partial resection, liver transplantation, iron liver, fatty liver and the like.

[0070] As already described, the representations of the study area used for training, validation and prediction are representations of the study area in frequency space (also called spatial frequency space or Fourier space or frequency domain or Fourier representation).

[0071] In magnetic resonance imaging, the raw data is usually generated as so-called k-space data due to the measurement method described above. This k-space data is a representation of an examination region in a frequency domain; i.e., such k-space data can be used for training, validation, and prediction. If representations are present in the spatial domain, such representations in the spatial domain can be converted (transformed) into a representation in the frequency domain using Fourier transformation; conversely, representations in the frequency domain can be converted (transformed) into a representation in the spatial domain, for example, using inverse Fourier transformation.

[0072] If a radiological image of an examination area is available in the form of a two-dimensional image in spatial space, this representation of the examination area can be converted into a two-dimensional representation of the examination area in frequency space by means of a 2D Fourier transformation.

[0073] A three-dimensional image (volume representation) of an examination area can be treated as a stack of two-dimensional images. It is also conceivable that the three-dimensional image is converted into a three-dimensional representation of the examination area in the frequency domain using a 3D Fourier transform.

[0074] It is also conceivable to use a transformation other than the Fourier transform to convert position-space representations into frequency-space representations. The three main properties that such a transformation must satisfy are: a) Existence of a unique inverse transformation (unambiguous relationship between spatial and frequency domain representation) b) Locality of contrast information c) Robustness against insufficient image registration

[0075] Details on the transformation from one representation to another are described in a variety of specialist books and publications (see, for example: W. Burger, MJ Burge: Digital Image Processing: An Algorithmic Introduction Using Java, Texts in Computer Science, 2nd edition, Springer-Verlag, 2016, ISBN: 9781447166849; W. Birkfellner: Applied Medical Image Processing, Second Edition: A Basic Course, Taylor & Francis, 2014, ISBN: 9781466555570; R. Bracewell: Fourier Analysis and Imaging, Springer Science & Business Media, 2004, ISBN: 9780306481871).

[0076] Fig. 2 shows an example and schematically the generation of representations of an investigation area in the spatial space and in the frequency space.

[0077] In Fig. 2 A timeline is shown. At three different points in time t 1 , t 2 and t 3, measurement-based representations of an examination area are generated. The examination area is the lung of a human. At the time t 1, a first representation is generated. This can be a representation (O1) of the examination area (lung) in spatial space or a representation (F1) of the examination area (lung) in frequency space. The representation (O1) of the examination area in spatial space can be converted to a Fourier transform. FT into a representation (F1) of the examination area in the frequency domain. The representation (F1) of the examination area in the frequency domain can be converted using inverse Fourier transformation iFT into a representation (O1) of the study area in spatial space. The two representations (O1) and (F1) contain the same information about the study area, only in a different representation. At the time t 2, a further representation is generated. This can be a representation (O2) of the examination area (lung) in spatial space or a representation (F2) of the examination area (lung) in frequency space. The representation (O2) of the examination area in spatial space can be converted to a Fourier transform. FT into a representation (F2) of the examination area in the frequency domain. The representation (F2) of the examination area in the frequency domain can be converted using inverse Fourier transformation iFT into a representation (O2) of the study area in spatial space. The two representations (O2) and (F2) contain the same information about the study area, only in a different representation. At the time t 3, a further representation is generated. This can be a representation (O3) of the examination area (lung) in spatial space or a representation (F3) of the examination area (lung) in frequency space. The representation (O3) of the examination area in spatial space can be converted to a Fourier transform. FT into a representation (F3) of the examination area in the frequency domain. The representation (F3) of the examination area in the frequency domain can be converted using inverse Fourier transformation iFT into a representation (O3) of the study area in spatial space. The two representations (O3) and (F3) contain the same information about the study area, only in a different representation.

[0078] For humans, representations (O1), (O2), and (O3) of the examination area in spatial space are the most common representations; humans can perceive these directly. Representations (O1), (O2), and (O3) show how contrast agent is dynamically distributed in the veins. The same information is contained in representations (F1), (F2), and (F3), but is more difficult for humans to perceive.

[0079] Fig. 3 shows schematically and exemplarily how the Fig. 2 The generated representations of the examination area (F1), (F2), and (F3) in the frequency domain can be used to train a prediction model (PM). The representations (F1), (F2), and (F3) form a set of training data for a study object. Training is performed using a plurality of training data sets from a plurality of study objects.

[0080] Representations (F1) and (F2) are a plurality (in this case, two) of first reference representations of the examination area in the frequency domain, at least a portion of which represents the examination area during a first time period after application of a contrast agent. Representation (F3) is a second reference representation of the examination area in the frequency domain, which represents the examination area during a second time period. The prediction model is Fig. 3 trained to predict the representation (F3) of the examination area in frequency space from the representations (F1) and (F2) of the examination area in frequency space. The representations (F1) and (F2) are input into the prediction model (PM), and the prediction model calculates a representation (F3*) from the representations (F1) and (F2). The asterisk (*) signals that the representation (F3*) is a predicted representation. The calculated representation (F3*) is compared with the representation (F3). The deviations can be used in a backpropagation process to train the prediction model and reduce the deviations to a defined minimum. If the prediction model has been trained using a large number of training data sets for a large number of examination objects, and the prediction has achieved a defined accuracy, the trained prediction model can be used for prediction.This is shown schematically and exemplarily in . Fig. 4 shown.

[0081] Fig. 4 shows that in Fig. 3 trained prediction model (PM). The prediction model is used to generate one or more second representations of the examination region in the frequency domain based on a plurality of first representations of the examination region in the frequency domain, at least some of which represent the examination region in a first time period after application of a contrast agent, wherein the one or more second representations represent the examination region in a second time period.

[0082] In the present example, two initial representations (F̃1) and (F̃2) of the examination area in the frequency domain are input into the prediction model (PM), and the prediction model (PM) generates (calculates) a second representation (F̃3*). The tilde (~) signals that the representations are representations of a new examination object, for which no representations are usually available that have been used in the training procedure for training the prediction model. The asterisk (*) signals that the representation (F̃3*) is a predicted representation. The representation (F̃3*) of the examination area in the frequency domain can, for example, be calculated using inverse Fourier transformation. iFT be transformed into a representation (Õ3*) of the study area in spatial space.

[0083] The use of representations of the study area in frequency space has advantages over the use of representations of the study area in spatial space (also image space or " image space When using representations of the study area in the frequency domain, contrast information, which is important for training and prediction, can be separated from detailed information (fine structures). It is therefore possible to focus on the information that the prediction model should learn during training and to also focus on the information that the prediction model should predict during prediction: contrast information.

