Enhancing visibility of a contrast agent

Abstract

A computer-implemented method of enhancing visibility of a contrast agent in computed tomography (CT) imaging is provided. The method comprises receiving CT data representing a CT image (110) comprising the contrast agent; generating, from the CT data, one or more corresponding contrast agent enhanced CT images (130), wherein the visibility of the contrast agent is enhanced as compared to the CT image (110) using a trained machine learning model (120); and outputting the one or more contrast agent enhanced CT images (130). A contrast agent dose used in the received CT data is lower than a contrast agent dose in a corresponding conventional CT image comprising the contrast agent.

Description

[0001] ENHANCING VISIBILITY OF A CONTRAST AGENT

[0002] Technical Field

[0003] The present disclosure relates to enhancing visibility of a contrast agent in computed tomography (CT) imaging.

[0004] Background

[0005] Contrast agents are routinely used to provide visibility of features in CT images. For instance, contrast agents are used to provide visibility of the vasculature and also organs, such as the kidneys. Contrast agents include substances such as Iodine, or Gadolinium, and which provide a distinction between media via their X-ray attenuation values. For instance, Iodine- based contrast agents are routinely injected into the vasculature in order to provide a distinction between blood and surrounding tissue. A contrast agent may be injected directly into a blood vessel of interest, or it may be injected at another position in the vasculature of a subject, and from which position the contrast agent flows into a vessel, or region, of interest. The contrast agent may be injected into the body in the form of a so-called “bolus”. The contrast agent is typically injected using a syringe. The syringe may be controlled manually, or using a contrast agent injector, often referred-to simply as an “injector”. The injected contrast agent results in a contrast agent distribution in the anatomy. The contrast agent distribution is then imaged using a CT imaging system. The resulting images may be used to perform a diagnosis, such as to identify a stenosis in the vasculature or to provide a blood flow measurement, or they may be used for other purposes, such as to navigate interventional devices or to perform a treatment procedure in the vasculature.

[0006] Despite the visibility of image features that is provided via the use of contrast agents, some image features remain difficult to visualise. Consequently, there remains a need to enhance the visibility of contrast agent in CT images.

[0007] Summary

[0008] According to one aspect of the present disclosure, a computer-implemented method of enhancing visibility of a contrast agent in CT imaging is provided. The method includes: receiving CT data representing a CT image comprising the contrast agent; generating, from the CT data, one or more corresponding contrast agent enhanced CT images, wherein the visibility of the contrast agent is enhanced, as compared to the CT image, using a trained machine learning model; and outputting the one or more contrast agent enhanced CT images; wherein a contrast agent dose used in the received CT data is lower than a contrast agent dose in CT data representing a corresponding conventional CT image comprising the contrast agent.

[0009] In this aspect, the trained machine learning model generates one or more contrast agent enhanced CT images from a CT image. The contrast agent dose used in the received CT data is lower than a contrast agent dose in CT data representing a corresponding conventional CT image comprising the contrast agent. In other words, the contrast agent dose used to acquire the CT data is lower than a contrast agent dose used to acquire CT data representing a corresponding conventional CT image comprising the contrast agent. Thus, the method facilitates the provision of contrast agent enhanced CT images from CT images that have been acquired using a lower contrast agent dose than that used to acquire conventional CT images. This is beneficial for patients because it reduces the side-effects associated with contrast agent usage.

[0010] In an example, the machine learning model is trained to generate the one or more contrast agent enhanced CT images using training data and ground truth data. The training data comprises a plurality of conventional CT images comprising contrast agent distributions, and the ground truth data comprising, for each conventional CT image, one or more corresponding contrast agent enhanced CT images. For each conventional CT image, the conventional CT image and the one or more corresponding contrast agent enhanced CT images are generated from the same spectral CT image.

[0011] Conventional CT images and also corresponding contrast agent enhanced CT images can be reliably generated from spectral CT images. For instance, material decomposition algorithms and trained neural networks can be used to reliably generate a conventional CT image and one or more corresponding contrast agent enhanced CT images from a spectral CT image. Examples include a MonoE40 image, i.e. a virtual monoenergetic image with an energy of 40keV, and a so-called Iodine image. Such images can be reliably generated from spectral CT images without the risk of introducing image artifacts, and which could otherwise be misinterpreted as real image features and lead to a misdiagnosis. Moreover, spectral CT images are typically acquired using a lower contrast agent dose than conventional CT images. Consequently, the correspondence between a conventional CT image and its one or more corresponding contrast agent enhanced CT images, and which is provided by generating these images from the same spectral CT image, is valid at lower doses of contrast agent than that used to acquire conventional CT images. Therefore, by training the machine learning model to generate the one or more contrast agent enhanced CT images using conventional CT images and one or more corresponding contrast agent enhanced CT images that are generated from the same spectral CT image, the machine learning algorithm is both reliable, and also valid at lower doses of contrast agent than that used to acquire conventional CT images. Thus, the method facilitates the generation of reliable contrast agent enhanced CT images from CT images that have been acquired using a lower contrast agent dose than that used to acquire conventional CT images.

[0012] Further aspects, features, and advantages of the present disclosure will become apparent from the following description of examples, which is made with reference to the accompanying drawings.

[0013] Brief Description of the Drawings

[0014] Fig. 1 is a flowchart illustrating an example of a computer-implemented method of enhancing visibility of a contrast agent in CT imaging, in accordance with some aspects of the present disclosure.

[0015] Fig. 2 is a schematic diagram illustrating an example of a system 300 for enhancing visibility of a contrast agent in CT imaging in accordance with some aspects of the present disclosure.

[0016] Fig. 3 is a schematic diagram illustrating an example of the operation of generating S 120 (Fig. 1) from CT data representing a CT image 110 and one or more corresponding contrast agent enhanced CT images 130 in accordance with some aspects of the present disclosure.

[0017] Fig. 4 is a schematic diagram illustrating an example of the training of a machine learning model 120 to generate a contrast agent enhanced CT image 130 using a conventional CT image 110Tand a corresponding contrast agent enhanced CT image 130GTthat are generated from the same spectral CT image 140Tin accordance with some aspects of the present disclosure.

[0018] Fig. 5 is an example of a) a conventional CT image, b) a corresponding virtual-noncontrast (VNC) image, c) a MonoE40 image, and d) a MonoE120 image in accordance with some aspects of the present disclosure.

[0019] Fig. 6 is an example of a) a conventional CT image 110Tfor training a machine learning model, b) a corresponding virtual-non-contrast (VNC) image 150’, c) a MonoE40 image 130”, and e) a contrast agent enhanced CT image 130GTfor use as ground truth data and which is generated by combining the MonoE40 image 130” with the VNC image 150’ in accordance with some aspects of the present disclosure.

[0020] Fig. 7 is a schematic diagram illustrating a first example of the provision of a reference database 160 relating contrast agent injector parameters 170j and patient parameters 180! to a quality of the contrast agent enhanced CT image generated by the trained machine learning model 120 in accordance with some aspects of the present disclosure.

[0021] Fig. 8 is a schematic diagram illustrating a second example of the provision of a reference database 160 relating contrast agent injector parameters 170j and patient parameters 180! to a quality of the contrast agent enhanced CT image generated by the trained machine learning model 120 in accordance with some aspects of the present disclosure.

[0022] Fig. 9 is a schematic diagram illustrating an example of obtaining, based on a reference database 160, based on the value of the one or more patient parameters 180 and based on a specified value of the quality metric 220, a value of one or more contrast agent injector parameters 170j for use in acquiring a CT image of the patient in accordance with some aspects of the present disclosure.

[0023] Fig. 10 is a schematic diagram illustrating an example of obtaining, from a reference database 160, based on the value of the one or more contrast agent injector parameters 170j, and based on the value of one or more patient parameters 180 a corresponding value of the quality metric 220 in accordance with some aspects of the present disclosure.

[0024] Fig. 11 is a schematic diagram illustrating an example of providing a reference database 250 relating contrast agent injector parameters 170j and image acquisition parameters 260k to a quality of a contrast agent enhanced CT image 130 generated by a trained machine learning model 120 in accordance with some aspects of the present disclosure.

[0025] Fig. 12 is a schematic diagram illustrating an example of obtaining, based on a reference database 250, based on the value of the one or more image acquisition parameters 260k, and based on a specified value of a quality metric 220, a value of one or more corresponding contrast agent injector parameters 170j for use in acquiring a CT image in accordance with some aspects of the present disclosure.

[0026] Fig. 13 is a schematic diagram illustrating an example of obtaining, based on a reference database 250, based on a value of the one or more contrast agent injector parameters 170j, and based on a value of one or more image acquisition parameters 260k, a corresponding value of a quality metric 220 in accordance with some aspects of the present disclosure.

[0027] Fig. 14 is a schematic diagram illustrating an example of providing a reference database 270 relating contrast agent injector parameters 170® used to acquire a spectral CT image 110sto contrast agent injector parameters 170Cj for acquiring a corresponding conventional CT image 110cin accordance with some aspects of the present disclosure.

[0028] Fig. 15 is a schematic diagram illustrating an example of obtaining, based on a reference database 270, based on a specified value of a quality metric 220, and using the received value of the one or more contrast agent injector parameters 170Cj used to acquire the conventional CT image as the values of the one or more contrast agent injector parameters 170Cj, a corresponding value of one or more contrast agent injector parameters 170® used to acquire the spectral CT images in accordance with some aspects of the present disclosure. Detailed Description

[0029] Examples of the present disclosure are provided with reference to the following description and figures. In this description, for the purposes of explanation numerous specific details of certain examples are set forth. Reference in the specification to “an example”, “an implementation” or similar language means that a feature, structure, or characteristic described in connection with the example is included in at least that one example. It is also to be appreciated that features described in relation to one example may also be used in another example, and that all features are not necessarily duplicated in each example for the sake of brevity. For instance, features described in relation to a computer-implemented method, may be implemented in a computer program product, and in a system, in a corresponding manner.

[0030] In the following description, reference is made to examples of a computer- implemented method of enhancing visibility of a contrast agent in CT imaging. Reference is made to examples in which the visibility of a contrast agent in cardiac CT images is enhanced. It is, however, to be appreciated that the computer-implemented method may be used to enhance the visibility of a contrast agent in CT images wherein regions other than the heart are imaged. For instance, the computer-implemented method may be used to enhance the visibility of a contrast agent in CT images wherein the brain, the lung, the abdomen, and so forth, is imaged. More generally, the computer-implemented method may be used to enhance the visibility of a contrast agent in CT images wherein any anatomical region is imaged.

[0031] It is noted that the computer-implemented methods disclosed herein may be provided as a non-transitory computer-readable storage medium including computer-readable instructions stored thereon, which, when executed by at least one processor, cause the at least one processor to perform the method. In other words, the computer-implemented methods may be implemented in a computer program product. The computer program product can be provided by dedicated hardware or hardware capable of running the software in association with appropriate software. When provided by a processor, the functions of the method features can be provided by a single dedicated processor, by a single shared processor, or by a plurality of individual processors, some of which can be shared. The functions of one or more of the method features may for instance be provided by processors that are shared within a networked processing architecture, such as a client / server architecture, a peer-to-peer architecture, the Internet, or the Cloud.

