Machine-learning based medical imaging

Abstract

A system may be configured to generate multiple machine-learning based volumetric images from initial measurement data. Slices of a machine-learning based volumetric image may correspond to ML-based versions of respective slices of the initial volumetric image. The system may calculate differences between the different machine-learning based volumetric images. The system may enable a user to select an image slice for viewing. Based on the selection the system may identify multiple machine-learning based image slices that correspond to the selected image slice, generate a display object based on the identified machine-learning based image slices, and display the display object.

Description

MACHINE-LEARNING BASED MEDICAL IMAGINGSTATEMENT REGARDING FEDERAL FUNDING

[0001] This invention was made with government support under R01 CA249538 awarded by the National Institutes of Health. The government has certain rights in the invention.RELATED APPLICATIONS

[0002] The present application claims priority to U.S. Prov. App. No. 63 / 726203, which is incorporated herein by reference for all purposes.BRIEF DESCRIPTION OF THE DRAWINGS

[0003] FIG. 1 is a block diagram illustrating an example of a system to generate and display machine-learning based volumetric images.

[0004] FIG. 2 illustrates an example process for training a generative diffusion model.

[0005] FIG. 3 illustrates an example of three image samples generated from the same measurement data.

[0006] FIG. 4 illustrates an example of concurrently displayed machine-learning-based image slices that correspond to the same image slice of a standard volumetric image.

[0007] FIGS. 5A-5D illustrate ML-based image slices displayed sequentially.

[0008] FIGS. 6A-6D illustrate an example of a display of image slices generated using different modalities.

[0009] FIG. 7 is a flow diagram illustrating an example of a process for machine-learning based medical imaging.DETAILED DESCRIPTION

[0010] In some cases, machine learning models are able to outperform classical approaches to medical imaging, such as, for example, in denoising, spatial resolution enhancement, reconstruction of undersampled data, and artifact reduction. In x-ray CT, the use of machine learning models has facilitated lower dose acquisitions to minimize risks, provide enhanced spatial resolution, minimize artifacts associated with metalimplants or patient motion, and reduce data acquisition requirements (e.g. through sparse data acquisitions). These successes are enabled through the capacity of machine learning models to capture rich prior information about the images to be estimated. Similarly, machine learning models are able to learn relationships between inputs and outputs that are difficult to derive from first principles / physics. Examples include the estimation of material densities from a single polyenergetic CT dataset, synthesis of a CT volume from MRI data, and fast x-ray scatter estimation.

[0011] Despite the improvements, the use of machine learning models may result in undesired outcomes. For example, the use of machine learning models outside of a particular context may result in misdiagnoses. Consider, as a non-limiting example, CT reconstruction from a single view. Such a reconstruction may have its place for navigation, attenuation correction for nuclear imaging, etc.; but it may not be appropriate for detection of subtle lesions. In some cases, machine learning model outputs can appear to be plausible CT volumes but can misrepresent specific details. The presence of such features that do not exist in reality may be referred to as hallucinations. Moreover, because a machine learning model may be so effective in reducing artifacts and noise, a resulting image may not have any indication that it is not trustworthy. For example, an image that would be identified as a “bad” or unreliable image by a medical professional (e.g., due to underlying issues with the imager or other hardware), may appear without any indication that the underlying data was bad or unreliable (e.g., the machine learning model may have removed the artifacts that would have given a medical professional reason to doubt the accuracy of the image). In contrast, when a filtered-back projection (FBP) yields an image with high regional noise, streaks, or other artifacts, a reader can moderate their confidence.

[0012] In some cases, the “false features” generated by a machine learning model may not even be considered a hallucination. For example, some consider a “hallucination” to be comprised of features inserted by a machine learning model that fall in that system null space (e.g., hence unsupported by the measurements). However, it is possible for a machine learning model to create false features in low SNR scenarios not in the null space. For example, consider a case where the specific details are incorrect, but thosedetails still convey information - e.g., a stochastic texture where the “realization” is incorrect, but the texture itself is diagnostic (e.g., a clearly spiculated lesion but the individual spiculations are misrepresented).

[0013] Uncertainty due to the use of machine learning models can arise from the measurements noise (aleatoric) as well as the machine learning model (epistemic). In some cases, aleatoric uncertainty may be predicted in FBP through variance and correlation prediction as well as in MBIR with appropriate local approximations. In certain cases, epistemic uncertainty may be determined using various approaches (e.g., bootstrap, delta / nonlinear regression, Monte Carlo dropout, etc.). Notwithstanding, it is unclear how to communicate uncertainty in a meaningful way to a medical professional.

[0014] The uncertainty and issues surrounding the use of machine learning models for imaging purposes may be further exacerbated based on the imaging scenario. For example, abdominal imaging can be noisy due to patient size and liver lesions are often low contrast. Accordingly, improved detection and characterization of liver lesions including tasks that influence the differentiation between common malignant lesions (hepatocellular carcinoma, cholangiocarcinoma, lymphoma, sarcoma, metastatic tumors, etc.) and common benign lesions (cysts, hemangioma, adenoma, focal nodular hyperplasia, etc.) can provide significant benefits.

[0015] In addition, improving imaging using machine learning models may also provide significant benefits to patients by reducing the amount of exposure to x-rays. For example, low-dose CT approaches have been widely recommended as a strategy to detect and diagnose pulmonary nodules. However, the images generated from low-dose CT may not be as high of quality as a high-dose CT. A machine learning model that can enable the use of low-dose CT in place of high-dose CT can result in lower exposure to x-rays by patients.

[0016] To address the aforementioned issues, an imaging system may be configured to generate, from measurement data and / or a volumetric image, multiple machine-learningbased (ML-based) volumetric images and display an ML-based image or set of ML-based images that indicate or highlight areas of uncertainty.

[0017] While examples described herein may rely on improved imaging for liver lesions using CT data, it will be understood that the concepts described herein may have broad application across imaging tasks and imaging modalities.

[0018] FIG. 1 is a block diagram illustrating an example of a system 100 to generate and display machine-learning based volumetric images. In the illustrated example, the system 100 includes an imager 102, machine learning model 104, and display system 106. It will be understood, however, that the system 100 may include fewer or more components. For example, the system 100 may include the display system 106, which receives machine-learning based volumetric images from a machine learning model 104 (e.g., that is part of a separate system) and / or the system 100 may include the machine-learning model 104 and display system 106, which receive the initial image data from the imager 102 (e.g., that is part of a separate system).

[0019] The imager 102 may be configured to generate measurement data corresponding to a volumetric image (such data may also be referred to herein as volumetric data and / or medical image data). For example, the imager 102 may be implemented as a CT imager configured to generate CT image data. As described herein, low dose CT data may result in low resolution or low-quality images (e.g., as compared to high radiation dose CT data) that can make it more difficult to differentiate between common malignant lesions (hepatocellular carcinoma, cholangiocarcinoma, lymphoma, sarcoma, metastatic tumors, etc.) and common benign lesions (cysts, hemangioma, adenoma, focal nodular hyperplasia, etc.).

