GENERATION OF SYNTHETIC RADIOLOGICAL RECORDINGS

Description

TECHNICAL AREA

[0001] The present invention relates to the technical field of radiology, in particular to supporting radiologists in radiological examinations using artificial intelligence methods. The present invention relates to training a machine learning model and using the trained model to predict representations of an examination area after the application of different amounts of a contrast agent. INTRODUCTION

[0002] WO2021 / 197996A1 discloses a method for generating radiological synthetic representations of an examination area of ​​an object under investigation. Based on measured radiological images of an examination area showing blood vessels in the examination area with decreasing contrast intensity over time, the method generates synthetic radiological images of the examination area showing blood vessels with constant contrast intensity.

[0003] EP3875979A1 discloses a method, a device, a system and a computer program product for determining an optimized time to start the recording and acquisition of an MRI scan after the application of a contrast agent.

[0004] WO2019 / 074938A1 discloses a method for reducing the amount of contrast agent used in the production of radiological images using an artificial neural network.

[0005] In a first step, a training dataset is generated. The training dataset includes, for each person in a large number of people, i) a native radiological image ( zero-contrast image ), ii) a radiological image after the administration of a small amount of contrast medium ( low-contrast image ) and iii) a radiological image after the administration of a standard amount of contrast medium ( full-contrast image ).

[0006] In a second step, an artificial neural network is trained to predict, for each person in the training dataset, an artificial radiological image showing an image area after the application of the standard amount of contrast agent, based on the native image and the image after administration of a small amount of contrast agent. The measured radiological image after the application of a standard amount of contrast agent serves as the reference during training. ground truth ).

[0007] In a third step, the trained artificial neural network can be used to generate a synthetic radiological image for a new person based on a native image and a radiological image after the application of a small amount of contrast agent, which shows the recorded area as it would look if a standard amount of contrast agent had been applied.

[0008] It has been shown that synthetic radiological images generated in this way can exhibit artifacts. Fine structures may not be accurately represented, or structures may appear in areas of the synthetic radiological image that are not present in the represented tissue (see, e.g., K. Schwarz et al.: On the Frequency Bias of Generative Models, https: / / doi.org / 10.48550 / arXiv.2111.02447). SUMMARY

[0009] This and other problems are addressed in the present revelation.

[0010] The first subject matter of the present disclosure is a computer-implemented method for generating a synthetic radiological image (predictive method). The predictive method comprises: Providing a trained machine learning model, ▪ wherein the trained machine learning model has been trained on training data, to generate a synthetic representation of the investigation area of ​​the investigation object based on at least one input representation of an investigation area of ​​an investigation object, wherein the at least one input representation represents the investigation area without contrast agent and / or after application of a first quantity of a contrast agent, and the synthetic representation represents the investigation area after application of a second quantity of the contrast agent, wherein the second quantity is larger than the first quantity, wherein the training data for each investigation object of a plurality of investigation objects i) at least one input representation of the investigation area of ​​the investigation object,ii) comprise a target representation of the domain of study and iii) a transformed target representation, wherein the transformed target representation represents at least a part of the domain of study in the frequency domain if the target representation represents the domain of study in the spatial domain, or in the spatial domain if the target representation represents the domain of study in the frequency domain, wherein training the machine learning model comprises reducing deviations between i) at least a part of the synthetic representation and at least a part of the target representation and ii) between at least a part of a transformed synthetic representation and at least a part of the transformed target representation, receiving at least one input representation of a domain of study of a new object of study,wherein the at least one input representation represents the investigation area of ​​the new investigation object without contrast agent and / or after application of an initial amount of a contrast agent, inputting the at least one input representation of the investigation area of ​​the new investigation object into the trained machine learning model, receiving a synthetic representation of the investigation area of ​​the new investigation object from the machine learning model, outputting and / or storing the synthetic representation of the investigation area of ​​the new investigation object and / or transmitting the synthetic representation of the investigation area of ​​the new investigation object to a separate computer system.

[0011] Another subject of the present disclosure is a computer system comprising a receiving unit, a control and processing unit, and an output unit, wherein the control and processing unit is configured to provide a trained machine learning model, wherein the trained machine learning model has been trained on training data to generate a synthetic representation of the investigation area based on at least one input representation of an investigation area of ​​an investigation object, wherein the training data for each investigation object of a plurality of investigation objects comprise i) an input representation of the investigation area of ​​the investigation object, ii) a target representation of the investigation area, and iii) a transformed target representation, wherein the at least one input representation represents the investigation area of ​​the respective investigation object without contrast agent or after application of an initial quantity of a contrast agent.▪ where the target representation represents the investigation area of ​​the respective investigation object after an application of a second quantity of the contrast agent, the second quantity being larger than the first quantity, ▪ where the transformed target representation represents at least a part of the investigation area of ​​the respective investigation object in the frequency domain if the target representation represents the investigation area of ​​the respective investigation object in spatial space, or in spatial space if the target representation represents the investigation area of ​​the respective investigation object in the frequency domain,o wherein training the machine learning model comprises reducing deviations between i) at least a part of the synthetic representation and at least a part of the target representation and ii) at least a part of a transformed synthetic representation and at least a part of the transformed target representation, wherein the control and computing unit is configured to cause the receiving unit to receive at least one input representation of an investigation area of ​​a new investigation object, wherein the at least one input representation represents the investigation area without and / or after an application of an initial amount of a contrast agent, wherein the control and computing unit is configured to input the received at least one input representation into the trained machine learning model, wherein the control and computing unit is configured,to receive a synthetic representation of the investigation domain of the new investigation object from the trained machine learning model, wherein the control and processing unit is configured to cause the output unit to output and / or store the synthetic representation of the investigation domain of the new investigation object and / or transmit it to a separate computer system.

[0012] Another subject of the present disclosure is a computer program product comprising a computer program that can be loaded into the main memory of a computer system and causes the computer system to perform the following steps: Providing a trained machine learning model, wherein the trained machine learning model has been trained on training data, to generate a synthetic representation of the investigation area of ​​the investigation object based on at least one input representation of an investigation area of ​​an investigation object, wherein the training data for each investigation object of a plurality of investigation objects comprises i) an input representation of the investigation area of ​​the investigation object, ii) a target representation of the investigation area of ​​the investigation object, and iii) a transformed target representation, wherein the at least one input representation represents the investigation area of ​​the respective investigation object without contrast agent or after application of an initial quantity of a contrast agent.▪ where the target representation represents the investigation area of ​​the respective investigation object after application of a second quantity of the contrast agent, the second quantity being larger than the first quantity, ▪ where the transformed target representation represents at least a part of the investigation area of ​​the investigation object in the frequency domain if the target representation represents the investigation area of ​​the investigation object in the spatial domain, or in spatial domain if the target representation represents the investigation area of ​​the investigation object in the frequency domain, o where training the machine learning model comprises reducing the deviations between i) at least a part of the synthetic representation and at least a part of the target representation and ii) between at least a part of a transformed synthetic representation and at least a part of the transformed target representation,Receiving at least one input representation of an investigation area of ​​a new investigation object, wherein the at least one input representation represents the investigation area of ​​the new investigation object without contrast agent and / or after application of an initial amount of a contrast agent; inputting the at least one input representation of the investigation area of ​​the new investigation object into the trained machine learning model; receiving a synthetic representation of the investigation area of ​​the new investigation object from the trained machine learning model; outputting and / or storing the synthetic representation of the investigation area of ​​the new investigation object and / or transmitting the synthetic representation of the investigation area of ​​the new investigation object to a separate computer system.

[0013] Another subject of the present disclosure is the use of a contrast agent in a radiological examination procedure, wherein the radiological examination procedure comprises: Providing a trained machine learning model, ∘ wherein the trained machine learning model has been trained on training data, to generate a synthetic representation of the investigation area of ​​the investigation object based on at least one input representation of an investigation area of ​​an investigation object, o wherein the training data for each investigation object of a plurality of investigation objects comprise i) an input representation of the investigation area of ​​the investigation object, ii) a target representation of the investigation area of ​​the investigation object, and iii) a transformed target representation, ▪ wherein the at least one input representation represents the investigation area of ​​the respective investigation object without or after an application of an initial amount of the contrast agent,▪ where the target representation represents the investigation area of ​​the respective investigation object after application of a second quantity of the contrast agent, the second quantity being larger than the first quantity, ▪ where the transformed target representation represents at least a part of the investigation area of ​​the investigation object in the frequency domain if the target representation represents the investigation area of ​​the investigation object in the spatial domain, or in spatial domain if the target representation represents the investigation area of ​​the investigation object in the frequency domain, o where training the machine learning model comprises reducing the deviations between i) at least a part of the synthetic representation and at least a part of the target representation and ii) between at least a part of a transformed synthetic representation and at least a part of the transformed target representation,Receiving at least one input representation of an investigation area of ​​a new investigation object, wherein the at least one input representation represents the investigation area of ​​the new investigation object without contrast agent and / or after application of an initial amount of a contrast agent; inputting the at least one input representation of the investigation area of ​​the new investigation object into the trained machine learning model; receiving a synthetic representation of the investigation area of ​​the new investigation object from the trained machine learning model; outputting and / or storing the synthetic representation of the investigation area of ​​the new investigation object and / or transmitting the synthetic representation of the investigation area of ​​the new investigation object to a separate computer system.

[0014] Another subject of the present disclosure is a kit comprising a contrast agent and a computer program that can be loaded into the working memory of a computer system and causes the computer system to perform the following steps: Providing a trained machine learning model, ∘ wherein the trained machine learning model has been trained on training data, to generate a synthetic representation of the investigation area of ​​the investigation object based on at least one input representation of an investigation area of ​​an investigation object, o wherein the training data for each investigation object of a plurality of investigation objects comprise i) an input representation of the investigation area of ​​the investigation object, ii) a target representation of the investigation area of ​​the investigation object, and iii) a transformed target representation, ▪ wherein the at least one input representation represents the investigation area of ​​the respective investigation object without or after an application of an initial amount of the contrast agent,▪ where the target representation represents the investigation area of ​​the respective investigation object after application of a second quantity of the contrast agent, the second quantity being larger than the first quantity, ▪ where the transformed target representation represents at least a part of the investigation area of ​​the investigation object in the frequency domain if the target representation represents the investigation area of ​​the investigation object in the spatial domain, or in spatial domain if the target representation represents the investigation area of ​​the investigation object in the frequency domain, o where training the machine learning model comprises reducing the deviations between i) at least a part of the synthetic representation and at least a part of the target representation and ii) between at least a part of a transformed synthetic representation and at least a part of the transformed target representation,Receiving at least one input representation of an investigation area of ​​a new investigation object, wherein the at least one input representation represents the investigation area of ​​the new investigation object without contrast agent and / or after application of an initial amount of a contrast agent; inputting the at least one input representation of the investigation area of ​​the new investigation object into the trained machine learning model; receiving a synthetic representation of the investigation area of ​​the new investigation object from the trained machine learning model; outputting and / or storing the synthetic representation of the investigation area of ​​the new investigation object and / or transmitting the synthetic representation of the investigation area of ​​the new investigation object to a separate computer system.

[0015] Further items and embodiments can be found in the following description, the patent claims and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Fig. 1 This shows, in an exemplary and schematic way, the process of training the machine learning model. Fig. 2 shows, in an exemplary and schematic way, the use of a trained machine learning model for prediction. Fig. 3 shows, by way of example and schematically, another embodiment of training the machine learning model. Fig. 4 shows, by way of example and schematically, another embodiment of training the machine learning model. Fig. 5 shows, by way of example and schematically, another embodiment of training the machine learning model. Fig. 6 shows, by way of example and schematically, another embodiment of training the machine learning model. Fig. 7 shows, by way of example and schematically, another embodiment of training the machine learning model. Fig. 8 shows, by way of example and schematically, another embodiment of training the machine learning model. Fig. 9 shows, by way of example and schematically, another embodiment of training the machine learning model. Fig. 10 schematically shows an example of using the trained machine learning model for prediction. Fig. 11 shows the result of a validation (a) of the according to Fig. 8 trained machine learning model and (b) the according to Fig. 9 trained machine learning model. Fig. 12 shows an exemplary and schematic computer system according to the present disclosure. Fig. 13 shows, by way of example and schematically, another embodiment of the computer system according to the present disclosure. Fig. 14 schematically shows, in the form of a flowchart, an embodiment of the procedure for training a machine learning model. Fig. 15 schematically shows, in the form of a flowchart, an embodiment of the procedure for generating a synthetic representation of an investigation area of ​​an investigation object using the trained machine learning model. DETAILED DESCRIPTION

[0017] The invention is explained in more detail below, without distinguishing between the subject matter of the invention (method, computer system, computer program, use, kit). Rather, the following explanations are intended to apply analogously to all subject matter (method, computer system, computer program, use, kit), regardless of the context in which they are made.