[0084] While contrast information in a spatial-space representation of an examination area is typically distributed throughout the entire representation (each pixel / voxel carries contrast information), the contrast information in a frequency-space representation of an examination area is encoded in and around the center of the frequency space. In other words, the low frequencies in a frequency-space representation are responsible for contrast, while the high frequencies contain information about fine structures.

[0085] This makes it possible to separate the contrast information, limit training and prediction to the contrast information, and feed back information about the fine structures after training / prediction.

[0086] In a preferred embodiment, the method according to the invention comprises the following steps: Receiving a plurality of first representations of an examination region of an examination object in the frequency domain, wherein at least a portion of the first representations represents the examination region during a first time period after an application of a contrast agent, specifying a region in the first representations, wherein the specified region comprises the center of the frequency domain, reducing the first representations to the specified region, wherein a plurality of reduced first representations are obtained, supplying the plurality of reduced first representations to a prediction model, receiving one or more second representations of the examination region in the frequency domain from the prediction model, wherein the one or more second representations represent the examination region during a second time period,Supplementing the one or more second representations with one or more regions of the received first representations that lie outside the specified region, thereby obtaining one or more supplemented second representations, transforming the one or more supplemented second representations into one or more representations of the examination region in the spatial space, outputting and / or storing the one or more representations of the examination region in the spatial space.

[0087] The region can be specified in the first representations, for example, by a user of the computer system according to the invention entering one or more parameters into the computer system according to the invention and / or selecting them from a list that define the shape and / or size of the region. However, it is also conceivable that the specification occurs automatically, for example, by the computer system according to the invention, which has been configured accordingly to select a predefined region in the representations of the examination region.

[0088] The specified range is usually smaller than the frequency space filled by the first representations, but in any case includes the center of the frequency space.

[0089] A region of the frequency space encompassing the center of the frequency space (also called the origin or zero point) contains the contrast information relevant for the method according to the invention. If the specified region is smaller than the frequency space filled by the first representations, the subsequent prediction requires less computational effort (this applies in particular to the training of the prediction model). The size of the region can therefore directly influence the computational effort.

[0090] In principle, it is also possible to specify a range corresponding to the entire frequency space filled by the first representations; in such a case, no reduction to a subrange of the frequency space takes place and the computational effort is maximized.

[0091] By specifying a region around the center of the frequency domain, the user of the computer system according to the invention can decide for themselves whether they want training and prediction based on the complete representations of the examination region in the frequency domain or whether they want to reduce the computational effort. The size of the specified region can directly influence the required computational effort.

[0092] The specified region typically has the same dimension as the frequency space: in the case of a 2D representation in a 2D frequency space, the specified region is typically a surface; in the case of a 3D representation in a 3D frequency space, the specified region is typically a volume.

[0093] The specified area can, in principle, have any shape; for example, it can be round and / or square, concave and / or convex. The area is preferably cuboid or cube-shaped in the case of a 3D frequency space in a Cartesian coordinate system, and rectangular or square in the case of a 2D frequency space in a Cartesian coordinate system. However, it can also be spherical or circular, or have another shape.

[0094] Preferably, the geometric center of gravity of the specified area coincides with the center of the frequency space.

[0095] The representations used for training, validation, and prediction are reduced to the specified range. The term "reducing" means that all parts of a representation that do not lie within the specified range are cut away (discarded) or covered by a mask. Masking involves covering the areas that lie outside the specified range with a mask, leaving only the specified area uncovered. When covering with a mask, for example, the color values ​​of the corresponding pixels / voxels can be set to zero (black).

[0096] The representations obtained in this way are also referred to as reduced representations in this description.

[0097] The first representations obtained after reduction (reduced first representations) are fed into the prediction model: The prediction model has been pre-trained in a training procedure to learn the dynamic influence of the amount of contrast agent on representations of the examination area in the frequency domain. Reduced representations (reduced first representations and reduced second representations) are also preferably used in training.

[0098] The prediction model has thus learned the dynamic influence of contrast agent on a representation of the examination area and can apply this learned "knowledge" to predict one or more (reduced) second representations based on the (reduced) first representations.

[0099] The one or more predicted second representations represent the region of interest in frequency space during a second time period.

[0100] The one or more predicted second representations are calculated and output by the prediction model based on the (reduced) first representations.

[0101] In the event that at least one predicted second representation has been generated on the basis of reduced first representations, it is now useful to add back the previously discarded (cut away or covered with a mask) parts in order to avoid losing any information on fine structures in the final artificially generated image.

[0102] To reuse the previously discarded (cut away or masked) portions, the at least one predicted second representation can be superimposed on at least one received first representation such that the at least one predicted second representation replaces the corresponding superimposed frequency ranges of the at least one originally received first representation. Preferably, the predicted second representation replaces the corresponding frequency ranges of the originally received first representation representing the examination region without contrast agent.

[0103] Replacing the superimposed frequency range corresponds to supplementing it with one or more ranges of the frequency space of the received first representations that were omitted when reducing the first representations to the specified range.

[0104] In other words: the frequency space of the at least one predicted second representation of the examination area is filled by those areas of at least one originally received first representation by which the at least one originally received first representation is larger than the predicted second representation.

[0105] By using frequency-domain representations of the study area, it is possible to separate contrast information from detail information, limit training and prediction to the contrast information, and then add the detail information back after training and / or prediction. As already described, this approach reduces the computational effort during training, validation, and prediction.

[0106] However, working in frequency space has another advantage over working in spatial space: the co-registration of the individual representations is less critical in frequency space than in spatial space. "Co-registration" (also called "image registration" in the state of the art) is an important process in digital image processing and serves to optimally match two or more images of the same scene, or at least similar scenes. One of the images is designated as the reference image, and the others are called object images. To optimally adapt these object images to the reference image, a compensating transformation is calculated. The images to be registered differ from one another because they were acquired from different positions, at different times, and / or with different sensors.

[0107] In the case of the present invention, the individual representations of the plurality of first representations of the examination area were, on the one hand, generated at different times; on the other hand, they differ with regard to the content and distribution of contrast agent in the examination area.

[0108] The advantage of using representations of the examination area in frequency space over representations of the examination area in spatial space is that the training, validation, and prediction procedures are more tolerant of image registration errors. In other words, if representations in frequency space are not precisely superimposed, this has less impact than if representations in spatial space are not precisely superimposed. This follows from the properties of the Fourier transform: as already described, the contrast information of Fourier-transformed images is always mapped near the origin (center) of Fourier space. Rotations or rotations in image space (spatial space) result in image information (e.g., a visible structure) being localized in a different area of ​​the image after the transformation.However, in Fourier space, these transformations do not change the region in which the contrast information relevant to the present invention is encoded.