[0020] As another example, the imager 102 may be implemented as an MRI imager, x- ray imager, ultrasound imager, or other imager that can generate volumetric (e.g., 3D) data. In some cases, the imager 102 may generate a volumetric image (based on measurement data).

[0021] Solely for illustration purposes, FIG. 1 shows a slice 103 of a volumetric image corresponding to (e.g., generated from) the measurement data generated by the imager 102. In the illustrated example, the slice 103 is a CT image of a spiculated liver lesion.

[0022] The machine-learning model 104 can be configured to generate ML-based volumetric images 105 and / or samples (also referred to herein as image samples or slice samples) based on (e.g., using) measurement data received from the imager 102. Such measurement data may be the same measurement data used to generate the slice 103 or may include the measurement data used to generate a set of slices corresponding to a volumetric image. In particular, the machine-learning model 104 may be trained to denoise, enhance spatial resolution, reconstruct under sampled data of, and / or reduce artifacts in the initial measurement data. As described herein, in performing these functions, the machine-learning model 104 may hallucinate or otherwise create false features in the underlying data. Moreover, the machine-learning model 104 may remove features (e.g., high regional noise, streaks, or other artifacts) from the volumetric image that would otherwise enable a viewer to determine that the image data is unreliable.

[0023] In some cases, the machine-learning model 104 may be implemented as a diffusion model and may be based on diffusion posterior sampling. In certain cases, the machine-learning model may be implemented using a U-net convolutional neural network architecture or other neural network architecture using convolutional neural networks, transformer neural networks, or the like.

[0024] In some cases, the machine learning model may be implemented as a generative diffusion model and may define a forward stochastic process using the following definition:where xt denotes a noisy and faded image based on the true image, xo, at time t with fading level and noise determined by the parameter at. reverse process conditioned on measurements, y, can be written asstandard Wiener process with reverse-time flows from 1 to 0, and dt is an infinitesimal negative timestep. This model can be trained with a degraded image (y) as a conditional network input, which allows a discretized version of the reverse process to be an implementation of the denoising / restoration algorithm.

[0025] As another example, the machine learning model may be implemented using Diffusion Posterior Sampling (DPS). This approach (see FIG. 2) may combine unsupervised training based on a typical stochastic diffusion process 202 with a conditional reverse stochastic process 208 where measurements are explicitly modeled. For example, forward diffusion steps 202A-D processes an image 201 by iteratively fading the image 201 and adding noise.

[0026] The conditional reverse stochastic process 208 generates a prior realization of image 201 starting with a noise input 207. The conditional reverse stochastic process 208 iteratively reduces noise at steps 208A-D to reproduce the image 201. Noise in the reverse process can be approximated by a neural network. The neural network may be trained conditionally with a degraded image input 204, providing an algorithm for image restoration. DPS adopts constraints on measurement data 206 so that the (unsupervised) degraded image input 204 is combined with a conditional reverse process based on measurements (through a likelihood model).

[0027] Mathematically, this approach arises from applying Bayes rule to (2), to write:logp(x + VXtlogp(y|xt))]dt + g(t)dwtt e [0,1] (3)

[0028] Estimating prior score, Vxtlogp(xt), by a neural network as se(xt,t), and making the following approximation - x0) (4)where x0denotes an estimate of xo from xt, and logp(y|xo) denotes the likelihood model for the measurements. Accordingly, in some cases, the machine learning model 104 can perform reconstruction from arbitrary measurements.

[0029] Returning to FIG. 1 , the machine learning model 104 can generate multiple ML- based volumetric images 105 from the initial measurement data (also referred to herein as volumetric data) (e.g., the measurement data generated by / received from the imager 102) and / or a volumetric image corresponding to (e.g., generated from) the measurement data. The volumetric data used by the machine learning model 104 to generate the ML- based volumetric images 105 may also be used to generate a standard volumetric image without using a machine learning model (e.g., using a standard non-machine-learning- based image generator). Accordingly, in some cases, the machine learning model 104 may generate the ML-based versions of a volumetric image using the same underlying measurement data used to generate a standard (e.g., non-machine-learning-based) volumetric image.

[0030] As described herein, the machine learning model 104 may be trained to denoise, enhance spatial resolution, reconstruct under sampled data of, and / or reduce artifacts in the initial measurement data. In certain cases, the machine learning model 104 may be trained to denoise, enhance spatial resolution, reduce artifacts, enhance, and / or modify a standard (non-ML-based) image. Thus, the volumetric images generated by the machine-learning model 104 may be ML-based versions of a standard volumetric image, such as, for example, versions with less noise, spatially enhanced resolution, reconstructed (from under sampled data), and / or with reduced artifacts.

[0031] In some cases, the machine learning model 104 may use a stochastic process to generate the ML-based volumetric images 105 and / or samples, where each ML-based volumetric image 105 may be different from each other and each sample generated from the same measurement data may be different.

[0032] FIG. 3 illustrates the potential complexity of uncertainty in posterior samples. For example, FIG. 3 illustrates example image samples 302, 304, and 306 generated from the same measurement data, where the initial measurement data included a spiculatedlesion. Each example image sample 302, 304, and 306 shows variability in the details of this spiculated lesion. While a variance image will suggest decreased certainty at the edges of the lesion, this ensemble of samples is all clearly spiculated.

[0033] In some cases, to generate ML-based volumetric images 105, the machine learning model 104 may generate multiple image slices using the standard volumetric image (e.g., break up the volumetric image into multiple slices) and generate multiple samples or image samples corresponding to each, some, or all image slices (e.g., using the measurement data corresponding to the image slices). As a non-limiting example, if the volumetric image is broken into one hundred slices and the machine learning model 104 generates twenty samples from each slice or its corresponding measurement data (e.g., the measurement data used to generate the slice), the machine learning model can generate twenty ML-based volumetric images 105 (each made up of one hundred ML- based image slices) or two thousand samples. Moreover, each slice from the ML-based volumetric images 105 can correspond to one of the slices from the standard volumetric image. In addition, as the machine learning model 104 may use a stochastic process to generate the ML-based volumetric images 105 and / or samples, each ML-based volumetric image 105 may be different from each other and each sample generated from the same standard image slice (or the same measurement data corresponding to the image slice) may be different.

[0034] In certain cases, the machine learning model 104 may generate the image samples and / or ML-based volumetric images 105 upon receipt of the initial volumetric data or volumetric image (e.g., before a user indicates a desire to view the images). Once generated, the machine learning model 104 may store (such as, in a data store) the generated image samples and / or ML-based volumetric images 105 until a user accesses them.