[0018] If the present description or the claims mention steps in a sequence, this does not necessarily mean that the invention is limited to that sequence. Rather, it is conceivable that the steps could also be carried out in a different sequence or even in parallel; unless one step builds upon another, which makes it essential that the building step be carried out subsequently (which will be clear in the specific case). The sequences mentioned thus represent preferred embodiments of the invention.

[0019] The invention is explained in more detail at several points with reference to the drawings. These drawings depict specific embodiments with specific features and combinations of features, primarily for illustrative purposes; the invention should not be understood as being limited to the features and combinations of features shown in the drawings. Furthermore, statements made in the description of the drawings with regard to features and combinations of features are intended to be generally applicable, that is, transferable to other embodiments and not limited to the embodiments shown.

[0020] With the aid of the present invention, representations of an investigation area of ​​an object under investigation can be predicted. Such a predicted representation is also referred to in this disclosure as a synthetic representation.

[0021] The "object of study" is usually a living being, preferably a mammal, and most preferably a human being.

[0022] The "area of ​​investigation" is a part of the object being examined, for example an organ or part of an organ such as the liver, brain, heart, kidney, lung, stomach, intestine, pancreas, thyroid gland, prostate, breast, or a part of the aforementioned organs or several organs, or another part of the body.

[0023] In one embodiment, the area under investigation comprises a liver or part of a liver, or the area under investigation is a liver or part of a liver of a mammal, preferably a human.

[0024] In another embodiment, the examination area comprises a brain or part of a brain, or the examination area is a brain or part of a brain of a mammal, preferably a human.

[0025] In another embodiment, the examination area comprises a heart or part of a heart, or the examination area is a heart or part of a heart of a mammal, preferably a human.

[0026] In another embodiment, the examination area comprises a thorax or part of a thorax, or the examination area is a thorax or part of a thorax of a mammal, preferably a human.

[0027] In another embodiment, the examination area comprises a stomach or part of a stomach, or the examination area is a stomach or part of a stomach of a mammal, preferably a human.

[0028] In another embodiment, the examination area comprises a pancreas or part of a pancreas, or the examination area is a pancreas or part of a pancreas of a mammal, preferably a human.

[0029] In a further embodiment, the examination area comprises a kidney or part of a kidney, or the examination area is a kidney or part of a kidney of a mammal, preferably a human.

[0030] In a further embodiment, the examination area comprises one or both lungs or part of a lung of a mammal, preferably a human.

[0031] In another embodiment, the examination area comprises a breast or part of a breast, or the examination area is a breast or part of a breast of a female mammal, preferably a female human.

[0032] In a further embodiment, the examination area comprises a prostate or part of a prostate, or the examination area is a prostate or part of a prostate of a male mammal, preferably a male human.

[0033] The examination area, also called the recording volume (English: field of view, The field of view (FOV) is a volume that is depicted in radiological images. The area under examination is typically selected by a radiologist, for example, on a panoramic radiograph (or overview image). localizer ). Alternatively or additionally, the scope of investigation can also be defined automatically, for example based on a selected protocol.

[0034] A "representation of the area under investigation" is usually the result of a radiological examination.

[0035] Radiology is the branch of medicine that deals with the application of primarily electromagnetic radiation and (including, for example, ultrasound diagnostics) mechanical waves for diagnostic, therapeutic, and / or scientific purposes. In addition to X-rays, other ionizing radiation such as gamma rays or electrons are also used. Since a key application is imaging, other imaging techniques such as sonography and magnetic resonance imaging (MRI) are also considered part of radiology, even though these techniques do not use ionizing radiation. The term "radiology" as used in the present invention therefore includes, in particular, the following examination methods: computed tomography, magnetic resonance imaging, and sonography.

[0036] In one embodiment of the present disclosure, the radiological examination is a magnetic resonance imaging examination.

[0037] In another embodiment, the radiological examination is a computed tomography examination.

[0038] In one embodiment, the radiological examination is an ultrasound examination.

[0039] Contrast agents are frequently used in radiological examinations to enhance contrast.

[0040] "Contrast agents" are substances or mixtures of substances that improve the visualization of the body's structures and functions during radiological examinations.

[0041] In computed tomography (CT), iodine-containing solutions are most 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 typically used as contrast agents. In sonography, liquids containing gas-filled microbubbles ( microbubbles ) enthalten, intravenös verabreicht. 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 / wpcontent / 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).

[0042] MRI contrast agents exert their effect in an MRI examination by altering the relaxation times of the structures that absorb the contrast agent. Two groups of substances can be distinguished: para- and superparamagnetic agents. Both groups possess unpaired electrons that induce a magnetic field around the individual atoms or molecules. Superparamagnetic agents lead to a predominantly T2 shortening, while paramagnetic contrast agents primarily 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 (SPIOs). superparamagnetic iron oxide Examples of paramagnetic contrast agents include gadolinium chelates such as gadopentetate dimeglumine (trade name: Magnevist®, etc.), gadoteric acid (Dotarem®, Dotagita®, Cyclolux®), gadodiamide (Omniscan®), gadoteridol (ProHance®), gadobutrol (Gadovist®), and gadoxetic acid (Primovist® / Eovist®).

[0043] A representation of the area of ​​investigation within the meaning of the present disclosure may be an MRI scan, a CT scan, an ultrasound image or the like, or the representation of the area of ​​investigation may be generated from one or more MRI scans, CT scans, ultrasound images or the like.

[0044] A representation of the investigation area within the meaning of the present disclosure can be a representation in spatial space (image space) or a representation in frequency space.

[0045] In a spatial representation, also referred to in this description as a spatial representation, the area under investigation is typically represented by a multitude of image elements (pixels or voxels), which may be arranged in a grid, for example, with each image element representing a portion of the area under investigation. A widely used format in radiology for storing and processing spatial representations is the DICOM format. DICOM (Digital Imaging and Communications in Medicine) is an open standard for storing and exchanging information in medical image data management.

[0046] In a frequency-space representation, also referred to in this description as a frequency-space diagram or frequency-space representation, the area under investigation is represented by a superposition of fundamental oscillations. For example, the area under investigation can be represented by a sum of sine and cosine functions with different amplitudes, frequencies, and phases. The amplitudes and phases can be plotted as a function of the frequencies, for example, in a two- or three-dimensional representation. Typically, the lowest frequency (origin) is placed at the center. The further one moves away from this center, the higher the frequencies. Each frequency can be assigned an amplitude, with which the frequency is represented in the frequency-space diagram, and a phase, which indicates how far the respective oscillation is shifted relative to a sine or cosine oscillation.

[0047] A representation in spatial space can be transformed into a representation in frequency space, for example, by a Fourier transform. Conversely, a representation in frequency space can be transformed into a representation in spatial space, for example, by an inverse Fourier transform.

[0048] Details about spatial representations and frequency space representations and their respective conversions are described in numerous publications, see e.g.: https: / / see.stanford.edu / materials / lsoftaee261 / book-fall-07.pdf.

[0049] Using a trained machine learning model, a synthetic representation of the domain of the object under investigation can be generated based on at least one representation of that domain. The at least one representation on which the (trained) machine learning model bases its synthetic representation is also referred to in this description as the input representation.

[0050] The at least one input representation represents the examination area of ​​a subject before and / or after the administration of a contrast agent. The synthetic representation represents the same examination area of ​​the same subject after the administration of a second amount of the contrast agent. The first and second amounts of the contrast agent differ from each other. Die second set ist larger than the first quantity.

[0051] The phrase "after the second dose of a contrast agent" should not be interpreted as meaning that the first and second doses are added together in the area under investigation. Rather, the phrase "the representation represents the area under investigation after the application of a (first or second) dose" should mean: "the representation represents the area under investigation with a (first or second) dose" or "the representation represents the area under investigation encompassing a (first or second) dose".

[0052] In a preferred embodiment, the first quantity of contrast agent is smaller than the standard quantity. The second quantity of contrast agent may be the same as the standard quantity; however, it may also be smaller or larger. The standard quantity is usually the quantity recommended by the manufacturer and / or distributor of the contrast agent and / or the quantity approved by a regulatory authority and / or the quantity listed in a package insert for the contrast agent. For example, the standard quantity of Primovist® is 0.025 mmol Gd-EOB-DTPA disodium per kg body weight.

[0053] In the aforementioned embodiment, the machine learning model of the present disclosure is configured and trained to generate, based on at least one input representation of an examination area of ​​an object being examined, a synthetic representation representing the examination area after the application of a second quantity of contrast medium, wherein the first quantity is preferably smaller than the second quantity. Such a trained machine learning model can be used, for example, as described in WO2019 / 074938A1, to reduce the amount of contrast medium used in radiological examinations.Such a trained machine learning model can be used to transform a radiological image produced after the administration of an initial (smaller) amount of contrast agent into a radiological image that, in terms of contrast distribution, resembles an image produced after the administration of a second (larger) amount of contrast agent. In other words, the machine learning model can be trained to generate contrast enhancement without increasing the amount of contrast agent.

[0054] The generation of the synthetic representation of the examination area, which represents the area after the application of a second amount of contrast agent, can be based, for example, on a first input representation representing the area without contrast agent and a second input representation representing the area after the application of a first amount of contrast agent. The first amount is smaller than the second amount. The second amount can be the standard amount of contrast agent. However, the second amount can also be larger than the standard amount of contrast agent.

[0055] The first quantity can be, for example, 90%, 80%, 70%, 60%, 50%, 40%, 30%, 20%, or 10% of the second quantity, or less than 10% of the second quantity, or any other percentage of the second quantity.

[0056] It is also possible that the at least one input representation comprises at least one first input representation and at least one second input representation. The at least one first input representation can represent the examination area after the application of a first quantity of contrast agent. The at least one second input representation can represent the examination area after the application of a second quantity of contrast agent. The first and second quantities are different; the first quantity is smaller than the second quantity. Based on the at least one first input representation and the at least one second input representation, a synthetic representation can be generated that represents the examination area after the application of a third quantity of contrast agent. The third quantity can differ from the first and second quantities, e.g.,The third set can be larger than the first and second sets.

[0057] It is also possible that the at least one input representation includes at least one first input representation, at least one second input representation, and at least one third input representation. The at least one first input representation can represent the examination area after the application of a first quantity of a contrast agent, where the first quantity can also be zero. The at least one second input representation can represent the examination area after the application of a second quantity of a contrast agent. The at least one third input representation can represent the examination area after the application of a third quantity of a contrast agent. The first, second, and third quantities can be different; for example, the first quantity can be smaller than the second quantity, and the second quantity can be smaller than the third quantity.Based on at least one first input representation, at least one second input representation, and at least one third input representation, a synthetic representation can be generated that represents the examination area after the application of a fourth quantity of the contrast agent. This fourth quantity can differ from the first, second, and third quantities; for example, the fourth quantity can be larger than the first, second, and third quantities.

[0058] It is also possible that further amounts of contrast agent will be administered.

[0059] The radiological examination procedure can be a magnetic resonance imaging (MRI) procedure, and the contrast agent can be an MRI contrast agent. That is, the at least one input representation can be at least one MRI scan, and the synthetic representation can be a synthetic MRI scan.

[0060] The radiological examination procedure can be a computed tomography examination procedure, and the contrast agent can be a CT contrast agent. That is, the at least one input representation can be at least one CT scan, and the synthetic representation can be a synthetic CT scan.