[0032] The explicit use of the terms “processor” or “controller” should not be interpreted as exclusively referring to hardware capable of running software, and can implicitly include, but is not limited to, digital signal processor “DSP” hardware, read only memory “ROM” for storing software, random access memory “RAM”, a non-volatile storage device, and the like. Furthermore, examples of the present disclosure can take the form of a computer program product accessible from a computer-usable storage medium, or a computer-readable storage medium, the computer program product providing program code for use by or in connection with a computer or any instruction execution system. For the purposes of this description, a computer-usable storage medium or a computer readable storage medium can be any apparatus that can comprise, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. The medium can be an electronic, magnetic, optical, electromagnetic, infrared, or a semiconductor system or device or propagation medium. Examples of computer-readable media include semiconductor or solid-state memories, magnetic tape, removable computer disks, random access memory “RAM”, read-only memory “ROM”, rigid magnetic disks, and optical disks. Current examples of optical disks include compact disk-read only memory “CD-ROM”, compact disk-read / write “CD-R / W”, Blu-Ray™ and DVD.

[0033] It is also noted that some operations that are described as being performed in the computer-implemented methods disclosed herein may be implemented using artificial intelligence techniques. Suitable techniques may include machine learning techniques, deep learning techniques, and neural networks. For instance, one or more neural networks, may be trained in a supervised, or in some cases unsupervised, manner, to implement the operations performed in the computer-implemented methods disclosed herein.

[0034] As mentioned above, there remains a need to enhance the visibility of contrast agent in CT images.

[0035] Fig. 1 is a flowchart illustrating an example of a computer-implemented method of enhancing visibility of a contrast agent in CT imaging in accordance with some aspects of the present disclosure. Fig. 2 is a schematic diagram illustrating an example of a system 300 for enhancing visibility of a contrast agent in CT imaging in accordance with some aspects of the present disclosure. It is noted that operations that are described in relation to the method illustrated in Fig. 1 may also be performed by the system 300 illustrated in Fig. 2. Likewise, operations that are described as being performed by the system 300 illustrated in Fig. 2 may also be performed in the method described with reference to Fig. 1. With reference to Fig. 1 and Fig. 2, the computer-implemented method of enhancing visibility of a contrast agent in computed tomography (CT) imaging, includes: receiving S 110 CT data representing a CT image 110 comprising the contrast agent; generating S 120, from the CT data, one or more corresponding contrast agent enhanced CT images 130, wherein the visibility of the contrast agent is enhanced as compared to the CT image 110 using a trained machine learning model 120; and outputting S 130 the one or more contrast agent enhanced CT images 130; wherein a contrast agent dose used in the received CT data is lower than a contrast agent dose in CT data representing a corresponding conventional CT image 110 comprising the contrast agent.

[0036] The trained machine learning model 120 generates one or more contrast agent enhanced CT images 130 from a CT image 110. The contrast agent dose used in the received CT data is lower than a contrast agent dose in CT data representing a corresponding conventional CT image comprising the contrast agent. In other words, the contrast agent dose used to acquire the CT data is lower than a contrast agent dose used to acquire CT data representing a corresponding conventional CT image comprising the contrast agent. Thus, the method facilitates the provision of contrast agent enhanced CT images from CT images that have been acquired using a lower contrast agent dose than that used to acquire conventional CT images. This is beneficial for patients because it reduces the side-effects associated with contrast agent usage.

[0037] The operations performed in the method illustrated in Fig. 1 are described in more detail below.

[0038] Referring initially to the operation SI 10 illustrated in Fig. 1, in this operation, CT data representing a CT image 110 comprising the contrast agent is received.

[0039] In general, the CT data that is received in the operation SI 10 may be raw data, i.e. CT data that has not yet been reconstructed into a CT image, or it may be image data, i.e. CT data that already been reconstructed into a CT image. In general, the CT image 110 represented by the CT data may include any region of interest from within the anatomy. The CT image 110 represented by the CT data may include a portion of the heart, for example. In other words, the CT image may be a cardiac image. The CT image 110 represented by the CT data may alternatively be a CT image that includes another region of interest from within the anatomy, such as the brain, the lung, the abdomen, and so forth. The CT image 110 comprises a contrast agent. In other words, the CT image 110 comprises a contrast agent distribution. The contrast agent distribution results from the injection of a contrast agent into the vasculature. The contrast agent may be Iodine, Gadolinium, or another contrast agent.

[0040] The CT data that is received in the operation S 110 may be acquired using a CT imaging system. In one example, the CT data is acquired using a conventional CT imaging system. In this example, the CT data may be referred to as CT data representing a conventional CT image 110. A conventional CT imaging system is distinguished from a spectral CT imaging system by its inability to generate X-ray attenuation data representing X- ray attenuation within multiple different energy intervals. A conventional CT imaging system is only capable of generating X-ray attenuation data representing X-ray attenuation within a single energy interval. By contrast, a spectral CT imaging system is capable of generating X- ray attenuation data representing X-ray attenuation within multiple different energy intervals. The X-ray attenuation data generated by a spectral CT imaging system may be processed, e.g. using various material decomposition algorithms, or a neural network, in order to distinguish between media that have similar X-ray attenuation values when measured within a single energy interval, and which would be indistinguishable in X-ray attenuation data obtained using a conventional CT imaging system. Thus, X-ray attenuation data generated by a spectral CT imaging system may be processed to provide projection images with improved specificity to materials, such as contrast agent, tissue, bone, and so forth.

[0041] The CT data that is received in the operation S 110 may be received via any form of data communication, including via wired, or wireless, or optical fiber communication. By way of some examples, when wired data communication is used, the communication may take place via electrical signals that are transmitted on an electrical cable. When wireless data communication is used, the communication may take place via RF or infrared signals. When an optical fiber data communication is used, the communication takes place via optical signals that are transmitted on an optical fiber.

[0042] Referring now to the operation S120 illustrated in Fig. 1, in this operation, one or more corresponding contrast agent enhanced CT images 130 are generated from the CT data using a trained machine learning model 120. The visibility of the contrast agent in the one or more corresponding contrast agent enhanced CT images 130 is enhanced as compared to the CT image 110.

[0043] The operation S 120 is illustrated in Fig. 2 via the inputting of the CT image 110 into the trained machine learning model 120, and which in this example is implemented by at least one processor 320. In response to the inputting, the trained machine learning model 120 generates one or more corresponding contrast agent enhanced CT images 130, as illustrated in Fig. 2. The operation S120 is also illustrated in Fig. 3, which is a schematic diagram illustrating an example of the operation of generating one or more corresponding contrast agent enhanced CT images 130 from CT data representing a CT image 110 in accordance with some aspects of the present disclosure.

[0044] Various examples of the trained machine learning model 120 illustrated in Fig. 2 and in Fig. 3 are described in detail below. In general, the trained machine learning model may be provided by various types of machine learning models. The machine learning model may be provided by a neural network or by a deep learning model, for example.

[0045] The visibility of the contrast agent in the one or more corresponding contrast agent enhanced CT images 130 that are generated in the operation S120 is enhanced as compared to the CT image 110. In one example, the one or more corresponding contrast agent enhanced CT images 130 include an image that corresponds to a MonoE40 image. In another example, an Iodine-based contrast agent is used to acquire the received CT data, and the one or more corresponding contrast agent enhanced CT images 130 include an image that corresponds to an Iodine image.

[0046] With continued reference to the operation S120 illustrated in Fig. 2, a contrast agent dose used in the received CT data is lower than a contrast agent dose in CT data representing a corresponding conventional CT image 110 comprising the contrast agent. In other words, the contrast agent dose used to acquire the CT data is lower than a contrast agent dose used to acquire CT data representing a corresponding conventional CT image comprising the contrast agent. A corresponding conventional CT image comprising the contrast agent refers to conventional CT image representing the same anatomical region and acquired using the same type of contrast agent.

[0047] In an example, a total contrast agent dose used to acquire the CT data is lower than a total contrast agent dose used to acquire CT data representing a corresponding conventional CT image comprising the contrast agent. A total contrast agent dose may refer to a total volume (e.g. millilitres, ml) of contrast agent (e.g. Iodine) that was used to acquire the CT data. For instance, the CT may be acquired subsequent to the injection of a contrast agent bolus into the vasculature, and the total contrast agent dose may refer to the total amount of contrast agent in the bolus. For example, contrast agent typically has a fixed concentration e.g. 300 mg / ml of Iodine. Thus, 100 ml of intravenous contrast agent would result in 30g of Iodine being administered. The maximum dose is then given as a volume in ml, e.g. 300 ml. Alternatively, if the contrast agent is injected over time, then the total contrast agent dose may be determined by integrating a flow rate (e.g. millilitres per second) over time.

[0048] Referring now to the operation S130 illustrated in Fig. 1, in this operation, the one or more contrast agent enhanced CT images 130 are outputted. The one or more contrast agent enhanced CT images 130 may be outputed in various ways, including to a display device, such as a monitor or a virtual / augmented reality display device, to a printer, to a computer- readable storage medium, to the Internet, or to the Cloud, and so forth.

[0049] Thus, the trained machine learning model 120 generates one or more contrast agent enhanced CT images 130 from a CT image. The contrast agent dose used in the received CT data is lower than a contrast agent dose in CT data representing a corresponding conventional CT image comprising the contrast agent. In other words, the contrast agent dose used to acquire the CT data is lower than a contrast agent dose used to acquire CT data representing a corresponding conventional CT image comprising the contrast agent. Thus, the method facilitates the provision of contrast agent enhanced CT images from CT images that have been acquired using a lower contrast agent dose than that used to acquire conventional CT images. This is beneficial for patients because it reduces the side-effects associated with contrast agent usage.

[0050] Further examples of the method illustrated in Fig. 1 are described below with reference to Fig. 4.

[0051] In an example, the machine learning model 120 is trained to generate the one or more contrast agent enhanced CT images 130 using training data and ground truth data. The training data comprises a plurality of conventional CT images 110Tcomprising contrast agent distributions. The ground truth data comprises - for each conventional CT image - one or more corresponding contrast agent enhanced CT images 130GT, and wherein for each conventional CT image 110T, the conventional CT image 110Tand the one or more corresponding contrast agent enhanced CT images 130GTare generated from the same spectral CT image 140T.

[0052] Conventional CT images and also corresponding contrast agent enhanced CT images can be reliably generated from spectral CT images. For instance, material decomposition algorithms, and likewise trained neural networks, can be used to reliably generate a conventional CT image, and one or more corresponding contrast agent enhanced CT images, from a spectral CT image. Examples of such contrast-agent enhanced CT images include a MonoE40 image, i.e. a virtual monoenergetic image with an energy of 40keV, and a so-called Iodine image. Such images can be reliably generated from spectral CT images without the risk of introducing image artifacts, and which could otherwise be mis-interpreted as real image features and lead to a misdiagnosis. Moreover, spectral CT images are typically acquired using a lower contrast agent dose than conventional CT images. Consequently, the correspondence between a conventional CT image, and its one or more corresponding contrast agent enhanced CT images, and which is provided by generating these images from the same spectral CT image is valid at lower doses of contrast agent than that used to acquire conventional CT images. Therefore, by training the machine learning model to generate the one or more contrast agent enhanced CT images using conventional CT images and one or more corresponding contrast agent enhanced CT images that are generated from the same spectral CT image, the machine learning algorithm is both reliable and also valid at lower doses of contrast agent than that used to acquire conventional CT images. Thus, the method facilitates the generation of reliable contrast agent enhanced CT images from CT images that have been acquired using a lower contrast agent dose than that used to acquire conventional CT images.