[0035] As described herein, generating image samples can result in thousands of generated image samples that may be stored indefinitely until a user requests them and / or a user deletes the generated image samples from storage. Generating and storing such image samples can increase the computing costs of the system 100. To address this issue, in some cases, the machine learning model 104 may wait to generate the ML-based volumetric images and / or image samples until receiving a user request (e.g., generate the ML-based images in real-time and responsive to the user’s request). For example, a user may review one or more standard volumetric images and indicate (e.g., select / hover with a user device, view the volumetric image for a period of time, and / or other potential indication) that the user would like to view ML-based image slices corresponding to the indicated image slice. In response, the machine learning model 104 may generate the ML-based volumetric images 105 after receiving the user request.

[0036] In some cases, the machine learning model 104 may generate an example ML- based volumetric image 105 to display to the user along with the standard volumetric image. For example, the machine learning model 104 may, in response to receipt of the initial volumetric data, generate a single ML-based volumetric image 105 (e.g., made up of multiple slices). The machine learning model 104 may generate additional ML-based volumetric images 105 based on a user request or indication of interest as described herein.

[0037] The display system 106 may be implemented using one or more computing devices (e.g., with a multiprocessor, memory, etc.) and may be configured to provide a graphical user interface 110 for a client computing device and / or a user to display the volumetric images, image slices, ML-based volumetric images 105, and / or ML-based image slices (e.g., slices of the ML-based volumetric images 105).

[0038] In some cases, the display system 106 (or another component of the system) may calculate uncertainty parameters 108 with respect to the ML-based volumetric images 105. As described herein, the machine-learning-based images may hallucinate or otherwise generate inaccurate data that is not easily detectable. Accordingly, the display system 106 may use the determined uncertainty parameters 108 to display the ML-based volumetric images 105 and / or ML-based image slices in a way that conveys the uncertainty of the underlying data (e.g., the underlying samples) and / or images to a user.

[0039] In some cases, the uncertainty parameters 108 may include differences between the images. For example, the display system 106 may compare individual or groups of pixels of one ML-based image slice from one ML-based volumetric image 105 with corresponding individual or groups of pixels from a ML-based image slice of a differentML-based volumetric image 105 to identify differences. In some cases, the display system 106 may compare some or all ML-based image slices from some or all ML-based volumetric images 105 to determine the differences between some or all of the ML-based volumetric images 105. The differences may be determined based on statistical measures of variability such as standard deviation, probability scores, etc. The determined differences may correspond to the uncertainty parameters 108.

[0040] The display system 106 may enable a user to view ML-based image slices (e.g., slice of a ML-based volumetric image). It will be understood that the slices used / displayed by the display system 106 may be the same or different from (e.g., “cut” differently than) the slices used by the machine learning model 104 to generate the samples and / or the ML-based volumetric images 105 and the samples. For example, the machine learning model 104 may use slices (or measurement data) along the sagittal plane to generate the samples and / or ML-based volumetric images 105, whereas, the display system 106 may enable a user to use slices along any plane (e.g., sagittal, coronal, transverse, oblique) or angle. As such, the slices generated by the machine learning model may be different than the image slice displayed by the display system. In some such cases, different portions of image slices generated using the machine learning model may be combined to form the image slice displayed by the display system. For example, the generated image slices may be generated along the saggital plane. To display an image slice along the coronal plane, portions of different image slices generated by the machine learning model that share the same coronal plane may be stitched or combined together to generate the slice along the coronal plane.

[0041] In some cases, the display system 106 may generate a display object for display to a user based on and / or using the ML-based volumetric images 105, ML-based image slices, and / or the uncertainty parameters 108. In certain cases, the display object may include a composite image generated from some or all of the ML-based volumetric images 105 and / or some or all of the ML-based image slices that correspond to the same measurement data or a particular slice of a standard volumetric image. In some such cases, a pixel or group of pixels in the composite image may be generated based on the value for a corresponding pixel or group of pixels in the ML-based image slices. Forexample, the system may determine the average color of the same pixel across the ML- based image slices and assign the average color to the corresponding pixel of a composite image slice. As another non-limiting example, if 51 % of the ML-based images (or some other threshold) have the same or similar value for a particular pixel, the system may assign that value (or color) to the corresponding pixel of the composite image slice. In some cases, the composite image or image slice may be based on (or generated using) a statistical measure of variability between the ML-based image slices and / or based on performance of a model observer for a particular task.

[0042] In certain cases, the display system 106 may generate a composite volumetric image from one or more of the ML-based volumetric images 105 (or a composite image slice from one or more of the ML-based image slices) and display the differences between the ML-based image slices on the composite image. For example, the display system 106 can display the pixels or areas of the image with greater variability (e.g., based on a calculated statistical measure such as standard deviation, probability, variance, etc.) across the different ML-based volumetric images 105 / ML-based image slices different from the pixels or areas of the image with less variability across the different ML-based volumetric images 105 / ML-based image slices.

[0043] In certain cases, the display system 106 can generate a heatmap, or other indicator to indicate the gradient showing the pixels or areas with less and greater variability between the ML-based volumetric images 105 and / or ML-based image slices (e.g., the showing pixel variation over multiple images), etc. The display system 106 may display the generated heatmap along with the composite volumetric image to indicate to the user specific portions of the volumetric image that are most reliable.

[0044] In some cases, the display system 106 may show some or all of a set of ML-based image slices that correspond to the same measurement data or image slice of a standard volumetric image sequentially. FIGS. 5A-5D illustrate an example of displaying ML-based image slices 502-508 sequentially with a bar 510 at the top enabling a user to cycle through and / or select different ML-based image slices that correspond to the same measurement data or same image slice of a standard volumetric image. In some cases, the image slices may be sequentially ordered based on their calculated variability (e.g.,standard deviation, probability scores, etc.), such that image slices with pixels associated with a lower variability are displayed before images with pixels associated with a higher variability.

[0045] In certain cases, the display object generated by the display system may include one of the ML-based image slices or a combination of the ML-based image slices. For example, the display system 106 may generate a video or video file from a set of ML- based image slices that correspond to the same image slice of the standard volumetric image. In certain cases, the display system 106 may loop the video continuously. FIGS. 5A-5D may also illustrate an example of displaying ML-based image slices as a video with bar 510 representing the view of the video at different times.

[0046] In certain cases, the display system 106 may show some or all of a set of ML- based image slices that correspond to the same measurement data (e.g,. generated form the same measurement data) or image slice of the standard volumetric image concurrently or at the same time. For example, the set of ML-based image slices may be displayed in a grid and / or tiled view with one or more columns and one or more rows. FIG. 4 illustrates an example of concurrently displaying some or all ML-based image slices that correspond to the same measurement data or same image slice of a standard volumetric image. The displayed ML-based image slices may be ordered based on their calculated variability. For example, image slices with pixels associated with a higher variance may be in the bottom rows, and image slices with pixels associated with a lower variance may be in the top rows.