[0109] Fig. 5 shows, by way of example and schematically, a step in training a prediction model according to a preferred embodiment of the present invention.

[0110] Two first representations (F1) and (F2) of an object under investigation in the frequency domain and a second representation (F3) of the object under investigation in the frequency domain are received. In the representations (F1), (F2) and (F3) the same area is A specified. The area A encompasses the center of the frequency space and, in this case, has a square shape, with the geometric center of gravity of the square coinciding with the center of the frequency space. The representations (F1), (F2), and (F3) are mapped to the respective specified range. Areduced: this results in three reduced representations (F1 red ), (F2 red ) and (F3 red ). The reduced representations are used for training. The prediction model is trained to predict the reduced representation (F3 red ) from the reduced representations (F1 red ) and (F2 red ). The reduced representations (F1 red ) and (F2 red ) are fed to the prediction model (PM), and the prediction model calculates a reduced representation (F3* red ) that should be as close as possible to the reduced representation (F3 red ).

[0111] Fig. 6 shows, by way of example and schematically, how Fig. 5 trained prediction model can be used for prediction.

[0112] In the present example, two first representations (F̃1) and (F̃2) of the examination area are received in the frequency domain and each mapped to a specified area Areduced. This results in two reduced first representations (F̃1 red ) and (F̃2 red ). The reduced first representations (F̃1 red ) and (F̃2 red ) are fed to the trained prediction model (PM). The trained prediction model (PM) calculates a reduced second representation (F̃3 red *) from the reduced first representations (F̃1 red ) and (F̃2 red ). In a further step, the reduced second representation (F̃3 red *) is supplemented by that area (F̃1 DI ) of the received first representation (F̃1) that was discarded when reducing the received first representation (F̃1) (the area that lies outside the specified area A As described, instead of or in addition to parts of the received first representation (F̃1), parts of the received second representation (F̃2) can also be added to the reduced third representation (F̃3 red *).

[0113] From the supplemented representation (F̃3 red *) + (F̃1 DI ) a representation of the investigation area in the position space (Õ3*) can be generated by inverse Fourier transformation.

[0114] It should be noted that other methods can also be used to transform a frequency-space representation into a position-space representation, such as iterative reconstruction methods.

[0115] The method according to the invention can be carried out using a computer system. A computer system configured (e.g., using the computer program according to the invention) to carry out the method according to the invention is a further subject of the present invention.

[0116] Fig. 7 shows schematically and by way of example an embodiment of the computer system according to the invention. The computer system (10) comprises a receiving unit (11), a control and computing unit (12), and an output unit (13).

[0117] A "computer system" is an electronic data processing system that processes data using programmable computing instructions. Such a system typically includes a control and processing unit, often referred to as a "computer," which includes a processor for performing logical operations and a RAM for loading a computer program, as well as peripherals.

[0118] In computer technology, "peripherals" refers to all devices connected to a computer that serve to control the computer and / or act as input and output devices. Examples include monitors (screens), printers, scanners, mice, keyboards, joysticks, drives, cameras, microphones, speakers, etc. Internal connectors and expansion cards are also considered peripherals in computer technology.

[0119] Today's computer systems are often divided into desktop PCs, portable PCs, laptops, notebooks, netbooks and tablet PCs and so-called handhelds (e.g., smartphones); all of these systems can be used to implement the invention.

[0120] Inputs to the computer system (for example, for control by a user) are made via input devices such as a keyboard, a mouse, a microphone, a touch-sensitive display, and / or the like. Outputs are made via the output unit (13), which can in particular be a monitor (screen), a printer, and / or a data storage device.

[0121] The computer system (10) according to the invention is configured to predict one or more second representations of the examination region of the examination object in the frequency domain from a plurality of first representations of an examination region of an examination object in the frequency domain, which represent the examination region during a first time period after an application of a contrast agent, wherein the one or more second representations represent / represent the examination region during a second time period.

[0122] The control and computing unit (12) serves to control the receiving unit (11) and the output unit (13), coordinate the data and signal flows between the various units, process representations of the examination area, and generate artificial radiological images. It is conceivable that multiple control and computing units may be present.

[0123] The receiving unit (11) serves to receive representations of an examination area. The representations can, for example, be transmitted from a magnetic resonance tomograph or a computed tomograph or read from a data memory. The magnetic resonance tomograph or the computed tomograph can be a component of the computer system according to the invention. However, it is also conceivable that the computer system according to the invention is a component of a magnetic resonance tomograph or a computed tomograph. The transmission of representations can take place via a network connection or a direct connection. The transmission of representations can take place via a radio connection (WLAN, Bluetooth, mobile radio and / or the like) and / or via a cable. It is conceivable that several receiving units are present. The data memory can also be a component of the computer system according to the invention or can be linked to it, for example.be connected via a network. It is conceivable that multiple data storage devices exist.

[0124] The receiving unit receives the representations and, if applicable, other data (such as information on the object under examination, recording parameters and / or the like) and transmits them to the control and computing unit.

[0125] The control and computing unit is configured to generate artificial radiological images based on the received data.

[0126] The output unit (13) can be used to display the artificial radiological images (e.g., on a monitor), output them (e.g., via a printer), or store them in a data storage device. Multiple output units are conceivable.

[0127] Fig. 8 shows, by way of example in the form of a flow chart, a preferred embodiment of the method according to the invention for training a prediction model.

[0128] The method (100) comprises the following steps: (110) Receiving training data, wherein the training data for each examination object of a plurality of examination objects comprises i) a plurality of first reference representations of an examination region in the frequency domain, at least a part of which represents the examination region during a first time period after an application of a contrast agent, and ii) one or more second reference representations of the examination region in the frequency domain, which represent the examination region during a second time period, (120) for each examination object: supplying the plurality of first reference representations to a prediction model, wherein the prediction model is trained to generate one or more second reference representations based on the plurality of first reference representations, wherein the training comprises minimizing an error function,wherein the error function quantifies deviations of the generated second reference representation(s) from the one or more received second reference representation(s), (130) outputting and / or storing the trained prediction model and / or supplying the trained prediction model to a method for predicting one or more representations of the examination area of ​​a new examination object. ,

[0129] Fig. 9 shows, by way of example in the form of a flow chart, a further preferred embodiment of the method according to the invention for training a prediction model.

