Resolution enhancement and noise reduction in computed tomography using physics-informed artificial intelligence

Abstract

Computed tomography (CT) images obtained using an energy-integrating detector (EID) are enhanced by increasing their spatial resolution, reducing noise, and / or reducing artifacts. The enhanced CT image data are generated by inputting the EID-CT image data to a suitably trained machine learning model, such as a neural network. The machine learning model is trained on training data that includes CT image pairs. Each CT image pair includes a high-resolution photon-counting detector CT (PCD-CT) image and a low-resolution CT image, which may be an EID-CT image, noise information from an EID-CT image, or a degraded PCD-CT image.

Description

RESOLUTION ENHANCEMENT AND NOISE REDUCTION IN COMPUTED TOMOGRAPHY USING PHYSICS-INFORMED ARTIFICIAL INTELLIGENCECROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Patent Application Serial No. 63 / 553,050, filed on February 13, 2024, and entitled “RESOLUTION ENHANCEMENT AND NOISE REDUCTION IN COMPUTED TOMOGRAPHY USING PHYSICS- INFORMED ARTIFICIAL INTELLIGENCE,” and of U.S. Provisional Patent Application Serial No. 63 / 554,774, filed on February 16, 2024, and entitled “RESOLUTION ENHANCEMENT AND NOISE REDUCTION IN COMPUTED TOMOGRAPHY USING PHYSICS-INFORMED ARTIFICIAL INTELLIGENCE,” both of which are herein incorporated by reference in their entirety.STATEMENT OF FEDERALLY SPONSORED RESEARCH

[0001] This invention was made with government support under EB028590 awarded by the National Institutes of Health. The government has certain rights in the invention.BACKGROUND

[0002] Coronary computed tomography angiography (cCTA) is a widely used non- invasive diagnostic exam for patients with suspected coronary7artery7disease (CAD), which is estimated to be present in 49% of adults over age 20. However, most clinical CT scanners are limited in spatial resolution from the use of energy -integrating-detectors (EIDs). Radiological evaluation of CAD is challenging, as coronary arteries are small (3-4 mm diameter) and calcifications within them are highly attenuating, leading to blooming artifacts on CT images. Blooming artifacts cause dense calcifications to appear larger than their actual physical size, which in turn can make the adjacent arterial lumen look smaller, leading to a potential overestimation of disease severity. As such, coronary7artery imaging is a task well suited for high spatial resolution imaging modalities.

[0003] Recently, photon-counting-detector (PCD) CT became commercially available, allowing for ultra-high resolution (UHR) data acquisition, decreased blooming artifact, and increased quantification accuracy of stenosis severity. However. PCD-CTs are very new and costly, restricting widespread accessibility. As of now7and the foreseeable future, most CT scanners are still based on the conventional EID w ith limited spatial resolution.SUMMARY OF THE DISCLOSURE

[0004] The present disclosure addresses the aforementioned drawbacks by providing a method for generating an enhanced computed tomography (CT) image. The method includes accessing energy -integrating detector CT (EID-CT) image data with a computer system, wherein the EID-CT image data have been acquired from a subject using an EID-CT system; accessing a neural network with the computer system, wherein the neural network has been trained on training data to generate enhanced CT image data from CT images acquired with an EID-CT system; inputting the EID-CT image data to the neural network using the computer system, generating enhanced CT image data as an output; and outputting the enhanced CT image data via the computer system.BRIEF DESCRIPTION OF THE DRAWINGS

[0005] FIG. 1 illustrates a schematic of example data preparation with paired low- and high-resolution images, patch extraction, and data augmentation (e.g., 90-degree rotation), training, and testing workflow in accordance with some embodiments described in the present disclosure.

[0006] FIG. 2 illustrates a schematic of an example machine learning model architecture (modified U-NET) used in some embodiments described in the present disclosure, where the input patch size was 128x128 pixels with 128 filters, 2 max pooling and 2 up- convolutional layers.

[0007] FIG. 3 is a flowchart setting forth the steps of an example method for generating enhanced CT image data by inputting CT images acquired with an energy-integrating detector (EID) to a suitably trained machine learning model, such as a neural netw ork.

[0008] FIG. 4 is a flowchart setting forth the steps of an example method for training a machine learning model, such as a neural netw ork, to generate enhanced CT images from EID- CT image data.

[0009] FIG. 5 shows representative images of network input (FIG. 5A), reference (FIG. 5B), and network output (FIG. 5C) demonstrating notable improvement in coronary artery visualization (Arrow'). WW / WL = 2800 / 540 HU. Line profile results across a representative coronary artery' (FIG. 5D, right) with corresponding locations from the input, reference, and network output image (FIG. 5D, left) indicate improved resolution with network output showing a sharper peak more closely matching the reference image.

[0010] FIG. 6 shows results of difference images between reference and input (FIG. 6A) and reference and network output (FIG. 6B). WW / WL = 1000 / 0 (HU).

[0011] FIGS. 7A and 7B show (FIG. 7A) multiplanar reformat (MPR) of the proximal left anterior descending (LAD) in a representative patient with input (left), reconstructed reference image (middle), and ILUMENATE output (right). Quantified percent diameter stenosis values are depicted in the bottom right for each image ty pe and the quantification of stenosis is reported in FIG. 7B.

[0012] FIGS. 8 A and 8B show a schematic of example data preparation and training workflow (FIG. 8A) and testing on EID-CT images (FIG. 8B) in accordance with some embodiments described in the present disclosure.

[0013] FIG. 9 shows representative images of three patient cases (A-C) comparing original energy-integrating detector CT images (left) with ILUMENATE images (right) demonstrating blooming artifact reduction and improved image sharpness using ILUMENATE with axial (top row) and coronal multi-planar reformats (MPRs) (bottom row). (WW / WL = 1500 / 250 HU).

[0014] FIG. 10 shows representative images of a circumflex artery with dense calcification on energy -integrating detector input image (left), ILUMENATE output (middle), and corresponding line profiles (right).

[0015] FIG. 11 shows results of percent diameter stenosis quantification for 36 lesions comparing energy -integrating detector (EID) input (blue) with ILUMENATE output (orange) with lines connecting individual lesions (A). Box and whisker plot of results of percent diameter stenosis measured with EID input (blue) and ILUMENATE output (orange) (B).

[0016] FIG. 12 shows violin plots of both readers' scores for overall image quality (A) (l=worst, 5=best). sharpness (B) (l=worst. 5=best), and image noise (C) (l=worst, 4=best), comparing energy -integrating detector input (blue) with ILUMENATE output (orange).

[0017] FIG. 13 is a block diagram of an example CT image enhancement system.

[0018] FIG. 14 is a block diagram of example components that can implement the system of FIG. 13.