[0061] It is also possible that the at least one input representation includes at least one MRI scan and at least one CT scan.

[0062] It is also possible that the radiological examination procedure is a computed tomography examination procedure and the contrast agent is an MRI contrast agent.

[0063] In a preferred embodiment, the contrast agent is an MRI contrast agent (regardless of whether it is used in a magnetic resonance imaging or computed tomography procedure).

[0064] The MRI contrast agent may be an extracellular contrast agent. Extracellular contrast agents are low-molecular-weight, water-soluble compounds that, after intravenous administration, distribute themselves in the blood vessels and interstitial space. They are excreted via the kidneys after a relatively short period of circulation in the bloodstream. Examples of extracellular MRI contrast agents include the gadolinium chelates gadobutrol (Gadovist®), gadoteridol (Prohance®), gadoteric acid (Dotarem®), gadopentetic acid (Magnevist®), and gadodiamide (Omnican®).

[0065] The MRI contrast agent may be an intracellular contrast agent. Intracellular contrast agents are partially absorbed into the cells of tissues and subsequently excreted. Intracellular MRI contrast agents based on gadoxetic acid, for example, are characterized by their partial specific uptake by liver cells (hepatocytes), accumulation in functional tissue (parenchyma), and enhancement of contrast in healthy liver tissue before being excreted via bile into the feces. Examples of such gadoxetic acid-based contrast agents are described in US 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®).

[0066] Gadoxetate disodium (GD, Primovist®) belongs to the group of intracellular contrast agents. It is approved for use in MRI of the liver to detect and characterize lesions in patients with known or suspected focal liver disease. GD, with its lipophilic ethoxybenzyl moiety, exhibits a biphasic distribution: first, distribution in the intravascular and interstitial spaces after bolus injection, followed by selective uptake by hepatocytes. GD is excreted unchanged from the body in approximately equal amounts via the kidneys and the hepatobiliary pathway (50:50 dual excretion mechanism). Due to its selective accumulation in healthy liver tissue, GD is also referred to as a hepatobiliary contrast agent.

[0067] GD is approved at a dose of 0.1 ml / kg body weight (BW) (0.025 mmol / kg BW Gd). The recommended administration of GD involves an undiluted intravenous bolus injection at a flow rate of approximately 2 ml / second, followed by flushing the IV cannula with physiological saline. A standard protocol for liver imaging using GD consists of several planning and pre-contrast sequences. Following IV bolus injection of the contrast agent, dynamic images are typically acquired during the arterial (approximately 30 seconds post-injection, pi), portal venous (approximately 60 seconds pi), and transition phases (approximately 2–5 minutes pi). The transition phase typically already shows some increase in liver signal intensity due to the incipient uptake of the agent by hepatocytes.Additional T2-weighted and diffusion-weighted (DWI) images can be acquired after the dynamic phase and before the late hepatobiliary phase.

[0068] In one embodiment, the contrast agent is gadoxetate disodium.

[0069] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium(III) 2-[4,7,10-tris(carboxymethyl)-1,4,7,10-tetrazacyclododec-1-yl]acetic acid (also known as gadolinium-DOTA or gadoteric acid).

[0070] In another embodiment, the contrast agent is an agent comprising gadolinium(III) ethoxybenzyldiethylenetriaminepentaacetic acid (Gd-EOB-DTPA); preferably, the contrast agent comprises the disodium salt of gadolinium(III) ethoxybenzyldiethylenetriaminepentaacetic acid (also known as gadoxetic acid).

[0071] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium(III) 2-[3,9-bis[1-carboxylato-4-(2,3-dihydroxypropylamino)-4-oxobutyl]-3,6,9,15-tetrazabicyclo[9.3.1]pentadeca-1(15),11,13-trien-6-yl]-5-(2,3-dihydroxypropylamino)-5-oxopentanoate (also known as gadopiclenol, see e.g. WO2007 / 042504 as well as WO2020 / 030618 and / or WO2022 / 013454).

[0072] In one embodiment of the present disclosure, the contrast agent is an agent comprising dihydrogen[(±)-4-carboxy-5,8,11-tris(carboxymethyl)-1-phenyl-2-oxa-5,8,11-triazatridecane-13-oato(5-)]gadolinate(2-) (also known as gadobenic acid).

[0073] In one embodiment of the present disclosure, the contrast agent comprises tetragadolinium-[4,10-bis(carboxylatomethyl)-7-{3,6,12,15-tetraoxo-16-[4,7,10-tris-(carboxylatomethyl)-1,4,7,10-tetraazacyclododecan-1-yl]-9,9-bis({[({2-[4,7,10-tris-(carboxylatomethyl)-1,4,7,10-tetraazacyclododecan-1-yl]propanoyl}amino)acetyl]-amino}methyl)-4,7,11,14-tetraazahepta-decan-2-yl}-1,4,7,10-tetraazacyclododecan-1-yl]acetate (also known as gadoquatrane) (see, e.g., J. Lohrke et al.: Preclinical Profile of Gadoquatrane: A Novel Tetrameric, Macrocyclic High Relaxivity Gadolinium-Based Contrast Agent. Invest Radiol., 2022, 1, 57(10): 629-638; WO2016193190).

[0074] In one embodiment of the present disclosure, the contrast agent is an agent comprising a Gd 3+< complex of a compound of formula (I) includes, whereby Ar a group selected from represents, where # represents the link to X, X represents a group selected from CH₂, (CH₂)₂, (CH₂)₃, (CH₂)₄ and *-(CH₂)₂-O-CH₂-#<, where * represents the link to Ar and #< represents the link to the acetic acid residue, R₁<, R₂< and R₃< independently represent a hydrogen atom or a group selected from C₁-C₃-alkyl, -CH₂OH, -(CH₂)₂OH and -CH₂OCH₃, R₄< a group selected from C₂-C₄-alkoxy, (H₃C-CH₂)-O-(CH₂)₂-O-, (H₃C-CH₂)-O-(CH₂)₂-O-(CH₂)₂-O- and (H 3 C-CH 2 )-O-(CH 2 ) 2 -O-(CH 2 ) 2 -O-(CH 2 ) 2 -O- represents, R 5< represents a hydrogen atom, and R 6< represents a hydrogen atom, or a stereoisomer, tautomer, hydrate, solvate or salt thereof, or a mixture thereof.

[0075] In one embodiment of the present disclosure, the contrast agent is an agent comprising a Gd 3+< complex of a compound of formula (II) includes, whereby Ar a group selected from represents, where # represents the link to X, X represents a group selected from CH 2 , (CH 2 ) 2 , (CH 2 ) 3 , (CH 2 ) 4 and *-(CH 2 ) 2 -O-CH 2 - #<, where * represents the link to Ar and #< represents the link to the acetic acid residue, R 7< represents a hydrogen atom or a group selected from C 1 -C 3 -alkyl, -CH 2 OH, -(CH 2 ) 2 OH and -CH 2 OCH 3; R 8< represents a group selected from C 2-C 4-alkoxy, (H 3 C-CH 2 O)-(CH 2 ) 2 -O-, (H 3 C-CH 2 O)-(CH 2 ) 2 -O-(CH 2 ) 2 -O- and (H 3 C-CH 2 O)-(CH 2 ) 2 -O-(CH 2 ) 2 -O-(CH 2 ) 2 -O-; R 9< and R 10< independently represent a hydrogen atom; or a stereoisomer, tautomer, hydrate, solvate or salt thereof, or a mixture thereof.

[0076] The term "C1-C3 alkyl" refers to a linear or branched, saturated, monovalent hydrocarbon group with 1, 2, or 3 carbon atoms, e.g., methyl, ethyl, n-propyl, and isopropyl. The term "C2-C4 alkyl" refers to a linear or branched, saturated, monovalent hydrocarbon group with 2, 3, or 4 carbon atoms.

[0077] The term "C 2 -C 4 -Alkoxy" means a linear or branched, saturated, monovalent group of the formula (C 2 -C 4 -Alkyl)-O-, in which the term "C 2 -C 4 -Alkyl" is defined as above, e.g. a methoxy, ethoxy, n-propoxy or isopropoxy group.

[0078] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium 2,2',2"-(10-{1-carboxy-2-[2-(4-ethoxyphenyl)ethoxy]ethyl}-1,4,7,10-tetraazacyclododecane-1,4,7-triyl)triacetate (see, e.g., WO2022 / 194777, Example 1).

[0079] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium 2,2',2"-{10-[1-carboxy-2-{4-[2-(2-ethoxyethoxy)ethoxy]phenyl}ethyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}triacetate (see, e.g., WO2022 / 194777, Example 2).

[0080] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium 2,2',2"-{10-[(1R)-1-carboxy-2-{4-[2-(2-ethoxyethoxy)ethoxy]phenyl}ethyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}triacetate (see, e.g., WO2022 / 194777, Example 4).

[0081] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium (2S,2'S,2"S)-2,2',2"-{10-[(1S)-1-carboxy-4-{4-[2-(2-ethoxyethoxy)ethoxy] phenyl}butyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}tris(3-hydroxypropanoate) (see e.g. WO2022 / 194777, Example 15).

[0082] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium 2,2',2"-{10-[(1S)-4-(4-butoxyphenyl)-1-carboxybutyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}triacetate (see, e.g., WO2022 / 194777, Example 31).

[0083] In one embodiment of the present disclosure, the contrast agent is an agent containing gadolinium-2,2',2"-{(2 S )-10-(carboxymethyl)-2-[4-(2-ethoxyethoxy)benzyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}triacetate.

[0084] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium-2,2',2"-[10-(carboxymethyl)-2-(4-ethoxybenzyl)-1,4,7,10-tetraazacyclododecane-1,4,7-triyl]triacetate.

[0085] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium(III) 5,8-bis(carboxylatomethyl)-2-[2-(methylamino)-2-oxoethyl]-10-oxo-2,5,8,11-tetraazadodecane-1-carboxylate hydrate (also known as gadodiamide).

[0086] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium(III) 2-[4-(2-hydroxypropyl)-7,10-bis(2-oxido-2-oxoethyl)-1,4,7,10-tetrazacyclododec-1-yl]acetate (also known as gadoteridol).

[0087] In one embodiment of the present disclosure, the contrast agent is an agent comprising gadolinium(III) 2,2',2"-(10-((2R,3S)-1,3,4-trihydroxybutan-2-yl)-1,4,7,10-tetraazacyclododecane-1,4,7-triyl)triacetate (also known as gadobutrol or Gd-DO3A-butrol).

[0088] A "machine learning model" can be understood as a computer-implemented data processing architecture. The model can receive input data and provide output data based on this input data and model parameters. Through training, the model can learn a relationship between the input data and the output data. During training, model parameters can be adjusted to produce a desired output for a given input.

[0089] When training such a model, it is presented with training data from which it can learn. The trained machine learning model is the result of the training process. In addition to input data, the training data includes the correct output data (target data) that the model is to generate based on the input data. During training, patterns are recognized that map the input data to the target data.

[0090] During the training process, the input data for the training data is fed into the model, and the model generates output data. This output data is then compared to the target data. Model parameters are modified to reduce the deviations between the output and target data to a (defined) minimum. To modify the model parameters with a view to reducing these deviations, an optimization method such as gradient descent can be used.

[0091] The deviations can be analyzed using an error function (English: loss function ) can be quantified. Such an error function can be used to determine an error (English: loss The goal of the training process is to calculate the error for a given pair of output and target data. This can involve modifying (adjusting) the parameters of the machine learning model to reduce the error to a (defined) minimum for all pairs in the training dataset.

[0092] If the output and target data are numbers, for example, the error function can be the absolute difference between these numbers. In this case, a large absolute error may mean that one or more model parameters need to be changed significantly.

[0093] For output data in the form of vectors, difference metrics between vectors such as the mean squared error, a cosine distance, a norm of the difference vector such as a Euclidean distance, a Chebyshev distance, an Lp norm of a difference vector, a weighted norm, or another type of difference metric of two vectors can be chosen as the error function.

