MACHINE LEARNING IN THE FIELD OF CONTRAST-ENHANCED RADIOLOGY

Description

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

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

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

[0004] All these procedures can utilize substances that facilitate the visualization or differentiation of certain structures within an object under investigation. These substances are called contrast agents.

[0005] Based on their distribution pattern in tissue, contrast agents can be roughly divided into the following categories: extracellular, intracellular and intravascular contrast agents.

[0006] 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®).

[0007] Intracellular contrast agents are partially absorbed into tissue cells and subsequently excreted. For example, gadoxetic acid-based intracellular MRI contrast agents are characterized by their partial uptake specifically 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 under brand names such as Primovist® and Eovist®. Another MRI contrast agent with lower hepatocyte uptake is gadobenate dimeglumine (Multihance®).

[0008] Intravascular contrast agents are characterized by a significantly longer residence time in the bloodstream compared to extracellular contrast agents. Gadofosveset, for example, is a gadolinium-based intravascular MRI contrast agent. It was used in its trisodium monohydrate form (Ablavar®). It binds to serum albumin, which results in the contrast agent's long residence time in the bloodstream (half-life in blood approximately 17 hours).

[0009] Ablavar® was withdrawn from the market in 2017. No other blood-pool contrast agent is commercially available for magnetic resonance imaging (MRI). Similarly, no blood-pool contrast agent is currently available for computed tomography (CT).

[0010] When generating radiological images with a comparatively long acquisition / scan time, for example images of the thorax and abdomen under free breathing to visualize the vascular system (e.g., pulmonary embolism diagnosis under free breathing in MRI), an extracellular contrast agent is excreted relatively quickly from the blood vessel system, so that the contrast decreases rapidly.

[0011] When a scan volume is scanned layer by layer to generate multiple tomograms, the contrast enhancement decreases continuously from tomogram to tomogram due to the washout of the contrast agent. Layers scanned at a later time point will appear with lower contrast in a corresponding tomogram than layers scanned at an earlier time point.

[0012] The present invention addresses this problem.

[0013] The publication "Prior-Guided Image Reconstruction for Accelerated Multi-Contrast MRI via Generative Adversarial Networks" (2020) to UH Dar Salman et al. presents a method for the recovery of undersampled multi-contrast MRI images based on generative adversarial networks (GAN).

[0014] A first object of the present invention is a method for generating an artificial, contrast-enhanced tomogram, comprising the steps of: • Receiving a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an object under investigation after the application of a contrast agent; • Receiving a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the object under investigation after the application of the contrast agent, wherein the second tomogram was generated temporally after the first tomogram; • Providing a prediction model, wherein the prediction model has been trained on a training dataset to transfer contrast information from at least one tomogram to another tomogram, thereby generating an artificial tomogram; • Feeding the first and second tomograms to the prediction model; • Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the second layer in the acquisition volume.where contrast information from the first tomogram has been transferred to the artificial tomogram, ∘ Output of the artificial tomogram.

[0015] Preferably, the first subject matter of the present invention relates to a method for generating an artificial, contrast-enhanced tomogram, comprising the steps of: • Receiving a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an object under investigation after the application of a contrast agent; • Receiving a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the object under investigation after the application of the contrast agent; • Receiving a third tomogram, wherein the third tomogram represents a third layer within the acquisition volume of the object under investigation after the application of the contrast agent, the third layer being located between the first and second layers, the third tomogram being generated temporally after the first and second tomograms; • Providing a prediction model, wherein the prediction model has been trained on a training dataset.To transfer contrast information from at least two tomograms to one tomogram, thereby generating an artificial tomogram; ∘ Feeding the first, second, and third tomograms to the prediction model; ∘ Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the third layer in the acquisition volume, and wherein contrast information from the first and second tomograms has been transferred into the artificial tomogram; ∘ Outputting the artificial tomogram.

[0016] Another object of the present invention is a method for training a predictive model, comprising the steps Receiving a training dataset, wherein the training dataset comprises a plurality of reference tomograms for each reference object of a plurality of reference objects, wherein the reference tomograms represent an acquisition volume of the reference object after the application of a contrast agent, wherein the reference tomograms for each reference object comprise at least two tomograms: a first reference tomogram and a second reference tomogram, wherein the first reference tomogram represents a first layer within the acquisition volume, wherein the second reference tomogram represents a second layer within the acquisition volume, and wherein the second reference tomogram was acquired temporally after the first reference tomogram; training the prediction model, wherein the first and second reference tomograms are supplied to the prediction model during training, and wherein the prediction model is trained.To transfer contrast enhancement from the first reference tomogram to the second reference tomogram and to generate an artificial tomogram representing the second layer.

[0017] Preferably, a further object of the present invention relates to a method for training a predictive model, comprising the steps Receiving a training dataset, wherein the training dataset comprises a plurality of reference tomograms for each reference object of a plurality of reference objects, ∘ wherein the reference tomograms represent an acquisition volume of the reference object after the application of a contrast agent, ∘ wherein the reference tomograms for each reference object comprise at least three tomograms: a first reference tomogram, a second reference tomogram, and a third reference tomogram, ▪ wherein the first reference tomogram represents a first layer within the acquisition volume, ▪ wherein the second reference tomogram represents a second layer within the acquisition volume, ▪ wherein the third reference tomogram represents a third layer within the acquisition volume, the third layer being located in the acquisition volume between the first layer and the second layer,wherein the third reference tomogram was generated temporally after the first and second reference tomograms, training the prediction model, wherein the first, second, and third reference tomograms are fed to the prediction model, wherein the prediction model is trained, transferring contrast enhancement from the first and second reference tomograms to the third reference tomogram and generating an artificial tomogram representing the third layer.

[0018] Another object of the present invention 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 cause the receiving unit to receive at least one first and one second tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an object under investigation after the application of a contrast agent, wherein the second tomogram represents a second layer within the acquisition volume of the object under investigation after the application of the contrast agent, wherein the second tomogram was generated temporally after the first tomogram, wherein the control and processing unit is configured to feed the first and the second tomogram to a prediction model, wherein the prediction model has been trained on a training dataset, to transfer contrast information from at least one tomogram to another tomogram, thereby generating an artificial tomogram.∘where the control and processing unit is configured to receive an artificial tomogram from the prediction model, wherein the artificial tomogram represents the second layer in the acquisition volume, wherein contrast information from the first tomogram has been transferred to the artificial tomogram, ∘where the control and processing unit is configured to cause the output unit to output the artificial tomogram.

[0019] Another object of the present invention 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: • Receiving a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an object under investigation after the application of a contrast agent; • Receiving a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the object under investigation after the application of the contrast agent, wherein the second tomogram was generated temporally after the first tomogram; • Feeding the first and second tomograms to a prediction model, wherein the prediction model has been trained on a training dataset, to transfer contrast information from at least one tomogram to another tomogram and thereby generate an artificial tomogram; • Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the second layer in the acquisition volume.where contrast information from the first tomogram has been transferred to the artificial tomogram, ∘ Output of the artificial tomogram.