[0019] FIGS. 15A and 15B illustrate an example CT system that can be implemented in accordance with some embodiments described in the present disclosure.DETAILED DESCRIPTION

[0020] Described here are systems and methods for enhancing computed tomography (CT) images that have been obtained using an energy-integrating detector (EID). In some instances, the CT images may be coronary CT angiography (cCTA) images or other CT images of a subject’s heart. In some other instances, the CT images may depict other vessels. In still other examples, the CT images may depict bony structures or may be acquired for other high- resolution bone imaging tasks. The CT images may be enhanced by increasing their spatial resolution, reducing noise in the images, and / or reducing the presence of artifacts in the images. In general, a machine learning model is trained on training data to generate these enhanced images. The machine learning model is trained, for example, using pairs of images: one high- resolution photon-counting detector (PCD) image as the target and one corresponding low- resolution image as the input. In some cases, the low-resolution image may be constructed using noise from a low-resolution EID-CT image overlaid on low-resolution PCD-CT image patches. Once trained, the machine learning model receives a low-resolution image obtained from a conventional EID as an input, and outputs an enhanced image, which may include a denoised image and / or a higher spatial resolution with reduced blooming artifact, such as may be comparable to that from a PCD. In some implementations, both training and inference are performed in the image domain, such that the disclosed systems and methods can be vendoragnostic. By enhancing images acquired with conventional EIDs, the disclosed systems and methods provide an improvement to currently existing CT systems, which may not have higher resolution PCDs available to them.

[0021] Advantageously, the disclosed systems and methods are capable of significantly improving the resolution of EID-CT images with use of a deep convolutional neural netw ork (CNN), which can allow for more accurate stenosis assessments. As noted above, the CNN is trained using high spatial resolution PCD-CT images in addition to lower resolution EID-CT images. The disclosed systems and methods can also work directly on the EID-CT images without the need for raw data or proprietary information from the vendor.

[0022] It is an advantage of the disclosed systems and methods to improve access to high-spatial resolution cardiac imaging. This can be accomplished by transforming conventional CT images to higher spatial resolution images, which have comparable spatial resolution to that of high-end PCD-CT scanner technology, which is not widely available in the clinical setting.

[0023] In some implementations, the CNN may be referred to as an ILUMENATE (Improved LUMEN visualization through Artificial super-resoluTion imagEs) network or model, to improve visualization of coronary arteries from low spatial resolution (LR) CT images by creating a high spatial resolution (HR) image simulating the resolution of PCD-CT. An example of the workflow is show n in FIG. 1. In the example illustrated in FIG. 1, training input (LR) CT images were reconstructed from patient images acquired using PCD-CT, degrading the resolution with a smoother reconstruction kernel to make it the same as what is used clinically with EID-CT. The network uses a modified U-Net architecture (FIG. 2) and learned the difference between these images and true high-resolution PCD-CT images reconstructed with a dedicated sharp kernel available with PCD-CT. The network was trained on PCD-CT images and applied to unseen EID-CT cases during inference.

[0024] In some implementations, physics knowledge and prior information can be built into the neural network to enhance performance. For instance, the modulation transfer function (MTF) is a physical metric in CT used to describe the spatial resolution, or “resolvability” in a given CT image dataset. A higher number of line pairs per centimeter, (Ip / cm) indicates higher spatial resolution. The disclosed systems and methods can use simulated EID-CT images by reconstructing PCD-CT data with a smooth kernel (e.g., Bv40 with a 10% MTF at 6.61 Ip / cm, which matches EID-CT) and a high resolution dataset (e.g., Bv72 with 10% MTF at 17.4 Ip / cm), from which the disclosed machine learning model(s) can leam the physical difference in spatial resolution to be applied to unseen CT images with low resolution. Additionally or alternatively, the CNN may include an attention mechanism to guide netw ork focusing on main targets. These may include, but are not limited to, masking out anatomies outside of a target anatomical region (e.g., the heart) and including an attention gate in the network architecture.

[0025] Referring now to FIG. 3. a flowchart is illustrated as setting forth the steps of an example method for generating classified feature data using a suitably trained neural network or other machine learning algorithm. As will be described, the neural netw ork or other machine learning algorithm takes CT images acquired with an energy-integrating detector as input data and generates enhanced CT image data as output data. As an example, the enhanced CT image data can include CT images that have been denoised, have had their spatial resolution increased, have had artifacts in the images reduced, or combinations thereof.

[0026] The method includes accessing EID-CT image data with a computer system, as indicated at step 302. Accessing the EID-CT image data may include retrieving such data from a memory or other suitable data storage device or medium. Additionally or alternatively.accessing the EID-CT image data may include acquiring such data with a CT system having EIDs and transferring or otherwise communicating the data to the computer system, which may be a part of the CT system. As mentioned above, the EID-CT image data generally includes CT images acquired using energy -integrating detectors.

[0027] A trained neural network (or other suitable machine learning algorithm) is then accessed with the computer system, as indicated at step 304. In general, the neural network is trained, or has been trained, on training data in order to generate enhanced CT image data from CT images acquired using energy -integrating detectors (e.g., EID-CT image data).

[0028] Accessing the trained neural network may include accessing network parameters (e.g., weights, biases, or both) that have been optimized or otherwise estimated by training the neural network on training data. In some instances, retrieving the neural network can also include retrieving, constructing, or otherwise accessing the particular neural network architecture to be implemented. For instance, data pertaining to the layers in the neural network architecture (e.g., number of layers, type of layers, ordering of layers, connections between layers, hyperparameters for layers) may be retrieved, selected, constructed, or otherwise accessed.

[0029] An artificial neural network generally includes an input layer, one or more hidden layers (or nodes), and an output layer. Ty pically, the input layer includes as many nodes as inputs provided to the artificial neural network. The number (and the type) of inputs provided to the artificial neural network may vary based on the particular task for the artificial neural network.

[0030] The input layer connects to one or more hidden layers. The number of hidden layers varies and may depend on the particular task for the artificial neural network. Additionally, each hidden layer may have a different number of nodes and may be connected to the next layer differently. For example, each node of the input layer may be connected to each node of the first hidden layer. The connection between each node of the input layer and each node of the first hidden layer may be assigned a weight parameter. Additionally, each node of the neural network may also be assigned a bias value. In some configurations, each node of the first hidden layer may not be connected to each node of the second hidden layer. That is, there may be some nodes of the first hidden layer that are not connected to all of the nodes of the second hidden layer. The connections between the nodes of the first hidden layers and the second hidden layers are each assigned different weight parameters. Each node of the hidden layer is generally associated with an activation function. The activation function defineshow the hidden layer is to process the input received from the input layer or from a previous input or hidden layer. These activation functions may vary and be based on the type of task associated with the artificial neural network and also on the specific type of hidden layer implemented.