[0047] In some cases, the display system 106 may show some or all of a set of ML-based image slices that correspond to the same measurement data or image slice of a standard volumetric image with (e.g., at the same time or concurrently with) the corresponding image slice of the standard volumetric image. FIGS. 6A-6D illustrates an example of displaying standard image slice concurrently with one or more ML-based image slices. FIG. 6A illustrates an example FBP image slice 602 of ribs generated using a high dose CT scan. Similarly, FIG. 6B illustrates an example FBP image slice 604 of the corresponding view of the ribs using a low dose CT scan.

[0048] FIG. 6C illustrates an example display object 606 generated from multiple ML- based image slices, where the ML-based images are generated from the FBP image slice 604. This display object 606 may enable a user to cycle through ML-based images, display the ML-based images in succession or display the ML-based images as a video (as described herein with reference to FIGS. 5A-D). FIG. 6D displays a display object 608 of a single ML-based image slice and / or composite of multiple ML-based image slices generated from the FBP image slice 604.

[0049] In certain cases, the display system 106 may enable a user to select an image slice corresponding to a slice of the standard volumetric image. The display system 106 may then enable a user to select how to view the ML-based volumetric images 105 corresponding to the selected image slice. As such, the display system 106 may enable the user to view the ML-based volumetric images 105 (and corresponding images slices) concurrently (at the same time), sequentially, as a video, and / or as a composite image slide or heatmap.

[0050] In some cases, the display system 106 may display one of the ML-based image slices (or an image slice corresponding to the standard volumetric image) and enable a user to select a portion of the image slice. Based on the selection of the portion of the image slice, the display system 106 may provide a zoomed in view of the selected portion and enable the user to select how to view the portion of the image slice and / or enable the user to cycle through corresponding portions of the selected portion in the ML-based image slices / ML-based volumetric image. In this way, the display system 106 can provide context for the ML-based image slices (or portions thereof) that are displayed.

[0051] By displaying some, all, or a combination of some / all of the ML-based volumetric images 105 and / or ML-based image slices, the display system 106 can facilitate understanding of the areas of the volumetric image associated with greater uncertainty by the machine learning model 104. In this way, the imaging system 100 can improve the resolution and fidelity of volumetric images using a machine-learning model, while increasing the likelihood that the right level of confidence is placed on the accuracy of the ML-based volumetric images 105 / ML-based image slices.

[0052] As described herein, the display system 106 may initially display one or more image slices of the standard volumetric image for the user to review before generating the ML-based image slices. For example, the display system 106 may display the standard volumetric image and / or image slices of the standard volumetric image. The user may select a portion of the standard volumetric image / image slices to indicate that the user would like to see more detailed ML-based volumetric images corresponding to the selected portion of the volumetric image / selected image slices. The machine learning model 104 may then generate the ML-based volumetric images 105 and / or ML-based image slices and cause the display system 106 to display the ML-based volumetric images 105 as described herein.

[0053] In further cases, the machine learning model 104 may generate an example slice of the ML-based volumetric image 105 for (or corresponding to) each standard volumetric image slice and cause the display system 106 to display the example ML-based volumetric image slices concurrently with the corresponding standard volumetric image slices (e.g., sequentially, in a grid pattern, etc.). The user may select one or more example ML-based volumetric image slices for further review, and the machine learning model 104 can generate additional ML-based image slices from the standard volumetric image 105 and / or from the measurement data used to generate the standard volumetric image 105.

[0054] FIG. 7 is a flow diagram illustrating an example of a process 700 for machinelearning based medical imaging. The process 700 may be executed, for example, using one or more processors or computing devices associated with the display system 106. For simplicity, the process 700 will be described as being performed by a system, such as the system 100.

[0055] At block 702, the system receives measurement data corresponding to a first volumetric image. As described herein, the measurement data may be received from an imager, such as imager 102. In some cases, the measurement data may include computer tomography (CT) data. In further cases, the measurement data may include x- ray imagery, ultrasound imagery, and / or any other form of medical imagery data.

[0056] The system may receive CT measurement data from the imager 102 in the form of low dose CT data. For example, the imager 102 may use a lower dose of radiation inperforming the CT imaging than conventional CT imaging. Such low dose CT data may result in low resolution and / or low-quality images compared to conventional imaging. Such low resolution images may make it difficult for users to properly diagnose imaged lesions (e.g., diagnosing a malignant lesion compared to diagnosing a benign lesion).

[0057] At block 704, the system generates (a plurality of) second volumetric images based on the received measurement data. In some cases, the system may utilize a machinelearning model (such as, machine learning model 104) to generate the plurality second volumetric images (or ML-based images).

[0058] The machine learning model 104 may generate the second volumetric images directly from the measurement data and / or indirectly from the measurement data. For example, the machine learning model 104 may receive and / or use the measurement data to generate one or more ML-based images and / or the machine learning model 104 may receive and / or use one or more standard images (or image slices) generated from the measurement data to generate the second volumetric images. Accordingly, the system may generate the second volumetric images based on at least one of the measurement data or the first volumetric image.

[0059] As described herein, in some cases, the machine learning model 104 may be implemented as a generative diffusion model and / or may use diffusion posterior sampling to generate the second volumetric images using the measurement data. In particular, the machine-learning model 104 may be trained to denoise, enhance spatial resolution, reconstruct under sampled data of, and / or reduce artifacts in the measurement data to generate the second volumetric images. As described herein, the machine learning model 104 may use a stochastic process to generate the second volumetric images such that each second volumetric image (or sample) may be different.

[0060] In certain cases, the machine learning model 104 may generate one or more second volumetric images by using the first volumetric image (or slices thereof) as an input to the machine learning model 104. In some cases, the machine learning model 104 may perform one or more sets of modifications (e.g., denoise, enhance, reconstruct, etc.) to the first volumetric image to generate each second volumetric image. For example, the machine learning model 104 may apply different denoising, spatialresolution, and / or artifact reductions based on stochastic processes to generate the different second volumetric images. Thus, each volumetric image of the second volumetric images may vary based on the different denoising, spatial resolution, and / or artifact reductions, or other revisions, performed by the machine learning model 104.

[0061] Additionally or alternatively, the system may generate multiple image slices from the measurement data. For example, the system may break up the first volumetric image into multiple first images slices and communicate the first image slices to the machine learning model 104. For each first image slice of the first volumetric image, the machine learning model 104 may generate multiple second (or ML-based) image slices as a set of second image slices. Each second image slice in the set of second slices may correspond to (e.g., be generated from) the same first image slice (or the measurement data corresponding to the same first image slice) but vary from each other based on the different denoising, spatial resolution, and / or artifact reductions, or other revisions, performed by the machine learning model 104.

[0062] As a non-limiting example, the system may break up a first volumetric image (e.g., from the measurement data) into one hundred first image slices and communicate the one hundred image slices to the machine learning model 104. The machine learning model 104 may perform twenty distinct revisions on each of the one hundred first image slices (and generate a set of twenty ML-based image slices for each of the one hundred first image slices). In some cases, the machine learning model 104 may use the different image slices to identify or retrieve corresponding measurement data and generate the twenty distinct samples or image slices based on the retrieved measurement data. Thus, each of the image slices in a set of twenty ML-based images may correspond to (e.g., be generated from) the same image slice (or corresponding measurement data) but vary from each other based on the different denoising, spatial resolution, and / or artifact reductions, or other revisions, performed by the machine learning model 104.