[0020] Another object of the present invention is the use of a contrast agent in a method for predicting an artificial contrast-enhanced tomogram, wherein the method comprises the following steps: • Generating a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an object under investigation after the application of a contrast agent; • Generating a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the object under investigation after the application of the contrast agent, wherein the second tomogram is generated temporally after the first tomogram; • Feeding the first and second tomograms to a prediction model, wherein the prediction model has been trained on a training dataset, to transfer contrast information from at least one tomogram to another tomogram and thereby generate an artificial tomogram; • Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the second layer in the acquisition volume.where contrast information from the first tomogram has been transferred to the artificial tomogram, ∘ Output of the artificial tomogram.

[0021] Another object is a contrast agent for use in a method for predicting an artificial contrast-enhanced tomogram, the method comprising the following steps: • Generating a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an object under investigation after the application of a contrast agent; • Generating a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the object under investigation after the application of the contrast agent, wherein the second tomogram is generated temporally after the first tomogram; • Feeding the first and second tomograms to a prediction model, wherein the prediction model has been trained on a training dataset, to transfer contrast information from at least one tomogram to another tomogram and thereby generate an artificial tomogram; • Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the second layer in the acquisition volume.where contrast information from the first tomogram has been transferred to the artificial tomogram, ∘ Output of the artificial tomogram.

[0022] Another item is a kit comprising a contrast agent and a computer program product according to the invention.

[0023] The invention is explained in more detail below without distinguishing between the subject matter of the invention (method, computer system, computer program product, use, contrast agent for use, kit). Rather, the following explanations are intended to apply analogously to all subject matter of the invention, regardless of the context in which they are made (method, computer system, computer program product, use, contrast agent for use, kit).

[0024] 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.

[0025] The present invention generates at least one artificial, contrast-enhanced tomogram for an object under investigation based on a plurality of real tomograms.

[0026] The "object of study" is usually a living being, preferably a mammal, and most preferably a human being.

[0027] The term "plural" means a number of at least two.

[0028] A "tomogram" is a representation of a layer within a scan volume of the object under investigation. Synonymous terms for "tomogram" include "section image," "layer image," and "layer scan."

[0029] A tomogram reproduces the internal structures as they would be after cutting open the object under investigation or after cutting out a (thin) slice.

[0030] A tomogram is usually the result of a radiological examination. Examples of such radiological examinations are computed tomography (CT), magnetic resonance imaging (MRI), and sonography.

[0031] The term "real" means that a tomogram is the result of a real measurement, i.e., a measurement actually performed on a test subject (preferably a living being). The term "real" serves to distinguish it from artificial tomograms, i.e., tomograms that are synthetically generated and therefore not the (direct) result of a real measurement, i.e., not the (direct) result of a measurement actually performed on a test subject. However, an artificial (synthetically generated) tomogram can be based on one or more real tomograms. Artificial tomograms generated according to the invention are based on at least two, preferably at least three, real tomograms.

[0032] To generate real tomograms, a portion of the object being examined—the examination area—is typically subjected to a radiological examination. The "examination area," also called the acquisition 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 (also known as a field radiograph). localizer The examination area can be defined manually. Alternatively or additionally, it can also be defined automatically, for example, based on a selected protocol. The examination area can be, for example, the liver or part of the liver, the lungs or part of the lungs, the heart or part of the heart, the aorta or part of the aorta, abdominal blood vessels, leg and pelvic blood vessels, the esophagus or part of the esophagus, the stomach or part of the stomach, the small intestine or part of the small intestine, the large intestine or part of the large intestine, the abdomen or part of the abdomen, the pancreas or part of the pancreas, and / or any other part of the object being examined.

[0033] The real tomograms used according to the invention to generate one or more artificial tomograms represent layers within an acquisition volume of an object under investigation after the application of a contrast agent. Preferably, the layers are planar layers with a defined layer thickness, arranged in a spatial sequence parallel to one another within the acquisition volume of the object under investigation.

[0034] In a first step, a contrast agent is administered to the subject (e.g., via an intravenous bolus injection), which distributes throughout the imaging volume, and multiple actual tomograms are acquired. It is also possible to acquire one or more tomograms before the contrast agent is administered; these so-called native images show the imaging volume without the administration of a contrast agent.

[0035] The tomograms produced after application of the contrast agent are usually generated sequentially: the imaging volume is scanned layer by layer and a tomogram is generated from each layer.

[0036] This is schematically shown in Fig. 1 shown. Fig. 1 The diagram schematically depicts a person whose upper body is undergoing a radiological examination. The upper body represents the scan volume. Seven tomograms are generated from this scan volume, which are then... Fig. 1 The tomograms are labeled with the reference symbols T1, T2, T3, T4, T5, T6, and T7. Each tomogram represents a layer within the acquisition volume. These layers are in Fig. 1 The layers are designated with the reference symbols S1, S2, S3, S4, S5, S6, and S7. Each of these layers has a defined thickness. The layers can be directly adjacent to one another, spaced apart, or partially overlapping. In any case, there are two layers that define the recording volume; these are layers S1 and S7. These layers are also referred to as the outer layers in this description. The remaining, inner layers S2, S3, S4, S5, and S6 each have two immediately adjacent layers (also referred to as nearest neighbors). S2's immediately adjacent layers are S1 and S3, S3's immediately adjacent layers are S2 and S4, and so on. The term "immediately adjacent" does not mean that the layers are directly adjacent to one another; as already explained, immediately adjacent layers can also overlap or be spaced apart.The term "immediately adjacent" is intended to indicate that there is no layer that is geographically closer to a layer under consideration than the immediately adjacent layers. Only layers from which a tomogram is generated are considered.

[0037] In general, there are n layers, whereby n is a natural number greater than or equal to 2. There are 2 outer layers and n-2 inner layers. The layers S1, ..., Sn are arranged parallel to each other in a spatial sequence, where the spatial sequence is represented by the indices 1, ... , n.

[0038] In conventional radiological examinations, the slices are usually scanned sequentially, i.e., in the order S1, S2, S3, S4, S5, S6, S7 or in the order S7, S6, S5, S4, S3, S2, S1. The terms "scan" or "scanning" are used synonymously in this description with the term "generating a tomogram".

[0039] Fig. 2 This schematically illustrates the generation of tomograms in a conventional radiological examination along a timeline. The abscissa (x-axis) represents time. t on. The time t 0 indicates the administration of a contrast agent; at the time t 0 A contrast agent is administered to the subject of the examination. The intensity is measured on the ordinate (y-axis). Ia signal attributable to the contrast agent in the imaging volume is plotted. The curve represents the contrast enhancement caused by the contrast agent in the imaging volume. At time t 0 There is no contrast agent in the imaging volume yet; it takes a certain amount of time for the contrast agent to reach the imaging volume after administration (e.g., as a bolus into an arm vein). At the time t 1 A significant contrast enhancement, caused by the contrast agent in the imaging volume, is already present. At the time t 1 Therefore, a first tomogram T1 is generated. The acquisition of a tomogram itself takes a certain amount of time. In the present example, the time to generate a tomogram Δ is t In the present example, the generation of the next tomogram begins immediately after the generation of the previous one. This occurs within the time between...t 1 and t 2 that is, the tomogram T1 was generated from layer S1. In the time between t 2 and t 3 The tomogram T2 is generated by layer S2; in the time between t 3 and t 4 The tomogram T3 is generated by layer S3, and so on.

[0040] As in Fig. 2 As can be seen by way of example, the contrast enhancement achieves I shortly before reaching that point t 2 Its contrast enhancement reaches its maximum and then decreases continuously. This decrease is caused by the gradual removal of the contrast agent from the imaging volume. As a result, tomograms acquired later in the time interval will have lower contrast enhancement than those acquired earlier.