[0031] Each hidden layer may perform a different function. For example, some hidden layers can be convolutional hidden layers which can. in some instances, reduce the dimensionality of the inputs. Other hidden layers can perform statistical functions such as max pooling, which may reduce a group of inputs to the maximum value; an averaging layer; batch normalization; and other such functions. In some of the hidden layers each node is connected to each node of the next hidden layer, which may be referred to then as dense layers. Some neural networks including more than, for example, three hidden layers may be considered deep neural networks.

[0032] The last hidden layer in the artificial neural network is connected to the output layer. Similar to the input layer, the output layer typically has the same number of nodes as the possible outputs.

[0033] The EID-CT image data are then input to the trained neural network, generating output as enhanced CT image data, as indicated at step 306. As one example, the enhanced CT image data may include higher resolution CT images. As another example, the enhanced CT image data may include denoised CT images. As yet another example, the enhanced CT image data may include CT images in which artifacts (e.g.. blooming artifacts) have been reduced. In still other instances, the enhanced CT image data may include CT images with a combination of increased spatial resolution, reduced noise, and / or reduced artifacts.

[0034] The enhanced CT image data generated by inputting the EID-CT image data to the trained neural network(s) can then be displayed to a user, stored for later use or further processing, or both, as indicated at step 308.

[0035] Referring now to FIG. 4, a flowchart is illustrated as setting forth the steps of an example method for training one or more neural networks (or other suitable machine learning algorithms) on training data, such that the one or more neural networks are trained to receive EID-CT image data (e.g., CT images acquired using energy -integrating detectors) as input data in order to generate enhanced CT image data as output data, where the enhanced CT image data include CT images with increased spatial resolution, reduced noise, and / or reduced image artifacts.

[0036] In general, the neural network(s) can implement any number of different neural network architectures. For instance, the neural network(s) could implement a convolutional neural network, a residual neural network, or the like. Alternatively, the neural network(s) could be replaced with other suitable machine learning or artificial intelligence algorithms, such as those based on supervised learning, unsupervised learning, deep learning, ensemble learning, dimensionality reduction, and so on.

[0037] The method includes accessing training data with a computer system, as indicated at step 402. Accessing the training data may include retrieving such data from a memory or other suitable data storage device or medium. Alternatively, accessing the training data may include acquiring such data with one or more CT systems and transferring or otherwise communicating the data to the computer system.

[0038] In general, the training data can include pairs of CT images. As one example, each pair of CT images can include a CT image acquired with a photon-counting detector and reconstructed with a dedicated high-resolution kernel as the target (or label) in the CT image pair and also reconstructed with a low-resolution kernel matching the resolution of EID-CT (or input) in the CT image pair. Thus, in some implementations, the image pair can come from the same PCD-CT (or EID-CT) system with two different resolutions. Additionally or alternatively, EID-CT image data may be included in training. In such instances, the high- resolution PCD-CT image can be selected as the target (or label) in the CT image pair and the corresponding low-resolution EID-CT image can be selected as the input in the CT image pair. Additionally or alternatively, the low-resolution image used in the training pair may be noise information that is spliced from a low-resolution EID-CT image. The noise information may be overlaid on a low-resolution PCD-CT image in some instances.

[0039] The method can include assembling training data from the CT image pairs using a computer system. This step may include assembling the CT image pairs into an appropriate data structure on which the neural network or other machine learning algorithm can be trained. Assembling the training data may include assembling CT image pairs and other relevant data. In some implementations, assembling the training data may include augmenting the CT image pairs using one or more data augmentation techniques to enhance the diversity and size of the training dataset. As one example, augmenting the CT image pairs may include applying various transformations to the images, such as rotations, flips, zooms, resizing, and patch extraction. For instance, each image in the CT image pair may be rotated by one or more rotation angles(e.g., 90 degrees). By introducing these variations, the augmented dataset becomes more robust, helping the model generalize better to different scenarios and reducing overfitting.

[0040] As another example, augmenting the training data may include adding noise to the PCD-CT image in the CT image pairs. For instance, EID noise can be added to PCD training data to provide target noise texture and enhance network performance. Noise could be obtained from phantom scans independent of patient images.

[0041] One or more neural networks (or other suitable machine learning algorithms) are trained on the training data, as indicated at step 404. In general, the neural network can be trained by optimizing network parameters (e.g., weights, biases, or both) based on minimizing a loss function. As one non-limiting example, the loss function may be a mean squared error loss function.

[0042] Training a neural network may include initializing the neural network, such as by computing, estimating, or otherwise selecting initial network parameters (e.g., weights, biases, or both). During training, an artificial neural network receives the inputs for a training example and generates an output using the bias for each node, and the connections between each node and the corresponding weights. For instance, training data can be input to the initialized neural network, generating output as enhanced CT image data. The artificial neural network then compares the generated output with the actual output of the training example in order to evaluate the quality of the enhanced CT image data. For instance, the enhanced CT image data can be passed to a loss function to compute an error. The current neural network can then be updated based on the calculated error (e.g., using backpropagation methods based on the calculated error). For instance, the current neural network can be updated by updating the network parameters (e.g., weights, biases, or both) in order to minimize the loss according to the loss function. In some implementations, a customized loss function that includes both data fidelity and feature loss can be used.

[0043] The training continues until a training condition is met. The training condition may correspond to, for example, a predetermined number of training examples being used, a minimum accuracy threshold being reached during training and validation, a predetermined number of validation iterations being completed, and the like. When the training condition has been met (e.g., by determining whether an error threshold or other stopping criterion has been satisfied), the current neural network and its associated network parameters represent the trained neural network. Different types of training processes can be used to adjust the bias values and the weights of the node connections based on the training examples. The trainingprocesses may include, for example, gradient descent, Newton's method, conjugate gradient, quasi-Newton, Levenberg-Marquardt, among others.

[0044] The artificial neural network can be constructed or otherwise trained based on training data using one or more different learning techniques, such as supervised learning, unsupervised learning, reinforcement learning, ensemble learning, active learning, transfer learning, or other suitable learning techniques for neural networks. As an example, supervised learning involves presenting a computer system with example inputs and their actual outputs (e.g., categorizations). In these instances, the artificial neural network is configured to leam a general rule or model that maps the inputs to the outputs based on the provided example inputoutput pairs.

[0045] The one or more trained neural networks are then stored for later use. as indicated at step 406. Storing the neural network(s) may include storing network parameters (e.g., weights, biases, or both), which have been computed or otherwise estimated by training the neural network(s) on the training data. Storing the trained neural network(s) may also include storing the particular neural network architecture to be implemented. For instance, data pertaining to the layers in the neural network architecture (e.g., number of layers, type of layers, ordering of layers, connections between layers, hyperparameters for layers) may be stored.

[0046] In an example study, the systems and methods described in the present disclosure were used to enhance coronary CT angiography (cCTA) images.