[0130] The method (200) comprises the following steps: (210) Receiving training data, wherein the training data for each examination object of a plurality of examination objects comprises i) a plurality of first reference representations of an examination region in the frequency domain, at least a part of which represents the examination region during a first time period after an application of a contrast agent, and ii) one or more second reference representations of the examination region in the frequency domain, which represent the examination region during a second time period, (220) Specifying a region in the first reference representations, wherein the specified region comprises the center of the frequency domain, (230) Reducing the reference representations to the specified region, wherein for each examination object a plurality of reduced first reference representations and one or more reduced second reference representation(s) are obtained,(240) for each examination object: feeding the plurality of reduced first reference representations to a prediction model, wherein the prediction model is trained to generate one or more reduced second reference representations based on the plurality of reduced first reference representations, wherein the training comprises minimizing an error function, wherein the error function quantifies deviations of the generated reduced second reference representation(s) from the one or more reduced second reference representation(s) of the training data, (250) outputting and / or storing the trained prediction model and / or feeding the trained prediction model to a method for predicting one or more representations of the examination area of ​​a new examination object.

[0131] Fig. 10 shows, by way of example in the form of a flow chart, a preferred embodiment of the method according to the invention for predicting one or more representations.

[0132] The method (300) comprises the following steps: (310) Providing a prediction model, wherein the prediction model has been trained according to the method (100) described above, (320) Receiving a plurality of first representations of an examination region of an examination object in the frequency domain, wherein at least a portion of the first representations represents the examination region during a first time period after an application of a contrast agent, (330) Supplying the plurality of first representations to the prediction model, (340) Receiving one or more predicted representations of the examination region in the frequency domain from the prediction model, wherein the one or more predicted representations represent the examination region during a second time period, (350) Transforming the one or more predicted representations into one or more representations of the examination region in the spatial domain,(360) Outputting and / or storing the one or more representations of the area of ​​investigation in spatial space. ,

[0133] Fig. 11 shows, by way of example in the form of a flow chart, a further preferred embodiment of the method according to the invention for predicting one or more representations.

[0134] The method (400) comprises the following steps: (410) Providing a prediction model, wherein the prediction model has been trained according to the method (200) described above, (420) Receiving a plurality of first representations of an examination region of an examination object in the frequency domain, wherein at least a portion of the first representations represents the examination region during a first time period after an application of a contrast agent, (430) Specifying a region in the first representations, wherein the specified region comprises the center of the frequency domain, (440) Reducing the first representations to the specified region, wherein a plurality of reduced first representations are obtained, (450) Feeding the plurality of reduced first representations to the prediction model, (460) Receiving one or more second representations of the examination region in the frequency domain from the prediction model,wherein the one or more second representations represent the examination area during a second period of time, (470) supplementing the one or more second representations with one or more areas of the received first representations that lie outside the specified area, thereby obtaining one or more supplemented second representations, (480) transforming the one or more supplemented second representations into one or more representations of the examination area in the spatial space, (490) outputting and / or storing the one or more representations of the examination area in the spatial space.

[0135] Below are some examples of how the present invention can be used to generate artificial radiological images. Beispiel 1

[0136] In one embodiment, the present invention is used to deliver an intravascular contrast agent (also known as a blood pool contrast agent, blood pool (contrast) agent referred to).

[0137] When generating radiological images with a comparatively long acquisition / scan time, such as free-breathing images of the thorax and abdomen to visualize the vascular system (e.g., free-breathing pulmonary embolism diagnostics in MRI), an extracellular contrast agent is excreted relatively quickly from the vascular system, resulting in a rapid decrease in contrast. However, it would be advantageous to be able to maintain the contrast for a longer period.

[0138] To solve this problem, in a first step, a plurality of first representations of an examination area are generated / received in the frequency domain, wherein at least some of the first representations represent the examination area after application of a contrast agent.

[0139] The applied contrast agent can be an extracellular and / or an intracellular contrast agent.

[0140] The contrast agent is preferably injected into a blood vessel of the subject being examined, such as a vein in the arm, in a weight-adjusted manner. From there, it moves with the blood along the circulatory system.

[0141] The "circulatory system" is the path that blood takes in the body of humans and most animals. It is the blood flow system formed by the heart and a network of blood vessels (cardiovascular system).

[0142] An extracellular contrast agent circulates in the bloodstream for a period of time that depends on the object being examined, the contrast agent, and the amount administered, while it is continuously excreted from the bloodstream via the kidneys.

[0143] During the distribution and / or circulation of the contrast agent in the vascular system of the subject, at least one first representation of the vascular system or a portion thereof is acquired. Multiple first representations may be acquired, representing different phases of the distribution of the contrast agent in the vascular system or a portion thereof (e.g., inflow phase, arterial phase, venous phase, and / or the like). Acquiring multiple images allows for subsequent differentiation of blood vessel types.

[0144] The measured representations represent the vascular system or a portion thereof with contrast enhancement compared to the surrounding tissue. Preferably, at least one representation shows arteries with contrast enhancement (arterial phase), while at least one other representation shows veins with contrast enhancement (venous phase).

[0145] Artificial representations are generated based on the measured representations. The artificial representations preferably show the same examination area as the measured representations. If a plurality of measured representations of the examination area were acquired at different times after the application of the contrast agent, the later representations in particular show blood vessels with a progressively decreasing contrast compared to the surrounding tissue, as the contrast agent is gradually excreted from the blood vessels. The artificial representations, on the other hand, show the blood vessels with a consistently high contrast compared to the surrounding tissue.

[0146] This is achieved by feeding the measured representations to the prediction model according to the invention, which has been previously trained to predict several representations that show a temporally constant contrast enhancement of blood vessels on the basis of measured representations that show a temporally varying contrast enhancement of blood vessels.

[0147] The reference data used to train and validate such a prediction model typically includes measured representations of the examination area after the application of an extracellular or intracellular contrast agent. The reference data may also include representations of the examination area after the application of a blood pool contrast agent. Such reference data may, for example, be determined in a clinical study. Ferumoxytol, for example, may be used as an intravascular contrast agent in such a clinical study. Ferumoxytol is a colloidal iron-carbohydrate complex approved for the parenteral treatment of iron deficiency in chronic kidney disease when oral therapy is not feasible. Ferumoxytol is administered as an intravenous injection. Ferumoxytol is commercially available as a solution for intravenous injection under the brand names Rienso® or Ferahme®.The iron-carbohydrate complex exhibits superparamagnetic properties and can therefore (. off label ) can be used for contrast enhancement in MRI examinations (see e.g.: LP Smits et al.: Evaluation of ultrasmall superparamagnetic iron-oxide (USPIO) enhanced MRI with ferumoxytol to quantify arterial wall inflammation, Atherosclerosis 2017, 263: 211-218).

[0148] It is also conceivable to use representations after application of the intravascular contrast agent Ablavar ®< as training data.