[0041] If the tomograms are acquired in the order of the slice sequence (S1, S2, S3, S4, S5, S6, S7), i.e., in the sequence T1, T2, T3, T4, T5, T6, T7, the contrast enhancement caused by the contrast agent is lower in the "lower" slices (S7 and above) than the contrast enhancement in the "upper" slices (S1 and below). Tomograms of slices that exhibit lower contrast enhancement due to increasing contrast agent loss from the acquisition volume compared to tomograms of slices less affected by loss are also referred to as low-contrast tomograms in this description. Conversely, those tomograms of slices less affected by loss are referred to as high-contrast tomograms. In the case of the in Fig. 2 In the exemplary radiological examination shown, the high-contrast tomograms focus on the upper torso of the subject, while the low-contrast tomograms focus on the lower torso. This results in poorer image quality in the lower torso compared to the upper torso, because the acquired tomograms have lower contrast.

[0042] When generating radiological images with a comparatively long scan time using a contrast agent, the sequential generation of tomograms leads to tomograms taken later in the sequence having lower contrast and therefore poorer image definition.

[0043] According to the invention, this problem is solved by transferring contrast information from a high-contrast tomogram to a low-contrast tomogram.

[0044] Furthermore, tomograms are preferably not produced in the order of the layer sequence, but in a sequence in which the layers of the high-contrast tomograms are distributed as evenly as possible in the recording volume and the layers of the low-contrast tomograms lie between the layers of the high-contrast tomograms.

[0045] In a preferred embodiment, when a number of n Layers S1, S2, ..., Sn first generate the tomograms T1, T3, T5, ... and then the tomograms T2, T4, T6, ... where one tomogram T i each layer S i represented, whereby i an index that ranges from 1 to n accepts nan odd number greater than 2. In other words, tomograms are generated from the layers with odd indices in ascending or descending order, and then tomograms from the layers with even indices are generated in ascending or descending order. This is shown schematically and by way of example in Fig. 3 depicted. As shown in Fig. 2 is in Fig. 3 in a graphical representation the intensity I The measurement signal of a radiological examination, resulting from the presence of contrast medium in the imaging volume, is displayed as a function of time. Tomograms T1, T2, ..., T7 are generated from the immediately adjacent slices in the imaging volume S1, S2, ..., S7. However, the tomograms are not generated in the order in which the respective slices are adjacent to each other in the imaging volume, but rather in the sequence T1, T3, T5, T7, T2, T4, T6.

[0046] An alternative sequence is T7, T5, T3, T1, T6, T4, T2.

[0047] Another alternative sequence is T1, T3, T5, T7, T6, T4, T2.

[0048] Another alternative sequence is T7, T5, T3, T1, T2, T4, T6.

[0049] Preferably, the number of tomograms produced is an odd number.

[0050] In another embodiment, if there is an odd number n on layers with the indices 1, 2, ... , n, The layers were scanned in an order according to the following rules: (i) Tomograms are first generated from the layers with odd-numbered indices, and then from the layers with even-numbered indices. A layer from which a tomogram has been generated is called a scanned layer. A scanned layer is assigned an identification number; the identification number indicates the position in the sequence at which the layer is scanned. (ii) Tomograms are first generated from the two outermost layers in the sequence S1, S2, S3, S4, S5, S6, S6, S7 ... n or Sn, S1 is generated. (iii) The gap with the greatest distance between two scanned layers is then identified. A gap is an area between two scanned layers containing unscanned layers. If there are multiple gaps with the same distance, the sum of the identifiers of the scanned layers that define each gap is calculated. The gap with the largest sum of identifiers is selected. If there are multiple such gaps with the same largest identifiers, the gap with the greatest distance to the last scanned layer is selected. If there are multiple gaps with the greatest distance to the last scanned layer, the gap with the greatest distance to the layer immediately preceding the last scanned layer is selected from among the corresponding gaps (and so on).(iv) In a gap, the layer that is most likely to halve the gap is scanned. If there are several such layers, the layer with the greatest distance to the last scanned layer is selected.

[0051] Rule (i) takes precedence over all other rules. Rule (ii) takes precedence over rules (iii) and (iv). Rule (iii) takes precedence over rule (iv).

[0052] Fig. 4 The procedure is illustrated using an example and a schematic diagram. Fig. 4 (a) Eleven consecutive layers S1, ..., S11 within a single acquisition volume are shown schematically. The layers are labeled 1 to 11. These indices indicate the spatial sequence in which the layers follow one another within the acquisition volume. None of the layers have been scanned yet. Unscanned layers are indicated by dashed lines.

[0053] According to rule (i), the odd-numbered layers are scanned first, before the even-numbered layers are scanned. In other words, as long as there is at least one unscanned odd-numbered layer, no even-numbered layer will be scanned.

[0054] According to rule (ii), the outer layers are scanned first; either S1 first and then S11 or vice versa.

[0055] In Fig. 4 (b) The diagram shows that in this example, S1 is scanned first. S1 has the identification number 1. Scanned layers are marked with a solid, bold line. After S1, S11 is scanned. S11 receives the identification number 2.

[0056] According to rule (iii), the gap with the greatest distance between two scanned layers is now selected. As in Fig. 4 (b) As can be seen, there are only two scanned layers (S1, S11) that enclose a single gap. The next layer is scanned within this gap.

[0057] According to rule (iv), the layer in a gap that is most likely to halve the gap is identified. S6 lies exactly in the middle of the gap; scanning S6 would halve the gap. However, layer S6 has an even index (6), and as long as there are still odd-numbered layers, these are preferred (rule (i)). According to S6, scanning layer S5 or layer S7 would most likely halve the gap. Thus, there are two possible layers (S5, S7) in the gap that could be scanned. According to rule (iv), the layer furthest from the last scanned layer is selected. The last scanned layer is S11. The distance from S5 to S11 is greater than the distance from S7 to S11. Accordingly, layer S5 is scanned next. It receives in Fig. 4 (c) the identification number 3.

[0058] In Fig. 4 (c) It can be seen that after scanning layer S5, two gaps remain. One gap is between layers S1 and S5, and another is between layers S5 and S11. The gap between S5 and S11 is the larger of the two; it is selected next according to rule (iii). Within the selected gap, the layer most likely to halve the gap is identified. S8 lies exactly halfway between S5 and S11; however, S8 is an even-indexed layer (8), which, according to rule (i), is not yet next because there are still unscanned odd-indexed layers. After S8, S7 and S9 would be most likely to halve the gap. S9 is further away from the last scanned layer (S5) and is therefore scanned next. S9 is in Fig. 4 (d) marked with a solid line and bears the identification number 4.

[0059] In Fig. 4 (d) It can be seen that after scanning layer S9, three gaps remain. One gap is between S1 and S5, one between S5 and S9, and one between S9 and S11. The gap between S9 and S11 is smaller than the other two and is therefore initially excluded (Rule iii). The gap between S1 and S5 has the same distance as the gap between S5 and S9. According to Rule (iii), the sum of the identifiers of the bounding layers is now calculated for each of these gaps. The gap between S1 and S5 is bounded by layers S1 and S5. Their identifiers are 1 (S1) and 3 (S5). The sum is four. The gap between S5 and S9 is bounded by layers S5 and S9. Their identifiers are 3 (S5) and 4 (S9). The sum is seven. The sum of the layer identifiers that define the gap between S5 and S9 is therefore greater than the sum of the layer identifiers that define the gap between S1 and S5.The gap between S5 and S9 is selected. Within the gap, the layer most likely to halve the gap is chosen (rule iv). This is layer S7. It is scanned next; it receives [value missing]. Fig. 4 (e) the identification number 5.