[0053] This example is illustrated in Fig. 4, which is a schematic diagram illustrating an example of the training of a machine learning model 120 to generate a contrast agent enhanced CT image 130 using a conventional CT image 110Tand a corresponding contrast agent enhanced CT image 130GTthat are generated from the same spectral CT image 140Tin accordance with some aspects of the present disclosure. On the left-hand side of Fig. 4, an example of a spectral CT image 140Tis illustrated. Moving towards the right-hand side in Fig. 4, the spectral CT image 140Tis used to generate a conventional CT image 110T, and one or more corresponding contrast agent enhanced CT images 130GT, as indicated by the two diverging arrows. This operation may be performed using known techniques, such as using a material decomposition algorithm, or using a neural network.

[0054] In one example, the one or more contrast agent enhanced CT images 130, 130GTinclude a virtual monoenergetic image wherein an X-ray photon energy represented in the virtual monoenergetic image is less than a mean energy of X-ray photons represented in the corresponding conventional CT image 110, 110T. For example, if a conventional CT imaging system having a peak energy of 120keV is used to acquire the CT data, the MonoE energy should be less than 120keV. An example of such an image is a so-called MonoE40 image. Alternatively, or additionally, the one or more contrast agent enhanced CT images 130, 130GTcomprise a contrast agent-specific image. Examples of such images include a so-called Iodine image or a Gadolinium image. Known material decomposition algorithms or trained neural networks may be used to generate images such as these from a spectral CT image 140T.

[0055] An example of the use of material decomposition algorithms to generate a contrast agent enhanced CT image, and a conventional CT image from a spectral CT image is illustrated in Fig. 5. Fig. 5 is an example of a) a conventional CT image, b) a corresponding virtual-non-contrast, VNC, image, c) a MonoE40 image, and d) a MonoE120 image in accordance with some aspects of the present disclosure. The images illustrated in Fig. 5 depict virtual monoenergetic images which approximate monoenergetic images as would be obtained with a monoenergetic X-ray beam. As may be seen in Fig. 5, the visibility of the contrast agent in the MonoE40 image illustrated in Fig. 5c) is enhanced as compared to the conventional CT image illustrated in Fig. 5a). The conventional CT image illustrated in Fig. 5a) and the MonoE40 image illustrated in Fig. 5c) may be used as a conventional CT image 110Tand as a contrast agent enhanced CT image 130GT, respectively, fortraining the machine learning model 120.

[0056] Another example of the use of material decomposition algorithms to generate a contrast agent enhanced CT image and a conventional CT image from a spectral CT image is illustrated in Fig. 6. Fig. 6 is an example of a) a conventional CT image 110Tfor training a machine learning model, b) a corresponding virtual-non-contrast (VNC) image 150’, c) a MonoE40 image 130”, and e) a contrast agent enhanced CT image 130GTfor use as ground truth data and which is generated by combining the MonoE40 image 130” with the VNC image 150’ in accordance with some aspects of the present disclosure. In the example illustrated in Fig. 6, a virtual-non-contrast (VNC) image 150’, i.e. image b) and a MonoE40 image 130”, i.e. image c), and which are generated from the same spectral CT image, are combined to provide a contrast agent enhanced CT image 130GT. The images in Fig. 6b) and Fig. 6c) may be combined by weighting their pixel intensities with predetermined weights and summing the resulting values. For instance, the pixel intensities in one of the two images may be weighted with a value R wherein R < 1, and the pixel intensities in the other of the two images may be weighted with a value (1 — / ?). and the pixel intensities summed to provide the contrast agent enhanced CT image 130GT. A VNC image approximates an image that would have been acquired if no contrast agent was administered. A VNC image may be computed using known material decomposition algorithms. The operation of combining the images may include performing a denoising operation on the resulting image in order to reduce the amount of noise in the resulting image. Using this technique to generate at least some of the contrast agent enhanced CT image 130GTin the ground truth data that are used to train the machine learning model reduces the effort associated with curating training data for training the machine learning model 120.

[0057] More generally, in this example, for at least one of the conventional CT images in the training data 110T, at least one of the one or more corresponding contrast agent enhanced CT images 130GTin the ground truth data are generated from the same spectral CT image 140Tby: generating, from the spectral CT image 140T, a first contrast agent enhanced CT image 130’ wherein the contrast agent distribution is enhanced as compared to the conventional CT image 110T; generating, from the spectral CT image 140T, a second CT image 150’ wherein the contrast agent distribution is depleted as compared to the conventional CT image 110T; combining the first contrast agent enhanced CT image 130” with the second CT image 150’ to provide the corresponding contrast agent enhanced CT image 130GT.

[0058] Referring now to the upper branch of Fig. 4, the conventional CT image 110Tfrom the training data is inputted into the machine learning model 120. During training, the machine learning model generates a corresponding contrast agent enhanced CT image 130’. A difference, A , between the contrast agent enhanced CT image 130’ that is generated by the machine learning model 120, and the contrast agent enhanced CT images 130GTfrom the ground truth data, is then evaluated, e.g. using a loss function, and used to adjust the parameters of the machine learning model 120, as indicated by the thin arrow directed towards the machine learning model 120. Training is performed for multiple conventional CT images 110Tfrom the training data, and their corresponding contrast agent enhanced CT images 130GTfrom the ground truth data, both being generated from the same spectral CT image 140T, until the machine learning model 120 generates accurate contrast agent enhanced CT images 130’, i.e. contrast agent enhanced CT images 130’ that are within a specified difference of their corresponding contrast agent enhanced CT images 130GT, as determined by the value of the loss function, A.

[0059] As mentioned above, the machine learning model 120 may be provided by various types of machine learning models. The machine learning model may be provided by a neural network or by a deep learning model, for example. When provided by a neural network, the neural network may have various architectures. For instance, the neural network may be provided by various regression or generative models. Examples of such models include (variational) autoencoders, and latent diffusion models. The corresponding loss functions that are used during training may enforce pixel correspondences between the images. Examples of suitable loss functions include the mean-squared error, LI loss, and so forth. Conditional variants of the above (e.g. conditioned LDMs or spatially adaptive GANs) may also be used. The predictions, and also the losses, may be evaluated across the entire volume represented by the CT data, or alternatively then may be evaluated only within a sub-volume selected from within the volume represented by the CT data, e.g. on a sub-region of interest, or a patch, or a slice, of the CT data.

[0060] In general, the machine learning model 120 may be trained to generate the one or more contrast agent enhanced CT images 130 by: inputting the conventional CT image 110Tfrom the training data into the machine learning model 120; predicting one or more corresponding contrast agent enhanced CT images 130’ using the machine learning model 120, in response to the inputting; and adjusting parameters of the machine learning model 120 based on a difference, A, between the one or more contrast agent enhanced CT images 130’, predicted by the machine learning model 120, and the one or more corresponding contrast agent enhanced CT images 130GTfrom the ground truth data; repeating the inputting, the predicting, and the adjusting, until a stopping criterion is met.

[0061] In the example mentioned above in which the machine learning model 120 is a neural network, the parameters, or more particularly the weights and biases, control the operation of activation functions in the neural network. In supervised learning, the training process automatically adjusts the weights and the biases, such that when presented with the input data, the neural network accurately provides the corresponding expected output data. In order to do this, the value of the loss functions, or errors, are computed based on a difference between predicted output data and the expected output data. The value of the loss function may be computed using functions such as the negative log-likelihood loss, the mean absolute error (or LI norm), the mean squared error, the root mean squared error (or L2 norm), the Huber loss, or the (binary) cross entropy loss. During training, the value of the loss function is typically minimized, and training is terminated when the value of the loss function satisfies a stopping criterion. Sometimes, training is terminated when the value of the loss function satisfies one or more of multiple criteria.

[0062] Various methods are known for solving the loss minimization problem, including gradient descent, Quasi-Newton methods, and so forth. Various algorithms have been developed to implement these methods and their variants, including but not limited to Stochastic Gradient Descent “SGD”, batch gradient descent, mini-batch gradient descent, Gauss-Newton, Levenberg Marquardt, Momentum, Adam, Nadam, Adagrad, Adadelta, RMSProp, and Adamax “optimizers”. These algorithms compute the derivative of the loss function with respect to the model parameters using the chain rule. This process is called backpropagation since derivatives are computed starting at the last layer or output layer, moving toward the first layer or input layer. These derivatives inform the algorithm how the model parameters must be adjusted in order to minimize the error function. That is, adjustments to model parameters are made starting from the output layer and working backwards in the network until the input layer is reached. In a first training iteration, the initial weights and biases are often randomized. The neural network then predicts the output data, which is likewise, random. Backpropagation is then used to adjust the weights and the biases. The training process is performed iteratively by making adjustments to the weights and biases in each iteration. Training is terminated when the error, or difference between the predicted output data and the expected output data, is within an acceptable range for the training data, or for some validation data. Subsequently the neural network may be deployed, and the trained neural network makes predictions on new input data using the trained values of its parameters. If the training process was successful, the trained neural network accurately predicts the expected output data from the new input data.

[0063] The training of a neural network is often performed using a Graphics Processing Unit “GPU” or a dedicated neural processor such as a Neural Processing Unit “NPU” or a Tensor Processing Unit “TPU”. Training often employs a centralized approach wherein cloud-based or mainframe-based neural processors are used to train a neural network. Following its training with the training dataset, the trained neural network may be deployed to a device for analyzing new input data during inference. The processing requirements during inference are significantly less than those required during training, allowing the neural network to be deployed to a variety of systems such as laptop computers, tablets, mobile phones and so forth. Inference may for example be performed by a Central Processing Unit “CPU”, a GPU, an NPU, a TPU, on a server, or in the cloud.

[0064] The trained machine learning model 120 may then be used to generate contrast agent enhanced CT images 130 from inputted CT images 110, as illustrated in Fig. 3. The machine learning model 120 illustrated in Fig. 4 may be implemented by one or more processors 320 in the system illustrated in Fig. 1, for example.

[0065] The spectral CT images that are used to generate the training data 110Tand ground truth data 130GTin the example illustrated in Fig. 4 may be acquired from various types of spectral CT imaging systems. These include, for example, so-called “dual energy” Spectral CT imaging systems, and Photon counting Spectral CT imaging systems. In general, a spectral CT system may be provided by various configurations of X-ray sources and an X-ray detectors. For instance, a spectral CT imaging system may include multiple monochromatic sources, or one or more polychromatic sources, and the X-ray detector may include: a common detector for detecting X-ray radiation across multiple different X-ray energy intervals or multiple detectors, where each detector detects X-ray radiation within a different X-ray energy interval, or a multi-layer detector in which X-ray radiation within each of multiple different X-ray energy intervals is detected by corresponding layers, or a photon counting detector that bins detected X-ray photons into one of multiple energy intervals based on their individual energies. In general, a discrimination between different X-ray energy intervals may be provided at the X-ray source by temporally switching the X-ray anode potential of a single X-ray source, i.e. by “rapid kVp switching”, or by temporally switching, or filtering, the emission of X-rays from multiple X-ray sources. In such configurations, a common X-ray detector may be used to detect X-rays across multiple different energy intervals, attenuation data for each of the energy intervals being generated in a time-sequential manner. Alternatively, a discrimination between different X-ray energy intervals may be provided at the X-ray detector by using a multi-layer detector, or a photon counting detector. Such detectors can detect X-rays from multiple X-ray energy intervals near-simultaneously, and thus there is no need to perform temporal switching at the X-ray source. In another configuration, the need to sequentially switch different X-ray sources emitting X-rays in different energy intervals may be obviated by mounting X-ray source-detector pairs to a gantry at rotationally-offset positions around the axis of rotation. In yet another configuration, the spectral CT data may be provided by intercepting the X-ray beam in a conventional X-ray imaging system with one or more spectral filters in order to temporally control the spectrum of X-rays detected by an X-ray detector.