[0063] With continued reference to the example, using the ML-based image slices, the machine learning model 104 can generate twenty distinct second volumetric images. Each second volumetric image may include one hundred ML-based image slices from different sets of ML-based image slices. For example, the first image slice from each ofthe one hundred sets of ML-based image slices may be combined or grouped to form one ML-based volumetric image (e.g., a second volumetric image). Similarly, the second image slice from each of the one hundred set of ML-based images slices may be combined or grouped to form another volumetric image (e.g., another second volumetric image).

[0064] In some cases, the system may, via the machine learning model 104, generate the plurality of second volumetric images (e.g., from the measurement data, first volumetric image and / or from the plurality of image slices) when the machine learning model 104 receives the measurement data, first volumetric image, and / or plurality of image slices. In some such cases, the system may receive the plurality of second volumetric images from the machine learning model 104 and store (e.g., in a data store) the plurality of second volumetric images until receiving a user request for display.

[0065] As described herein, generating a plurality of second volumetric images from a plurality of image slices may result in thousands of generated images, which may be computationally expensive. Storing the thousands of generated images indefinitely may further increase computing costs. To address this issue, the system may wait to receive an indication from a user device to generate the plurality of second volumetric images. For example, by waiting until a request to display a particular image slice of the first volumetric image, the system can avoid storing the generated plurality of second volumetric images until requested by a user.

[0066] In some cases, the system may generate the second volumetric images for (only) a subset of image slices or measurement data. For example, the system may (only) generate a plurality of second image slices that correspond to an image slice (or corresponding measurement data) that has been requested by a user (e.g., by selecting, hovering, clicking on, an image slice from the first volumetric data etc.). Accordingly, the system may reduce computing costs by generating a portion of the second image slices instead of generating hundreds or thousands of second image slices that correspond to the entire first volumetric image. Put another away, the system may generate a subset of the second volumetric image (e.g., the portion or portions corresponding to slices or measurement data requested by a user).

[0067] In certain cases, the system may, using the machine learning model 104, initially generate one or more second volumetric images based on the measurement data. For example, the system may generate a single (or small number) of second volumetric images (e.g., when the system receives the measurement data and / or when the system receives a user request for display of a particular image slice of the first volumetric image). In some such cases, the system may display the one or more second volumetric images along with the first volumetric image jointly. Responsive to a user request, the system may generate additional second volumetric images using the first volumetric image (or generate additional second image slices corresponding to a first image slice). In this way, the system may reduce the computational resources used to generate and store the second volumetric data until requested by a user.

[0068] At block 706, the system receives an indication to display an image slice generated based on at least a portion of the measurement data. For example, the image slice may be generated directly from the measurement data (e.g., using the measurement data) and / or indirectly (e.g., using an image that was generated from the measurement data, such as from and / or using a standard (first) image). The indication to display the image slice may include a user selection of the image slice or an indication of interest by the user (e.g., by clicking on, touching the screen, hovering over, or otherwise selecting the image slice).

[0069] In some cases, the indication to display an image slice may correspond to a user selection or indication of interest by the user of an image slice of a standard (first) volumetric image. For example, the user may review a standard (e.g., non-ML-based) volumetric image and indicate a desire to view one of the slices of the volumetric image and / or select an image file that corresponds to a standard image slice.

[0070] In certain cases, the indication to display an image slice may correspond to a user selection or indication of interest by the user of an ML-based image (e.g., of a second volumetric image). For example, in certain cases, the system may display ML-based image slices or ML-based volumetric images to a user. In some such cases, the system may receive a user selection of one of the ML-based image slices.

[0071] In some cases, the user may not know whether an image is an ML-based image (second volumetric image) or a standard image (first volumetric image). As such, the indication to display an image slice may correspond to a request to view a volumetric image or image slice. For example, the user may select a CT image or CT image slice without knowing whether the CT image was generated using an ML model or not.

[0072] At block 708, the system identifies a plurality of second image slices from the plurality of second volumetric images. In some cases, the system may identify the plurality of second image slices responsive to the user indication to display the image slice.

[0073] In certain cases, the system may identify the second image slices that correspond to the image slice indicated for display. For example, if the image slide indicated for display is an ML-based image slice, the system may identify other ML-based image slices that correspond to the indicated ML-based image slice (e.g., ML-based image slices generated from the same measurement data as the indicated image slice). If the image slice is a standard or first image slice, the system may identify the ML-based (second) image slices generated from or using the same measurement data as the first image slice or the ML-based (second) image slices generated from or using the first image slice.

[0074] In some cases, the indicated image slice may be along a different plane than the ML-based (second) image slices. For example, the system may use the sagittal plane to generate the plurality of second image slices, whereas the user may select (or indicate) a slice of the first volumetric image along another plane (e.g., coronal, transverse, oblique, etc.) and / or angle. In some such cases, the system may identify the plurality of second image slices that correspond to the same measurement data as the indicated image slice (e.g., the second image slices that cover the same area as the indicated image slice).

[0075] In certain cases, the system may identify the plurality of second image slices by generating them. As described herein, in some cases, the system may, in response to receiving an indication to display an image slice, generate multiple second image slices using the indicated (or first) image slice or measurement data corresponding to the indicated image slice. For example, the system may use the machine learning model and different denoising, spatial resolution, and / or artifact reductions based on stochastic processes to generate the second image slices from a first (normal) image slice. In suchcases, once the system receives the indication to display the particular image slice (via the user selection), the system may communicate with the machine learning model 104 to generate the plurality of second image slices that correspond to the particular image slice.

[0076] At block 710, the system causes a display system or display device to display a display object (e.g., via a graphical user interface). As described herein, the display object may be based on the identified plurality of second image slices. For example, the display object may include or be one of the second image slices, include some or all of second image slices (e.g., displayed as a list or grid, etc.), be a composite image of one or more of the second image slices, and / or be generated using one or more of the second image slices (e.g., a heatmap or variability map, etc.) .

[0077] In some cases, as described herein, displaying the display object may include displaying the plurality of second image slices as a grid or in a sequence (e.g., displaying each second image slice together one after the other) so that the user can efficiently cycle through the second image slices. In further cases, the system may generate a video based on the plurality of second image slices, and displaying the display object may include displaying the video on the GUI.

[0078] In certain cases, as described herein, the display object may include or may be a composite image. For example, the system may generate a composite (e.g., combined) image of some or all of the plurality of second image slices. In some cases, the value of an individual pixel or group of pixels in the composite image may be generated based on the value of the corresponding pixel or group of pixels in the plurality of second image slices. In certain cases, the system may use an average (or other statistical measure) or threshold to determine the content of an individual pixel or group of pixels. For example, if 51 % of the plurality of second image slices include a particular value for a pixel (or the average of the plurality of second image slices is the particular value), then the composite image may include that value for the corresponding pixel or group of pixels. It will be understood that other thresholds and / or statistical measures may be used to determine the value of pixels in the composite image.