[0060] In Fig. 4 (e) It can be seen that there is a largest gap, namely the gap between S1 and S5. This is selected next (Rule (iii)). Within the gap, layer S3 is the layer most likely to halve the gap (Rule iv). This layer is scanned next; it receives in Fig. 4 (f) the identification number 6.

[0061] In Fig. 4 (f) It can be seen that in the present example, all layers with an odd-numbered index have already been scanned. Therefore, the layers with even-numbered indices will now be scanned. According to rule (iii), the gap with the greatest distance between the layers bounding the gap is identified. However, the gaps are all equally spaced. Therefore, according to rule (iii), the gap is chosen where the sum of the identifiers of the layers bounding the gap is greatest. The sum of the identifiers of the bounding layers is: 7 in the case of S1 / S3, 9 in the case of S3 / S5, 8 in the case of S5 / S7, 9 in the case of S7 / S9, and 6 in the case of S9 / S11. Thus, there are two possible gaps. Of these, according to rule (iii), the gap with the greatest distance to the last scanned layer is chosen. This is the S7 / S9 gap, because the last scanned layer (S3) is closer to layer S3 / S5 than to layer S7 / S9.Within the gap between S7 and S9 there is only one unscanned layer (S8); this is selected; it receives . Fig. 4 (g) the identification number 7.

[0062] In Fig. 4 (g) The gaps remaining are between S1 and S3, S3 and S5, S5 and S7, and S9 and S11. From the previous step, it is already known that the gap between S3 and S5 now has the largest sum of identification numbers (9). Within the gap between S3 and S5, there is only one unscanned layer (S4); this layer is selected and assigned a value in Fig. 4 (h) the identification number 8.

[0063] In Fig, 4 (h) The gaps remaining are between S1 and S3, S5 and S7, and S9 and S11. From a previous step, it is already known that the gap between S5 and S7 now has the largest sum of identification numbers (8). Within the gap between S5 and S7, there is only one unscanned layer (S6); this layer is selected and assigned a value in Fig. 4 (i) the identification number 9.

[0064] In Fig, 4 (i) The gaps between S1 and S3, and between S9 and S11, remain. It is already known from a previous step that the gap between S1 and S3 now has the largest sum of identification numbers (7). Within the gap between S1 and S3, there is only one unscanned layer (S2); this layer is selected and assigned a value in Fig. 4 (j) the identification number 10; which then receives the last remaining shift (S10) in Fig. 4 (j) the identification number 11.

[0065] In this example, the scan sequence is therefore S1, S11, S5, S9, S7, S3, S8, S4, S6, S2, S10.

[0066] If layer S11 is the first layer scanned instead of layer S1 (see rule ii), the following sequence results according to the rules mentioned above: S11, S1, S7, S3, S5, S9, S4, S8, S6, S10, S2.

[0067] Other rules and sequences are conceivable. Preferably, a sequence or rules for a sequence are defined at the beginning of the training of the prediction model, and this sequence / rules is / are maintained during training, validation, and prediction using the trained prediction model.

[0068] According to the invention, the sequence in which the layers in a scanning volume are scanned is determined such that there is at least one tomogram of a layer that is immediately adjacent to two previously scanned layers. In other words, there is at least one tomogram of a layer that was generated later than the tomograms of the immediately adjacent layers.

[0069] Preferably, there are multiple tomograms T j , where each tomogram of the plurality of tomograms represents a layer S j represented, where each tomogram T j was produced later than the tomograms T j -1 and T j +1 of the one directly related to shift S j adjacent layers S j -1 and S j +1 , where j is an index indicating the position of the layer in an arrangement of layers in a receiving volume.

[0070] Each tomogram T j a layer S j is lower contrast than the tomograms T j -1 and T j +1 of the immediately adjacent layers S j -1 and S j +1 . According to the invention, for each low-contrast tomogram T j an artificial, contrast-enhanced tomogram T j * generated.

[0071] In one embodiment of the invention, the artificial, contrast-enhanced tomogram T j * based on the low-contrast tomogram T j of layer S j and based on the tomograms T j -1 and T j+1 of the immediately adjacent layers S j -1 and S j +1 generated.

[0072] However, it is also conceivable to use tomograms of more distant layers, in addition to tomograms of the immediately adjacent layers, to generate an artificial tomogram.

[0073] In a further embodiment of the invention, the artificial, contrast-enhanced tomogram T j * based on the low-contrast tomogram T j layer S j and based on a plurality of tomograms T j -k and T j +k of the adjacent layers S j -k and S j +k is generated, where k is an index ranging from 1 to m, where m is an integer. Preferably, m is equal to 2 or 3.

[0074] The prediction model according to the invention is configured (trained) to transfer contrast information from a (high-contrast) tomogram to a (low-contrast) tomogram. Typically, the high-contrast tomogram is generated before the low-contrast tomogram; the low-contrast tomogram is usually more affected by washout. washout ) affected as the high-contrast tomogram. In a preferred embodiment, the high-contrast and the low-contrast tomograms represent immediately adjacent layers in an acquisition volume, i.e., the high-contrast tomogram represents a first layer after application of a contrast agent and the low-contrast tomogram represents a second layer after application of the contrast agent, and the first and second layers are immediately adjacent, and the high-contrast tomogram was generated temporally before the low-contrast tomogram.

[0075] For more than two layers / two tomograms, the prediction model is preferably configured (trained) with the contrast enhancement of tomograms T. j -k and T j +k adjacent layers S j -k and S j +k on the tomogram T j layer S j to transfer, where k is an index ranging from 1 to m, where m is an integer. Preferably, m is equal to 1, 2, or 3.

[0076] The predictive model is configured (trained) to generate an artificial, contrast-enhanced tomogram of a slice in an acquisition volume based on at least two, preferably at least three, measured tomograms. The at least two measured tomograms comprise: • A first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an object under investigation after the application of a contrast agent; • A second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the object under investigation after the application of the contrast agent, wherein the second tomogram is / was generated temporally after the first tomogram. Preferably, the first layer and the second layer are immediately adjacent to each other.

[0077] In the case of at least three measured tomograms, these include: • A first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an object under investigation after the application of a contrast agent; • A second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the object under investigation after the application of the contrast agent; • A third tomogram, wherein the third tomogram represents a third layer within the acquisition volume of the object under investigation after the application of the contrast agent, the third layer preferably being located between the first and second layers. Preferably, the first and second layers are immediately adjacent to the third layer. The third tomogram is / was generated after the first and second tomograms.

[0078] According to the invention, an artificial, contrast-enhanced tomogram is generated on the basis of the at least two measured tomograms, which represents a layer with a higher contrast enhancement than the corresponding real, measured tomogram.

[0079] Preferably, an artificial, contrast-enhanced tomogram is generated based on at least three measured tomograms, which represents the third layer with a higher contrast enhancement than the third (real, measured) tomogram.