[0094] For higher-dimensional outputs, such as two-dimensional, three-dimensional, or even higher-dimensional outputs, an element-wise difference metric can be used. Alternatively or additionally, the output data can be transformed before calculating an error value, for example, into a one-dimensional vector.

[0095] In the present case, the machine learning model is trained using training data to generate a synthetic representation of a study area of ​​a study object based on at least one input representation of the study area of ​​the study object.

[0096] The training data includes a set of input and target data for each of a large number of study objects.

[0097] The term "much" means at least ten, preferably more than one hundred.

[0098] Each set of input data and target data includes at least one input representation of an area of ​​investigation of the object of investigation as input data, as well as a target representation of the area of ​​investigation of the object of investigation and a transformed target representation of the area of ​​investigation of the object of investigation as target data.

[0099] The area of ​​investigation is usually the same for all objects under investigation.

[0100] Each input representation represents the investigation area of ​​the respective object before or after the application of an initial amount of a contrast agent.

[0101] Each target representation represents the area of ​​investigation of the respective subject after the application of a second amount of the contrast agent.

[0102] Each transformed target representation represents at least a part of the investigation area of ​​the respective object of study. i) in the frequency domain if the target representation represents the area of ​​investigation of the object under investigation in spatial space, or ii) in spatial space if the target representation represents the area of ​​investigation of the object under investigation in the frequency domain.

[0103] Each transformed target representation represents the area of ​​investigation of the respective subject after the application of a second amount of the contrast agent.

[0104] A transformed target representation in the frequency domain can be generated from a target representation in the spatial domain, for example by Fourier transformation.

[0105] A transformed target representation in spatial space can be generated from a target representation in frequency space, for example by inverse Fourier transformation.

[0106] When training the machine learning model, for each object under investigation, at least one input representation of the domain of investigation is fed into the machine learning model (i.e., fed into the machine learning model). Based on this input representation and model parameters, the machine learning model generates a synthetic representation.

[0107] In the event that the at least one input representation supplied to the model represents the investigation area in spatial space, the synthetic representation preferably (but not necessarily) also represents the investigation area in spatial space.

[0108] In the event that the at least one input representation supplied to the model represents the investigation domain in the frequency domain, the synthetic representation preferably (but not necessarily) also represents the investigation domain in the frequency domain.

[0109] The machine learning model can be configured and trained to generate, based on at least one input representation of the domain of study, a synthetic representation of the domain of study that i) represents the domain of study in the frequency domain if the at least one representation represents the domain of study in the spatial domain, or ii) represents the domain of study in the spatial domain if the at least one representation represents the domain of study in the frequency domain. In other words, the machine learning model can be configured and trained to perform (among other things) a transformation from the spatial domain to the frequency domain or vice versa.

[0110] A transformed synthetic representation is generated from or to the synthetic representation. If the synthetic representation represents the domain of investigation in spatial space, the transformed synthetic representation represents at least a part of the domain of investigation in frequency space.

[0111] If the synthetic representation represents the domain of investigation in the second state in the frequency domain, the transformed synthetic representation represents at least a part of the domain of investigation in the spatial domain. The generation of the transformed synthetic representation can be achieved by transforming the synthetic representation itself, and / or the machine learning model can be configured and trained to generate a transformed synthetic representation based on the at least one input representation.

[0112] Using an error function, the deviations i) between at least a part of the synthetic representation and at least a part of the target representation and ii) between at least a part of the transformed synthetic representation and at least a part of the transformed target representation are quantified.

[0113] The error function can have two terms: a first term to quantify the deviations between at least a part of the synthetic representation and at least a part of the target representation, and a second term to quantify the deviations between at least a part of the transformed synthetic representation and at least a part of the transformed target representation. The terms can be added together in the error function. The terms can also be weighted differently in the error function. The following equation gives an example of a (total) error function.L to quantify the deviations: L = λ 1 ⋅ L 1 + λ 2 ⋅ L 2

[0114] This is L the (overall) error function, L 1 a term that represents the deviations between the synthetic representation and the target representation, L 2 is a term that quantifies the deviations between the transformed synthetic representation and the transformed target representation, and λ1 and λ2 are weighting factors that can take values ​​between 0 and 1, for example, and give the two terms different weights in the error function. It is possible for the weighting factors to be kept constant or varied during the training of the machine learning model.

[0115] Examples of fault functions that can be used to carry out the present invention are the L1 fault function (L1 loss ), L2 error function (L2 loss ), Lp loss function, structural similarity index measure ( structural similarity index measure (SSIM)), VGG loss, perceptual loss ( perceptual loss ) or a combination of the above-mentioned functions or other error functions. Further details on error functions can be found, for example, in the scientific literature (see, e.g.: R. Mechrez et al.: The Contextual Loss for Image Transformation with Non-Aligned Data, 2018, arXiv:1803.02077v4; H. Zhao et al.: Loss Functions for Image Restoration with Neural Networks, 2018, arXiv:1511.08861v3; D. Fuoli et al.: Fourier Space Losses for Efficient Perceptual Image Super-Resolution, arXiv:2106.00783v1).

[0116] Fig. 1 This diagram illustrates, in a schematic and exemplary manner, the process of training the machine learning model. The training is performed using training data. Fig. 1 Training data TD for a study object is shown. The training data TD comprises, as input data, a first input representation R1 of a study area of ​​the study object and a second input representation R2 of the study area of ​​the study object. The first input representation R1 and the second input representation R2 represent the study area in spatial space. The first input representation R1 represents the study area, for example, without contrast agent or after the application of an initial amount of contrast agent. The second input representation R2 represents the study area, for example, after the application of a second amount of contrast agent.

[0117] The training data TD also includes a target representation TR as target data. The target representation TR also represents the investigation area in spatial space. The target representation TR represents the investigation area, for example, after the administration of a third amount of the contrast agent.

[0118] The machine learning model MLM is trained to predict the target representation TR based on the first input representation R1 and the second input representation R2, as well as on model parameters MP. The first input representation R1 and the second input representation R2 are fed into the machine learning model as input data. The machine learning model is configured to generate a synthetic representation SR of the domain of the object under investigation.

[0119] The synthetic representation SR is used in the Fig. 1 The example shown uses a transformation. T (e.g., a Fourier transform) a transformed synthetic representation SR T< generated. Similarly, the target representation TR is transformed. T a transformed goal representation TR T< generated. The transformed synthetic representation SR T< and the transformed goal representation TR T< These are frequency space representations of the examination area after the application of the third amount of contrast agent.

[0120] A first error function L The first function is used to quantify the deviations between the synthetic representation SR and the target representation TR. A second error function L 2 is used to compare the differences between the transformed synthetic representation SR T< and the transformed goal representation TR T< to quantify. The error functions L 1 and L The two calculated errors are combined in a total error function. L summarized into a total error (e.g. by addition with or without weighting).

[0121] Model parameters MP are modified with a view to reducing the overall error. This reduction can be achieved using an optimization method, for example, a gradient descent method.

[0122] The process is repeated for a large number of test objects until the total error reaches a predefined (desired) minimum and / or until the total error can no longer be reduced by modifying model parameters.

[0123] The trained machine learning model can then be used for prediction. This is illustrated schematically in [reference to relevant example]. Fig. 2 As shown: At least one input representation R1*, R2* of the domain of a new object under investigation is fed to the trained machine learning model MLM t<. The trained machine learning model MLM t< generates a synthetic representation SR* of the domain of the new object under investigation. The term "new" means that input data from the new object under investigation has not typically been used previously in training and / or validating the machine learning model. The generated synthetic representation SR* can be output, stored, and / or transmitted to a separate computer system.

[0124] In one embodiment of the present disclosure, the quantification of the deviations between the transformed synthetic representation and the transformed target representation is based solely on proportions of said representations. This is illustrated by way of example and schematically in Fig. 3 depicted.

[0125] The in Fig. 3 The process depicted corresponds to the process that occurs in Fig. 1 as shown, with the following differences: The transformed goal representation TR T< will be applied to a defined proportion of TR T,P< reduced. The representation TR T,P< is part of the transformed target representation TR. The representation TR T,P< can be achieved by a function P from the transformed target representation TR T< This can be generated, for example, by setting all frequency values ​​outside the predefined part to zero. Similarly, the transformed synthetic representation SR is created. T< to a defined proportion SR T,P< reduced. The proportions to which the transformed target representation TR applies T< and the transformed synthetic representation SR T< Reduced frequencies usually correspond to each other, i.e., they usually affect the same frequencies. In the Fig. 3 The process shown is both the transformed target representation TR T< as well as the transformed synthetic representation SR T< The frequencies have been reduced to a portion that includes the low frequencies in an area (e.g., rectangle or square) around the center (see the dashed white frames). This area primarily encodes contrast information.

[0126] In the Fig. 3 In the example shown, therefore, not the entire frequency range is used in the calculation using the error function. L 2. The representations in the frequency domain are reduced to a range of low frequencies; the higher frequencies are discarded. The low-frequency range predominantly encodes contrast information, while the higher-frequency range predominantly encodes information about fine structures. This means that the machine learning model in the Fig. 3 In the example shown, in addition to generating the synthetic representation SR, the system is trained to accurately reproduce low frequencies in particular. This places a focus on contrast information during training.

[0127] It should be noted that Fig. 3 It should not be understood that the transformed target representation TR T< and the transformed synthetic representation SR T< They must be trimmed to reduce them to the defined TR ratio. T,P< or the defined proportion SR T,P< to reduce. It is also possible to quantify the deviations between the proportion SR T,P< and the TR share T,P< using the error function L 2 only the areas SR T,P< and TR T,P< within the SR representations T< and TR T< to be taken into account. The term "reduce" should therefore be understood to mean that, for the determination of the deviations between the transformed synthetic representation SR, T< and the transformed goal representation TR T< only the areas SR T,P< and TR T,P< within the SR representations T< and TR T< This must be taken into account. This also applies analogously to Fig. 4 , Fig. 5 , Fig. 6 , Fig. 7 , Fig. 8 and Fig. 9 .

[0128] Fig. 4 This shows another example where not the entire frequency range is considered, but rather a focus is placed on a defined frequency range. The one in Fig. 4 The process depicted corresponds to the process that occurs in Fig. 3 is shown, with the difference that the function P ensures that the transformed target representation TR T< and the transformed synthetic representation SR T< Each is reduced to a proportion with higher frequencies, while low frequencies are discarded (the low frequency values ​​are set to zero in this example, represented by the color black). This means that the machine learning model in the Fig. 4 In the example shown, in addition to generating the synthetic representation SR, the system is trained to correctly reproduce higher frequencies in particular, thus focusing on the accurate reproduction of fine structures. As already mentioned in relation to Fig. 3 As explained, the low frequency values ​​do not necessarily have to be set to zero; likewise, it is possible that when quantifying the deviations between the transformed synthetic representation SR T< and the transformed goal representation TR T< using the error function L 2. Only the areas not shown in black are taken into account. This also applies analogously to... Fig. 8 and Fig. 9 .

[0129] It should also be noted that the part to which the frequency domain is reduced in the error calculation includes other parts besides those in Fig. 3 , Fig. 4 and Fig. 8 The frequency space can take on the shapes and sizes shown. Furthermore, multiple parts can be selected, and the frequency space representations can be restricted (reduced) to several parts (areas). For example, it is possible to divide the frequency space into different areas and generate frequency space representations of the investigation area for several or all areas, in which only the frequencies of the respective area appear. In this way, different frequency ranges can be weighted / considered differently during training.

[0130] In the Fig. 3 and Fig. 4 In the examples shown, representations R1, R2, TR, and SR are spatial representations. It is also possible that one, several, or all of these representations are frequency-space representations.

[0131] Are the transformed synthetic representations SR T< and the transformed goal representation TR T< For representations in spatial space, a part (or several parts) can also be selected, and the representations can be reduced to this part (these parts). In such a case, the error calculation focuses on features in spatial space that are found in the selected part (or parts).