[0047] Study participants underwent cCTA exams performed on a clinical PCD-CT scanner using the UHR mode with 120 x 0.2 mm detector collimation and a retrospective electrocardiogram (ECG) gated spiral acquisition mode. Images for each of the patients were reconstructed to yield 2 datasets: one low resolution (LR) and one high resolution (HR). All images were reconstructed using a quantum iterative noise reduction (QIR) algorithm with strength of 4, a 200 mm field of view (FOV), 1024 matrix size, 0.6 mm slice thickness, and 0.3 mm increment. LR images were reconstructed using a Bv40 kernel (10% MTF at 6.61 Ip / cm) which is the routine kernel utilized for clinical cCTA on EID-CT. Corresponding HR images were reconstructed using a Bv72 kernel (10% MTF at 17.4 Ip / cm) to maximize the benefit of PCD-CT UHR acquisition mode. Reconstruction locations of the LR and HR were identically matched.

[0048] Patient images were separated into training and validation (8 patients), and testing (5 patients) data sets. To prepare the data for training and validation, 34,400 patches of size 128 x 128 pixels were randomly extracted from LR and HR training and validation patientdata. Patches were matched in location across LR and HR images. Next, corresponding LR and HR patches were rotated by 90 degrees for data augmentation to increase the input for a total of 68,800 LR and HR patch pairs (90% for training (61,920), 10% reserved for validation (6,880)). LR patches served as the input, and HR patches served as the label. Five patient cases were reserved for testing, with inputs reconstructed following the LR reconstruction parameters, and corresponding reference images following reconstruction parameters of the HR labels (Figure 1A).

[0049] The ILUMENATE network employed a modified U-Net architecture. The model included 2 max pooling and 2 up-convolutional layers (128 filters at the first layer, doubling at each max pooling layer), rectified linear unit (ReLU) activation function, and Adam optimizer with a learning rate of 0.001 (FIG. 2). Mean-squared-error (MSE) served as the loss function, defined as:

[0050] where n is the number of paired data points, Y, is the HR label, and Xtis the prediction for the ith training example. Training was performed using a GPU (Nvidia Titan RTX) with TensorFlow and Keras.

[0051] Model performance was evaluated both qualitatively and quantitatively. Overall image quality improvement was assessed through visual inspection. Spatial resolution was assessed via line profiles draw n through the left anterior descending coronary artery for a test patient case comparing network input, reconstructed PCD reference, and model output images. Region-of-interest (ROI) measurements were made in three locations: bone, soft tissue, and a contrast-enhance vessel to assess CT number stability and image noise of the test cases.

[0052] Structural similarity index (SSIM) between the reference images and network output was computed for all slices and averaged for the test cases. Difference images were generated for the input and reference, and the network output and reference. To assess the impact of resolution on stenosis quantification, coronary artery stenoses with dense calcification were identified from test cases and processed using syngio.via CT (VB70). an investigational version of commercial softw are, to quantify the percent diameter stenosis and compared among input, reference, and netw ork output images.

[0053] Visual inspection demonstrated that ILUMENATE improved overall image quality of the input low resolution images, with sharper edges (e.g., coronaries and calcifications) and better delineation of anatomy, closely matching the reference highresolution images, as shown in FIGS. 5A-5C. ILUMENATE maintained stable CT numbers with less than 4% difference (0.2% in contrast enhanced vessel, 3.3% in bone, and 1.1% in soft tissue) with respect to reference image ROIs. Interestingly, a denoising effect was observed with respect to the reference image, with the output image having an average of 28% lower image noise (68 HU vs 94 HU).

[0054] Line profiles were drawn across a small calcification in the left anterior descending artery in the same location on the image for the input, reference, and network output, as shown in FIG. 5D. The network output demonstrated sharper edges than the input across the calcium and soft tissue boundary, more closely matching the reference image.

[0055] Difference images between the reference and input (FIG. 6A) and network output and reference image (FIG. 6B) show increased overall image similarity between network output and reference. Mean SSIM results from the test datasets between network output and reference image was 0.70.

[0056] Of the 5 test cases. 5 calcified stenosis were identified. ILUMENATE output reduced the percent diameter stenosis in each lesion measured as compared to the network input. Overall, the network input measured the largest percent diameter stenosis, with an average percent of 7% higher than the reference and network output.

[0057] FIGS. 7A and 7B show multiplanar reformat (MPR) of the proximal left anterior descending (LAD) artery in a representative patient with input (left), reconstructed reference image (middle), and ILUMENATE output (right). Quantified percent diameter stenosis values are depicted in the bottom right for each image type (FIG. 7 A) and the quantification of stenosis is reported in FIG. 7B.

[0058] ILUMENATE improved overall image quality and produced virtually reconstructed HR PCD-CT images from unseen LR images. ILUMENATE output images preserved CT number accuracy when compared to reference images, allowed for increased coronary artery lumen visualization, preserved structural similarity with respect to reference images, and decreased the percent diameter stenosis quantified.

[0059] The proposed method improved visualization of cardiac anatomy and coronary artery lumen patency in the presence of calcifications, potentially sparing patients from unnecessary interventions. Future work for this project aims to implement ILUMENATE on true EID-CT images to further increase widespread accessibility to UHR coronary artery CT images.

[0060] When applied to unseen LR patient images, ILUMENATE produced images closely matching UHR PCD-CT references, improving the images qualitatively and quantitatively. These improvements decreased blooming artifact and reduced percent diameter stenosis for each of the lesions assessed in this study, potentially changing CAD patient management.

[0061] In another example study, the systems and methods described in the present disclosure were used to enhance cCTA images. In this example, the trained CNN was trained with noise texture information from EID-CT images, and the deployed CNN was observed to substantially improve image sharpness and visualization of coronary artery7lumens of EID-CT, bringing the resolution closer to ultra-high-resolution (UHR) PCD-CT. Percent diameter stenosis of coronary arteries was significantly reduced by an average of 4.42% with the CNN compared to EID-CT input images (p < 0.001). In a blinded, randomized reader study, readers preferred CNN images in all 22 testing cases. Both readers scored the CNN images significantly superior (p < 0.05) for each category7. Reader 1 scored image quality 3.54 / 3.95, sharpness 2.91 / 3.95, and noise 3.05 / 3.91. and reader 2 scored image quality 3.86 / 4.73, sharpness 3.14 / 4.91, and noise 3.36 / 4 for original EID-CT / CNN images, respectively.