[0080] In order for the prediction model according to the invention to make the predictions described here, it must first be configured accordingly.

[0081] The prediction model is preferably developed using a self-learning algorithm in a supervised or unsupervised machine learning process (English: unsupervised learning The system is trained (configured). Training data is used for learning. This training data comprises, from a variety of study objects, a plurality of tomograms of a study area for each object. The study area (the acquisition volume) is usually the same for all study objects (e.g., a part of a human body, an organ, or part of an organ). The tomograms of the training dataset are also referred to as reference representations in this description.

[0082] For each subject under investigation, the training data shall include at least i) a first reference tomogram, wherein the first reference tomogram represents a first layer within the acquisition volume after application of a contrast agent, and ii) a second reference tomogram, wherein the second reference tomogram represents a second layer within the acquisition volume after application of the contrast agent, wherein the second reference tomogram represents the second layer at a later time than the first reference tomogram represents the first layer and is therefore less contrasty than the first reference tomogram.

[0083] Preferably, the training data for each subject of investigation comprise at least i) a first reference tomogram, wherein the first reference tomogram represents a first layer within the acquisition volume after application of a contrast agent, ii) a second reference tomogram, wherein the second reference tomogram represents a second layer within the acquisition volume after application of the contrast agent, and iii) a third reference tomogram, wherein the third reference tomogram represents a third layer within the acquisition volume after application of the contrast agent.

[0084] The third layer in the imaging volume lies between the first layer and the second layer, and the third reference tomogram was generated after the first reference tomogram and the second reference tomogram.

[0085] When using two reference tomograms for each object under investigation, the predictive model is trained to generate an artificial, contrast-enhanced tomogram of the second layer for each object. To do this, the first and second reference tomograms are input into the predictive model, which is then trained to output an artificial tomogram representing the second layer but with increased contrast enhancement compared to the second reference tomogram. The predictive model is trained to transfer contrast information from the first reference tomogram to the second reference tomogram.

[0086] When using three reference tomograms for each object under investigation, the predictive model is trained to generate an artificial, contrast-enhanced tomogram of the third layer for each object. To do this, the first, second, and one of the third reference tomograms are input into the predictive model, which is then trained to output an artificial tomogram representing the third layer but with increased contrast enhancement compared to the measured third reference tomogram. The predictive model is trained to transfer contrast information from the first and second reference tomograms to the third reference tomogram.

[0087] When using more than three reference tomograms per object, the predictive model is trained to generate an artificial, contrast-enhanced tomogram of a defined slice for each object. The reference tomograms include one reference tomogram of the defined slice, which typically has lower contrast enhancement than the other reference tomograms of other slices. The measured reference tomograms are fed into the predictive model, and the predictive model is trained to output an artificial tomogram representing the defined slice but with increased contrast enhancement compared to the measured reference tomogram of that slice. The predictive model is trained to transfer contrast information from the other reference tomograms to the measured reference tomogram of the defined slice.

[0088] In machine learning, the self-learning algorithm generates a statistical model based on the training data. This means it doesn't simply memorize examples, but rather "recognizes" patterns and regularities in the training data. This allows the predictive model to evaluate even unknown data. Validation data can be used to test the accuracy of the evaluation of unknown data.

[0089] Self-learning systems, which are trained using supervised or unsupervised learning, are described in various ways in the state of the art (see, for example, G. Bonaccorso: Hands-On Unsupervised Learning with Python, Packt Publishing, 2019, ISBN: 978-1789348279).

[0090] Preferably, the prediction model is an artificial neural network or includes at least one such network.

[0091] An artificial neural network comprises at least three layers of processing elements: a first layer with input neurons (kneads), 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.

[0092] The input neurons are used to receive measured tomograms. The output neurons are used to output artificial tomograms.

[0093] The processing elements of the layers between the input neurons and the output neurons are interconnected in a predetermined pattern with predetermined connection weights.

[0094] Training a neural network can be performed, for example, using a backpropagation method. The goal is for the network to map given input data to given output data as reliably as possible. The quality of the mapping 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.

[0095] In the trained state, the connection weights between the processing elements contain information regarding the relationship between measured and artificially generated tomograms, which can be used to generate artificial tomograms based on new tomograms not used during training.

[0096] 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 prediction accuracy of the trained network when applied to unknown data.

[0097] In a particularly preferred embodiment, the model used for training comprises a generative, opponent network (English: generative adversial network, Abbreviation: GAN). Such a GAN typically comprises two artificial neural networks, a first network and a second network. The first network is often also called the generator. The second network is often also called the discriminator. The generator and the discriminator are trained together and compete: the generator creates an artificial tomogram, and the discriminator must distinguish the artificial tomogram from a real tomogram. The generator is trained to create artificial tomograms that the discriminator cannot distinguish from real tomograms, and the discriminator is trained to increasingly distinguish the increasingly better artificial tomograms from real tomograms.

[0098] The principle is schematically represented in Fig. 5 The prediction model is shown. For training, it comprises a first artificial neural network, the generator G, and a second artificial neural network, the discriminator D. The training data TD for training the two networks includes P for a large number of people (in Fig. 5 (For clarity, only a single person is shown schematically.) A first tomogram T1, a second tomogram T2, and a third tomogram T3. The first tomogram T1 represents a first layer S1, the second tomogram T2 represents a second layer S2, and the third tomogram T3 represents a third layer S3 after the administration of a contrast agent. Layer S3 lies between layers S1 and S2. The tomograms were generated by scanning the layers in a radiological examination procedure after the administration of a contrast agent. Layers S1 and S2 were scanned before layer S3. In other words, tomogram T3 was generated after tomograms T1 and T2. Because tomogram T3 was generated later than tomograms T1 and T2, it exhibits lower contrast enhancement than tomograms T1 and T2 due to the more advanced washout of the contrast agent.In other words, T3 is low in contrast compared to T1 and T2; T1 and T2 are high in contrast compared to T3.

[0099] The first tomogram T1, the second tomogram T2, and the third tomogram T3 are fed to generator G. The generator is configured to produce an artificial tomogram Tk, representing the third layer, based on this input data. Generator G is trained to produce an artificial tomogram Tk with higher contrast enhancement than the third tomogram T3. Generator G is trained to transfer contrast information from the first tomogram T1 and the second tomogram T2 to tomogram T3, thereby generating the artificial, contrast-enhanced tomogram Tk. The discriminator receives real, high-contrast tomograms T1 and T2, as well as artificially generated, contrast-enhanced tomograms Tk, and is configured to indicate whether a received tomogram is an artificial or a real tomogram.The discriminator thus performs a classification; it assigns each received tomogram to one of two classes, a first class with real tomograms and a second class with artificially generated tomograms.

[0100] The result of the classification is a classification result R. Since it is known for each tomogram fed to the discriminator whether it is a real or an artificially generated tomogram, the result of the classification can be evaluated. The evaluation is performed using a loss function LF (English: loss function LF). loss function ) .The result of the evaluation then feeds into the training of both the generator G and the discriminator D, both of which are trained to deliver an improved result: in the case of the generator G, to produce an artificial tomogram that the discriminator cannot distinguish from a real tomogram, and in the case of the discriminator, to distinguish artificial tomograms from real tomograms.