[0132] Fig. 5 , Fig. 6 , Fig. 7 , Fig. 8 and Fig. 9 show further embodiments of the training method of the present disclosure.

[0133] In the Fig. 5 In the illustrated embodiment, the training data for each subject comprises at least one input representation R1 of a subject area and one target representation TR of the subject area. The at least one input representation R1 represents the subject area without and / or after the application of a first amount of contrast agent. The synthetic representation SR and the target representation TR represent the subject area after the application of a second amount of contrast agent.

[0134] The at least one input representation R1 and the target representation TR can each be a representation in spatial space or a representation in frequency space. In the Fig. 5 In the example shown, the input representation R1 and the target representation TR are representations in spatial space. The machine learning model MLM is trained to predict the target representation TR based on at least one input representation R1. This is achieved through a transformation. T A transformed input representation R2 is generated based on the input representation R1. Similarly, the transformation... T a transformed goal representation TR T< generated based on the target representation TR. In the Fig. 5 The example shown is the transformed input representation R2 and the transformed target representation TR. T< Representations of the investigation area in the frequency domain. The transformed input representation R2 represents the investigation area (like the input representation R1) without and / or after the application of the first amount of contrast agent; the transformed target representation TR T< represents the area under investigation (like the target representation TR) after the application of the second amount of contrast agent.

[0135] In Fig. 5 It is shown that the input representation R1 is transformed by an inverse transformation. T -1< can be obtained based on the transformed input representation R2. Similarly, a target representation TR can be obtained through the inverse transformation. T -1< based on the transformed target representation TR T< to be won. The inverse transformation T -1< is the value corresponding to the transformation TInverse transformation, indicated by the superscript -1. The reference to the inverse transformation is intended to clarify that the training data must have at least one input representation (R1 or R2) and one target representation (TR or TR). T< ) must contain; the other one (R2 or R1, or TR). T< or TR) can be obtained from the given representation by transformation or inverse transformation. This applies generally and not only to the representation in Fig. 5 depicted embodiment.

[0136] In the Fig. 5 In the illustrated embodiment, both the input representation R1 and the transformed input representation R2 are fed into the machine learning model MLM. The machine learning model MLM is configured to accept both a synthetic representation SR and a transformed synthetic representation SR. T< to produce.

[0137] By means of a first fault function L 1. The deviations between at least a part of the synthetic representation SR and at least a part of the target representation TR are quantified using a second error function. L 2. The deviations between at least a part of the transformed synthetic representation SR will be determined. T< and at least a part of the transformed target representation TR T< quantified. The in Fig. 5 The dashed white frames drawn in the transformed representations are intended to express that the error calculation using the error function, as in relation to Fig. 3 and Fig. 4 As described, the transformation does not necessarily have to be based on the entire frequency space representation, but rather the transformed representations can be reduced to a frequency range (or several frequency ranges). Similarly, the spatial representations SR and TR can be reduced to a part (or several parts) during error calculation (also shown by the dashed white frames). This applies analogously to the other embodiments as well, and not only to those described in [reference missing]. Fig. 5 depicted embodiment.

[0138] The error functions L 1 and L The two calculated errors are combined in a total error function. L summarized into a total error (e.g. by addition with or without weighting).

[0139] Model parameters MP are modified with a view to reducing the overall error. This can be done using an optimization method, for example a gradient descent method.

[0140] The process is repeated for a large number of test objects until the total error reaches a predefined (desired) minimum and / or until the total error can no longer be reduced by modifying model parameters.

[0141] The synthetic representation SR and the transformed synthetic representation SR T< They must be convertible into each other in the same way as the target representation TR and the transformed target representation TR. T< It is possible to introduce another error function that evaluates the quality of such a transformation. Thus, it is possible to... T from the synthetic representation SR a transformed synthetic representation SR T< to generate and to determine the deviations between this transformation-generated synthetic representation and the transformed synthetic representation generated by the machine learning model using a third error function. L 3. To quantify. Alternatively or additionally, it is possible to do so using the inverse transformation. T -1< from the transformed synthetic representation SR T< to generate a synthetic representation SR and to determine the deviations between this synthetic representation generated by inverse transformation and the synthetic representation generated by the machine learning model using a third error function. L 3 to quantify. This is schematically shown in Fig. 6 using the example of generating a transformed synthetic representation SR T #< through transformation T shown from the synthetic representation SR. In the Fig. 6 The example shown quantifies the error function.L 3 the deviations between the transformed synthetic representation generated by transformation SR T #< and the transformed synthetic representation SR generated by the machine learning model MLM T< The error functions L 1 , L 2 and L 3 are included in a total error function L combined (e.g., by addition with or without weighting). In the total error function L The individual terms can carry different weights, and these weights can be kept constant or varied during the training.

[0142] A variation of the ones in Fig. 5 and Fig. 6 The embodiment shown differs in that the machine learning model MLM is not fed both input representations R1 and R2, but only one of the two. The machine learning model can be configured and trained to generate the other one.

[0143] Fig. 7 Figure 1 schematically shows another embodiment of the training procedure. In this embodiment, two machine learning models are used: a first model MLM1 and a second model MLM2. The first machine learning model, MLM1, is trained to generate a target representation TR and / or a transformed target representation TR from an input representation R1 and / or R2. T< to predict. The second machine learning model, MLM2, is trained from a predicted target representation SR and / or a predicted transformed target representation SR. T< to predict (reconstruct) the original input representation R1 and / or R2 again. The machine learning models thus perform a cycle that improves the prediction quality of the first model MLM1 (which is used in the subsequent prediction).

[0144] The first machine learning model, MLM1, is fed at least one input representation R1 and / or at least one transformed input representation R2 as input data. The first model, MLM1, is configured to use a synthetic representation SR and / or a transformed synthetic representation SR. T< to generate MP 1 based on the input data and model parameters. A transformed synthetic representation SR T< can also be achieved through transformation T The synthetic representation SR can be generated. A synthetic representation SR can also be generated by inverse transformation. T-1< of the transformed synthetic representation SR T< be generated.

[0145] Deviations between the synthetic representation SR and the target representation TR can be corrected using an error function. L 1 1< can be quantified. Deviations between the transformed synthetic representation SR T< and the transformed goal representation TR T< can be achieved using a fault function L 2 1< can be quantified.

[0146] Deviations between a transformation resulting from an inverse transformation T -1< The synthetic representation SR obtained and the synthetic representation SR generated by the model MLM1 can be compared using an error function. L 3 1< be quantified (in Fig. 7 (not shown). Alternatively or additionally, deviations between a transformation and a T obtained transformed synthetic representation SR T< and the transformed synthetic representation SR generated by the model MLM1 T< using a fault function L 4 1< be quantified (in Fig. 7 (not shown).

[0147] The error functions L 1 1< , L 2 1< and, if available, L 3 1< and / or L 4 1< can be combined to form a total error function (in Fig. 7 (not shown).

[0148] Model parameter MP1 can be modified with a view to reducing the overall error.

[0149] The synthetic representation SR and / or the transformed synthetic representation SR T< The input(s) will be fed to the second machine learning model, MLM2. The second model, MLM2, is configured to reconstruct (predict) the first input representation, R1, and / or the second input representation, R2. The second model, MLM2, is configured to generate a predicted first input representation, R1#, and / or a predicted second input representation, R2#, based on the synthetic representation, SR, and / or the transformed synthetic representation, SR. T< and to generate MP2 based on model parameters. A second input representation R2# can also be created by transformation. T the first input representation R1# can be generated and / or a first input representation R1# can also be created by inverse transformation T -1< of the second input representation R2# is generated.

[0150] Deviations between the predicted first input representation R1# and the first input representation R1 can be corrected using an error function. L 1 2< can be quantified. Deviations between the predicted second input representation R2# and the second input representation R2 can be accounted for using an error function. L 2 2< can be quantified.

[0151] Deviations between a transformation resulting from an inverse transformation T -1< obtained input representation and the input representation generated by the MLM2 model can be compared using an error function. L 3 2< be quantified (in Fig. 7 (not shown). Alternatively or additionally, deviations between an input representation obtained through transformation and the input representation generated by the MLM2 model can be corrected using an error function. L 4 2< be quantified (in Fig. 6 (not shown).

[0152] The error functions L 1 2< , L 2 2< and, if available, L 3 2< and / or L 4 2< can be combined to form a total error function (in Fig. 7 (not shown).

[0153] Model parameter MP2 can be modified with a view to reducing the overall error.

[0154] Fig. 8 and Fig. 9 schematically show further examples of the procedure for training the machine learning model. Fig. 10 schematically shows another example of using the trained machine learning model for prediction.

[0155] In the Fig. 8 In the example shown, the machine learning model MLM is trained on one or more input representations R1,2, representing an area of ​​investigation of a subject before and / or after the application of an initial amount of contrast agent. Based on model parameters MP, the model is trained to generate a synthetic representation SR of the subject's area of ​​investigation. This synthetic representation SR represents the area of ​​investigation after the application of a second amount of contrast agent, where the second amount is larger than the first. In other words, the machine learning model is trained to predict a representation of the area of ​​investigation after the application of a larger amount of contrast agent than was actually applied.A target representation TR represents the area of ​​investigation of the object after the application of the second (larger) amount of contrast agent.

[0156] In the present example, the domain of investigation is a human brain. The at least one input representation R1,2, the target representation TR, and the synthetic representation SR represent the brain in place space.

[0157] The at least one input representation R1,2 is fed into the machine learning model MLM. The machine learning model generates the synthetic representation SR.

[0158] Through a transformation T The synthetic representation SR is transformed into a synthetic transformation SR T< transformed. The transformed synthetic transformation SR T< is reduced to a part using a function P; this results in a partial transformed synthetic transformation SR. T , P < .

[0159] Similarly, the target representation SR is transformed using the transformation. T a transformed goal representation TR T< and using the function P from the transformed target representation TR T< a partially transformed target representation TR T,P< generated.

[0160] The transformed synthetic representation SR T< and the transformed goal representation TR T< represent the investigation area in the frequency domain. The proportionally transformed synthetic representation SR T,P< SR is a transformed synthetic representation reduced to a frequency range with higher frequencies. T< The proportionally transformed target representation TR T,P< TR is a transformed target representation reduced to the frequency range with higher frequencies. T< . The MLM machine learning model is therefore trained to focus on fine-structure information encoded in the higher frequencies.

[0161] The machine learning model is implemented using an error function. L trained. The error function L quantifies the deviations i) between the synthetic representation SR and the target representation TR using a first term L 1 and ii) between the proportionally transformed synthetic representation SR T,P< and the proportionally transformed target representation TR T,P< by means of a second term L 2. The first term L 1 and the second term LFor example, in the error function, 2 can each be multiplied by a weight factor and the resulting products added (as described by equation Eq. 1).

[0162] In an optimization procedure, e.g. a gradient descent method, the model parameters MP can be adjusted with a view to reducing the value by means of the function. L The calculated error will be modified.

[0163] The described process is repeated for further test objects. The training can be terminated when the error function is used. L The calculated error reaches a defined minimum, i.e., the prediction quality reaches a desired level.

[0164] Fig. 9 This shows in relation to Fig. 8 The training methods discussed include the following extensions / changes: The training process uses two machine learning models: a first model MLM1 and a second model MLM2. The first machine learning model, MLM1, performs the functions that are relevant to the MLM machine learning model in relation to... Fig. 8 The second machine learning model, MLM2, is configured and trained to reproduce at least one input representation R1,2. In other words, the second machine learning model, MLM2, is configured and trained to generate a predicted input representation R1,2# based on the synthetic representation SR and model parameters MP2. The error function L has a third term L3, which quantifies the deviations between the input representation R1,2 and the predicted input representation R1,2#. In an optimization procedure, e.g., a gradient descent method, the model parameters MP1 and MP2 can be adjusted with respect to reducing the input representation using the function. L The calculated error will be modified.

[0165] Fig. 10 shows the use of the according to Fig. 8 or Fig. 9 trained machine learning model MLM or MLM1, in Fig. 10 Marked with MLM t<.