[0079] In some cases, as described herein, the system may generate the composite image based on a statistical measure of variability between the plurality of second image slices (or uncertainty parameters). For example, pixels with the least variability among the plurality of second image slices may be selected to be included in the composite image.

[0080] In certain cases, as described herein, for each pixel in the plurality of second image slices, the system may utilize a statistical measurement (e.g., standard deviation, probability, variability, etc.) to generate a heatmap to display in addition to or separately from the composite image and / or one or more of the plurality of second image slices. The heatmap may indicate a gradient showing the pixels or areas with less and greater variability between the second volumetric images and / or second image slices. In some cases, a higher variance (e.g., larger standard deviation value) for a pixel may indicate that there is a higher level of uncertainty as to the content of the pixel.

[0081] It will be understood that fewer, more, or different blocks may be used as part of the process 700. For example, the process 700 may begin at block 706 or by displaying a volumetric image (before moving to block 706). Moreover, the blocks of the process 700 may be performed in a different order or concurrently. For example, blocks 704 and 706 may be performed in reverse order and / or block 704 may be repeated after block 706, or be part of block 708, etc.Computer Implementations

[0082] A software program or algorithm, when referred to as "implemented in a computer- readable storage medium," includes computer-readable instructions stored in a memory device (e.g., memory device(s)). A processor is "configured to execute a software program" when at least one value associated with the software program is stored in a register that is readable by the processor. In some embodiments, routines executed to implement the disclosed techniques may be implemented as part of OS software (e.g., MICROSOFT WINDOWS® and LINUX®) or a specific software application, algorithm component, program, object, module, or sequence of instructions referred to as "computer programs."

[0083] Computer programs typically comprise one or more instructions set at various times in various memory devices of a computing device, which, when read and executed by at least one processor, will cause a computing device to execute functions involving the disclosed techniques. In some embodiments, a carrier containing the aforementioned computer program product is provided. The carrier may be one of an electronic signal, an optical signal, a radio signal, or a non-transitory computer-readable storage medium.

[0084] Any or all of the features and functions described above can be combined with each other, except to the extent it may be otherwise stated above or to the extent that any such embodiments may be incompatible by virtue of their function or structure, as will be apparent to persons of ordinary skill in the art. Unless contrary to physical possibility, it is envisioned that (i) the methods / steps described herein may be performed in any sequence and / or in any combination, and (ii) the components of respective embodiments may be combined in any manner.

[0085] Although the subject matter has been described in language specific to structural features and / or acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as examples of implementing the claims, and other equivalent features and acts are intended to be within the scope of the claims.

[0086] Conditional language, such as, among others, “can,” “could,” “might,” or “may,” unless specifically stated otherwise, or otherwise understood within the context as used, is generally intended to convey that certain embodiments include, while other embodiments do not include, certain features, elements and / or steps. Thus, such conditional language is not generally intended to imply that features, elements and / or steps are in any way required for one or more embodiments or that one or more embodiments necessarily include logic for deciding, with or without user input or prompting, whether these features, elements and / or steps are included or are to be performed in any particular embodiment.

[0087] Unless the context clearly requires otherwise, throughout the description and the claims, the words “comprise,” “comprising,” and the like are to be construed in an inclusivesense, as opposed to an exclusive or exhaustive sense, i.e., in the sense of “including, but not limited to.” As used herein, the terms "connected," "coupled," or any variant thereof means any connection or coupling, either direct or indirect, between two or more elements; the coupling or connection between the elements can be physical, logical, or a combination thereof. Additionally, the words “herein,” “above,” "below," and words of similar import, when used in this application, refer to this application as a whole and not to any particular portions of this application. Where the context permits, words using the singular or plural number may also include the plural or singular number respectively. The word "or" in reference to a list of two or more items, covers all of the following interpretations of the word: any one of the items in the list, all of the items in the list, and any combination of the items in the list. Likewise the term “and / or” in reference to a list of two or more items, covers all of the following interpretations of the word: any one of the items in the list, all of the items in the list, and any combination of the items in the list.

[0088] Conjunctive language such as the phrase “at least one of X, Y and Z,” unless specifically stated otherwise, is otherwise understood with the context as used in general to convey that an item, term, etc. may be either X, Y or Z, or any combination thereof. Thus, such conjunctive language is not generally intended to imply that certain embodiments require at least one of X, at least one of Y and at least one of Z to each be present. Further, use of the phrase “at least one of X, Y or Z” as used in general is to convey that an item, term, etc. may be either X, Y or Z, or any combination thereof.

[0089] In some embodiments, certain operations, acts, events, or functions of any of the algorithms described herein can be performed in a different sequence, can be added, merged, or left out altogether (e.g., not all are necessary for the practice of the algorithms). In certain embodiments, operations, acts, functions, or events can be performed concurrently, e.g., through multi-threaded processing, interrupt processing, or multiple processors or processor cores or on other parallel architectures, rather than sequentially.

[0090] Systems and modules described herein may comprise software, firmware, hardware, or any combination(s) of software, firmware, or hardware suitable for the purposes described. Software and other modules may reside and execute on servers,workstations, personal computers, computerized tablets, PDAs, and other computing devices suitable for the purposes described herein. Software and other modules may be accessible via local computer memory, via a network, via a browser, or via other means suitable for the purposes described herein. Data structures described herein may comprise computer files, variables, programming arrays, programming structures, or any electronic information storage schemes or methods, or any combinations thereof, suitable for the purposes described herein. User interface elements described herein may comprise elements from graphical user interfaces, interactive voice response, command line interfaces, and other suitable interfaces.

[0091] Further, processing of the various components of the illustrated systems can be distributed across multiple machines, networks, and other computing resources. Two or more components of a system can be combined into fewer components. Various components of the illustrated systems can be implemented in one or more virtual machines, rather than in dedicated computer hardware systems and / or computing devices. Likewise, the data may be stored in physical and / or logical data storage, including, e.g., storage area networks or other distributed storage systems. Moreover, in some embodiments the connections between the components shown represent possible paths of data flow, rather than actual connections between hardware. While some examples of possible connections are shown, any of the subset of the components shown can communicate with any other subset of components in various implementations.

[0092] Embodiments are also described above with reference to flow chart illustrations and / or block diagrams of methods, apparatus (systems) and computer program products. Each block of the flow chart illustrations and / or block diagrams, and combinations of blocks in the flow chart illustrations and / or block diagrams, may be implemented by computer program instructions. Such instructions may be provided to a processor of a general purpose computer, special purpose computer, specially-equipped computer (e.g., comprising a high-performance database server, a graphics subsystem, etc.) or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor(s) of the computer or other programmable data processing apparatus, create means for implementing the acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions may also be stored in a non-transitory computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the acts specified in the flow chart and / or block diagram block or blocks. The computer program instructions may also be loaded to a computing device or other programmable data processing apparatus to cause operations to be performed on the computing device or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computing device or other programmable apparatus provide steps for implementing the acts specified in the flow chart and / or block diagram block or blocks.