[0101] Once the system has been comprehensively trained, including both the generator and the discriminator, and the two networks achieve a predefined (desired) accuracy, the prediction model according to the invention can be derived from the system. For this purpose, the system can be reduced to the generator; the discriminator is no longer required for prediction. The generator has learned to produce an artificial tomogram based on at least two, preferably at least three, tomograms, which is indistinguishable or only barely distinguishable from a real tomogram.

[0102] Fig. 6 Figure 1 shows an exemplary and schematic representation of the prediction model according to the invention. This is the model described in Figure 2. Fig. 5 The generator G shown, which is based on a large number of reference tomograms in an unsupervised learning procedure as in relation to Fig, 5 has been described and trained.

[0103] In the present example, at least three tomograms are fed to the generator G for prediction: a first tomogram T1, a second tomogram T2, and a third tomogram T3. The first tomogram T1 represents a first layer S1 in an acquisition volume of person Pi, the second tomogram T2 represents a second layer S2 in the acquisition volume of person Pi, and the third tomogram T3 represents a third layer S3 in the acquisition volume of person Pi. Layer S3 lies between layers S1 and S2. The first tomogram T1 was acquired at a time t The second tomogram, T2, was produced after the application of a contrast agent. t 2 after administration of the contrast agent. The third tomogram, T3, was produced at a time point. t 3 after application of the contrast agent. The time t 3 is the time frame t 2 and t 1. Subsequent in time.

[0104] Generator G creates an artificial tomogram Tk from the received tomograms, representing the third layer S3. The artificial tomogram exhibits a higher contrast enhancement than the third tomogram T3.

[0105] In a preferred embodiment, a system of artificial neural networks based on a CycleGAN or a Pix2Pix architecture is used to train the prediction model.

[0106] Further information on generative adversarial networks can be found in publications on this topic (see, for example, NK Manaswi: Generative Adversarial Networks with Industrial Use Cases, BPB PUBN Verlag, 2020, ISBN: 9789389423853; WO2020 / 246996, WO2020 / 242572, EP3785231, WO2021 / 049784, EP3767590, US20190295302). The method according to the invention can be carried out using a computer system.

[0107] Fig. 7 Figure 10 schematically and exemplarily shows an embodiment of the computer system according to the invention. The computer system (10) comprises a receiving unit (11), a control and processing unit (12), and an output unit (13).

[0108] A "computer system" is a system for electronic data processing that processes data using programmable instructions. Such a system typically includes a control and processing unit, often also called a "computer," which comprises a processor for performing logical operations and main memory for loading a computer program, as well as peripherals.

[0109] 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, joysticks, drives, cameras, microphones, speakers, etc. Internal ports and expansion cards are also considered peripherals in computer technology.

[0110] Today's computer systems are often divided into desktop PCs, portable PCs, laptops, notebooks, netbooks and tablet PCs, and so-called handhelds (e.g., smartphones); all these systems can be used to carry out the invention.

[0111] Input to the computer system (for example, for control by a user) is provided via input devices such as a keyboard, a mouse, a microphone, a touch-sensitive display, and / or the like. Output is provided via the output unit (13), which may be, in particular, a monitor (screen), a printer, and / or a data storage device.

[0112] The computer system (10) according to the invention is configured to generate an artificial, contrast-enhanced tomogram from at least two, preferably at least three, real tomograms of an examination area of ​​an object under investigation, which represent the examination area after application of a contrast agent.

[0113] The control and processing unit (12) serves to control the receiving unit (11) and the output unit (13), to coordinate the data and signal flows between the different units, to process representations of the investigation area, and / or to generate artificial tomograms. It is conceivable that several control and processing units are present.

[0114] The receiving unit (11) serves to receive actual tomograms. The tomograms can be transmitted, for example, from a magnetic resonance imaging (MRI) scanner, a computed tomography (CT) scanner, or read from a data storage device. The MRI scanner or the CT scanner can be a component of the computer system according to the invention. It is also conceivable that the computer system according to the invention is a component of a MRI scanner or a CT scanner. The transmission of tomograms can take place via a network connection or a direct connection. Transmission of tomograms can occur via a wireless connection (WLAN, Bluetooth, mobile network, and / or the like) and / or via a wired connection. It is conceivable that several receiving units are present. The data storage device can also be a component of the computer system according to the invention or be connected to it, for example, via a network.It is conceivable that multiple data storage devices exist.

[0115] The receiving unit receives the tomograms and, if applicable, other data (such as information about the object under examination, recording parameters and / or the like) and transmits them to the control and computing unit.

[0116] The control and computing unit is configured to generate artificial tomograms based on the received data.

[0117] The output unit (13) allows the artificial tomograms to be displayed (for example, on a monitor), printed (e.g., via a printer), or stored in a data storage device. It is conceivable that multiple output units are present.

[0118] Fig. 8 Figure 1 shows, by way of example and schematically, a preferred embodiment of the inventive method for training a prediction model in the form of a flowchart.

[0119] The procedure (100), comprising the steps: (110) Receiving a training dataset, wherein the training dataset comprises a plurality of real reference tomograms for each reference object of a plurality of reference objects, wherein the reference tomograms represent an acquisition volume of the reference object after the application of a contrast agent, wherein the reference tomograms for each reference object comprise at least three tomograms: a first reference tomogram, a second reference tomogram, and a third reference tomogram, wherein the first tomogram represents a first layer within the acquisition volume, wherein the second tomogram represents a second layer within the acquisition volume, and wherein the third tomogram represents a third layer within the acquisition volume, the third layer being located in the acquisition volume between the first layer and the second layer.wherein the third tomogram was generated temporally after the first and second tomograms, (120) Training the prediction model, wherein the prediction model for training comprises two artificial neural networks, a first network and a second network, ∘ wherein the first network is configured to receive the first, second and third tomograms for each reference object and to generate an artificial tomogram based on the received tomograms, wherein the artificial tomogram represents the third layer in the recording volume ∘ wherein the second network is configured to receive the third tomogram and / or the artificial tomogram and to make a statement for the received tomogram as to whether the received tomogram is a real tomogram or an artificially generated tomogram, ∘ wherein the first network and the second network are trained together in an unsupervised learning procedure, wherein the first network is trained,(130) to generate artificial tomograms that are not recognized as artificial by the second network and the second network is trained to distinguish artificial tomograms from real tomograms, (130) storing at least the first network and / or feeding the first network to a method for predicting an artificial tomogram for a test object.

[0120] Fig. 9 Figure 1 shows, by way of example and schematically, a preferred embodiment of the inventive method for generating an artificial, contrast-enhanced tomogram in the form of a flowchart.