[0166] The trained machine learning model MLM t< is fed at least one input representation R1,2*, representing the brain of a new study subject. The term "new" means that the at least one input representation R1,2* has not already been used in training. The at least one input representation R1,2* represents the brain of the new study subject before and / or after the administration of an initial amount of a contrast agent.

[0167] The trained machine learning model generates a synthetic representation SR* based on at least one input representation R1,2*, which represents the brain after the application of a second amount of the contrast agent.

[0168] Fig. 11 shows the result of a validation (a) of the according to Fig. 8 trained machine learning model and (b) the according to Fig. 9 trained machine learning models. Each of the models was treated as if in relation to Fig. 10 The described process involves providing at least one input representation of a new object of study. Each model generates a synthetic representation SR*. A target representation TR exists for the object of study. Fig. 11 The generation of a difference representation is shown, where for each model a difference representation ΔR is generated by subtracting the respective synthetic representation SR* from the target representation TR. In the case of a perfect model, each of the synthetic representations SR* corresponds to the target representation TR, and the difference representation ΔR is completely black (all values ​​are zero). Fig. 11 It can be seen that in the case of the according to Fig. 8 trained machine learning model ( Fig. 11 (a) ) more pixels are different from zero than in the case according to Fig. 9 trained machine learning model ( Fig. 11 (b) The prediction quality of the according to Fig. 9 The trained machine learning model is therefore higher in the present example than the one trained according to Fig. 8 trained machine learning model.

[0169] The machine learning models according to the present disclosure may, for example, be an artificial neural network, or may comprise one or more such artificial neural networks.

[0170] An artificial neural network comprises at least three layers of processing elements: a first layer with input neurons (nodes), an N-th layer with at least one output neuron (node), and N-2 inner layers, where N is a natural number and greater than 2.

[0171] The input neurons are used to receive the at least one input representation. Typically, there is one input neuron for each pixel or voxel of an input representation if the representation is a spatial representation in the form of a raster graphic, or one input neuron for each frequency present in the input representation if the representation is a frequency-space representation. Additional input neurons may be present for additional input values ​​(e.g., information about the study area, the object under investigation, the conditions that prevailed when the input representation was generated, information about the state represented by the input representation, information about the time or time period during which the input representation was generated, and / or information about the amount of contrast agent administered).

[0172] The output neurons can be used to output a synthetic representation.

[0173] The processing elements of the layers between the input neurons and the output neurons are interconnected in a predetermined pattern with predetermined connection weights.

[0174] Preferably, the artificial neural network is a so-called Convolutional Neural Network (CNN) or it includes one.

[0175] A CNN typically consists essentially of filters (convolutional layer) and aggregation layers (pooling layer) that alternate and repeat, and ultimately of one or more layers of "normal" fully connected neurons (dense / fully connected layer).

[0176] Training a neural network can be performed, for example, using a backpropagation method. The goal is to achieve the most reliable possible mapping of the at least one input representation to the synthetic representation. The quality of the prediction is described by an error function. The aim is to minimize this error function. In the backpropagation method, the artificial neural network is trained by changing the connection weights.

[0177] In the trained state, the connection weights between the processing elements contain information regarding the dynamics of the relationship between the input representation(s) and the synthetic representation. This information can be used to predict a synthetic representation of the domain of a new object of study based on one or more existing representations of the domain of study. The term "new" here means that representations of the domain of study of the new object of study were not previously used in training the machine learning model.

[0178] A cross-validation method can be used to split the data into training and validation datasets. The training dataset is used for backpropagation training of the network weights. The validation dataset is used to verify the predictive accuracy with which the trained network can be applied to unknown (new) data.

[0179] The artificial neural network can have an autoencoder architecture; for example, the artificial neural network can have an architecture like the U-Net exhibit (see e.g. O. Ronneberger et al.: U-net: Convolutional networks for biomedical image segmentation, International Conference on Medical image computing and computer-assisted intervention, pages 234-241, Springer, 2015, https: / / doi.org / 10.1007 / 978-3-319-24574-4_28).

[0180] The artificial neural network can be a Generative Adversarial Network (GAN).

[0181] The artificial neural network can be a Regularized Generative Adversarial Network (see e.g. Q. Li et al.: RegGAN: An End-to-End Network for Building Footprint Generation with Boundary Regularization, Remote Sens. 2022, 14, 1835).

[0182] The artificial neural network can be a Conditional Adversarial Network be (see e.g. P. Isola et al.: Image-to-Image Translation with Conditional Adversarial Networks, arXiv:1611.07004 [cs.CV]).

[0183] The artificial neural network can be a transformer network (see e.g. D. Karimi et al.: Convolution-Free Medical Image Segmentation using Transformers, arXiv:2102.13645 [eess.IV]).

[0184] Fig. 12 Figure 1 shows an exemplary and schematic computer system according to the present disclosure, which can be used to train the machine learning model and / or to use the trained machine learning model for prediction.

[0185] A "computer system" is a system for electronic data processing that processes data using programmable instructions. Such a system typically comprises a "computer," the unit containing a processor for performing logical operations, as well as peripherals.

[0186] In computer technology, "peripherals" refers to all devices connected to a computer that are used to control the computer and / or as input and output devices. Examples include monitors (screens), printers, scanners, mice, keyboards, drives, cameras, microphones, speakers, etc. Internal ports and expansion cards are also considered peripherals in computer technology.

[0187] The in Fig. 12 The computer system shown (10) comprises a receiving unit (11), a control and computing unit (12) and an output unit (13).

[0188] The control and computing unit (12) serves to control the computer system (10), to coordinate the data flows between the units of the computer system (10) and to perform calculations.

[0189] The control and computing unit (12) is configured: to provide a trained machine learning model, to cause the receiving unit (11) to receive at least one input representation of an investigation area of ​​an investigation object, wherein the at least one input representation represents the investigation area without and / or after an application of a first amount of a contrast agent, to input the received at least one input representation into the trained machine learning model, to receive from the machine learning model a synthetic representation of the investigation area of ​​the investigation object, to cause the output unit (13) to output and / or store and / or transmit the synthetic representation of the investigation area of ​​the investigation object to a separate computer system.

[0190] Fig. 13 Figure 1 shows, by way of example and schematically, a further embodiment of the computer system according to the invention. The computer system (1) comprises a processing unit (21) which is connected to a memory (22). The processing unit (21) and the memory (22) form a control and arithmetic unit, as described in Fig. 12 shown.

[0191] The processing unit (21) (English: processing unit The processing unit (21) may comprise one or more processors alone or in combination with one or more memories. The processing unit (21) may be ordinary computer hardware capable of processing information such as digital images, computer programs, and / or other digital information. The processing unit (21) typically consists of an arrangement of electronic circuits, some of which may be implemented as an integrated circuit or as several interconnected integrated circuits (an integrated circuit is sometimes referred to as a "chip"). The processing unit (21) may be configured to execute computer programs, which may be stored in a working memory of the processing unit (21) or in the memory (22) of the same or another computer system.

[0192] The memory (22) can be ordinary computer hardware capable of storing information such as digital images (e.g., representations of the study area), data, computer programs, and / or other digital information, either temporarily and / or permanently. The memory (22) can be volatile and / or non-volatile and can be permanently installed or removable. Examples of suitable memory include RAM (Random Access Memory), ROM (Read-Only Memory), a hard disk, flash memory, a removable computer disk, an optical disc, a magnetic tape, or a combination of the above. Optical discs can include read-only compact discs (CD-ROM), read / write compact discs (CD-R / W), DVDs, Blu-ray discs, and similar media.

[0193] In addition to the memory (22), the processing unit (21) can also be connected to one or more interfaces (11, 12, 31, 32, 33) to display, transmit, and / or receive information. The interfaces can include one or more communication interfaces (32, 33) and / or one or more user interfaces (11, 12, 31). The one or more communication interfaces can be configured to send and / or receive information, for example, to and / or from an MRI scanner, a CT scanner, an ultrasound camera, other computer systems, networks, data storage devices, or the like. The one or more communication interfaces can be configured to transmit and / or receive information via physical (wired) and / or wireless communication links.The one or more communication interfaces may include one or more interfaces for connecting to a network, e.g., using technologies such as cellular, Wi-Fi, satellite, cable, DSL, fiber optic, and / or the like. In some examples, the one or more communication interfaces may include one or more near-field communication interfaces configured to connect devices using near-field communication technologies such as NFC, RFID, Bluetooth, Bluetooth LE, ZigBee, infrared (e.g., IrDA), or similar technologies.

[0194] The user interfaces may include a display (31). A display (31) may be configured to show information to a user. Suitable examples include a liquid crystal display (LCD), a light-emitting diode (LED) display, a plasma display (PDP), or the like. The user input interface(s) (11, 12) may be wired or wireless and may be configured to receive information from a user into the computer system (1), for example, for processing, storage, and / or display. Suitable examples of user input interfaces include a microphone, an image or video recording device (e.g., a camera), a keyboard or keypad, a joystick, a touch-sensitive surface (separate from or integrated into a touchscreen), or the like.In some examples, the user interfaces may include automatic identification and data capture (AIDC) technology for machine-readable information. This could include barcodes, radio frequency identification (RFID), magnetic stripes, optical character recognition (OCR), integrated circuit cards (ICC), and similar technologies. The user interfaces may also include one or more interfaces for communication with peripheral devices such as printers and the like.

[0195] One or more computer programs (40) can be stored in memory (22) and executed by the processing unit (21), which is programmed to perform the functions described in this description. The retrieval, loading, and execution of instructions from the computer program (40) can be sequential, with one instruction being retrieved, loaded, and executed at a time. However, the retrieval, loading, and / or execution can also be performed in parallel.

[0196] The memory (22) can also store the machine learning model according to the invention.

[0197] The computer system according to the invention can be designed as a laptop, notebook, netbook and / or tablet PC; it can also be a component of an MRI scanner, a CT scanner or an ultrasound diagnostic device.

[0198] Fig. 14 Figure 100 schematically shows, in the form of a flowchart, an embodiment of the procedure for training a machine learning model. The training procedure comprises the following steps: (110) Receiving and / or providing training data, wherein the training data for each of a plurality of investigation objects comprises a set of input data and target data, wherein each set comprises at least one input representation of an investigation area of ​​the investigation object as input data and a target representation of the investigation area of ​​the investigation object and a transformed target representation as target data, wherein the at least one input representation represents the investigation area without and / or after an application of a first quantity of a contrast agent and the synthetic representation represents the investigation area after the application of a second quantity of the contrast agent, wherein the transformed target representation represents at least a part of the investigation area of ​​the investigation object in the frequency domain.if the target representation represents the domain of investigation of the object of investigation in spatial space, or ▪ represents in spatial space if the target representation represents the domain of investigation of the object of investigation in frequency space, (120) Receiving and / or providing a machine learning model, wherein the machine learning model is configured to generate a synthetic representation of the domain of investigation of the object of investigation based on at least one input representation of a domain of investigation of an object of investigation and model parameters, (130) Training the machine learning model, wherein the training for each of the plurality of objects of investigation comprises: (131) Feeding the at least one input representation to the machine learning model,(132) Receiving a synthetic representation of the domain of the object of study from the machine learning model, (133) Generating and / or receiving a transformed synthetic representation based on and / or to the synthetic representation, wherein the transformed synthetic representation represents at least part of the domain of the object of study ▪ in the frequency domain if the synthetic representation represents the domain of the object of study in the spatial domain, or ▪ in the spatial domain if the synthetic representation represents the domain of the object of study in the frequency domain,(134) Quantifying the deviations i) between at least a part of the synthetic representation and at least a part of the target representation and ii) between at least a part of the transformed synthetic representation and at least a part of the transformed target representation using an error function, (135) Reducing the deviations by modifying model parameters, (140) Outputting and / or storing the trained machine learning model and / or the model parameters and / or transmitting the trained machine learning model and / or the model parameters to a separate computer system and / or using the trained machine learning model to generate a synthetic radiological image for a new subject.