[0066] In another example, the method illustrated in Fig. 1 includes: determining a total contrast agent dose for acquiring the received CT data; and outputting the total contrast agent dose; wherein the total contrast agent dose for acquiring the received CT data is determined based a relationship between a total contrast agent dose and a quality of contrast agent enhancement provided by the machine learning model.

[0067] The total contrast agent dose that is provided by this example may be used to acquire CT data such that when the CT data is inputted into the trained machine learning model 120, the trained machine learning model 120 provides a desired quality of contrast agent enhancement.

[0068] In a related example, the relationship between the total contrast agent dose and a quality of contrast agent enhancement provided by the machine learning model is determined by: receiving test data 190, the test data comprising a plurality of CT images 110, wherein for each CT image 110, the test data comprises a corresponding value of a total contrast agent dose used to acquire the CT image 110; for each of a plurality of the CT images 110 in the test data 190: inputting the CT image 110 into the trained machine learning model 120; generating, from the inputted CT image 110, a corresponding contrast agent enhanced CT image 130, using the trained machine learning model 120; evaluating a quality metric 220 for the contrast agent enhanced CT image 130; and storing the value of the quality metric 220 and the corresponding value of the total contrast agent dose to provide a database representing the relationship.

[0069] The database representing the relationship may then be used to determine a recommended total contrast agent dose that should be used to acquire CT data, and which when inputted into the trained machine learning model, results in a desired value of the quality metric. This helps to reduce the total contrast agent dose that is used to acquire the CT data and, in-tum, reduces the side-effects associated with contrast agent usage. The database may also be used to determine parameters of a model relating the value of the total contrast agent dose to the value of the quality metric 220. The model may be a classifier model, for example. The model representing the relationship may then similarly be used to determine a recommended total contrast agent dose that should be used to acquire CT data and which, when inputted into the trained machine learning model, results in a desired value of the quality metric. Again, this helps to reduce the total contrast agent dose that is used to acquire the CT data and, in-tum, reduces the side-effects associated with contrast agent usage.

[0070] In another example, the trained machine learning model 120 is used to provide a reference database 160 relating contrast agent injector parameters 170j and patient parameters 180! to a quality of the contrast agent enhanced CT image generated by the trained machine learning model 120. The reference database 160 is generated by inputting test data into the trained machine learning model 120. The reference database 160 may be used to determine the contrast agent injector parameter(s) 170j that are required in order to obtain a desired quality of contrast agent enhanced CT image from the trained machine learning model 120 given known patient parameter(s) 180!. The reference database 160 may alternatively be used to determine the quality of contrast agent enhanced CT image that can be expected from the trained machine learning model 120 given known patient parameter(s) 180tand known contrast agent injector parameter(s) 170j . In this example, the method illustrated in Fig. 1 includes: receiving test data 190, the test data comprising a plurality of CT images 110, wherein for each CT image 110, the test data comprises i) a corresponding value of one or more contrast agent injector parameters 170j used to acquire the CT image 110, and ii) a corresponding surview scan 210; for each of a plurality of the CT images 110 in the test data 190: inputting the CT image 110 into the trained machine learning model 120; generating, from the inputted CT image 110, a corresponding contrast agent enhanced CT image 130, using the trained machine learning model 120; evaluating a quality metric 220 for the contrast agent enhanced CT image 130; extracting, from the corresponding surview scan 210, a value of one or more patient parameters 180! for a patient represented in the surview scan 210; and storing the value of the quality metric 220, the corresponding value of the one or more contrast agent injector parameters 170j, and the corresponding extracted value of one or more patient parameters 180!, to provide a reference database 160 relating contrast agent injector parameters 170j and patient parameters 180tto a quality of the contrast agent enhanced CT image generated by the trained machine learning model 120.

[0071] This example is illustrated in Fig. 7, which is a schematic diagram illustrating a first example of the provision of a reference database 160 relating contrast agent injector parameters 170j and patient parameters 180tto a quality of the contrast agent enhanced CT image generated by the trained machine learning model 120 in accordance with some aspects of the present disclosure. In this example, the test data 190 is illustrated on the left-hand side of Fig. 7. The test data 190 includes a plurality of CT images 110, and for each CT image 110, i) a corresponding value of one or more contrast agent injector parameters 170j used to acquire the CT image 110, and ii) a corresponding surview scan 210. The contrast agent injector parameter(s) may include parameters such as a total contrast agent dose (e.g. volume and concentration), a contrast agent flow rate, a duration of the contrast agent injection, an indication of whether a saline chaser bolus was used, a type (e.g. manufacturer, or size) of injection needle, a location of the injection needle used perform the contrast agent injection, and so forth. A surview scan, also known as a scout scan, includes one or more projection images that are acquired using a relatively lower X-ray dose and / or image resolution than a diagnostic CT image. A surview scan is often acquired for the purpose of localising an anatomical region from which CT data is acquired. Typical projection directions include the anterior-posterior “AP” or lateral “LAT” direction.

[0072] As illustrated in the upper branch of Fig. 7, a CT image 110 from the test data 190 is inputted into the trained machine learning model 120. In response, the trained machine learning model 120 generates a corresponding contrast agent enhanced CT image 130. A quality metric 220 is then evaluated for the contrast agent enhanced CT image 130. The quality metric 220 may be evaluated based on various factors. For instance, the quality metric 220 may be evaluated based on a ratio of the average contrast agent attenuation within a region that is expected to include a contrast agent in the contrast agent enhanced CT image 130 to the average contrast agent attenuation within a region that is not expected to include a contrast agent in the contrast agent enhanced CT image 130. Alternatively or additionally, the quality metric 220 may be evaluated based on characteristics of the contrast agent enhanced CT image 130 such as image sharpness. Alternatively or additionally, the quality metric 220 may be evaluated based on the performance of post-processing steps, e.g. a heart segmentation, and which typically include a metric representing their quality. For instance, a mesh-based segmentation typically searches for specific edges that it needs to segment, e.g. the contrast-enhanced transition between blood-pool and myocardium. If the edge detection fails to detect these edges, the quality metric would be relatively lower than if the edges are detected. Similarly, data-driven Al models may be trained to yield a quality metric. For example, a classifier network that has been trained to distinguish contrast-enhanced, from non-enhanced scans, may output a probability of the correctness of its classification. The quality metric 220 may be evaluated using known image processing techniques. In one example, the quality metric 220 is evaluated selectively within a region of interest comprising at least a portion of the contrast agent distribution. Contrast agent, and consequently contrast agent enhancement, is only expected to occur in specific anatomical structures, such as a patient’s vasculature. The quality metric 220 may therefore be evaluated selectively within a region of interest such as a portion of the vasculature, for example, without the value of the quality metric being impacted by other regions in the contrast agent enhanced CT image 130. For instance, a vessel segmentation operation may be performed on the contrast agent enhanced CT image 130 in order to identify vessels for which the quality metric is selectively evaluated. In the example illustrated in Fig. 7, the quality metric 220 is evaluated using a quality metric determination unit 230. The quality metric determination unit 230 may be implemented by the at least one processor 320 illustrated in Fig. 2.

[0073] With reference to the lower branch in Fig. 7, a value of one or more patient parameters 180! for a patient represented in the surview scan 210 are extracted from the surview scan 210 that corresponds to the CT image 110 that was inputted into the trained machine earning model 120. Examples of patient parameters 180! that may be extracted in this operation include a total X-ray absorption of the patient, patient dimensions (e.g. a patient thickness), the patient weight, the anatomical region of the patient represented in the CT image 110, and so forth. Such parameters may be extracted from the surview scan 210 for the relevant portion of the CT image 110. Known image analysis techniques, (model based) segmentation techniques, and so forth, may be used to extract such patient parameters. In the example illustrated in Fig. 7, the patient parameters 180, are extracted using a parameter extraction unit 240. The parameter extraction unit 240 may be implemented by the at least one processor 320 illustrated in Fig. 2.

[0074] With reference to the lower right-hand portion of Fig. 7, the value of the quality metric 220, the corresponding value of the one or more contrast agent injector parameters 170j, and the corresponding extracted value of one or more patient parameters 180 are then stored to provide the reference database 160. The reference database 160 may be stored to a computer-readable storage medium, or to the Internet, or to the Cloud, and so forth. In general, the contrast agent injector parameters 170j, and the patient parameters 180! may represent a multidimensional space. The multidimensional space may include m injector parameters, and n patient parameters, as indicated via the respective parameter indices in Fig. 7. For ease of illustration, the example in Fig. 7 illustrates the value of the quality metric 220 for two dimensions, i.e. a single injector parameter, and a single patient parameter, and the value of the quality metric 220 is indicated via the shading of the symbols for the two parameters. In the example illustrated in Fig. 7, relatively higher values of the quality metric 220 are indicated as relatively lighter shading and relatively lower values of the quality metric 220 are indicated as relatively darker shading. In this example, a dashed line illustrates a separation between the region of the multiparametric space corresponding to relatively higher values of the quality metric 220, and a region of the multiparametric space corresponding to relatively lower values of the quality metric 220.

[0075] In another example, the trained machine learning model 120 is similarly used to provide a reference database 160 relating contrast agent injector parameters 170j and patient parameters 180! to a quality of the contrast agent enhanced CT image generated by the trained machine learning model 120. However, in this example, the test data that is used to provide the reference database 160 includes spectral CT images. This example is illustrated in Fig. 8, which is a schematic diagram illustrating a second example of the provision of a reference database 160 relating contrast agent injector parameters 170j and patient parameters 180tto a quality of the contrast agent enhanced CT image generated by the trained machine learning model 120, in accordance with some aspects of the present disclosure. The example illustrated in Fig. 8 shares many of the same features as the example illustrated in Fig. 7. However, in the Fig. 8 example, the CT images 110 in the test data 190 are spectral CT images 110s. In this example, the method also includes: for each of a plurality of the spectral CT images 110sin the test data 190: generating a corresponding conventional CT image 110scand a corresponding contrast agent enhanced CT image 130SCE; and Moreover, in this example, the operation of inputting the CT image 110 into the trained machine learning model 120, comprises inputting the conventional CT image 110scinto the trained machine learning model 120; and the operation of generating a corresponding contrast agent enhanced CT image 130, using the trained machine learning model 120, is performed using the inputted conventional CT image 110sc; and the operation of evaluating a quality metric 220 for the contrast agent enhanced CT image 130 is performed based on a difference, A, between the contrast agent enhanced CT image 130 generated using the trained machine learning model 120 and the contrast agent enhanced CT image 130SCEgenerated from the spectral CT image 110s.

[0076] Thus, in this example, the quality metric 220 is evaluated using the contrast agent enhanced CT image 130SCEthat is generated from the spectral CT image 110s. As described above, the conventional CT image 110scand the corresponding contrast agent enhanced CT image 130SCEmay be generated from the spectral CT images 110sin the test data 190 using known techniques, such as by using a material decomposition algorithm, or a trained neural network. As also described above, the contrast agent enhanced CT images 130SCEmay be virtual monoenergetic images wherein an X-ray photon energy represented in the virtual monoenergetic image is less than a mean energy of X-ray photons represented in the corresponding conventional CT image 110, 110T, such as a MonoE40 image, for example, or they may be contrast agent-specific images, such as an Iodine image, or a Gadolinium image, or example.