[0121] The procedure (200) comprises the following steps: (210) Receiving a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an object under investigation after the application of a contrast agent, (220) Receiving a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the object under investigation after the application of the contrast agent, (230) Receiving a third tomogram, wherein the third tomogram represents a third layer within the acquisition volume of the object under investigation after the application of the contrast agent, the third layer being located between the first and second layers, wherein the third tomogram is generated temporally after the first and second tomograms, (240) Providing a predictive model, wherein the predictive model has been trained in an unsupervised learning procedure using a training dataset,▪ wherein the training dataset for a plurality of reference objects comprises a first reference tomogram, a second reference tomogram, and a third reference tomogram, wherein the reference tomograms represent layers in an acquisition volume of the reference object, wherein the first reference tomogram represents a first layer within the acquisition volume, wherein the second tomogram represents a second layer within the acquisition volume, wherein the third tomogram represents a third layer within the acquisition volume, ∘ wherein the third layer lies in the acquisition volume between the first layer and the second layer, ∘ wherein the third tomogram was generated temporally after the first and second tomograms, ▪ wherein the predictive training model comprises two artificial neural networks, a first network and a second network, wherein the first network is configured to use the first tomogram for each reference object,(250) to receive the second and third tomograms and to generate an artificial tomogram based on the received tomograms, wherein the artificial tomogram represents the third layer in the acquisition volume, wherein the second network is configured to receive the third tomogram and / or the artificial tomogram and to make a statement about the received tomogram as to whether the received tomogram is a real tomogram or an artificially generated tomogram, wherein the first network and the second network have been jointly trained in an unsupervised learning procedure, wherein the first network has been trained to generate artificial tomograms that are not judged as artificial by the second network and the second network is trained to distinguish artificial tomograms from real tomograms, (250) feeding the first, second and third tomograms to the first neural network, (260) receiving an artificial tomogram,where the artificial tomogram represents the third layer in the acquisition volume, (270) Output of the artificial tomogram.

Claims

1. Computer-implemented method for training a prediction model, comprising the steps - Receiving a training dataset, wherein the training dataset for each reference object of a plurality of reference objects comprises a plurality of reference tomograms, ∘ wherein the reference tomograms represent an acquisition volume of the reference object after the application of a contrast agent, ∘ wherein the reference tomograms for each reference object comprise at least two tomograms: a first reference tomogram and a second reference tomogram, ▪ wherein the first reference tomogram represents a first layer within the acquisition volume, ▪ wherein the second reference tomogram represents a second layer within the acquisition volume, • wherein the second reference tomogram was generated after the first reference tomogram, ▪ wherein the first and the second layer are arranged parallel to each other in a spatial sequence, - Training the prediction model, wherein the prediction model is fed the first and the second reference tomogram during training, wherein the prediction model is trained to transfer contrast enhancement from the first reference tomogram to the second reference tomogram and to generate an artificial tomogram that represents the second layer.

2. Method according to claim 1, further comprising - Storing and / or outputting the trained prediction model and / or using the trained prediction model to predict an artificial tomogram.

3. Method according to claim 1 or 2, wherein the prediction model comprises two artificial neural networks, a first network and a second network, ∘ wherein the first network is configured to receive the first and the second tomogram for each reference object and to generate an artificial tomogram based on the received tomograms, wherein the artificial tomogram represents the second layer in the acquisition volume, ∘ wherein the second network is configured to receive the second tomogram and / or the artificial tomogram and to make a statement for the received tomogram as to whether the received tomogram is a real tomogram or an artificially generated tomogram, ∘ wherein the first network and the second network are jointly trained in a supervised or unsupervised learning process, wherein the first network is trained to generate artificial tomograms that are not classified as artificial by the second network and the second network is trained to distinguish artificial tomograms from real tomograms.

4. Method according to claim 1, 2 or 3, further comprising the step: - Storing at least the first network and / or feeding the first network to a method for predicting an artificial contrast-enhanced tomogram for an examination object.

5. Computer-implemented method for generating an artificial, contrast-enhanced tomogram, comprising the steps: ∘ Receiving a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an examination object after the application of a contrast agent, ∘ Receiving a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the examination object after the application of the contrast agent, wherein the second tomogram was generated after the first tomogram, ▪ wherein the first and the second layer are arranged parallel to each other in a spatial sequence, ∘ Providing a prediction model, wherein the prediction model was trained based on a training dataset to transfer contrast information from at least one tomogram to another tomogram and thereby generate an artificial tomogram, ∘ Feeding the first and second tomogram to the prediction model, ∘ Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the second layer in the acquisition volume, wherein contrast information from the first tomogram has been transferred in the artificial tomogram, ∘ Outputting the artificial tomogram.

6. Method according to claim 5 comprising the steps: ∘ Receiving a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an examination object after the application of a contrast agent, ∘ Receiving a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the examination object after the application of the contrast agent, ∘ Receiving a third tomogram, wherein the third tomogram represents a third layer within the acquisition volume of the examination object after the application of the contrast agent, wherein the third layer is located between the first and the second layer, wherein the third tomogram is generated after the first and the second tomogram, ∘ Providing a prediction model, wherein the prediction model was trained based on a training dataset to transfer contrast information from at least two tomograms to one tomogram and thereby generate an artificial tomogram, ∘ Feeding the first, second and third tomogram to the prediction model, ∘ Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the third layer in the acquisition volume, wherein contrast information from the first and from the second tomogram has been transferred in the artificial tomogram, ∘ Outputting the artificial tomogram.

7. Method according to claim 6 , wherein the providing of the prediction model comprises the following steps: - Receiving a training dataset, wherein the training dataset for each reference object of a plurality of reference objects comprises a plurality of reference tomograms, ∘ wherein the reference tomograms represent an acquisition volume of the reference object after the application of a contrast agent, ∘ wherein the reference tomograms for each reference object comprise at least three tomograms: a first reference tomogram, a second reference tomogram and a third reference tomogram, ▪ wherein the first reference tomogram represents a first layer within the acquisition volume, ▪ wherein the second reference tomogram represents a second layer within the acquisition volume, ▪ wherein the third reference tomogram represents a third layer within the acquisition volume, • wherein the third layer in the acquisition volume is located between the first layer and the second layer, • wherein the third reference tomogram was generated after the first reference tomogram and the second reference tomogram, - Training the prediction model, wherein the prediction model is fed the first, the second and the third reference tomogram, wherein the prediction model is trained to transfer contrast enhancement from the first and the second reference tomogram to the third reference tomogram and to generate an artificial tomogram that represents the third layer.

8. Process according to one of claims 6 or 7, wherein the prediction model during training comprises two artificial neural networks, a first network and a second network, • wherein the first network is configured to receive the first, the second and the third tomogram for each reference object and to generate an artificial tomogram based on the received tomograms, wherein the artificial tomogram represents the third layer in the acquisition volume, • wherein the second network is configured to receive the third tomogram and / or the artificial tomogram and to make a statement for the received tomogram as to whether the received tomogram is a real tomogram or an artificially generated tomogram, • wherein the first network and the second network are or have been jointly trained in an unsupervised learning process, wherein the first network is or has been trained to generate artificial tomograms that are not classified as artificial by the second network and the second network is or has beentrained to distinguish artificial tomograms from real tomograms, wherein the prediction of the artificial tomogram is carried out by means of the first network.

9. Process according to one of claims 5 to 8, comprising the steps: ∘ Receiving a number n of tomograms T1, ..., Tn, wherein n is an odd integer greater than 2, ▪ wherein each tomogram T1, ..., Tn represents a layer S1, ..., Sn within an acquisition volume of an examination object after the application of a contrast agent, • wherein the layers S1, ..., Sn are arranged parallel to each other in a spatial sequence, ∘ wherein the spatial sequence is indicated by the indices 1, ..., n, ▪ wherein the tomograms have been generated in a sequence in which first the tomograms T2k+1 and then the tomograms T2k are generated, wherein k is an index that runs through the values from 0 to (n-1) / 2, ∘ Providing a prediction model, wherein the prediction model has been trained on a training dataset to transfer contrast information from in each case two immediately adjacent tomograms that were generated at an earlier time to tomograms that were generated at a later time, thereby generating artificial contrast-enhanced tomograms, ∘ Feeding the received tomograms to the prediction model, ∘ Receiving a number (n-1) / 2 of artificial tomograms KT, wherein each artificial tomogram KT2i+1 represents the respective layer S2i+1 and has been generated based on the tomograms T2i and T2i+2, wherein i is an index that can take the values from 0 to k, ∘ Outputting at least one of the artificial tomograms.