[0093] Any patents and applications and other references noted above, including any that may be listed in accompanying filing papers, are incorporated herein by reference. Aspects of the invention can be modified, if necessary, to employ the systems, functions, and concepts of the various references described above to provide yet further implementations of the invention. These and other changes can be made to the invention in light of the above Detailed Description. While the above description describes certain examples of the invention, the invention can be practiced in many ways. Details of the system may vary considerably in its specific implementation, while still being encompassed by the invention disclosed herein. As noted above, particular terminology used when describing certain features or aspects of the invention should not be taken to imply that the terminology is being redefined herein to be restricted to any specific characteristics, features, or aspects of the invention with which that terminology is associated. In general, the terms used in the following claims should not be construed to limit the invention to the specific examples disclosed in the specification, unless the above Detailed Description section explicitly defines such terms. Accordingly, the actual scope of the invention encompasses not only the disclosed examples, but also all equivalent ways of practicing or implementing the invention under the claims.

[0094] To reduce the number of claims, certain aspects of the invention are presented below in certain claim forms, but the applicant contemplates other aspects of the inventionin any number of claim forms. For example, while only one aspect of the invention may be recited as a means-plus-function claim under 35 U.S.C sec. 112(f) (AIA), other aspects may likewise be embodied as a means-plus-function claim, or in other forms, such as being embodied in a computer-readable medium. Any claims intended to be treated under 35 U.S.C. §112(f) will begin with the words “means for,” but use of the term “for” in any other context is not intended to invoke treatment under 35 U.S.C. §112(f). Accordingly, the applicant reserves the right to pursue additional claims after filing this application, in either this application or in a continuing application.Examples

[0095] Various example embodiments of the disclosure can be described by the following clauses:

[0096] Clause 1. A method comprising receiving, from an imager, measurement data corresponding to a first volumetric image; generating, using a machine-learning model, a plurality of second volumetric images based on the measurement data; receiving an indication to display a particular image slice generated based on at least a portion of the measurement data; identifying, based on the indication, a plurality of second image slices from the plurality of second volumetric images, wherein the plurality of second image slices correspond to the particular image slice; and causing a graphical user interface to display a display object based on the plurality of second image slices.

[0097] Clause 2. The method of Clause 1 , wherein the measurement data is computed tomography (CT) data.

[0098] Clause 3. The method of any of Clauses 1 -2, wherein generating the plurality of second volumetric images comprises generating a plurality of image slices from the measurement data; communicating each of the plurality of image slices to the machinelearning model; for each of the plurality of image slices, receiving a set of second image slices; and generating the plurality of second volumetric images based on the plurality of sets of second image slices, wherein at least one second volumetric image of the plurality of second volumetric images includes at least one second image slice from each of the plurality of sets of second image slices.

[0099] Clause 4. The method of Clause 3, wherein the particular image slice of the first volumetric image is along a different plane than the plurality of image slices.

[0100] Clause 5. The method of Clause 3, wherein each second image slice of the set of second image slices is different.

[0101] Clause 6. The method of any of Clauses 3-5, wherein the machine-learning model generates each of the set of second image slices using a stochastic process.

[0102] Clause 7. The method of any of the preceding Clauses, wherein each of the plurality of second image slices is different.

[0103] Clause 8. The method of any of the preceding Clauses, wherein causing a display to display a display object based on the plurality of second image slices comprises displaying a sequence of the plurality of second image slices.

[0104] Clause 9. The method of any of the preceding Clauses, wherein causing a display to display a display object based on the plurality of second image slices comprises generating a video based on the plurality of second image slices and causing the display to display the video.

[0105] Clause 10. The method of any of the preceding Clauses, further comprising determining a plurality of differences between the plurality of second image slices, wherein causing a display to display a display object based on the plurality of second image slices comprises generating a composite image based on the plurality of second image slices and causing the display to display the composite image and an indication of at least one difference of the plurality of differences on the composite image.

[0106] Clause 11 . The method of any of the preceding Clauses, wherein causing a display to display a display object based on the plurality of second volumetric images comprises generating a composite image based on the plurality of second volumetric images and causing the display to display the composite image.

[0107] Clause 12. A system comprising at least one processor configured to receive, from an imager, measurement data corresponding to a first volumetric image; generate, using a machine-learning model, a plurality of second volumetric images based on themeasurement data; receive an indication to display a particular image slice; identify, based on the indication, a plurality of second image slices from the plurality of second volumetric images, wherein the plurality of second image slices correspond to the particular image slice; and cause a display to display a display object based on the plurality of second image slices.

[0108] Clause 13. The system of Clause 12, wherein the measurement data is computed tomography (CT) data.

[0109] Clause 14. The system of any of Clauses 12-13, wherein to generate the plurality of second volumetric images, the at least one processor is configured to generate a plurality of image slices from the measurement data; communicate each of the plurality of image slices to the machine-learning model; for each of the plurality of image slices, receive a set of second image slices; and generate the plurality of second volumetric images based on the plurality of sets of second image slices, wherein at least one second volumetric image of the plurality of second volumetric images includes at least one second image slice from each of the plurality of sets of second image slices.

[0110] Clause 15. The system of Clause 14, wherein the particular image slice is along a different plane than the plurality of image slices.

[0111] Clause 16. The system of Clause 14, wherein each second image slice of the set of second image slices is different.

[0112] Clause 17. The system of Clause 14, wherein the machine-learning model generates each of the set of second image slices using a stochastic process.

[0113] Clause 18. The system of any of Clauses 12-17, wherein each of the plurality of second image slices is different.

[0114] Clause 19. The system of any of Clauses 12-18, wherein the at least one processor is further configured to cause a display to display a display object based on the plurality of second image slices by displaying a sequence of the plurality of second image slices.

[0115] Clause 20. The system of any of Clauses 12-19, wherein the at least one processor is further configured to cause a display to display a display object based on the pluralityof second image slices by generating a video based on the plurality of second image slices and causing the display to display the video.

[0116] Clause 21 . The system of any of Clauses 12-20, wherein the at least one processor is further configured to determine a plurality of differences between the plurality of second image slices, and wherein to cause the display to display the display object based on the plurality of second image slices, the at least one processor is configured to generate a composite image based on the plurality of second image slices and cause the display to display the composite image and an indication of at least one difference between at least two of the plurality of differences on the composite image.

[0117] Clause 22. The system of any of Clauses 12-21 , wherein to cause the display to display the display object based on the plurality of second image slices, the at least one processor is configured to generate a composite image based on the plurality of second volumetric images and cause the display to display the composite image.