[0199] Fig. 15Figure 200 schematically shows, in the form of a flowchart, an embodiment of the procedure for generating a synthetic representation of a domain of investigation of a new object of study using the trained machine learning model. The prediction procedure (200) comprises the following steps: (210) Providing a trained machine learning model, wherein the trained machine learning model has been trained on training data, to generate a synthetic representation of the area of ​​investigation of the object of investigation based on at least one input representation of an area of ​​investigation of an object of investigation, wherein the at least one input representation represents the area of ​​investigation without and / or after an application of a first quantity of a contrast agent and the synthetic representation represents the area of ​​investigation after the application of a second quantity of the contrast agent, wherein the training data for each object of investigation of a plurality of objects of investigation comprise i) at least one input representation of the area of ​​investigation, ii) a target representation of the area of ​​investigation and iii) a transformed target representation of the area of ​​investigation,wherein the transformed target representation represents at least a part of the domain of study in the frequency space if the target representation represents the domain of study in the spatial space, or represents it in the spatial space if the target representation represents the domain of study in the frequency space, wherein training the machine learning model comprises reducing discrepancies between i) at least a part of the synthetic representation and at least a part of the target representation and ii) between at least a part of a transformed synthetic representation and at least a part of the transformed target representation, (220) receiving at least one input representation of a domain of study of a new object of study,(230) Input representation of the new investigation area without and / or after application of an initial quantity of a contrast agent, (240) Input representation of the new investigation area into the trained machine learning model, (250) Receiving a synthetic representation of the new investigation area from the machine learning model, (250) Output and / or store the synthetic representation of the new investigation area and / or transmit the synthetic representation of the new investigation area to a separate computer system.

Claims

1. Computer-implemented method for generating a synthetic radiological image, comprising: - providing a trained machine-learning model (MLMt), o the trained machine-learning model (MLMt) having been trained by means of training data (TD) to generate on the basis of at least one input representation (R1, R2) of an examination region of an examination object a synthetic representation (SR) of the examination region of the examination object, o the training data (TD) comprising for each examination object of a multiplicity of examination objects i) an input representation (R1, R2) of the examination region of the examination object, ii) a target representation (TR) of the examination region of the examination object and iii) a transformed target representation (TRT), • the at least one input representation (R1, R2) representing the examination region of the respective examination object without or after administration of a first amount of the contrast agent, • the target representation (TR) representing the examination region of the respective examination object after administration of a second amount of the contrast agent, the second amount being larger than the first amount, • the transformed target representation (TRT) representing at least part of the examination region of the examination object in frequency space, if the target representation (TR) represents the examination region of the examination object in real space, or in real space, if the target representation (TR) represents the examination region of the examination object in frequency space, o the training of the machine-learning model (MLMt) comprising reducing the differences between i) at least part of the synthetic representation (SR) and at least part of the target representation (TR) and ii) between at least part of a transformed synthetic representation (SRT) and at least part of the transformed target representation (TRT), - receiving at least one input representation (R1*, R2*) of an examination region of a new examination object, the at least one input representation (R1*, R2*) representing the examination region of the new examination object without contrast agent and / or after administration of a first amount of a contrast agent, - inputting the at least one input representation (R1*, R2*) of the examination region of the new examination object into the trained machine-learning model (MLMt), - receiving a synthetic representation (SR*) of the examination region of the new examination object from the trained machine-learning model (MLMt), - outputting and / or storing the synthetic representation (SR*) of the examination region of the new examination object and / or transmitting the synthetic representation (SR*) of the examination region of the new examination object to a separate computer system.

2. Method according to Claim 1, wherein the training of the machine-learning model (MLMt) comprises: - receiving and / or providing the training data (TD), the training data (TD) comprising a set of input data and target data for each examination object of the multiplicity of examination objects, ∘ each set comprising the at least one input representation (R1, R2) of the examination region of the examination object as input data and the target representation (TR) of the examination region of the examination object and the transformed target representation (TRT) as target data, ∘ the at least one input representation (R1, R2) representing the examination region without contrast agent and / or after administration of the first amount of the contrast agent and the target representation (TR) representing the examination region after the administration of the second amount of the contrast agent, ∘ the transformed target representation (TRT) representing at least part of the examination region of the examination object • in frequency space, if the target representation (TR) represents the examination region of the examination object in real space, or • in real space, if the target representation (TR) represents the examination region of the examination object in frequency space, - receiving and / or providing a machine-learning model (MLM), the machine-learning model (MLM) being configured to generate on the basis of at least one input representation (R1, R2) of an examination region of an examination object and model parameters (MP) a synthetic representation (SR) of the examination region of the examination object, - training the machine-learning model (MLM), the training comprising for each examination object of the multiplicity of examination objects: ∘ feeding the at least one input representation (R1, R2) of the examination region of the examination object to the machine-learning model (MLM), ∘ receiving a synthetic representation (SR) of the examination region of the examination object from the machine-learning model (MLM), ∘ generating and / or receiving a transformed synthetic representation (SRT) on the basis of the synthetic representation (SR) and / or in relation to the synthetic representation (SR), the transformed synthetic representation (SRT) representing at least part of the examination region of the examination object • in frequency space, if the synthetic representation (SR) represents the examination region of the examination object in real space, or • in real space, if the synthetic representation (SR) represents the examination region of the examination object in frequency space, ∘ quantifying the differences i) between at least part of the synthetic representation (SR) and at least part of the target representation (TR) and ii) between at least part of the transformed synthetic representation (SRT) and at least part of the transformed target representation (TRT) by means of a loss function (L), ∘ reducing the differences by modifying model parameters (MP), - outputting and / or storing the trained machine-learning model (MLMt) and / or transmitting the trained machine-learning model (MLMt) to a separate computer system.

3. Method according to Claim 2, wherein the receiving and / or providing of the training data (TD) comprises: - generating a partial transformed target representation (TRT,P), the partial transformed target representation (TRT,P) being reduced to one part or multiple parts of the transformed target representation (TRT), wherein the generating and / or receiving of a transformed synthetic representation (SRT) on the basis of and / or in relation to the synthetic representation (SRT) comprises: - generating a partial transformed synthetic representation (SRT,P), the partial transformed synthetic representation (SRT,P) being reduced to one part or multiple parts of the transformed synthetic representation (SRT), wherein the quantifying of the differences between the transformed synthetic representation (SRT) and the transformed target representation (TRT) comprises: - quantifying the differences between the partial transformed synthetic representation (SRT,P) and the partial transformed target representation (TRT,P).

4. Method according to any of Claims 1 to 3, wherein the training comprises for each examination object of the multiplicity of examination objects: ∘ feeding the at least one input representation (R1, R2) to the machine-learning model, ∘ receiving the synthetic representation (SR) and a first transformed synthetic representation (SRT) of the examination region of the examination object from the machine-learning model (MLM), the first transformed synthetic representation (SRT) representing at least part of the examination region of the examination object • in frequency space, if the synthetic representation (SR) represents the examination region of the examination object in real space, or • in real space, if the synthetic representation (SR) represents the examination region of the examination object in frequency space, ∘ generating a second transformed synthetic representation (SRT) on the basis of the synthetic representation (SR) by means of a transform (T), the second transformed synthetic representation (SRT#) representing at least part of the examination region of the examination object • in frequency space, if the synthetic representation (SR) represents the examination region of the examination object in real space, or • in real space, if the synthetic representation (SR) represents the examination region of the examination object in frequency space, - quantifying the differences i) between at least part of the synthetic representation (SR) and at least part of the target representation (TR), ii) between at least part of the first transformed synthetic representation (SRT) and at least part of the transformed target representation (TRT) and iii) between at least part of the first transformed synthetic representation (SRT) and at least part of the second transformed synthetic representation (SRT) by means of a loss function (L), - reducing the differences by modifying model parameters (MP),5. Method according to any of Claims 1 to 4, wherein the machine-learning model (MLM) undergoing training comprises a first machine-learning model (MLM1) and a second machine-learning model (MLM2), wherein the first machine-learning model (MLM1) is configured to generate on the basis of at least one input representation (R1, R2) and model parameters (MP1) a synthetic representation (SR) of the examination region of the examination object, wherein the second machine-learning model (MLM2) is configured to reconstruct on the basis of the synthetic representation (SR) of the examination region of the examination object and model parameters (MP2) at least one input representation (R1, R2), wherein the training comprises for each examination object of the multiplicity of examination objects: ∘ generating a transformed input representation (R2) on the basis of at least one input representation (R1) by means of a transform (T), the transformed input representation (R2) representing at least part of the examination region of the examination object • in frequency space, if the at least one input representation (R1) represents the examination region of the examination object in real space, or • in real space, if the at least one input representation (R1) represents the examination region of the examination object in frequency space, ∘ feeding the at least one input representation (R1) and / or the transformed input representation (R2) to the first machine-learning model, ∘ receiving a synthetic representation (SR) of the examination region of the examination object from the first machine-learning model (MLM1), ∘ generating and / or receiving a transformed synthetic representation (SRT) on the basis of and / or in relation to the synthetic representation (SR), the transformed synthetic representation (SRT) representing at least part of the examination region of the examination object • in frequency space, if the synthetic representation (SR) represents the examination region of the examination object in real space, or • in real space, if the synthetic representation (SR) represents the examination region of the examination object in frequency space, ∘ feeding the synthetic representation (SR) and / or the transformed synthetic representation (SRT) to the second machine-learning model (MLM2), ∘ receiving a predicted input representation (R1#) from the second machine-learning model (MLM2), ∘ generating and / or receiving a transformed predicted input representation (R2#) on the basis of and / or in relation to the predicted input representation (R1#), the transformed predicted input representation (R2#) representing at least part of the examination region of the examination object • in frequency space, if the predicted input representation (R1#) represents the examination region of the examination object in real space, or • in real space, if the predicted input representation (R1#) represents the examination region of the examination object in frequency space, ∘ quantifying the differences i) between at least part of the synthetic representation (SR) and at least part of the target representation (TR), ii) between at least part of the transformed synthetic representation (SRT) and at least part of the transformed target representation (TRT), iii) between at least part of the input representation (R1) and at least part of the predicted input representation (R1#) and iv) between at least part of the transformed input representation (R2) and at least part of the transformed predicted input representation (R2#) by means of a loss function (L), o reducing the differences by modifying model parameters.

6. Method according to any of Claims 1 to 5, wherein the examination object is a mammal, preferably a human.

7. Method according to any of Claims 1 to 6, wherein the examination region is or includes a liver, brain, heart, kidney, lung, stomach, intestine, pancreas, thyroid gland, prostate or breast of a human.

8. Method according to any of Claims 1 to 7, wherein each input representation of the at least one input representation (R1, R2, R1*, R2*) is a representation of the examination region in real space, the target representation (TR) is a representation of the examination region in real space, and the synthetic representation (SR, SR*) is a representation of the examination region in real space.

9. Method according to any of Claims 3 to 8, wherein the partial transformed synthetic representation (SRT,P) represents the examination region in frequency space, the partial transformed synthetic representation (SRT,P) being reduced to a frequency range of the transformed synthetic representation (SRT), contrast information being encoded in the frequency range.

10. Method according to any of Claims 3 to 8, wherein the partial transformed synthetic representation (SRT,P) represents the examination region in frequency space, the partial transformed synthetic representation (SRT,P) being reduced to a frequency range of the transformed synthetic representation (SRT), information about fine structures being encoded in the frequency range.