[0077] As mentioned above, the reference database 160 may be used to determine the contrast agent injector parameter(s) 170j that are required in order to obtain a desired quality of contrast agent enhanced CT image from the trained machine learning model 120 given known patient parameter(s) 180!. Thus, in one example, the method illustrated in Fig. 1 includes: receiving a surview scan 210; extracting, from the surview scan 210, a value of one or more patient parameters 180! for a patient represented in the surview scan 210; and obtaining, based on the reference database 160, based on the value of the one or more patient parameters 180 and based on a specified value of the quality metric 220, a value of one or more contrast agent injector parameters 170j for use in acquiring a CT image of the patient; and outputing the value of one or more contrast agent injector parameters 170j to provide contrast agent parameters for use in acquiring CT data for use with the trained machine learning model 120.

[0078] This example is illustrated in Fig. 9, which is a schematic diagram illustrating an example of obtaining, based on a reference database 160, based on the value of the one or more patient parameters 180 and based on a specified value of the quality metric 220, a value of one or more contrast agent injector parameters 170j for use in acquiring a CT image of the patient, in accordance with some aspects of the present disclosure. In the example illustrated in Fig. 9, the patient parameters 180! are extracted from the surview scan 210 in the same manner as described above with reference to Fig. 7. The reference database 160 may be used as a lookup table to provide the value of the contrast agent injector parameter(s) 170j for which a specified value of the quality metric 220 may be obtained, given the value(s) of the patient parameters that are extracted from the surview scan. In the illustrated example, a relatively higher (light-shaded) value of the quality metric is desired, and the corresponding, value of the contrast agent injector parameter(s) 170j is obtained from the reference database 160. Instead of using the reference database 160 as a lookup table, a model, such as a classification model, may be generated using the reference database 160, and the model then used to provide the value of the contrast agent injector parameter(s) 170j .

[0079] As also mentioned above, the reference database 160 may alternatively be used to determine the quality of contrast agent enhanced CT image that can be expected from the trained machine learning model 120 given known patient parameter(s) 180! and known contrast agent injector parameter(s) 170j . Thus, in another example, the method illustrated in Fig. 1 includes: receiving a value of one or more contrast agent injector parameters 170j ; receiving a surview scan 210; extracting, from the surview scan 210, a value of one or more patient parameters 180tfor a patient represented in the surview scan 210; and obtaining, based on the reference database 160, based on the value of the one or more contrast agent injector parameters 170j, and based on the value of the one or more patient parameters 180 a corresponding value of the quality metric 220; and outputing the value of the quality metric 220 to provide an assessment of a suitability of contrast agent injector parameters 170j for acquiring a CT image 130 for use in the trained machine learning model 120.

[0080] This example is illustrated in Fig. 10, which is a schematic diagram illustrating an example of obtaining, from a reference database 160, based on the value of the one or more contrast agent injector parameters 170j, and based on the value of one or more patient parameters 180 a corresponding value of the quality metric 220, in accordance with some aspects of the present disclosure. In the example illustrated in Fig. 10, the patient parameters 180! are extracted from the surview scan 210 in the same manner as described above with reference to Fig. 7. The extracted patient parameters 180! are used in combination with the contrast agent injector parameters 170j to look-up the corresponding value of the quality metric 220 that is expected to be obtained from inputting into the trained machine learning model 120 a CT image 110 corresponding to the surview scan 210, and which has been acquired using the contrast agent injector parameters 170j . The value of the quality metric 220 therefore provides an assessment of a suitability of contrast agent injector parameters 170j for acquiring a CT image 130 for use in the trained machine learning model 120. As was described with reference to the Fig. 9 example, instead of using the reference database 160 as a lookup table, a model, such as a classification model, may be generated using the reference database 160, and the model then used to provide the value of the quality metric 220.

[0081] In another example, the trained machine learning model 120 is used to provide a reference database 250 relating contrast agent injector parameters 170j and image acquisition parameters 260k to a quality of a contrast agent enhanced CT image 130 generated by the trained machine learning model 120. The reference database 250 is generated by inputting test data into the trained machine learning model 120. The reference database 250 may be used to determine the contrast agent injector parameter(s) 170j that are required in order to obtain a desired quality of contrast agent enhanced CT image from the trained machine learning model 120 given known image acquisition parameter(s) 260k. The reference database 250 may alternatively be used to determine the quality of contrast agent enhanced CT image that can be expected from the trained machine learning model 120 given known image acquisition parameter(s) 260k and known contrast agent injector parameter(s) 170j .

[0082] In this example, the method illustrated in Fig. 1 includes: receiving test data 190, the test data comprising a plurality of CT images 110, wherein for each CT image 110, the test data comprises i) a corresponding value of one or more contrast agent injector parameters 170j used to acquire the CT image 110, and ii) corresponding image acquisition parameters 260k used to acquire the CT image 110; for each of a plurality of the CT images 110 in the test data 190: inputting the CT image 110 into the trained machine learning model 120; generating, from the inputted CT image 110, a corresponding contrast agent enhanced CT image 130, using the trained machine learning model 120; evaluating a quality metric 220 for the contrast agent enhanced CT image

[0083] 130; and storing the value of the quality metric 220, the corresponding value of the one or more contrast agent injector parameters 170j, and the corresponding image acquisition parameters 260k, to provide a reference database 250 relating contrast agent injector parameters 170j and image acquisition parameters 260k to a quality of a contrast agent enhanced CT image 130 generated by the trained machine learning model 120.

[0084] This example is illustrated in Fig. 11, which is a schematic diagram illustrating an example of providing a reference database 250 relating contrast agent injector parameters 170j and image acquisition parameters 260k to a quality of a contrast agent enhanced CT image 130 generated by a trained machine learning model 120, in accordance with some aspects of the present disclosure. In this example, the test data 190 is illustrated on the left-hand side of Fig. 7. The test data 190 includes a plurality of CT images 110, and for each CT image 110, i) a corresponding value of one or more contrast agent injector parameters 170j used to acquire the CT image 110, and ii) corresponding image acquisition parameters 260k used to acquire the CT image 110. The image acquisition parameter(s) 260k may include parameters such as a tube voltage (kV), a tube current, an accumulation time, an indication of the use of modulation, an indication of the use of a filter (e.g. copper, or aluminium) to modify the X- ray spectrum, a type of CT imaging system used to acquire the CT data, and so forth.

[0085] As illustrated in the upper branch of Fig. 11, a CT image 110 from the test data 190 is inputted into the trained machine learning model 120. In response, the trained machine learning model 120 generates a corresponding contrast agent enhanced CT image 130. A quality metric 220 is then evaluated for the contrast agent enhanced CT image 130. The quality metric 220 may be evaluated using the techniques described with reference to Fig. 7.

[0086] With reference to the central, and lower branches in Fig. 11, value(s) of the injector parameter(s) 170j that were used to acquire the CT image 110, the value(s) of the one or more image acquisition parameters 260k that were used to acquire the CT image 110 and the corresponding value of the quality metric 220, are then stored to provide the reference database 250 illustrated in the lower right-hand portion of Fig. 11. The reference database 160 may be stored to a computer-readable storage medium, or to the Internet, or to the Cloud, and so forth. In general, the contrast agent injector parameters 170j, and the image acquisition parameters 260k may represent a multidimensional space, as described with reference to Fig. 7. In the example illustrated in Fig. 11, relatively higher values of the quality metric 220 are indicated as relatively lighter shading and relatively lower values of the quality metric 220 are indicated as relatively darker shading. In this example, a dashed line illustrates a separation between the region of the multiparametric space corresponding to relatively higher values of the quality metric 220, and a region of the multiparametric space corresponding to relatively lower values of the quality metric 220.

[0087] In another example, the trained machine learning model 120 is similarly used to provide a reference database 250 relating contrast agent injector parameters 170j and image acquisition parameters 260k to a quality of the contrast agent enhanced CT image generated by the trained machine learning model 120. However, in this example, the test data that is used to provide the reference database 160 includes spectral CT images. This example corresponds to the example illustrated in Fig. 8, with the difference that the test data includes image acquisition parameters 260k, rather than a surview scan 210. In this example, the method also includes: for each of a plurality of the spectral CT images 110sin the test data 190: generating a corresponding conventional CT image 110scand a corresponding contrast agent enhanced CT image 130SCE; and

[0088] Moreover, in this example, the operation of inputting the CT image 110 into the trained machine learning model 120, comprises inputting the conventional CT image 110scinto the trained machine learning model 120; and the operation of generating a corresponding contrast agent enhanced CT image 130, using the trained machine learning model 120, is performed using the inputted conventional CT image 110sc; and the operation of evaluating a quality metric 220 for the contrast agent enhanced CT image 130 is performed based on a difference, A, between the contrast agent enhanced CT image 130 generated using the trained machine learning model 120 and the contrast agent enhanced CT image 130SCEgenerated from the spectral CT image 110s.

[0089] As mentioned above, the reference database 250 may be used to determine the contrast agent injector parameter(s) 170j that are required in order to obtain a desired quality of contrast agent enhanced CT image from the trained machine learning model 120 given known image acquisition parameter(s) 260k.

[0090] Thus, in one example, the method illustrated in Fig. 1 includes: receiving a value of one or more image acquisition parameters 260k for acquiring a CT image; and obtaining, based on the reference database 250, based on the value of the one or more image acquisition parameters 260k, and based on a specified value of the quality metric 220, a value of one or more corresponding contrast agent injector parameters 170j for use in acquiring the CT image; and outputing the value of the one or more contrast agent injector parameters 170j .

[0091] This example is illustrated in Fig. 12, which is a schematic diagram illustrating an example of obtaining, based on a reference database 250, based on the value of the one or more image acquisition parameters 260k, and based on a specified value of a quality metric 220, a value of one or more corresponding contrast agent injector parameters 170j for use in acquiring a CT image, in accordance with some aspects of the present disclosure. The reference database 250 may be used as a lookup table to provide the value of the contrast agent injector parameter(s) 170j for which a specified value of the quality metric 220 may be obtained, given the value(s) of the image acquisition parameters 260k. Instead of using the reference database 250 as a lookup table, a model, such as a classification model, may be generated using the reference database 250, and the model then used to provide the value of the contrast agent injector parameter(s) 170j .

[0092] The reference database 250 may alternatively be used to determine the quality of contrast agent enhanced CT image that can be expected from the trained machine learning model 120 given known image acquisition parameter(s) 260k and known contrast agent injector parameter(s) 170j.

[0093] Thus, in another example, the method illustrated in Fig. 1 includes: receiving a value of the one or more contrast agent injector parameters 170j for acquiring a CT image; receiving a value of one or more image acquisition parameters 260k for acquiring the CT image; and obtaining, based on the reference database 250, based on the value of the one or more contrast agent injector parameters 170j, and based on the value of the one or more image acquisition parameters 260k, a corresponding value of the quality metric 220; and outputing the value of the quality metric 220 to provide an assessment of a suitability of the contrast agent injector parameters 170j and the image acquisition parameters 260k for acquiring a CT image for use in the trained machine learning model 120.