10. Process according to one of claims 5 to 9, comprising the step: - Generating an odd number n of tomograms T1, ..., Tn, wherein the tomograms are generated according to the following rules: (i) Tomograms with odd indices 1, 3, ..., n are generated before tomograms with even indices 2, 4, ..., (n-1). (ii) Generation of the tomograms from the two outermost layers in the order T1, Tn or Tn, T1. (iii) Identification of a gap with the largest distance between two scanned layers, wherein the gap is an area between two scanned layers in which there are still unscanned layers. If there are multiple gaps with the same distance, the sum of the identification numbers of the scanned layers that bound the gap is calculated for each of these gaps. The gap with the largest sum of identification numbers is chosen. If there are multiple such gaps with the same largest identification numbers, the gap with the largest distance to the most recently scanned layer is chosen. If there are multiple gaps with the largest distance to the most recently scanned layer, the gap with the largest distance to the layer that was scanned immediately before the most recently scanned layer is selected (and so on). (iv) Generation of the tomogram of the layer in a gap that is most likely to halve the gap. If there are multiple such layers, the layer with the largest distance to the most recently scanned layer is chosen. wherein rule (i) takes precedence over all other rules, and wherein rule (ii) takes precedence over rule (iii) and (iv), and wherein rule (iii) takes precedence over rule (iv).

11. System comprising • a receiving unit, • a control and computing unit, and • an output unit, ∘ wherein the control and computing unit is configured to cause the receiving unit to receive at least a first and a second tomogram, - wherein the first tomogram represents a first layer within an acquisition volume of an examination object after the application of a contrast agent, - wherein the second tomogram represents a second layer within the acquisition volume of the examination object after the application of the contrast agent, - wherein the second tomogram was generated temporally after the first tomogram, - wherein the first and second layers are arranged parallel to each other in a spatial sequence, ∘ wherein the control and computing unit is configured to feed the first and second tomogram to a prediction model, wherein the prediction model has been trained on a training dataset to transfer contrast information from at least one tomogram to another tomogram, thereby generating an artificial tomogram, ∘ wherein the control and computing unit is configured to receive an artificial tomogram from the prediction model, wherein the artificial tomogram represents the second layer in the acquisition volume, wherein in the artificial tomogram contrast information from the first tomogram has been transferred, ∘ wherein the control and computing unit is configured to cause the output unit to output the artificial tomogram.

12. System comprising according to claim 11, ∘ wherein the control and computing unit is configured to cause the receiving unit to receive at least a first, a second, and a third tomogram, - wherein the first tomogram represents a first layer within an acquisition volume of an examination object after the application of a contrast agent, - wherein the second tomogram represents a second layer within the acquisition volume of the examination object after the application of the contrast agent, - wherein the third tomogram represents a third layer within the acquisition volume of the examination object after the application of the contrast agent, wherein the third layer lies between the first and the second layer, wherein the third tomogram is generated temporally after the first and the second tomogram, ∘ wherein the control and computing unit is configured to feed the first, the second, and the third tomogram to a prediction model, wherein the prediction model has been trained on a training dataset to transfer contrast information from at least two tomograms to one tomogram, thereby generating an artificial tomogram, ∘ wherein the control and computing unit is configured to receive an artificial tomogram from the prediction model, wherein the artificial tomogram represents the third layer in the acquisition volume, wherein in the artificial tomogram contrast information from the first and the second tomogram has been transferred, ∘ wherein the control and computing unit is configured to cause the output unit to output the artificial tomogram.

13. Computer program product comprising a computer program that can be loaded into a working memory of a computer system and there causes the computer system to execute the following steps: ∘ Receiving a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an examination object after the application of a contrast agent, ∘ Receiving a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the examination object after the application of the contrast agent, wherein the second tomogram was generated temporally after the first tomogram, wherein the first and the second layer are arranged parallel to each other in a spatial sequence, ∘ Feeding the first and second tomogram to a prediction model, wherein the prediction model has been trained on a training dataset to transfer contrast information from at least one tomogram to another tomogram, thereby generating an artificial tomogram, ∘ Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the second layer in the acquisition volume, wherein in the artificial tomogram contrast information from the first tomogram has been transferred, ∘ Outputting the artificial tomogram.

14. Computer program product according to claim 13, wherein the computer program causes the computer system to execute the following steps: ∘ Receiving a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an examination object after the application of a contrast agent, ∘ Receiving a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the examination object after the application of the contrast agent, ∘ Receiving a third tomogram, wherein the third tomogram represents a third layer within the acquisition volume of the examination object after the application of the contrast agent, wherein the third layer lies between the first and the second layer, wherein the third tomogram is generated temporally after the first and the second tomogram, ∘ Feeding the first, second, and third tomogram to a prediction model, wherein the prediction model has been trained on a training dataset to transfer contrast information from at least two tomograms to one tomogram and thereby generate an artificial tomogram, ∘ Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the third layer in the acquisition volume, wherein in the artificial tomogram contrast information from the first and second tomogram has been transferred, ∘ Outputting the artificial tomogram.

15. Use of a contrast agent in a process for predicting an artificial contrast-enhanced tomogram, wherein the process comprises the following steps: ∘ Generating a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an examination object after the application of a contrast agent, ∘ Generating a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the examination object after the application of the contrast agent, wherein the second tomogram is generated temporally after the first tomogram, ▪ Wherein the first and the second layer are arranged parallel to each other in a spatial sequence, ∘ Feeding the first and second tomogram to a prediction model, wherein the prediction model has been trained on a training dataset to transfer contrast information from at least one tomogram to another tomogram and thereby generate an artificial tomogram, ∘ Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the second layer in the acquisition volume, wherein in the artificial tomogram contrast information from the first tomogram has been transferred, ∘ Outputting the artificial tomogram.

16. Use according to claim 15, wherein the process comprises the following steps: ∘ Generating a first tomogram, wherein the first tomogram represents a first layer within an acquisition volume of an examination object after the application of a contrast agent, ∘ Generating a second tomogram, wherein the second tomogram represents a second layer within the acquisition volume of the examination object after the application of the contrast agent, ∘ Generating a third tomogram, wherein the third tomogram represents a third layer within the acquisition volume of the examination object after the application of the contrast agent, wherein the third layer lies between the first and the second layer, wherein the third tomogram is generated temporally after the first and the second tomogram, ∘ Feeding the first, second, and third tomogram to a prediction model, wherein the prediction model has been trained on a training dataset to transfer contrast information from at least two tomograms to one tomogram and thereby generate an artificial tomogram, ∘ Receiving an artificial tomogram from the prediction model, wherein the artificial tomogram represents the third layer in the acquisition volume, wherein in the artificial tomogram contrast information from the first and second tomogram has been transferred, ∘ Outputting the artificial tomogram.

17. Kit comprising a contrast agent and a computer program product according to one of claims 13 or 14.