[0149] It is also conceivable to synthetically generate the reference representations, which show blood vessels in the examination area with a contrast enhancement that remains constant over time, e.g., using segmentation methods based on the first representations. A wide variety of segmentation methods are described in the literature. The following publications are examples: F. Conversano et al.: Hepatic Vessel Segmentation for 3D Planning of Liver Surgery, Acad Radiol 2011, 18: 461-470; S. Moccia et al.: Blood vessel segmentation algorithms - Review of methods, datasets and evaluation metries, Computer Methods and Programs in Biomedicine 158 (2018) 71-91; M. Marcan et al.: Segmentation of hepatic vessels from MRI images for planning of electroporation-based treatments in the liver, Radiol Oncol 2014; 48(3): 267-281; ​​TA Hope et al.: Improvement of Gadoxetate Arterial Phase Capture With a High Spatio-Temporal Resolution Multiphase Three-Dimensional SPGR-Dixon Sequence, Journal of Magnetic Resonance Imaging 38: 938-945 (2013); WO2009 / 135923A1, US6754376B1, WO2014 / 162273A1, WO2017 / 139110A1, WO2007 / 053676A2, EP2750102A1).

[0150] The trained prediction model then generates second representations based on the supplied first representations, which show a temporally constant contrast enhancement of the blood vessels. Beispiel 2

[0151] In a preferred embodiment, the present invention is used to generate (predict) one or more artificial MRI images in dynamic contrast-enhanced magnetic resonance imaging.

[0152] The term "recording" is used in the following text. A "recording" is a representation within the meaning of the present invention. A recording can be a representation in the spatial domain or a representation in the frequency domain. Representations in the frequency domain are always used for training the prediction model and for the prediction; for example, k-space data. However, if representations in the spatial domain are generated by measurement, these can be converted into representations in the frequency domain, for example, using Fourier transformation, before being fed into the training and / or prediction.

[0153] The examination area is placed in a basic magnetic field. The examination area is subjected to an MRI procedure, during which a plurality of MRI images are generated that show the examination area during a first period of time. These MRI images, generated by measurement during the first period of time, are also referred to in this description as the first MRI images.

[0154] The term plurality means that at least two (first) MRI images, preferably at least three (first), most preferably at least four (first) MRI images are generated.

[0155] The subject is administered a contrast agent, which is distributed throughout the examination area. The contrast agent is preferably administered intravenously as a weight-adjusted bolus (for example, into a vein in the arm).

[0156] The contrast agent is preferably a hepatobiliary contrast agent such as Gd-EOB-DTPA or Gd-BOPTA. In a particularly preferred embodiment, the contrast agent is a substance or mixture of substances containing gadoxetic acid or a salt of gadoxetic acid as the contrast-enhancing agent. Most preferably, the contrast agent is the disodium salt of gadoxetic acid (Gd-EOB-DTPA disodium).

[0157] The first time period preferably comprises the flooding of the examination area with the contrast agent. The first time period preferably comprises the arterial phase and / or the portal venous phase and / or the late phase in dynamic contrast-enhanced magnetic resonance imaging of a liver or part of a liver of an examination subject. These phases are defined and described, for example, in the following publications: J. Magn. Reson. Imaging, 2012, 35(3): 492-511, doi:10.1002 / jmri.22833; Clujul Medical, 2015, Vol. 88 no. 4: 438-448, DOI: 10.15386 / cjmed-414; Journal of Hepatology, 2019, Vol. 71: 534-542, http: / / dx.doi.org / 10.1016 / j.jhep.2019.05.005).

[0158] Fig. 12 shows schematically the temporal course of the concentrations of contrast agent in hepatic arteries (A), hepatic veins (V), and healthy liver cells (P) after the application of a hepatobiliary contrast agent into a human arm vein. The concentrations are represented in the form of the signal intensities I in the named areas (hepatic arteries, hepatic veins, liver cells) during magnetic resonance imaging as a function of time t. During an intravenous bolus injection, the concentration of the contrast agent in the hepatic arteries (A) increases first (dashed curve). The concentration passes through a maximum and then decreases. The concentration in the hepatic veins (V) increases more slowly than in the hepatic arteries and reaches its maximum later (dotted curve). The concentration of the contrast agent in the healthy liver cells (P) increases slowly (solid curve) and only reaches its maximum at a much later time (the maximum is in the Fig. 12 not shown). Several characteristic time points can be defined: At time TP0, contrast agent is administered intravenously as a bolus. At time TP1, the concentration (signal intensity) of the contrast agent in the hepatic arteries reaches its maximum. At time TP2, the signal intensity curves at the hepatic arteries and the hepatic veins intersect. At time TP3, the concentration (signal intensity) of the contrast agent in the hepatic veins reaches its maximum. At time TP4, the signal intensity curves at the hepatic arteries and the liver cells intersect. At time T5, the concentrations in the hepatic arteries and the hepatic veins have decreased to a level at which they no longer cause measurable contrast enhancement.

[0159] In a preferred embodiment, the first time period is selected so that such MRI images are generated from the liver or part of the liver of an examination subject, (i) showing the examination area without contrast agent, (ii) showing the examination area during the arterial phase in which the contrast agent spreads through the arteries in the examination area, (iii) showing the examination area during the portal venous phase in which the contrast agent enters the examination area via the portal vein, and (iv) showing the examination area during the late phase in which the concentration of contrast agent in the arteries and veins decreases and the concentration of contrast agent in the liver cells increases.

[0160] Preferably, the first period begins one minute to one second before the administration of the contrast agent, or with the administration of the contrast agent, and lasts from the administration of the contrast agent for a period of 2 minutes to 15 minutes, preferably 2 minutes to 13 minutes, and even more preferably 3 minutes to 10 minutes. Since the contrast agent is excreted very slowly via the renals and biliary tract, the second period can last up to two hours or more after the administration of the contrast agent.

[0161] In a preferred embodiment, the first time period comprises at least the times TP0, TP1, TP2, TP3 and TP4.

[0162] In a preferred embodiment, at least MRI images of all the following phases are generated (measured): in a time period before TP0, in the time period from TP0 to TP1, in the time period from TP1 to TP2, in the time period from TP2 to TP3 and in the time period TP3 to TP4.

[0163] It is conceivable that one or more MRI images (measurement-technical) are generated in the time periods before TP0, from TP0 to TP1, from TP1 to TP2, from TP2 to TP3 and from TP3 to TP4.

[0164] Based on the (first) MRI images generated (measured) during the first time period, a second MRI image or multiple second MRI images showing the examination area during a second time period are predicted. MRI images predicted for the second time period are also referred to as second MRI images in this description.

[0165] In a preferred embodiment of the present invention, the second time period follows the first time period.

[0166] The second period is preferably a period within the hepatobiliary phase; preferably a period beginning at least 10 minutes after administration of the contrast agent, preferably at least 20 minutes after administration of the contrast agent.