[0094] This example is illustrated in Fig. 13, which is a schematic diagram illustrating an example of obtaining, based on a reference database 250, based on a value of the one or more contrast agent injector parameters 170j, and based on a value of one or more image acquisition parameters 260k, a corresponding value of a quality metric 220, in accordance with some aspects of the present disclosure. In the example illustrated in Fig. 13, image acquisition parameters 260k are used in combination with the contrast agent injector parameters 170j to look-up the corresponding value of the quality metric 220 that is expected to be obtained from inputing into the trained machine learning model 120 a CT image that has been acquired using the contrast agent injector parameters 170j and the image acquisition parameters 260k. The value of the quality metric 220 therefore provides an assessment of a suitability of the contrast agent injector parameters 170j and the image acquisition parameters 260k for acquiring a CT image for use in the trained machine learning model 120. As was described with reference to the Fig. 9 example, instead of using the reference database 160 as a lookup table, a model, such as a classification model, may be generated using the reference database 160, and the model then used to provide the value of the quality metric 220.

[0095] In another example, the trained machine learning model 120 is used to provide a reference database 270 relating contrast agent injector parameters 170® used to acquire a spectral CT image 110sto contrast agent injector parameters 170Cj for acquiring a corresponding conventional CT image 110c. This example may be used to provide recommended contrast agent injector parameters 170CLDj for use in acquiring the conventional CT image 110 for use in the trained machine learning model 120. Thus, it helps to reduce the total contrast agent dose that is used to acquire the CT data, and in-tum reduces the sideeffects associated with contrast agent usage. In this example, the method illustrated in Fig. 1 includes: receiving test data 190, the test data comprising a plurality of spectral CT images 110s, wherein for each spectral CT image 110s, the test data comprises i) a corresponding value of one or more contrast agent injector parameters 170sused to acquire the spectral CT image 110s, and ii) a corresponding value of one or more contrast agent injector parameters 170Cj for acquiring a conventional CT image 110c; for each of a plurality of the spectral CT images 110sin the test data: generating a corresponding conventional CT image 110scand a corresponding contrast agent enhanced CT image 130SCE; inputting the conventional CT image 110scinto the trained machine learning model 120; generating, from the inputted conventional CT image 110sc, a corresponding contrast agent enhanced CT image 130, using the trained machine learning model 120; evaluating a quality metric 220 for the contrast agent enhanced CT image 130, wherein the quality metric 220 is evaluated based on a difference, A, between the contrast agent enhanced CT image 130 generated using the trained machine learning model 120 and the contrast agent enhanced CT image generated from the spectral CT image 130SCE; and subject to the value of the quality metric 220 exceeding a threshold value, storing the corresponding value of the one or more contrast agent injector parameters 170® used to acquire the spectral CT image 110s, and the corresponding value of one or more contrast agent injector parameters 170Cj for acquiring a conventional CT image 110c, to provide a reference database 270 relating contrast agent injector parameters 170sused to acquire a spectral CT image 110sto contrast agent injector parameters 170Cj for acquiring a corresponding conventional CT image 110c.

[0096] This example is illustrated in Fig. 14, which is a schematic diagram illustrating an example of providing a reference database 270 relating contrast agent injector parameters 170sused to acquire a spectral CT image 110sto contrast agent injector parameters 170Cj for acquiring a corresponding conventional CT image 110c, in accordance with some aspects of the present disclosure. The upper portion of the example illustrated in Fig. 14 has similarities with the upper portion of the example illustrated in Fig. 8. Thus, the operations of generating a corresponding conventional CT image 110scand a corresponding contrast agent enhanced CT image 130SCEfrom a spectral CT image 110sin the test data; inputting the conventional CT image 110scinto the trained machine learning model 120; generating a corresponding contrast agent enhanced CT image 130 using the trained machine learning model 120; and evaluating a quality metric 220 for the contrast agent enhanced CT image 130, may be performed in the same manner as described above with reference to Fig. 8

[0097] The example illustrated in Fig. 14 differs from the example illustrated in Fig. 8 in that the latter includes in the test data 190, a corresponding value of one or more contrast agent injector parameters 170Cj for acquiring a conventional CT image 110c; rather than a corresponding surview scan. This is illustrated in the lower left-hand portion of Fig. 14. The contrast agent injector parameters 170Cj for acquiring a conventional CT image 110c, are parameters that are deemed to be suitable for acquiring a conventional CT image 110c. These parameters may be provided by a user. For instance, the user may know, based on typical protocols used at a medical facility, the values of the parameters that would typically be used to acquire a conventional CT image 110cof the anatomical region represented in the spectral CT image 110s. With reference to the lower right-hand portion of Fig. 14, the value(s) of the spectral CT injector parameter(s) 170sthat were used to acquire the spectral CT image 110s, the value(s) of the contrast agent injector parameters 170Cj for acquiring a conventional CT image 110cand the corresponding value of the quality metric 220, are then stored to provide the reference database 270. This operation may be performed as described above with reference to Fig. 8. As mentioned above, the reference database 270 may be used to provide recommended contrast agent injector parameters for use in acquiring a conventional CT image 110 for use in the trained machine learning model 120. Thus, in one example, the method illustrated in Fig. 1 includes: receiving a value of one or more contrast agent injector parameters 170Cj used to acquire a conventional CT image; obtaining, based on the reference database 270, based on a specified value of the quality metric 220, and using the received value of the one or more contrast agent injector parameters 170Cj used to acquire the conventional CT image as the values of the one or more contrast agent injector parameters 170Cj, a corresponding value of the one or more contrast agent injector parameters 170® used to acquire the spectral CT images; and outputting the value of the one or more contrast agent injector parameters 170® used to acquire the spectral CT images as recommended contrast agent injector parameters 170CLDj for use in acquiring the conventional CT image 110 for use in the trained machine learning model 120.

[0098] This example is illustrated in Fig. 15, which is a schematic diagram illustrating an example of obtaining, based on a reference database 270, based on a specified value of a quality metric 220, and using the received value of the one or more contrast agent injector parameters 170Cj used to acquire the conventional CT image as the values of the one or more contrast agent injector parameters 170Cj, a corresponding value of one or more contrast agent injector parameters 170Sj used to acquire the spectral CT images, in accordance with some aspects of the present disclosure. As illustrated in Fig. 15, in this example, the reference database 270, or a model derived from the reference database 270, is used to determine contrast agent injector parameters 170Sj that were used to acquire a spectral CT image from inputted contrast agent injector parameters 170Cj used to acquire a conventional CT image. The contrast agent injector parameters 170Sj that were used to acquire the spectral CT image, are then outputted as recommended contrast agent injector parameters 170CLDj for use in acquiring the conventional CT image 110 for use in the trained machine learning model 120. In so doing, this example facilitates a reduction in the total contrast agent dose that is used to acquire the CT data, and in-tum reduces the side-effects associated with contrast agent usage.

[0099] In another example, a computer program product is provided. The computer program product comprises instructions which when executed by at least one processor, cause the at least one processor to carry out a method of enhancing visibility of a contrast agent in computed tomography, CT, imaging. The method comprises: receiving S 110 CT data representing a CT image 110 comprising the contrast agent; generating S 120, from the CT data, one or more corresponding contrast agent enhanced CT images 130, wherein the visibility of the contrast agent is enhanced as compared to the CT image 110 using a trained machine learning model 120; and outputting S 130 the one or more contrast agent enhanced CT images 130; wherein a contrast agent dose used in the received CT data is lower than a contrast agent dose in CT data representing a corresponding conventional CT image 110 comprising the contrast agent.

[0100] In this example, the machine learning model 120 may be trained to generate the one or more contrast agent enhanced CT images 130 using training data and ground truth data, the training data comprising a plurality of conventional CT images 110Tcomprising contrast agent distributions, and the ground truth data comprising for each conventional CT image, one or more corresponding contrast agent enhanced CT images 130GT, and wherein for each conventional CT image 110T, the conventional CT image 110Tand the one or more corresponding contrast agent enhanced CT images 130GTare generated from the same spectral CT image 140T

[0101] In another example, a system 300 for enhancing visibility of a contrast agent in computed tomography, CT, imaging, is provided. The system comprises: a memory 310 that stores a plurality of instructions; and at least one processor 320 coupled to the memory and configured to execute the plurality of instructions to: receive SI 10 CT data representing a CT image 110 comprising the contrast agent; generate S120, from the CT data, one or more corresponding contrast agent enhanced CT images 130, wherein the visibility of the contrast agent is enhanced as compared to the CT image 110 using a trained machine learning model 120; and output S130 the one or more contrast agent enhanced CT images 130; wherein a contrast agent dose used in the received CT data is lower than a contrast agent dose in CT data representing a corresponding conventional CT image 110 comprising the contrast agent.

[0102] This example is illustrated in Fig. 1. In this example, the machine learning model 120 may be trained to generate the one or more contrast agent enhanced CT images 130 using training data and ground truth data, the training data comprising a plurality of conventional CT images 110Tcomprising contrast agent distributions, and the ground truth data comprising for each conventional CT image, one or more corresponding contrast agent enhanced CT images 130GT, and wherein for each conventional CT image 110T, the conventional CT image 110Tand the one or more corresponding contrast agent enhanced CT images 130GTare generated from the same spectral CT image 140T.

[0103] It is noted that the system 300 may also include one or more of: a CT imaging system for providing the CT data; an injector for injecting a contrast agent; a display, such as a monitor, or a virtual / augmented reality display device, for displaying the contrast agent enhanced CT images 130, other outputs generated by the at least one processor 320, and so forth; and a user input device configured to receive user input in relation to the operations performed by the at least one processor 320, such as a keyboard, a mouse, a touchscreen, and so forth.

[0104] It is noted that variations of some of the examples described above are also contemplated. For instance, in one example a computer-implemented method of enhancing visibility of a contrast agent distribution in computed tomography, CT, images, is provided. The method includes: receiving S 110 CT data representing a conventional CT image 110 comprising a contrast agent distribution; inputting the CT data into a trained machine learning model 120; and generating S120, from the inputted CT data, one or more corresponding contrast agent enhanced CT images 130 wherein the visibility of the contrast agent distribution is enhanced as compared to the conventional CT image 110, using the trained machine learning model 120; and outputting S 130 the one or more contrast agent enhanced CT images 130; and wherein the machine learning model 120 is trained to generate the one or more contrast agent enhanced CT images 130 using training data and ground truth data, the training data comprising a plurality of conventional CT images 110Tcomprising contrast agent distributions, and the ground truth data comprising for each conventional CT image, one or more corresponding contrast agent enhanced CT images 130GT, and wherein for each conventional CT image 110T, the conventional CT image 110Tand the one or more corresponding contrast agent enhanced CT images 130GTare generated from the same spectral CT image 140T.

[0105] This example corresponds to the example described above with reference to Fig. 1 - Fig. 4, with the additional requirement that the received CT data represents a conventional CT image 110 comprising a contrast agent distribution, with the additional requirement of training data and ground truth data, and without the requirement that a contrast agent dose used in the received CT data is lower than a contrast agent dose in CT data representing a corresponding conventional CT image 110 comprising the contrast agent. In another example, a computer-implemented method of providing a reference database 160 relating contrast agent injector parameters 170j and patient parameters 180! to a quality of a contrast agent enhanced CT image generated by the trained machine learning model 120, is provided. The method includes: receiving test data 190, the test data comprising a plurality of CT images 110, wherein for each CT image 110, the test data comprises i) a corresponding value of one or more contrast agent injector parameters 170j used to acquire the CT image 110, and ii) a corresponding surview scan 210; for each of a plurality of the CT images 110 in the test data 190: inputting the CT image 110 into the trained machine learning model 120; generating, from the inputted CT image 110, a corresponding contrast agent enhanced CT image 130, using the trained machine learning model 120; evaluating a quality metric 220 for the contrast agent enhanced CT image 130; extracting, from the corresponding surview scan 210, a value of one or more patient parameters 180! for a patient represented in the surview scan 210; and storing the value of the quality metric 220, the corresponding value of the one or more contrast agent injector parameters 170j, and the corresponding extracted value of one or more patient parameters 180 to provide the reference database 160.