[0062] In the example study, thirty7patients undergoing clinically indicated cCTA were scanned with EID-CT and subsequently with UHR PCD-CT on the same day. EID-CT data were acquired using a routine clinical protocol that selected the scan mode as ECG prospectively -tn ggered high pitch FLASH or sequential mode, or a retrospectively gated spiral mode based on patient’s heart rhythm. Automatic tube potential selection was set to a reference of 120 kV and slider bar at 8 (optimized for soft tissue with contrast). Automatic exposure control was set to a quality reference tube-current-time product of 120 mAs / rotation. A 0.25 s rotation time and 192 x 0.6 mm detector collimation (with flying focal spot, physical collimation of 96 x 0.6 mm) were used. All PCD-CT data were acquired with UHR mode using a retrospective ECG-gated spiral acquisition mode w ith 0.25 s rotation time. Fourteen of the 30 patients w ere scanned using a UHR only (120 x 0.2 mm detector collimation) mode, with task-based automatic keV selection used with settings of 120 kV and 50 IQ level for patients > 90 kg, and 90 kV and 100 IQ level for patients < 90 kg, optimized for vascular tasks. The remaining 16 patients were acquired using a UHR and multi-energy7(UHR-ME) mode with 96 x 0.2 mm detector collimation and IQ level set to 50. For this study, only the UHR component of the data was used, and no spectral information was evaluated. As such, polychromatic (T3D) images were reconstructed for the 16 UHR-ME patients for fair comparison to UHR onlypatients. Exclusion criteria included pregnancy, inability to provide written informed consent, renal contraindication (eGFR<60), or reaction to medication administered in the clinical EID- CT exam. A total of thirty patients with dense coronary' artery calcifications were included in this study.

[0063] Eight patients scanned on PCD-CT were allocated to the training dataset. Each patient was reconstructed using a low resolution (LR) smooth kernel of Qr40 (10% MTF at 6.60 Ip / cm) matching the kernel used in the clinical practice for cCTA using EID-CT systems, and a 0.2 mm slice thickness with 0.1 mm increment. Adjacent slices were averaged to simulate the thicker slices used in the clinical practice for EID-CT. Additionally, input images were reconstructed with quantum iterative reconstruction (QIR) strength 4, 1024 matrix size, and 200 mm field of view (FOV). A corresponding high resolution (HR) dataset was reconstructed with a sharp kernel of Qr72 (10% MTF at 20.42 Ip / cm) with 0.2 mm slice thickness and 0.1 mm increment, QIR strength 4, 1024 matrix size, and a 200 mm FOV.

[0064] The ILUMINATE network was trained to leam in- and through- plane resolution improvements by having the thick slice LR input correspond to the central thin slice HR training label. Testing data included 22 patients scanned on EID-CT and reconstructed with Qr40 kernel, 0.6 mm slice thickness and 0.3 mm increment, ADMIRE strength 5, 1024 matrix size, and 200 mm FOV.

[0065] Patches of size 128 x 128 pixels were extracted from the training dataset in random but corresponding locations across LR and HR PCD-CT data, forming 33,945 patch pairs. Data augmentation included 90-degree rotation of patch pairs, leading to a total of 67,890 paired patches (90% fortraining (61,101), 10% reserved for validation (6,789)). To address the discrepancy in noise texture between EID-CT and PCD-CT scanners, patches of 128 x 128 pixels were extracted from images of a 30-cm water tank scanned on EID-CT while matching acquisition and reconstmction parameters of the testing dataset. The mean CT number was subtracted, and noise was added to the LR input patches to resolve the train-test mismatch (FIG. 8A). LR patches with added EID-CT noise served as netw ork inputs with corresponding HR patches as labels. Twenty-two patients scanned on EID-CT were reserved for testing (FIG. 8B).

[0066] As described above, the ILUMENATE model employed a modified U-Net architecture with 2 max pooling and 2 up-convolutional layers (128 filters at the first layer doubling at each max pooling layer), and rectified linear unit (ReLU) activation function, and Adam optimizer with a learning rate of 0.001 (FIG. 2). MSE served as the loss function tominimize the magnitude of the difference between the label and prediction defined above. Training included of 100 epochs and was performed using a GPU (Nvidia Titan RTX) with TensorFlow version 2. 10. 1 and Keras version 2. 10.0.

[0067] Spatial resolution was evaluated using line profiles and percent diameter stenosis quantified. Two experienced radiologists, blinded to image type, selected preferred series and scored images for overall quality, sharpness, and noise comparing original EID-CT and ILUMENATE output. Visual inspection demonstrated that ILUMENATE improved the overall image qual i ty of EID-CT images and reduced blooming artifact in dense calcifications (FIG. 10). ILUMENATE images demonstrated superior spatial resolution with line profiles drawn through the dense calcification showing a sharper peak for calcium and trough for iodinated lumen (FIG. 11). ILUMENATE maintained stable CT numbers with an average of 3.68% difference (0.89% in aorta, 6.71% in bone, and 3.43% in soft tissue) with respect to EID-CT ROI’s. Of the 22 patients used in testing, 36 stenotic lesions were identified. The mean percent diameter stenosis for EID-CT lesions was 33.94% (std: 15.37%, range 4% - 64%) which reduced to 29.53% (std: 15.28%, range 4% - 63%)) with ILUMENATE. Overall, 27 of the 36 lesions decreased in percent diameter stenosis with ILUMENATE, with 5 lesions remaining the same, and 4 lesions increasing. The percent diameter stenosis was significantly reduced by an average of 4.42% + / - 4.82% with ILUMENATE compared to EID-CT images (p<0.001) (FIG. 12). Among these, 9 lesions decreased in stenosis grade with ILUMENATE compared to EID-CT. Of these, one lesion decreased from moderate to mild. 8 from mild to minimal, and one lesion increased from minimal to mild with ILUMENATE.

[0068] Both readers preferred ILUMENATE in 22 of 22 cases presented. The average score for the original EID-CT / ILUMENATE images respectively for reader 1 was overall image quality: 3.54 / 3.95 (SD = 0.50 / 0.21, p=0.0021), sharpness: 2.91 / 3.95 (SD = 0.42 / 0.64 p<0.001), and noise: 3.05 / 3.91 (SD = 0.21 / 0.29, p<0.001); and for reader 2 overall image quality: 3.86 / 4.73 (SD = 0.34 / 0.54, pO.001), sharpness: 3.14 / 4.91 (SD = 0.46 / 0.29, p<0.001), and noise: 3.36 / 4 (SD = 0.48 / 0, p<0.001) (FIG. 12).

[0069] In this study, a supervised deep learning framework was developed and trained using UHR PCD-CT data and inferenced on EID-CT images for improved visualization of coronary arterial lumen in the presence of dense calcified stenotic lesions. ILUMENATE demonstrated CT number stability and improved spatial resolution of EID-CT images, shown by the line profile with sharper peaks and lumen boundaries. The percent diameter stenosis reduction was small (4.42%) but significant with ILUMENATE compared to EID-CT. Ofthese, 9 of the 36 lesions decreased in stenosis severity score and one lesion increased with use of ILUMENATE. ILUMENATE images were preferred by both blinded readers in every testing case presented. ILUMENATE images were scored significantly favorably by each reader for image quality, sharpness, and noise.