[0167] The prediction of the at least one second representation representing the examination region during the second time period is carried out with the aid of the prediction model according to the invention. The prediction model has been trained in advance to predict one or more MRI images showing the examination region during the second time period based on a plurality of first MRI images showing an examination region during the first time period.

[0168] The example described here is also shown schematically in Fig. 1(b) shown. Beispiel 3

[0169] In a further preferred embodiment of the present invention, the present invention is used to differentiate lesions in the liver from blood vessels. In T1-weighted MRI images, Primovist® leads to a significant signal enhancement in healthy liver parenchyma < 10-20 minutes after injection (in the hepatobiliary phase), while lesions containing no or only a few hepatocytes, e.g., metastases or moderately to poorly differentiated hepatocellular carcinomas (HCCs), appear as darker areas. However, in the hepatobiliary phase, the blood vessels also appear as dark areas, so that in the MRI images generated during the hepatobiliary phase, differentiation of liver lesions and blood vessels based on contrast alone is not possible.

[0170] The present invention can be used to generate artificial MRI images of a liver or part of a liver of an examination subject in which the contrast between the blood vessels in the liver and the liver cells is artificially minimized in order to make liver lesions more visible.

[0171] A "recording" is a representation within the meaning of the present invention. A recording can be a representation in the spatial domain or a representation in the frequency domain. Frequency-domain representations are always used for training the prediction model and for the prediction. However, representations in the spatial domain can be generated using measurement technology, which are then converted, for example, using Fourier transformation, into frequency-domain representations before being fed into the training and / or prediction.

[0172] The plurality of first representations comprises at least one representation of the examination area in which blood vessels can be identified, which are preferably contrast-enhanced by a contrast agent (blood vessel representation).

[0173] When using a paramagnetic contrast agent, the blood vessels in such a representation are characterized by high signal intensity due to the contrast enhancement (high-signal representation). Those (connected) structures within such a representation that exhibit a signal intensity within an empirically determinable range can be assigned to blood vessels. This representation thus provides information about where blood vessels are depicted in a spatial representation and which structures in a spatial representation are attributable to blood vessels (arteries and / or veins).

[0174] The plurality of first representations further comprises at least one representation of the examination area in which healthy liver cells are shown with contrast enhancement (liver cell representation), e.g. a representation of the examination area that was recorded during the hepatobiliary phase.

[0175] The information about the blood vessels from the at least one blood vessel representation is combined with the information from the at least one liver cell representation. At least one representation is (artificially) generated (calculated) in which the contrast difference between structures attributable to blood vessels and structures attributable to healthy liver cells is leveled.

[0176] The term "leveling" here means "alignment" or "minimization." The goal of leveling is to make the boundaries between blood vessels and healthy liver cells disappear in the artificially generated representation, and to make blood vessels and healthy liver cells appear as a uniform tissue in the artificially generated representation, from which liver lesions stand out structurally through a different contrast.

[0177] Typically, a (number = 1) artificial representation is predicted based on a (number = 1) blood vessel representation and a (number = 1) liver cell representation.

[0178] It is conceivable that in addition to at least one blood vessel representation and at least one liver cell representation, at least one native representation is also used to predict the at least one artificial representation.

[0179] In one embodiment, generating the artificial representation comprises the following steps: Feeding the at least one blood vessel representation and the at least one liver cell representation to a prediction model, wherein the prediction model has been trained on the basis of reference representations by means of supervised learning to generate at least one artificial representation from at least one reference blood vessel representation and at least one reference liver cell representation, wherein in the at least one artificial representation the contrast difference between structures attributable to blood vessels and structures attributable to healthy liver cells is leveled, receiving at least one artificial representation as output from the prediction model. Beispiel 4

[0180] In a further embodiment, the present invention is used to generate a native MRI image of the liver. One or more artificial MRI images of a liver or part of a liver of an examination subject are generated, showing the liver or part of the liver without contrast enhancement induced by a contrast agent. The artificial MRI image(s) is / are created based on MRI images that were all acquired with contrast enhancement induced by a contrast agent.

[0181] A "recording" is a representation within the meaning of the present invention. A recording can be a representation in the spatial domain or a representation in the frequency domain. Frequency-domain representations are always used for training the prediction model and for the prediction. However, representations in the spatial domain can be generated using measurement technology, which are then converted, for example, using Fourier transformation, into frequency-domain representations before being fed into the training and / or prediction.

[0182] The examination area is placed in a basic magnetic field. The examination subject is administered a contrast agent, which is distributed throughout the examination area. The contrast agent is preferably administered intravenously (for example, into an arm vein) as a bolus, adjusted to the patient's weight. The contrast agent is preferably a hepatobiliary contrast agent such as Gd-EOB-DTPA or Gd-BOPTA. In a particularly preferred embodiment, the contrast agent is a substance or a mixture of substances containing gadoxetic acid or a salt of gadoxetic acid as the contrast-enhancing agent. Most preferably, it is the disodium salt of gadoxetic acid (Gd-EOB-DTPA disodium).

[0183] A plurality of first representations of the examination region are generated, representing the examination region in a first time period after the application of the contrast agent. Preferably, the plurality of first representations are T1-weighted images.

[0184] Preferably, the plurality of first representations comprises at least one representation of the examination region that represents the examination region during the dynamic phase, e.g., at least one representation that represents the examination region during the arterial phase, the venous phase and / or during the late phase (see, e.g., Fig. 12 and the explanations in Example 2). When using a paramagnetic contrast agent, the blood vessels in such representations are characterized by high signal intensity due to contrast enhancement (high-signal representation).

[0185] Preferably, the plurality of first representations further comprises at least one representation of the examination region, which represents the examination region during the hepatobiliary phase. During the hepatobiliary phase, the healthy liver tissue (parenchyma) is displayed with contrast enhancement.

[0186] MRI examinations of the dynamic and hepatobiliary phases take a comparatively long time. During this time, patient movement should be avoided to minimize motion artifacts in radiological images. The prolonged restriction of movement can be uncomfortable for the patient. For this reason, shortened MRI scanning procedures are now becoming established. In these procedures, a contrast agent is administered to the subject a certain time period (i.e., 10 to 20 minutes) before the MRI scan in order to be able to acquire MRI images directly during the hepatobiliary phase. MRI images of the dynamic phase are then acquired during the same MRI scan after the administration of a second dose of contrast agent. Compared to a conventional MRI scan, the length of time a patient or subject remains in the MRI scanner is therefore significantly shorter.Therefore, according to the invention, at least one representation of the liver or part of the liver is preferably recorded in the hepatobiliary phase after a (first) application of a first contrast agent to the examination subject, and at least one further representation of the same liver or part of the same liver is recorded in the dynamic phase after application of a second contrast agent or a second application of the first contrast agent to the same examination subject. The first contrast agent is a hepatobiliary, paramagnetic contrast agent. The second contrast agent can also be an extracellular, paramagnetic contrast agent.