[0106] This example corresponds to the example described above with reference to Fig. 7.

[0107] In another example, a computer-implemented method of providing a reference database 250 relating contrast agent injector parameters 170j and image acquisition parameters 260k to a quality of a contrast agent enhanced CT image 130 provided by the trained machine learning model 120, is provided. The method includes: receiving test data 190, the test data comprising a plurality of CT images 110, wherein for each CT image 110, the test data comprises i) a corresponding value of one or more contrast agent injector parameters 170j used to acquire the CT image 110, and ii) corresponding image acquisition parameters 260k used to acquire the CT image 110; for each of a plurality of the CT images 110 in the test data 190: inputting the CT image 110 into the trained machine learning model 120; generating, from the inputted CT image 110, a corresponding contrast agent enhanced CT image 130, using the trained machine learning model 120; evaluating a quality metric 220 for the contrast agent enhanced CT image

[0108] 130; and storing the value of the quality metric 220, the corresponding value of the one or more contrast agent injector parameters 170j, and the corresponding image acquisition parameters 260k, to provide the reference database 250.

[0109] This example corresponds to the example described above with reference to Fig. 11.

[0110] In another example, a computer-implemented method of providing contrast agent injector parameters 170j for use in acquiring CT data for use with the trained machine learning model 120, is provided. The method includes: receiving a surview scan 210; extracting, from the surview scan 210, a value of one or more patient parameters 180! for a patient represented in the surview scan 210; and obtaining, based on the reference database 160, based on the value of the one or more patient parameters 180!, and based on a specified value of the quality metric 220, a value of one or more contrast agent injector parameters 170j for use in acquiring a CT image of the patient; and outputting the value of one or more contrast agent injector parameters 170j .

[0111] This example corresponds to the example described above with reference to Fig. 9.

[0112] In another example, a computer-implemented method of assessing a suitability of contrast agent injector parameters 170j for acquiring a CT image 130 for use in the trained machine learning model 120, is provided. The method includes: receiving a value of one or more contrast agent injector parameters 170j ; receiving a surview scan 210; extracting, from the surview scan 210, a value of one or more patient parameters 180tfor a patient represented in the surview scan 210; and obtaining, based on the reference database 160, based on the value of the one or more contrast agent injector parameters 170j, and based on the value of the one or more patient parameters 180 a corresponding value of the quality metric 220; and outputting the value of the quality metric 220.

[0113] This example corresponds to the example described above with reference to Fig. 10.

[0114] In another example, a computer-implemented method of providing contrast agent injector parameters 170j for use in acquiring CT data for use in the trained machine learning model 120, is provided. The method includes: receiving a value of one or more image acquisition parameters 260k for acquiring a CT image; and obtaining, based on the reference database 250, based on the value of the one or more image acquisition parameters 260k, and based on a specified value of the quality metric 220, a value of one or more corresponding contrast agent injector parameters 170j for use in acquiring the CT image; and outputting the value of the one or more contrast agent injector parameters 170j .

[0115] This example corresponds to the example described above with reference to Fig. 12.

[0116] In another example, a computer-implemented method of assessing a suitability of contrast agent injector parameters 170j and image acquisition parameters 260k for acquiring a CT image for use in the trained machine learning model 120, is provided. The method includes: receiving a value of the one or more contrast agent injector parameters 170j for acquiring a CT image; receiving a value of one or more image acquisition parameters 260k for acquiring the CT image; and obtaining, based on the reference database 250, based on the value of the one or more contrast agent injector parameters 170j, and based on the value of the one or more image acquisition parameters 260k, a corresponding value of the quality metric 220; and outputting the value of the quality metric 220.

[0117] This example corresponds to the example described above with reference to Fig. 13.

[0118] In another example, a computer-implemented method of providing a reference database 270 relating contrast agent injector parameters 170® used to acquire a spectral CT image 110sto contrast agent injector parameters 170Cj for acquiring a corresponding conventional CT image 110c, is provided. The method includes: receiving test data 190, the test data comprising a plurality of spectral CT images 110s, wherein for each spectral CT image 110s, the test data comprises i) a corresponding value of one or more contrast agent injector parameters 170sused to acquire the spectral CT image 110s, and ii) a corresponding value of one or more contrast agent injector parameters 170Cj for acquiring a conventional CT image 110c; for each of a plurality of the spectral CT images 110sin the test data: generating a corresponding conventional CT image 110scand a corresponding contrast agent enhanced CT image 130SCE; inputting the conventional CT image 110scinto the trained machine learning model 120; generating, from the inputted conventional CT image 110sc, a corresponding contrast agent enhanced CT image 130, using the trained machine learning model 120; evaluating a quality metric 220 for the contrast agent enhanced CT image 130, wherein the quality metric 220 is evaluated based on a difference, A, between the contrast agent enhanced CT image 130 generated using the trained machine learning model 120 and the contrast agent enhanced CT image generated from the spectral CT image 130SCE; and subject to the value of the quality metric 220 exceeding a threshold value, storing the corresponding value of the one or more contrast agent injector parameters 170® used to acquire the spectral CT image 110s, and the corresponding value of one or more contrast agent injector parameters 170Cj for acquiring a conventional CT image 110c, to provide the reference database 270.

[0119] This example corresponds to the example described above with reference to Fig. 14. In this example, the one or more contrast agent injector parameters used to acquire the spectral CT image 170sin the test data 190 may represent a relatively lower total contrast agent dose than the corresponding one or more contrast agent injector parameters 170Cj for acquiring the conventional CT image.

[0120] In another example, a computer-implemented method of providing recommended contrast agent injector parameters 170CLDj for use in acquiring a conventional CT image 110 for use in the trained machine learning model 120, is provided. The method includes: receiving a value of the one or more contrast agent injector parameters 170Cj used to acquire a conventional CT image; obtaining, based on the reference database 270, based on a specified value of the quality metric 220, and using the received value of the one or more contrast agent injector parameters 170Cj used to acquire the conventional CT image as the values of the one or more contrast agent injector parameters 170Cj, a corresponding value of the one or more contrast agent injector parameters 170sused to acquire the spectral CT images; and outputting the value of the one or more contrast agent injector parameters 170sused to acquire the spectral CT images as recommended parameters 170CLDj for use in acquiring the conventional CT image 110 for use in the trained machine learning model.

[0121] This example corresponds to the example described above with reference to Fig. 15.

[0122] In another example, a computer program product is provided. The computer program product comprises instructions which when executed by at least one processor, cause the at least one processor to carry out a method of enhancing visibility of a contrast agent distribution in computed tomography, CT, images. The method includes: receiving S 110 CT data representing a conventional CT image 110 comprising a contrast agent distribution; inputting the CT data into a trained machine learning model 120; and generating S120, from the inputted CT data, one or more corresponding contrast agent enhanced CT images 130 wherein the visibility of the contrast agent distribution is enhanced as compared to the conventional CT image 110, using the trained machine learning model 120; and outputting S 130 the one or more contrast agent enhanced CT images 130; and wherein the machine learning model 120 is trained to generate the one or more contrast agent enhanced CT images 130 using training data and ground truth data, the training data comprising a plurality of conventional CT images 110Tcomprising contrast agent distributions, and the ground truth data comprising for each conventional CT image, one or more corresponding contrast agent enhanced CT images 130GT, and wherein for each conventional CT image 110T, the conventional CT image 110Tand the one or more corresponding contrast agent enhanced CT images 130GTare generated from the same spectral CT image 140T

[0123] This example corresponds to the example described above with reference to Fig. 1 - Fig. 4.

[0124] In another example, a system for enhancing visibility of a contrast agent distribution in computed tomography, CT, images, is provided. The system includes: a memory that stores a plurality of instructions; and at least one processor coupled to the memory and configured to execute the plurality of instructions to: receive SI 10 CT data representing a conventional CT image 110 comprising a contrast agent distribution; input the CT data into a trained machine learning model 120; and generate S120, from the inputted CT data, one or more corresponding contrast agent enhanced CT images 130 wherein the visibility of the contrast agent distribution is enhanced as compared to the conventional CT image 110, using the trained machine learning model 120; and output S130 the one or more contrast agent enhanced CT images 130; and wherein the machine learning model 120 is trained to generate the one or more contrast agent enhanced CT images 130 using training data and ground truth data, the training data comprising a plurality of conventional CT images 110Tcomprising contrast agent distributions, and the ground truth data comprising for each conventional CT image, one or more corresponding contrast agent enhanced CT images 130GT, and wherein for each conventional CT image 110T, the conventional CT image 110Tand the one or more corresponding contrast agent enhanced CT images 130GTare generated from the same spectral CT image 140T

[0125] This example corresponds to the example described above with reference to Fig. 1 - Fig. 4 The above examples are to be understood as illustrative of the present disclosure, and not restrictive. Further examples are also contemplated. For instance, the examples described in relation to a computer-implemented method, may also be provided by the computer program product, or by the computer-readable storage medium, or by the system 300, in a corresponding manner. It is to be understood that a feature described in relation to any one example may be used alone, or in combination with other described features, and may be used in combination with one or more features of another of the examples, or a combination of other examples. Furthermore, equivalents and modifications not described above may also be employed without departing from the scope of the invention, which is defined in the accompanying claims. In the claims, the word “comprising” does not exclude other elements or operations, and the indefinite article “a” or “an” does not exclude a plurality. The mere fact that certain features are recited in mutually different dependent claims does not indicate that a combination of these features cannot be used to advantage. Any reference signs in the claims should not be construed as limiting their scope.

Claims

CLAIMS1. A computer-implemented method of enhancing visibility of a contrast agent in computed tomography (CT) imaging, the method comprising: receiving (SI 10) CT data representing a CT image (110) comprising the contrast agent; generating (S120), from the CT data, one or more corresponding contrast agent enhanced CT images (130), wherein the visibility of the contrast agent is enhanced as compared to the CT image (110) using a trained machine learning model (120); and outputting (S130) the one or more contrast agent enhanced CT images (130), wherein a contrast agent dose used in the received CT data is lower than a contrast agent dose in a corresponding conventional CT image comprising the contrast agent.

2. The computer-implemented method according to claim 1, wherein the machine learning model (120) is trained to generate the one or more contrast agent enhanced CT images (130) using training data and ground truth data, the training data comprising a plurality of conventional CT images (110T) comprising contrast agent distributions, and the ground truth data comprising, for each conventional CT image, one or more corresponding contrast agent enhanced CT images ( 130GT), and wherein for each conventional CT image (110T), the conventional CT image (110T) and the one or more corresponding contrast agent enhanced CT images ( 130GT) are generated from the same spectral CT image (140T).

3. The computer-implemented method according to claim 1 or claim 2, wherein the method further comprises: determining a total contrast agent dose for acquiring the received CT data; and outputting the total contrast agent dose, wherein the total contrast agent dose for acquiring the received CT data is determined based a relationship between a total contrast agent dose and a quality of contrast agent enhancement provided by the machine learning model.