[0070] Advantageously, the disclosed systems and methods are not vendor-specific, meaning that access to proprietary projection space data is no needed. Rather, the disclosed systems and methods use image data and circumvent the need for sinogram information.

[0071] The proposed super-resolution technique takes advantage of the ultra-high- resolution of PCD-CT, with training data from patients scanned on PCD-CT and LR input reconstructed using smooth kernels matching the resolution of clinical EID-CT systems, and HR labels reconstructed with a sharp kernel. As both inputs and labels are created from the same exam performed on the same scanner (with the only difference being the resolution), there is no need for image registration. To address the domain shift required for deploying images to a different system than used in training, noise texture (e.g., from a water tank scanned on EID-CT with matched parameters as the testing data) may be incorporated. This way, to translate the model to different EID-CT systems (e.g., different models or manufacturers), there is no need to collect new' patient data as many other algorithms require. Instead, only a single scan of uniform water tank, a phantom, or the like is needed to obtain noise texture. This increases the potential of the discloses systems and methods to be used in a wide range of scanners by eliminating the labor-intensive collection of a large volume of patient data.

[0072] The example study demonstrated the advantages of the disclosed, dedicated super resolution deep learning-based method for UHR EID-CT cCTA imaging, informed by clinical UHR PCD-CT. The neural network was trained using commercially available image data, avoiding vendor proprietary information. The deployed network demonstrated improved quantitative and qualitative performance in a blinded reader study, enhancing the resolution of EID-CT images. This reduction in blooming artifact and improved image quality' allows for improved noninvasive assessment of CAD, potentially sparing patients from unnecessary interventions. The proposed technique has the potential to bring the benefits of advanced CT technologies (i.e., PCD-CT) to the current installed base of EID-CT systems.

[0073] FIG. 13, shows an example of a system 1300 for generating enhanced CT image data in accordance with some embodiments of the systems and methods described in the present disclosure. As shown in FIG. 13, a computing device 1350 can receive one or more types of data (e.g., EID-CT image data) from data source 1302. In some embodiments.computing device 1350 can execute at least a portion of a CT image enhancement system 1304 to enhance CT images acquired with energy -integrating detectors received from the data source 1302.

[0074] Additionally or alternatively, in some embodiments, the computing device 1350 can communicate information about data received from the data source 1302 to a server 1352 over a communication network 1354. which can execute at least a portion of the CT image enhancement system 1304. In such embodiments, the server 1352 can return information to the computing device 1350 (and / or any other suitable computing device) indicative of an output of the CT image enhancement system 1304.

[0075] In some embodiments, computing device 1350 and / or server 1352 can be any suitable computing device or combination of devices, such as a desktop computer, a laptop computer, a smartphone, a tablet computer, a wearable computer, a server computer, a virtual machine being executed by a physical computing device, and so on. The computing device 1350 and / or server 1352 can also reconstruct images from the data.

[0076] In some embodiments, data source 1302 can be any suitable source of data (e.g., measurement data, images reconstructed from measurement data, processed image data), such as a CT system, another computing device (e.g., a server storing measurement data, images reconstructed from measurement data, processed image data), and so on. In some embodiments, data source 1302 can be local to computing device 1350. For example, data source 1302 can be incorporated with computing device 1350 (e.g.. computing device 1350 can be configured as part of a device for measuring, recording, estimating, acquiring, or otherwise collecting or storing data). As another example, data source 1302 can be connected to computing device 1350 by a cable, a direct wireless link, and so on. Additionally or alternatively, in some embodiments, data source 1302 can be located locally and / or remotely from computing device 1350, and can communicate data to computing device 1350 (and / or server 1352) via a communication network (e.g., communication network 1354).

[0077] In some embodiments, communication network 1354 can be any suitable communication network or combination of communication networks. For example, communication network 1354 can include a Wi-Fi network (which can include one or more wireless routers, one or more switches, etc.), a peer-to-peer network (e.g., a Bluetooth network), a cellular network (e.g., a 3G network, a 4G network, etc., complying with any suitable standard, such as CDMA, GSM, LTE, LTE Advanced. WiMAX, etc.), other types of wireless network, a wired network, and so on. In some embodiments, communication network1354 can be a local area network, a wide area network, a public netw ork (e.g., the Internet), a private or semi -private network (e.g., a corporate or university intranet), any other suitable type of network, or any suitable combination of networks. Communications links shown in FIG. 13 can each be any suitable communications link or combination of communications links, such as wired links, fiber optic links, Wi-Fi links, Bluetooth links, cellular links, and so on.

[0078] Referring now to FIG. 14. an example of hardware 1400 that can be used to implement data source 1302, computing device 1350, and server 1352 in accordance with some embodiments of the systems and methods described in the present disclosure is shown.

[0079] As show n in FIG. 14, in some embodiments, computing device 1350 can include a processor 1402, a display 1404, one or more inputs 1406, one or more communication systems 1408, and / or memory 1410. In some embodiments, processor 1402 can be any suitable hardware processor or combination of processors, such as a central processing unit (“CPU’’), a graphics processing unit (“GPU”), and so on. In some embodiments, display 1404 can include any suitable display devices, such as a liquid crystal display (“LCD”) screen, a light-emitting diode (“LED”) display, an organic LED (“OLED”) display, an electrophoretic display (e.g., an “e-ink” display), a computer monitor, a touchscreen, a television, and so on. In some embodiments, inputs 1406 can include any suitable input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, and so on.

[0080] In some embodiments, communications systems 1408 can include any suitable hardware, firmware, and / or software for communicating information over communication network 1354 and / or any other suitable communication networks. For example, communications systems 1408 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 1408 can include hardware, firmware, and / or software that can be used to establish a Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0081] In some embodiments, memory’ 1410 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 1402 to present content using display71404, to communicate with server 1352 via communications system(s) 1408, and so on. Memory 1410 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 1410 can include random-access memory (“RAM”), read-only memory(“ROM”), electrically programmable ROM (“EPROM”), electrically erasable ROM (“EEPROM”), other forms of volatile memory, other forms of non-volatile memory, one or more forms of semi-volatile memory', one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory' 1410 can have encoded thereon, or otherwise stored therein, a computer program for controlling operation of computing device 1350. In such embodiments, processor 1402 can execute at least a portion of the computer program to present content (e.g., images, user interfaces, graphics, tables), receive content from server 1352, transmit information to server 1352, and so on. For example, the processor 1402 and the memory 1410 can be configured to perform the methods described herein (e.g., the method of FIG. 3, the method of FIG. 4).