[0187] The first representations of the examination region are then fed to the prediction model according to the invention. The prediction model has been trained in advance, based on the received first representations, to predict one or more second representations that show the liver or a part of the liver of the examination subject without contrast enhancement induced by a contrast agent. The prediction model was preferably created using a self-learning algorithm in supervised machine learning. Training data is used for learning, which includes a plurality of representations of the examination region during the dynamic phase and the hepatobiliary phase of the liver or a part of the liver from a plurality of examination subjects. Furthermore, the training data also includes representations of the examination region for which no contrast enhancement was present, i.e.which were generated without the application of a contrast agent.

[0188] The example described here is also shown schematically in Fig. 1(c). Beispiel 5

[0189] In a further preferred embodiment, the present invention is used to reduce the examination time of a patient in dynamic contrast-enhanced magnetic resonance imaging of the liver.

[0190] Contrast medium is administered in the form of two boluses. The first application takes place at a time when the subject to be examined is not yet in the MRI scanner. During the first application, a first contrast medium is administered. The first contrast medium is preferably administered intravenously as a bolus, adjusted for weight (e.g. into a vein in the arm). The first contrast medium is preferably a hepatobiliary contrast medium such as Gd-EOB-DTPA or Gd-BOPTA. In a particularly preferred embodiment, the first contrast medium is a substance or a mixture of substances containing gadoxetic acid or a salt of gadoxetic acid as the contrast-enhancing agent. Very particularly preferably, it is the disodium salt of gadoxetic acid (Gd-EOB-DTPA disodium).

[0191] After the application of the first contrast agent, a period of time can be waited before the object to be examined is placed in the MRI scanner and a first MRI image is generated at a first time.

[0192] A "recording" is a representation within the meaning of the present invention. A recording can be a representation in the spatial domain or a representation in the frequency domain. Frequency-domain representations are always used for training the prediction model and for the prediction. However, representations in the spatial domain can be generated using measurement technology, which are then converted, for example, using Fourier transformation, into frequency-domain representations before being fed into the training and / or prediction.

[0193] The time period between the first application and the generation of the first MRI image is preferably in the range of 5 minutes to 1 hour, more preferably in the range of 10 minutes to 30 minutes, most preferably in the range of 8 minutes to 25 minutes.

[0194] The first MRI image represents the liver or part of the liver of the subject during the hepatobiliary phase after the first contrast agent is applied. Healthy liver cells are contrast-enhanced in the first MRI image due to the application of the first contrast agent.

[0195] The hepatobiliary phase, in which the first MRI image is generated, is also referred to in this description as the first hepatobiliary phase. The first contrast agent has reached the healthy liver cells and leads to contrast enhancement; in the case of a paramagnetic contrast agent, this leads to signal enhancement in the healthy liver cells. No MRI images are generated during the arterial phase, the portal venous phase, and the late phase, which occur after the application of the first contrast agent. The arterial phase, the portal venous phase, and the late phase, which occur after the application of the first contrast agent, are also referred to in this description as the first arterial phase, the first portal venous phase, and the first late phase.

[0196] It is conceivable that multiple MRI images will be acquired during the first hepatobiliary phase.

[0197] After one or more initial MRI images have been acquired during the first hepatobiliary phase, contrast medium is administered a second time. This second time, a second contrast medium is administered. The second contrast medium can be the same as the first contrast medium; however, it can also be a different contrast medium, preferably an extracellular contrast medium. The second contrast medium is also preferably administered intravenously as a weight-adjusted bolus (e.g., into an arm vein).

[0198] The application of the first contrast agent is also referred to as the first application in this description; the application of the second contrast agent is also referred to as the second application in this description. If the first contrast agent and the second contrast agent are the same, then a first application of a hepatobiliary contrast agent takes place, followed by a second application of the hepatobiliary contrast agent at a later time. If the first contrast agent and the second contrast agent are different, then a first application of a first contrast agent takes place, with the first contrast agent being a hepatobiliary contrast agent, followed by a second application of a second (different) contrast agent at a later time.

[0199] At the time of the second application (or at the time of application of the second contrast agent), the examination subject is preferably already in the MRI scanner. After the application of the second contrast agent, an arterial phase, a portal venous phase, and a late phase are again run through. This arterial phase, portal venous phase, and late phase are also referred to in this description as the second arterial phase, second portal venous phase, and second late phase. In the second arterial phase and / or in the second portal venous phase and / or in the second late phase, one or more MRI images are generated. These MRI images are referred to as second, third, fourth, etc., in the order in which they are acquired.

[0200] In a preferred embodiment, a second MRI image is generated during the second arterial phase, a third MRI image is generated during the second portal venous phase, and a fourth MRI image is generated during the second late phase. Such a second MRI image shows, in particular, arteries with contrast enhancement; such a third MRI image shows, in particular, veins with contrast enhancement.

[0201] It is also conceivable that more than one MRI image is generated during the aforementioned phases.

[0202] Artificial MRI images can then be calculated from the MRI images generated during one or more phases after the application of the first and second contrast agents.

[0203] The goal of generating artificial MRI images from measured MRI images is to increase the contrast between healthy liver tissue and other areas. When a hepatobiliary paramagnetic contrast agent is used as the first contrast agent, the signal intensity of healthy liver tissue is still increased during the second arterial phase, the second portal venous phase, and the second late phase as a result of the application of the first contrast agent. The second contrast agent, which spreads during these second phases, also leads to an increased signal in the tissue in which it spreads. Thus, there is only a low contrast in the MRI images between the healthy liver tissue and the remaining tissue contrast-enhanced by the (second) contrast agent.To increase this contrast, at least one artificial MRI image is generated using a predictive model that would show the examination area as it appeared in the dynamic phase after application of the first contrast agent, or as it would appear if only the second contrast agent had been applied: the blood vessels are shown with enhanced contrast due to the application of the second contrast agent, but healthy liver cells are not shown with enhanced contrast due to the application of the first contrast agent. In other words, an artificial MRI image is generated that looks like the second MRI image, with the difference that the contrast enhancement of the healthy liver cells caused by the application of the first contrast agent is subtracted (eliminated) from the second MRI image.

[0204] The example described here is also shown schematically in Fig. 1(d).