11. Computer system (10) for generating a synthetic radiological image, 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 provide a trained machine-learning model (MLMt), ∘ the trained machine-learning model (MLMt) having been trained by means of training data (TD) to generate on the basis of at least one input representation (R1, R2) of an examination region of an examination object a synthetic representation (SR) of the examination region of the examination object, ∘ the training data (TD) comprising for each examination object of a multiplicity of examination objects i) an input representation (R1, R2) of the examination region of the examination object, ii) a target representation (TR) of the examination region and iii) a transformed target representation (TRT), • the at least one input representation (R1, R2) representing the examination region of the respective examination object without contrast agent or after administration of a first amount of a contrast agent, • the target representation (TR) representing the examination region of the respective examination object after administration of a second amount of the contrast agent, the second amount being larger than the first amount, • the transformed target representation (TRT) representing at least part of the examination region of the respective examination object in frequency space, if the target representation (TR) represents the examination region of the respective examination object in real space, or in real space, if the target representation (TR) represents the examination region of the respective examination object in frequency space, ∘ the training of the machine-learning model (MLMt) comprising reducing differences between i) at least part of the synthetic representation (SR) and at least part of the target representation (TR) and ii) at least part of a transformed synthetic representation (SRT) and at least part of the transformed target representation (TRT), - wherein the control and calculation unit (12) is configured to cause the receiving unit (11) to receive at least one input representation (R1*, R2*) of an examination region of a new examination object, the at least one input representation (R1*, R2*) representing the examination region without and / or after administration of a first amount of a contrast agent, - wherein the control and calculation unit (12) is configured to input the received at least one input representation (R1*, R2*) into the trained machine-learning model (MLMt), - wherein the control and calculation unit (12) is configured to receive from the trained machine-learning model (MLMt) a synthetic representation (SR*) of the examination region of the new examination object, - wherein the control and calculation unit (12) is configured to cause the output unit (13) to output the synthetic representation (SR*) of the examination region of the new examination object and / or to store it and / or to transmit it to a separate computer system.

12. Computer program product for generating a synthetic radiological image, comprising a computer program (40) that can be loaded into a working memory (22) of a computer system (1), where it causes the computer system (1) to execute the following steps - providing a trained machine-learning model (MLMt), ∘ the trained machine-learning model (MLMt) having been trained by means of training data (TD) to generate on the basis of at least one input representation (R1, R2) of an examination region of an examination object a synthetic representation (SR) of the examination region of the examination object, ∘ the training data (TD) comprising for each examination object of a multiplicity of examination objects i) an input representation (R1, R2) of the examination region of the examination object, ii) a target representation (TR) of the examination region of the examination object and iii) a transformed target representation (TRT), • the at least one input representation (R1, R2) representing the examination region of the respective examination object without or after administration of a first amount of a contrast agent, • the target representation (TR) representing the examination region of the respective examination object after administration of a second amount of the contrast agent, the second amount being larger than the first amount, • the transformed target representation (TRT) representing at least part of the examination region of the examination object in frequency space, if the target representation (TR) represents the examination region of the examination object in real space, or in real space, if the target representation (TR) represents the examination region of the examination object in frequency space, ∘ the training of the machine-learning model (MLMt) comprising reducing the differences between i) at least part of the synthetic representation (SR) and at least part of the target representation (TR) and ii) between at least part of a transformed synthetic representation (SRT) and at least part of the transformed target representation (TRT), - receiving at least one input representation (R1*, R2*) of an examination region of a new examination object, the at least one input representation (R1*, R2*) representing the examination region of the new examination object without contrast agent and / or after administration of a first amount of a contrast agent, - inputting the at least one input representation (R1*, R2*) of the examination region of the new examination object into the trained machine-learning model (MLMt), - receiving a synthetic representation (SR*) of the examination region of the new examination object from the trained machine-learning model (MLMt), - outputting and / or storing the synthetic representation (SR*) of the examination region of the new examination object and / or transmitting the synthetic representation (SR*) of the examination region of the new examination object to a separate computer system.

13. Use of a contrast agent in a radiological examination method, the radiological examination method comprising: - providing a trained machine-learning model (MLMt), ∘ the trained machine-learning model (MLMt) having been trained by means of training data (TD) to generate on the basis of at least one input representation (R1, R2) of an examination region of an examination object a synthetic representation (SR) of the examination region of the examination object, ∘ the training data (TD) comprising for each examination object of a multiplicity of examination objects i) an input representation (R1, R2) of the examination region of the examination object, ii) a target representation (TR) of the examination region of the examination object and iii) a transformed target representation (TRT), • the at least one input representation (R1, R2) representing the examination region of the respective examination object without or after administration of a first amount of the contrast agent, • the target representation (TR) representing the examination region of the respective examination object after administration of a second amount of the contrast agent, the second amount being larger than the first amount, • the transformed target representation (TRT) representing at least part of the examination region of the examination object in frequency space, if the target representation (TR) represents the examination region of the examination object in real space, or in real space, if the target representation (TR) represents the examination region of the examination object in frequency space, ∘ the training of the machine-learning model (MLMt) comprising reducing the differences between i) at least part of the synthetic representation (SR) and at least part of the target representation (TR) and ii) between at least part of a transformed synthetic representation (SRT) and at least part of the transformed target representation (TRT), - receiving at least one input representation (R1*, R2*) of an examination region of a new examination object, the at least one input representation (R1*, R2*) representing the examination region of the new examination object without contrast agent and / or after administration of a first amount of a contrast agent, - inputting the at least one input representation (R1*, R2*) of the examination region of the new examination object into the trained machine-learning model (MLMt), - receiving a synthetic representation (SR*) of the examination region of the new examination object from the trained machine-learning model (MLMt), - outputting and / or storing the synthetic representation (SR*) of the examination region of the new examination object and / or transmitting the synthetic representation (SR*) of the examination region of the new examination object to a separate computer system.

14. Kit comprising a contrast agent and a computer program (40) which can be loaded into a working memory (22) of a computer system (1), where it causes the computer system (1) to execute the following steps: - providing a trained machine-learning model (MLMt), ∘ the trained machine-learning model (MLMt) having been trained by means of training data (TD) to generate on the basis of at least one input representation (R1, R2) of an examination region of an examination object a synthetic representation (SR) of the examination region of the examination object, ∘ the training data (TD) comprising for each examination object of a multiplicity of examination objects i) an input representation (R1, R2) of the examination region of the examination object, ii) a target representation (TR) of the examination region of the examination object and iii) a transformed target representation (TRT), • the at least one input representation (R1, R2) representing the examination region of the respective examination object without or after administration of a first amount of the contrast agent, • the target representation (TR) representing the examination region of the respective examination object after administration of a second amount of the contrast agent, the second amount being larger than the first amount, • the transformed target representation (TRT) representing at least part of the examination region of the examination object in frequency space, if the target representation (TR) represents the examination region of the examination object in real space, or in real space, if the target representation (TR) represents the examination region of the examination object in frequency space, ∘ the training of the machine-learning model (MLMt) comprising reducing the differences between i) at least part of the synthetic representation (SR) and at least part of the target representation (TR) and ii) between at least part of a transformed synthetic representation (SRT) and at least part of the transformed target representation (TRT), - receiving at least one input representation (R1*, R2*) of an examination region of a new examination object, the at least one input representation (R1*, R2*) representing the examination region of the new examination object without contrast agent and / or after administration of a first amount of a contrast agent, - inputting the at least one input representation (R1*, R2*) of the examination region of the new examination object into the trained machine-learning model (MLMt), - receiving a synthetic representation (SR*) of the examination region of the new examination object from the trained machine-learning model (MLMt), - outputting and / or storing the synthetic representation (SR*) of the examination region of the new examination object and / or transmitting the synthetic representation (SR*) of the examination region of the new examination object to a separate computer system.

15. Kit according to Claim 14, wherein the contrast agent is or comprises one or more contrast agents selected from the following list: gadoxetate disodium, gadolinium(III) 2-[4,7,10-tris(carboxymethyl)-1,4,7,10-tetrazacyclododec-1-yl]acetic acid, gadolinium(III) ethoxybenzyldiethylenetriaminepentaacetic acid, gadolinium(III) 2-[3,9-bis[1-carboxylato-4-(2,3-dihydroxypropylamino)-4-oxobutyl]-3,6,9,15-tetrazabicyclo[9.3.1]pentadeca-1(15),11,13-trien-6-yl]-5-(2,3-dihydroxypropylamino)-5-oxopentanoate, dihydrogen [(±)-4-carboxy-5,8,11-tris(carboxymethyl)-1-phenyl-2-oxa-5,8,11-triazatridecan-13-oato(5-)]gadolinate(2-), tetragadolinium [4,10-bis(carboxylatomethyl)-7-{3,6,12,15-tetraoxo-16-[4,7,10-tris-(carboxylatomethyl)-1,4,7,10-tetraazacyclododecan-1-yl]-9,9-bis({ [({2-[4,7,10-tris-(carboxylatomethyl)-1,4,7,10-tetraazacyclododecan-1-yl]propanoyl}amino)acetyl]amino}methyl)-4,7,11,14-tetraazaheptadecan-2-yl}-1,4,7,10-tetraazacyclododecan-1-yl]acetate, gadolinium 2,2',2"-(10-{1-carboxy-2-[2-(4-ethoxyphenyl)ethoxy]ethyl}-1,4,7,10-tetraazacyclododecane-1,4,7-triyl)triacetate, gadolinium 2,2',2"-{10-[1-carboxy-2-{4-[2-(2-ethoxyethoxy)ethoxy]phenyl}ethyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}triacetate, gadolinium 2,2',2' '-{10-[(1R)-1-carboxy-2-{4-[2-(2-ethoxyethoxy)ethoxy]phenyl}ethyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}triacetate, gadolinium (2S,2'S,2"S)-2,2',2"-{10-[(1S)-1-carboxy-4-{4-[2-(2-ethoxyethoxy)ethoxy] phenyl}butyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}tris(3-hydroxypropanoate), gadolinium 2,2',2"-{10-[(1S)-4-(4-butoxyphenyl)-1-carboxybutyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}triacetate, gadolinium-2,2',2"-{(2S)-10-(carboxymethyl)-2-[4-(2-ethoxyethoxy)benzyl]-1,4,7,10-tetraazacyclododecane-1,4,7-triyl}triacetate, gadolinium-2,2',2"-[10-(carboxymethyl)-2-(4-ethoxybenzyl)-1,4,7,10-tetraazacyclododecane-1,4,7-triyl]triacetate, gadolinium(III) 5,8-bis(carboxylatomethyl)-2-[2-(methylamino)-2-oxoethyl]-10-oxo-2,5,8,11-tetraazadodecane-1-carboxylate hydrate, gadolinium(III) 2-[4-(2-hydroxypropyl)-7,10-bis(2-oxido-2-oxoethyl)-1,4,7,10-tetrazacyclododec-1-yl]acetate, gadolinium(III) 2,2',2"-(10-((2R,3S)-1,3,4-trihydroxybutan-2-yl)-1,4,7,10-tetraazacyclododecane-1,4,7-triyl)triacetate, a Gd3+ complex of a compound of the formula (I) where Ar is a group selected from where # is the linkage to X, X is a group selected from CH2, (CH2)2, (CH2)3, (CH2)4 and *-(CH2)2-O-CH2-#, where * is the linkage to Ar and # is the linkage to the acetic acid residue, R1, R2 and R3 are each independently a hydrogen atom or a group selected from C1-C3 alkyl, -CH2OH, -(CH2)2OH and-CH2OCH3, R4 is a group selected from C2-C4 alkoxy, (H3C-CH2)-O-(CH2)2-O-, (H3C-CH2)-O-(CH2)2-O-(CH2)2-O- and (H3C-CH2)-O-(CH2)2-O-(CH2)2-O-(CH2)2-O-, R5 is a hydrogen atom, and R6 is a hydrogen atom, or a stereoisomer, tautomer, hydrate, solvate or salt thereof, or a mixture thereof, a Gd3+ complex of a compound of the formula (II) where Ar is a group selected from where # is the linkage to X, X is a group selected from CH2, (CH2)2, (CH2)3, (CH2)4 and *-(CH2)2-O-CH2-#, where * is the linkage to Ar and # is the linkage to the acetic acid residue, R7 is a hydrogen atom or a group selected from C1-C3 alkyl, -CH2OH, -(CH2)2OH and -CH2OCH3; R8 is a group selected from C2-C4 alkoxy, (H3C-CH2O)-(CH2)2-O-, (H3C-CH2O)-(CH2)2-O-(CH2)2-O- and (H3C-CH2O)-(CH2)2-O-(CH2)2-O-(CH2)2-O-; R9 and R10 are each independently a hydrogen atom; or a stereoisomer, tautomer, hydrate, solvate or salt thereof, or a mixture thereof.