[0082] In some embodiments, server 1352 can include a processor 1412, a display 1414, one or more inputs 1416, one or more communications systems 1418, and / or memory 1420. In some embodiments, processor 1412 can be any suitable hardware processor or combination of processors, such as a CPU, a GPU, and so on. In some embodiments, display 1414 can include any suitable display devices, such as an LCD screen, LED display. OLED display, electrophoretic display, a computer monitor, a touchscreen, a television, and so on. In some embodiments, inputs 1416 can include any suitable input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, and so on.

[0083] In some embodiments, communications systems 1418 can include any suitable hardware, firmware, and / or software for communicating information over communication network 1354 and / or any other suitable communication networks. For example, communications systems 1418 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a more particular example, communications systems 1418 can include hardware, firmware, and / or software that can be used to establish a Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0084] In some embodiments, memory’ 1420 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 1412 to present content using display 1414, to communicate with one or more computing devices 1350, and so on. Memory71420 can include any suitable volatile memory, non-volatile memory’, storage, or any suitable combination thereof. For example, memory 1420 can include RAM, ROM. EPROM. EEPROM, other types of volatile memory.other types of non-volatile memory, one or more types of semi-volatile memory, one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 1420 can have encoded thereon a server program for controlling operation of server 1352. In such embodiments, processor 1412 can execute at least a portion of the server program to transmit information and / or content (e.g., data, images, a user interface) to one or more computing devices 1350, receive information and / or content from one or more computing devices 1350, receive instructions from one or more devices (e.g., a personal computer, a laptop computer, a tablet computer, a smartphone), and so on.

[0085] In some embodiments, the server 1352 is configured to perform the methods described in the present disclosure. For example, the processor 1412 and memory 1420 can be configured to perform the methods described herein (e.g.. the method of FIG. 3, the method of FIG 4).

[0086] In some embodiments, data source 1302 can include a processor 1422, one or more data acquisition systems 1424, one or more communications systems 1426, and / or memory 1428. In some embodiments, processor 1422 can be any suitable hardware processor or combination of processors, such as a CPU, a GPU, and so on. In some embodiments, the one or more data acquisition systems 1424 are generally configured to acquire data, images, or both, and can include a CT system, such as a CT system with energy -integrating detectors. Additionally or alternatively, in some embodiments, the one or more data acquisition systems 1424 can include any suitable hardware, firmware, and / or software for coupling to and / or controlling operations of a CT system. In some embodiments, one or more portions of the data acquisition system(s) 1424 can be removable and / or replaceable.

[0087] Note that, although not shown, data source 1302 can include any suitable inputs and / or outputs. For example, data source 1302 can include input devices and / or sensors that can be used to receive user input, such as a keyboard, a mouse, a touchscreen, a microphone, a trackpad, a trackball, and so on. As another example, datasource 1302 can include any suitable display devices, such as an LCD screen, an LED display, an OLED display, an electrophoretic display, a computer monitor, a touchscreen, a television, etc., one or more speakers, and so on.

[0088] In some embodiments, communications systems 1426 can include any suitable hardware, firmware, and / or software for communicating information to computing device 1350 (and, in some embodiments, over communication network 1354 and / or any other suitable communication networks). For example, communications systems 1426 can include one or more transceivers, one or more communication chips and / or chip sets, and so on. In a moreparticular example, communications systems 1426 can include hardware, firmware, and / or software that can be used to establish a wired connection using any suitable port and / or communication standard (e.g., VGA, DVI video, USB, RS-232, etc.), Wi-Fi connection, a Bluetooth connection, a cellular connection, an Ethernet connection, and so on.

[0089] In some embodiments, memory' 1428 can include any suitable storage device or devices that can be used to store instructions, values, data, or the like, that can be used, for example, by processor 1422 to control the one or more data acquisition systems 1424, and / or receive data from the one or more data acquisition systems 1424; to generate images from data; present content (e.g., data, images, a user interface) using a display; communicate with one or more computing devices 1350; and so on. Memory 1428 can include any suitable volatile memory, non-volatile memory, storage, or any suitable combination thereof. For example, memory 1428 can include RAM, ROM, EPROM, EEPROM, other types of volatile memory, other ty pes of non-volatile memory, one or more ty pes of semi-volatile memory, one or more flash drives, one or more hard disks, one or more solid state drives, one or more optical drives, and so on. In some embodiments, memory 1428 can have encoded thereon, or otherwise stored therein, a program for controlling operation of data source 1302. In such embodiments, processor 1422 can execute at least a portion of the program to generate images, transmit information and / or content (e.g., data, images, a user interface) to one or more computing devices 1350, receive information and / or content from one or more computing devices 1350, receive instructions from one or more devices (e.g.. a personal computer, a laptop computer, a tablet computer, a smartphone, etc.), and so on.

[0090] In some embodiments, any suitable computer-readable media can be used for storing instructions for performing the functions and / or processes described herein. For example, in some embodiments, computer-readable media can be transitory or non-transitory. For example, non-transitory computer-readable media can include media such as magnetic media (e.g., hard disks, floppy disks), optical media (e.g., compact discs, digital video discs, Blu-ray discs), semiconductor media (e.g., RAM, flash memory', EPROM, EEPROM), any suitable media that is not fleeting or devoid of any semblance of permanence during transmission, and / or any suitable tangible media. As another example, transitory computer- readable media can include signals on networks, in wires, conductors, optical fibers, circuits, or any suitable media that is fleeting and devoid of any semblance of permanence during transmission, and / or any suitable intangible media.

[0091] As used herein in the context of computer implementation, unless otherwise specified or limited, the terms “component / ’ “system,” “module,” “framework,” and the like are intended to encompass part or all of computer-related systems that include hardware, software, a combination of hardware and software, or software in execution. For example, a component may be, but is not limited to being, a processor device, a process being executed (or executable) by a processor device, an object, an executable, a thread of execution, a computer program, or a computer. By way of illustration, both an application running on a computer and the computer can be a component. One or more components (or system, module, and so on) may reside within a process or thread of execution, may be localized on one computer, may be distributed between two or more computers or other processor devices, or may be included within another component (or system, module, and so on).

[0092] In some implementations, devices or systems disclosed herein can be utilized or installed using methods embodying aspects of the disclosure. Correspondingly, description herein of particular features, capabilities, or intended purposes of a device or system is generally intended to inherently include disclosure of a method of using such features for the intended purposes, a method of implementing such capabilities, and a method of installing disclosed (or otherwise known) components to support these purposes or capabilities. Similarly, unless otherwise indicated or limited, discussion herein of any method of manufacturing or using a particular device or system, including installing the device or system, is intended to inherently include disclosure, as embodiments of the disclosure, of the utilized features and implemented capabilities of such device or system.