Claims

1. Computer-implemented method comprising the steps of - receiving a plurality of first representations (F̃1, F̃2) of an examination region of an examination object in frequency space, at least some of the first representations (F̃1, F̃2) representing the examination region during a first time span after administration of a contrast agent, - feeding the plurality of first representations (F̃1, F̃2) to a prediction model (PM), the prediction model (PM) having been trained on the basis of first reference representations (F1, F2) of the examination region of a multiplicity of examination objects to generate from the first reference representations (F1, F2), of which at least some represent the examination region during a first time span after administration of a contrast agent in frequency space, one or more second reference representations (F3*) which represent the examination region during a second time span in frequency space, the second time span coming after the first time span, - receiving one or more predicted representations (F̃3*) of the examination region in frequency space from the prediction model (PM), the one or more predicted representations (F̃3*) representing the examination region during a second time span, the second time span coming after the first time span, - transforming the one or more predicted representations (F̃3*) into one or more representations (Õ3*) of the examination region in real space, - outputting the one or more representations (Õ3*) of the examination region in real space.

2. Method according to Claim 1, wherein the plurality of first representations (F̃1, F̃2) - comprises at least one representation (F̃1) of the examination region in frequency space that represents the examination region before administration of the contrast agent and - comprises at least one representation (F̃2) of the examination region in frequency space that represents the examination region in the first time span after administration of the contrast agent, and wherein the one or more predicted representations (F̃3*) represent the examination region in the second time span, the second time span coming after the first time span.

3. Method according to Claim 1, wherein the multiple predicted representations (F̃3*) represent the examination region in the second time span with a contrast enhancement that is constant over time.

4. Method according to any of Claims 1 to 3, wherein the prediction model (PM) is an artificial neural network or comprises such a network.

5. Method according to any of Claims 1 to 4, wherein the first representations (F̃1, F̃2) of the examination region in frequency space are k-space data of a magnetic resonance imaging examination.

6. Method according to any of Claims 1 to 4, wherein a plurality of radiological images of the examination region in real space are received in a first step and these received radiological images are converted by Fourier transform into the first representations (F̃1, F̃2) of the examination region in frequency space.

7. Method according to any of Claims 1 to 6, comprising the steps of: - receiving a plurality of first representations (F̃1, F̃2) of an examination region of an examination object in frequency space, at least some of the first representations (F̃1, F̃2) representing the examination region during a first time span after administration of a contrast agent, - specifying a region (A) in the first representations, the specified region (A) comprising the centre of the frequency space, - reducing the first representations (F̃1, F̃2) to the specified region, - feeding the plurality of reduced first representations (F̃1red, F̃2red) to a prediction model (PM), - receiving one or more second representations (F̃3*red) of the examination region in frequency space from the prediction model (PM), the one or more second representations (F̃3*red) representing the examination region during a second time span, the second time span coming after the first time span, - supplementing the one or more second representations (F̃3*red) by one or more regions (F̃1DI) of the received first representations (F̃1, F̃2) that lie outside the specified region (A), - transforming the one or more supplemented second representations into one or more representations (Õ3*) of the examination region in real space, - outputting the one or more representations (Õ3*) of the examination region in real space.

8. Method according to any of Claims 1 to 7, wherein the one or more supplemented second representations are transformed into one or more representations of the examination region (Õ3*) in real space by means of inverse Fourier transform (iFT).

9. System (10) comprising • a receiving unit (11), • a control and calculation unit (12) and • an output unit (13), wherein the control and calculation unit (12) is configured - to cause the receiving unit (11) to receive a plurality of first representations (F̃1, F̃2) of an examination region of an examination object in frequency space, at least some of the first representations (F̃1, F̃2) representing the examination region during a first time span after administration of a contrast agent, - to feed the plurality of first representations (F̃1, F̃2) to a prediction model (PM), the prediction model (PM) having been trained on the basis of first reference representations (F1, F2) of the examination region of a multiplicity of examination objects to generate from the first reference representations (F1, F2), of which at least some represent the examination region during a first time span after administration of a contrast agent in frequency space, one or more second reference representations (F3*) which represent the examination region during a second time span in frequency space, the second time span coming after the first time span, - to receive one or more predicted representations (F̃3*) of the examination region in frequency space from the prediction model (PM), the one or more predicted representations (F̃3*) representing the examination region during a second time span, the second time span coming after the first time span, - to transform the one or more predicted representations (F̃3*) into one or more representations (Õ3*) of the examination region in real space, - to cause the output unit (13) to output the one or more representations (Õ3*) of the examination region in real space.

10. Computer program product comprising a computer program that can be loaded into a working memory of a computer system (10), where it causes the computer system (10) to execute the following steps: - receiving a plurality of first representations (F̃1, F̃2) of an examination region of an examination object in frequency space, at least some of the first representations (F̃1, F̃2) representing the examination region during a first time span after administration of a contrast agent, - feeding the plurality of first representations (F̃1, F̃2) to a prediction model (PM), the prediction model (PM) having been trained on the basis of first reference representations (F1, F2) of the examination region of a multiplicity of examination objects to generate from the first reference representations (F1, F2), of which at least some represent the examination region during a first time span after administration of a contrast agent in frequency space, one or more second reference representations (F3*) which represent the examination region during a second time span in frequency space, the second time span coming after the first time span, - receiving one or more predicted representations (F̃3*) of the examination region in frequency space from the prediction model (PM), the one or more predicted representations (F̃3*) representing the examination region during a second time span, the second time span coming after the first time span, - transforming the one or more predicted representations (F̃3*) into one or more representations (Õ3*) of the examination region in real space, - outputting the one or more representations (Õ3*) of the examination region in real space.

11. Use of a contrast agent in a method for predicting at least one radiological image, the method comprising the following steps: - administering the contrast agent, the contrast agent spreading in an examination region of an examination object, - generating a plurality of first representations (F̃1, F̃2) of the examination region of the examination object in frequency space, at least some of the first representations (F̃1, F̃2) representing the examination region during a first time span after administration of the contrast agent, - feeding the plurality of first representations (F̃1, F̃2) to a prediction model (PM), the prediction model (PM) having been trained on the basis of first reference representations (F1, F2) of the examination region of a multiplicity of examination objects to generate from the first reference representations (F1, F2), of which at least some represent the examination region during a first time span after administration of a contrast agent in frequency space, one or more second reference representations (F3*) which represent the examination region during a second time span in frequency space, the second time span coming after the first time span, - receiving one or more predicted representations (F̃3*) of the examination region in frequency space from the prediction model (PM), the one or more predicted representations (F̃3*) representing the examination region during a second time span, the second time span coming after the first time span, - transforming the one or more predicted representations (F̃3*) into one or more representations (Õ3*) of the examination region in real space, - outputting the one or more representations (Õ3*) of the examination region in real space.