4. The computer-implemented method according to claim 3, wherein the relationship between a total contrast agent dose and a quality of contrast agent enhancement provided by the machine learning model is determined by:receiving test data (190), the test data comprising a plurality of CT images (110), wherein for each CT image (110) the test data comprises a corresponding value of a total contrast agent dose used to acquire the CT image (110); for each of a plurality of the CT images (110) in the test data (190): inputting the CT image (110) into the trained machine learning model (120); generating, from the inputted CT image (110), a corresponding contrast agent enhanced CT image (130), using the trained machine learning model (120); evaluating a quality metric (220) for the contrast agent enhanced CT image (130); and storing the value of the quality metric (220) and the corresponding value of the total contrast agent dose to provide a database representing the relationship.

5. The computer-implemented method according to claim 4, further comprising: determining parameters of a model relating the value of the total contrast agent dose to the value of the quality metric (220) to provide a model representing the relationship.

6. The computer-implemented method according to claim 2, wherein for at least one of the conventional CT images in the training data (110T), at least one of the one or more corresponding contrast agent enhanced CT images ( 130GT) in the ground truth data are generated from the same spectral CT image (140T) by: generating, from the spectral CT image (140T), a first contrast agent enhanced CT image (130’), wherein the contrast agent distribution is enhanced as compared to the conventional CT image (110T); generating, from the spectral CT image (140T), a second CT image (150’), wherein the contrast agent distribution is depleted as compared to the conventional CT image (110T); combining the first contrast agent enhanced CT image (130”) with the second CT image (150’) to provide the corresponding contrast agent enhanced CT image ( 130GT).

7. The computer-implemented method according to claim 1, further comprising: receiving test data (190), the test data comprising a plurality of CT images (110), wherein for each CT image (110), the test data comprises i) a corresponding value of one or more contrast agent injector parameters ( 170j) used to acquire the CT image (110), and ii) a corresponding surview scan (210); for each of a plurality of the CT images (110) in the test data (190): inputting the CT image (110) into the trained machine learning model (120);generating, from the inputted CT image (110), a corresponding contrast agent enhanced CT image (130) using the trained machine learning model (120); evaluating a quality metric (220) for the contrast agent enhanced CT image (130); extracting, from the corresponding surview scan (210), a value of one or more patient parameters (1800 for a patient represented in the surview scan (210); and storing the value of the quality metric (220), the corresponding value of the one or more contrast agent injector parameters ( 170j), and the corresponding extracted value of one or more patient parameters (1800 to provide a reference database (160) relating contrast agent injector parameters ( 170j) and patient parameters (1800 to a quality of the contrast agent enhanced CT image generated by the trained machine learning model (120).

8. The computer-implemented method according to claim 1, further comprising: receiving test data (190), the test data comprising a plurality of CT images (110), wherein for each CT image (110), the test data comprises i) a corresponding value of one or more contrast agent injector parameters ( 170j) used to acquire the CT image (110), and ii) corresponding image acquisition parameters (260k) used to acquire the CT image (110); for each of a plurality of the CT images (110) in the test data (190): inputting the CT image (110) into the trained machine learning model (120); generating, from the inputted CT image (110), a corresponding contrast agent enhanced CT image (130) using the trained machine learning model (120); evaluating a quality metric (220) for the contrast agent enhanced CT image (130); and storing the value of the quality metric (220), the corresponding value of the one or more contrast agent injector parameters ( 1700, and the corresponding image acquisition parameters (260k) to provide a reference database (250) relating contrast agent injector parameters ( 170j) and image acquisition parameters (260k) to a quality of a contrast agent enhanced CT image (130) generated by the trained machine learning model (120).

9. The computer-implemented method according to claim 7 or claim 8, wherein the CT images (110) in the test data (190) are spectral CT images (110s), and wherein the method further comprises: for each of a plurality of the spectral CT images (110s) in the test data (190): generating a corresponding conventional CT image (110sc) and a corresponding contrast agent enhanced CT image (130SCE);wherein the inputting the CT image (110) into the trained machine learning model (120) comprises inputting the conventional CT image (110sc) into the trained machine learning model (120); wherein the generating a corresponding contrast agent enhanced CT image (130) using the trained machine learning model (120) is performed using the inputted conventional CT image (110sc); and wherein the evaluating a quality metric (220) for the contrast agent enhanced CT image (130) is performed based on a difference (A) between the contrast agent enhanced CT image (130) generated using the trained machine learning model (120) and the contrast agent enhanced CT image ( 130SCE) generated from the spectral CT image (110s).

10. The computer-implemented method according to claim 1, further comprising: receiving a surview scan (210); extracting, from the surview scan (210), a value of one or more patient parameters (1800 for a patient represented in the surview scan (210); obtaining, based on the reference database (160), based on the value of the one or more patient parameters (1800, and based on a specified value of the quality metric (220), a value of one or more contrast agent injector parameters ( 170j) for use in acquiring a CT image of the patient; and outputting the value of one or more contrast agent injector parameters ( 170j) to provide contrast agent parameters for use in acquiring CT data for use with the trained machine learning model (120).

11. The computer-implemented method according to claim 1, further comprising: receiving a value of one or more contrast agent injector parameters ( 170j); receiving a surview scan (210); extracting, from the surview scan (210), a value of one or more patient parameters (1800 for a patient represented in the surview scan (210); and obtaining, based on the reference database (160), based on the value of the one or more contrast agent injector parameters ( 170j), and based on the value of the one or more patient parameters (1800, a corresponding value of the quality metric (220); and outputting the value of the quality metric (220) to provide an assessment of a suitability of contrast agent injector parameters ( 170j) for acquiring a CT image (130) for use in the trained machine learning model (120).

12. The computer-implemented method according to claim 1, further comprising: receiving a value of one or more image acquisition parameters (260k) for acquiring aCT image; and obtaining, based on the reference database (250), based on the value of the one or more image acquisition parameters (260k), and based on a specified value of the quality metric (220), a value of one or more corresponding contrast agent injector parameters ( 170j) for use in acquiring the CT image; and outputting the value of the one or more contrast agent injector parameters ( 170j) .

13. The computer-implemented method according to claim 1, further comprising: receiving a value of the one or more contrast agent injector parameters ( 170j) for acquiring a CT image; receiving a value of one or more image acquisition parameters (260k) for acquiring the CT image; and obtaining, based on the reference database (250), based on the value of the one or more contrast agent injector parameters ( 170j), and based on the value of the one or more image acquisition parameters (260k), a corresponding value of the quality metric (220); and outputting the value of the quality metric (220) to provide an assessment of a suitability of the contrast agent injector parameters ( 170j) and the image acquisition parameters (260k) for acquiring a CT image for use in the trained machine learning model (120).

14. The computer-implemented method according to claim 1, further comprising: receiving test data (190), the test data comprising a plurality of spectral CT images(110s), wherein for each spectral CT image (110s), the test data comprises i) a corresponding value of one or more contrast agent injector parameters (170s) used to acquire the spectral CT image (110s), and ii) a corresponding value of one or more contrast agent injector parameters ( 170Cj) for acquiring a conventional CT image (110c); for each of a plurality of the spectral CT images (110s) in the test data: generating a corresponding conventional CT image (110sc) and a corresponding contrast agent enhanced CT image (130SCE); inputting the conventional CT image (110sc) into the trained machine learning model (120);generating, from the inputted conventional CT image (110sc), a corresponding contrast agent enhanced CT image (130), using the trained machine learning model (120); evaluating a quality metric (220) for the contrast agent enhanced CT image (130), wherein the quality metric (220) is evaluated based on a difference (A) between the contrast agent enhanced CT image (130) generated using the trained machine learning model (120) and the contrast agent enhanced CT image generated from the spectral CT image (130SCE); and subject to the value of the quality metric (220) exceeding a threshold value, storing the corresponding value of the one or more contrast agent injector parameters (170s) used to acquire the spectral CT image (110s), and the corresponding value of one or more contrast agent injector parameters ( 170Cj) for acquiring a conventional CT image (110c), to provide a reference database (270) relating contrast agent injector parameters (170s) used to acquire a spectral CT image (110s) to contrast agent injector parameters ( 170Cj) for acquiring a corresponding conventional CT image (110c).

15. The computer-implemented method according to claim 1, further comprising: receiving a value of one or more contrast agent injector parameters ( 170Cj) used to acquire a conventional CT image; obtaining, based on the reference database (270) according to claim 14, based on a specified value of the quality metric (220), and using the received value of the one or more contrast agent injector parameters ( 170Cj) used to acquire the conventional CT image as the values of the one or more contrast agent injector parameters ( 170Cj), a corresponding value of the one or more contrast agent injector parameters (170s) used to acquire the spectral CT images; and outputting the value of the one or more contrast agent injector parameters (170s) used to acquire the spectral CT images as recommended contrast agent injector parameters (170CLDJ) for use in acquiring the conventional CT image (110) for use in the trained machine learning model (120).

16. The computer-implemented method according to any one of claims 7 - 15, wherein the evaluating a quality metric (220) is performed selectively within a region of interest comprising at least a portion of the contrast agent distribution.

17. A computer program product comprising instructions which, when executed by at least one processor, cause the at least one processor to carry out a method of enhancing visibility of a contrast agent in computed tomography (CT) imaging, the method comprising: receiving (SI 10) CT data representing a CT image (110) comprising the contrast agent; generating (S120), from the CT data, one or more corresponding contrast agent enhanced CT images (130), wherein the visibility of the contrast agent is enhanced as compared to the CT image (110) using a trained machine learning model (120); and outputting (S130) the one or more contrast agent enhanced CT images (130), wherein a contrast agent dose used in the received CT data is lower than a contrast agent dose in a corresponding conventional CT image comprising the contrast agent.

18. The computer program product according to claim 17, wherein the machine learning model (120) is trained to generate the one or more contrast agent enhanced CT images (130) using training data and ground truth data, the training data comprising a plurality of conventional CT images (110T) comprising contrast agent distributions, and the ground truth data comprising for each conventional CT image, one or more corresponding contrast agent enhanced CT images ( 130GT), and wherein for each conventional CT image (110T), the conventional CT image (110T) and the one or more corresponding contrast agent enhanced CT images ( 130GT) are generated from the same spectral CT image (140T).

19. A system (300) for enhancing visibility of a contrast agent in computed tomography (CT) imaging, comprising: a memory (310) that stores a plurality of instructions; and at least one processor (320) coupled to the memory and configured to execute the plurality of instructions to: receive (SI 10) CT data representing a CT image (110) comprising the contrast agent; generate (S120), from the CT data, one or more corresponding contrast agent enhanced CT images (130), wherein the visibility of the contrast agent is enhanced as compared to the CT image (110) using a trained machine learning model (120); and output (S130) the one or more contrast agent enhanced CT images (130), wherein a contrast agent dose used in the received CT data is lower than a contrast agent dose in a corresponding conventional CT image comprising the contrast agent.

20. The system according to claim 19, wherein the machine learning model (120) is trained to generate the one or more contrast agent enhanced CT images (130) using training data and ground truth data, the training data comprising a plurality of conventional CT images (110T) comprising contrast agent distributions, and the ground truth data comprising for each conventional CT image, one or more corresponding contrast agent enhanced CT images (130GT), and wherein for each conventional CT image (110T), the conventional CT image (110T) and the one or more corresponding contrast agent enhanced CT images ( 130GT) are generated from the same spectral CT image (140T).

Images