[0093] Referring particularly now to FIGS. 15 A and 15B, an example of an x-ray computed tomography (“CT”) imaging system 1500 is illustrated. The CT system includes a gantry 1502, to which at least one x-ray source 1504 is coupled. The x-ray source 1504 projects an x-ray beam 1506, which may be a fan-beam or cone-beam of x-rays, towards a detector array 1508 on the opposite side of the gantry 1502. The detector array 1508 includes a number of x-ray detector elements 1510. Together, the x-ray detector elements 1510 sense the projected x-rays 1506 that pass through a subject 1512, such as a medical patient or an object undergoing examination, that is positioned in the CT system 1500. In some implementation, the x-ray detector elements 1510 are energy-integrating detectors that produce an electrical signal that may represent the intensity of an impinging x-ray beam and, hence, the attenuation of the beam as it passes through the subject 1512. In some configurations, each x-ray detector 1510 may be a photon-counting detector that is capable of counting the number of x-ray photons thatimpinge upon the detector 1510. During a scan to acquire x-ray projection data, the gantry 1502 and the components mounted thereon rotate about a center of rotation 1514 located within the CT system 1500.

[0094] The CT system 1500 also includes an operator workstation 1516, which typically includes a display 1518: one or more input devices 1520, such as a keyboard and mouse; and a computer processor 1522. The computer processor 1522 may include a commercially available programmable machine running a commercially available operating system. The operator workstation 1516 provides the operator interface that enables scanning control parameters to be entered into the CT system 1500. In general, the operator workstation 1516 is in communication with a data store server 1524 and an image reconstruction system 1526. By way of example, the operator workstation 1516. data store sever 1524. and image reconstruction system 1526 may be connected via a communication system 1528, which may include any suitable network connection, whether wired, wireless, or a combination of both. As an example, the communication system 1528 may include both proprietary or dedicated networks, as well as open networks, such as the internet.

[0095] The operator workstation 1516 is also in communication with a control system 1530 that controls operation of the CT system 1500. The control system 1530 generally includes an x-ray controller 1532, a table controller 1534, a gantry7controller 1536, and a data acquisition system 1538. The x-ray controller 1532 provides power and timing signals to the x-ray source 1504 and the gantry controller 1536 controls the rotational speed and position of the gantry 1502. The table controller 1534 controls a table 1540 to position the subject 1512 in the gantry71502 of the CT system 1500.

[0096] The DAS 1538 samples data from the detector elements 1510 and converts the data to digital signals for subsequent processing. For instance, digitized x-ray data is communicated from the DAS 1538 to the data store server 1524. The image reconstruction system 1526 then retrieves the x-ray data from the data store server 1524 and reconstructs an image therefrom. The image reconstruction system 1526 may include a commercially available computer processor, or may be a highly parallel computer architecture, such as a system that includes multiple-core processors and massively parallel, high-density computing devices. Optionally, image reconstruction can also be performed on the processor 1522 in the operator workstation 1516. Reconstructed images can then be communicated back to the data store server 1524 for storage or to the operator workstation 1516 to be displayed to the operator or clinician.

[0097] The CT system 1500 may also include one or more networked workstations 1542. By way of example, a networked workstation 1542 may include a display 1544: one or more input devices 1546, such as a keyboard and mouse: and a processor 1548. The networked workstation 1542 may be located within the same facility as the operator workstation 1516, or in a different facility, such as a different healthcare institution or clinic.

[0098] The networked workstation 1542, whether within the same facility or in a different facility as the operator workstation 1516, may gain remote access to the data store server 1524 and / or the image reconstruction system 1526 via the communication system 1528. Accordingly, multiple networked workstations 1542 may have access to the data store server 1524 and / or image reconstruction system 1526. In this manner, x-ray data, reconstructed images, or other data may be exchanged between the data store server 1524. the image reconstruction system 1526, and the networked workstations 1542, such that the data or images may be remotely processed by a networked workstation 1542. This data may be exchanged in any suitable format, such as in accordance with the transmission control protocol (“TCP"’), the internet protocol ("IP”), or other known or suitable protocols.

[0099] The present disclosure has described one or more preferred embodiments, and it should be appreciated that many equivalents, alternatives, variations, and modifications, aside from those expressly stated, are possible and within the scope of the invention.

Claims

CLAIMS1. A method for generating an enhanced computed tomography (CT) image, the method comprising: accessing energy -integrating detector CT (EID-CT) image data with a computer system, wherein the EID-CT image data have been acquired from a subject using an EID-CT system; accessing a neural network with the computer system, wherein the neural network has been trained on training data to generate enhanced CT image data from CT images acquired with an EID-CT system; inputting the EID-CT image data to the neural network using the computer sy stem, generating enhanced CT image data as an output; and outputting the enhanced CT image data via the computer system.

2. The method of claim 1 , wherein the enhanced CT image data comprise a CT image with increased spatial resolution relative to the EID-CT image data.

3. The method of any one of claims 1 or 2, wherein the enhanced CT image data comprise a CT image with reduced noise relative to the EID-CT image data.

4. The method of any one of claims 1-3, wherein the enhanced CT image data comprise a CT image with reduced image artifacts.

5. The method of claim 4, wherein the reduced image artifacts comprise reduced blooming artifacts.

6. The method of claim 1, wherein the training data comprise CT image pairs, wherein each CT image pair comprises a high-resolution CT image paired with a low- resolution CT image.

7. The method of claim 6, wherein the high-resolution CT image in each CT image pair comprises a photon-counting detector CT (PCD-CT) image acquired wi th a PCD- CT system.

8. The method of any one of claims 6 or 7, wherein the low-resolution CT image in each CT image pair comprises an EID-CT image acquired with an EID-CT system.

9. The method of claim 8, wherein the EID-CT image has noise texture that is spliced into low-resolution training inputs.

10. The method of any one of claims 6 or 7, wherein the low-resolution CT image in each CT image pair comprises a degraded copy of the high-resolution CT image in the CT image pair.

11. The method of any one of claims 6 or 7, further comprising adding noise to the high-resolution image in each CT image pair.

12. The method of claim 11, wherein the noise comprises noise from a target image type.

13. The method of claim 11, wherein the noise added to the high-resolution image is obtained from a phantom scan image.

14. The method of any one of claims 6 or 7, further comprising processing the high-resolution CT image in each CT image pair with a noise reduction algorithm.

15. The method of claim 1. wherein the neural network comprises a convolutional neural network.

16. The method of claim 1, wherein the neural network is trained on the training data using a loss function comprising both data fidelity and feature loss.

17. The method of claim 1, wherein the EID-CT image data have been acquired from the subject using coronary' CT angiography (cCTA).

18. The method of claim 1. wherein the neural network implements an attention mechanism.

19. The method of claim 18, wherein the attention mechanism comprises masking out anatomies outside of a target anatomical region.

20. The method of claim 19, wherein the target anatomical region comprises a heart.

21. The method of claim 18, wherein the attention mechanism comprises an attention gate in the neural network.

Images