[0118] Clause 23. A computer-readable medium comprising computer-executable instructions that, when executed by at least one processor, cause the at least one processor to receive, from an imager, measurement data corresponding to a first volumetric image; generate, using a machine-learning model, a plurality of second volumetric images based on the measurement data; receive an indication to display a particular image slice; identify, based on the indication, a plurality of second image slices from the plurality of second volumetric images, wherein the plurality of second image slices correspond to the particular image slice; and cause a display to display a display object based on the plurality of second image slices.

[0119] Clause 24. A system comprising at least one processor and at least one computer- readable medium storing computer-executable instructions that, when executed by the at least one processor, cause the at least one processor to receive, from a machine-learning model, a plurality of first volumetric images, wherein the plurality of first volumetric images are generated by the machine-learning model from a second volumetric image; receive an indication to display a particular image slice of the second volumetric image; identify, based on the indication, a plurality of first image slices from the plurality of first volumetricimages, wherein the plurality of first image slices correspond to the particular image slice; and cause a display to display a display object based on the plurality of first image slices.

[0120] Clause 25. The system of Clause 24, wherein the measurement data is computed tomography (CT) data.

[0121] Clause 26. The system of any of Clauses 24-25, wherein each of the plurality of first image slices is different.

[0122] Clause 27. The system of any of Clauses 24-26, wherein to cause the display to display the display object based on the plurality of first image slices, the computerexecutable instructions cause the at least one processor to display a sequence of the plurality of first image slices.

[0123] Clause 28. The system of any of Clauses 24-27, wherein to cause the display to display the display object based on the plurality of first image slices, the computerexecutable instructions cause the at least one processor to generate a video based on the plurality of first image slices and cause the display to display the video.

[0124] Clause 29. The system of any of Clauses 24-28, wherein the computer-executable instructions further cause the at least one processor to determine a plurality of differences between the plurality of first image slices, and wherein to cause the display to display the display object based on the plurality of first image slices, the computer-executable instructions cause the at least one processor to generate a composite image based on the plurality of first image slices and cause the display to display the composite image and an indication of at least one difference of the plurality of differences on the composite image.

[0125] Clause 30. The system of any of Clauses 24-29, wherein to cause the display to display the display object based on the plurality of first image slices, the computerexecutable instructions cause the at least one processor to generate a composite image based on the plurality of first volumetric images and cause the display to display the composite image.

Claims

WHAT IS CLAIMED1 . A method, comprising: receiving, from an imager, measurement data corresponding to a first volumetric image; generating, using a machine-learning model, a plurality of second volumetric images based on the measurement data, receiving an indication to display a particular image slice generated based on at least a portion of the measurement data; identifying, based on the indication, a plurality of second image slices from the plurality of second volumetric images, wherein the plurality of second image slices correspond to the particular image slice; and causing a display to display a display object based on the plurality of second image slices.

2. The method of claim 1 , wherein the measurement data is computer tomography (CT) data.

3. The method of claim 1 , wherein generating the plurality of second volumetric images comprises: generating a plurality of image slices from the measurement data, communicating each of the plurality of image slices to the machine-learning model, for each of the plurality of image slices, receiving a set of second image slices, and generating the plurality of second volumetric images based on the plurality of sets of second image slices, wherein at least one second volumetric image of the plurality of second volumetric images includes at least one second image slice from each of the plurality of sets of second image slices.

4. The method of claim 3, wherein the particular image slice is along a different plane than the plurality of image slices.

5. The method of claim 3, wherein each second image slice of the set of second image slices is different.

6. The method of claim 3, wherein the machine-learning model generates each of the set of second image slices using a stochastic process.

7. The method of claim 1 , wherein each of the plurality of second image slices are different.

8. The method of claim 1 , causing a display to display a display object based on the plurality of second image slices comprises displaying a sequence of the plurality of second image slices.

9. The method of claim 1 , causing a display to display a display object based on the plurality of second image slices comprises generating a video based on the plurality of second image slices, and causing the display to display the video.

10. The method of claim 1 , further comprising: determining a plurality of differences between the plurality of second image slices, wherein causing a display to display a display object based on the plurality of second image slices comprises generating a composite image based on the plurality of second image slices, and causing the display to display the composite image and an indication of at least one difference of the plurality of differences on the composite image.11 . The method of claim 1 , wherein causing a display to display a display object based on the plurality of second image slices comprises generating a composite image based on the plurality of second image slices, and causing the display to display the composite image.

12. A system, comprising: at least one processor configured to:receive, from an imager, measurement data corresponding to a first volumetric image; generate, using a machine-learning model, a plurality of second volumetric images based on the measurement data, receive an indication to display a particular image slice generated based on at least a portion of the measurement data; identify, based on the indication, a plurality of second image slices from the plurality of second volumetric images, wherein the plurality of second image slices correspond to the particular image slice; and cause a display to display a display object based on the plurality of second image slices.

13. The system of claim 12, wherein the measurement data is computer tomography (CT) data.

14. The system of claim 12, wherein to generate the plurality of second volumetric images, the at least one processor is configured to: generate a plurality of image slices from the measurement data, communicate each of the plurality of image slices to the machine-learning model, for each of the plurality of image slices, receive a set of second image slices, and generate the plurality of second volumetric images based on the plurality of sets of second image slices, wherein at least one second volumetric image of the plurality of second volumetric images includes at least one second image slice from each of the plurality of sets of second image slices.

15. The system of claim 14, wherein the particular image slice is along a different plane than the plurality of image slices.

16. The system of claim 14, wherein each second image slice of the set of second image slices is different.

17. The system of claim 14, wherein the machine-learning model generates each of the set of second image slices using a stochastic process.

18. The system of claim 12, wherein each of the plurality of second image slices are different.

19. The system of claim 12, wherein the at least one processor is further configured to cause a display to display a display object based on the plurality of second image slices comprises displaying a sequence of the plurality of second image slices.

20. The system of claim 12, wherein the at least one processor is further configured to cause a display to display a display object based on the plurality of second image slices comprises generating a video based on the plurality of second image slices, and causing the display to display the video.21 . The system of claim 12, wherein the at least one processor is further configured to: determine a plurality of differences between the plurality of second image slices, wherein to cause the display to display the display object based on the plurality of second image slices, the at least one processor is configured to generate a composite image based on the plurality of second image slices, and cause the display to display the composite image and an indication of at least one difference between at least two of the plurality of differences on the composite image.

22. The system of claim 12, wherein to cause the display to display the display object based on the plurality of second image slices, the at least one processor is configured to generate a composite image based on the plurality of second image slices, and cause the display to display the composite image.

23. A computer-readable medium comprising computer-executable instructions that when executed by at least one processor cause the at least one processor to: receive, from an imager, measurement data corresponding to a first volumetric image; generate, using a machine-learning model, a plurality of second volumetric images based on the measurement data, receive an indication to display a particular image slice generated based on at least a portion of the measurement data;identify, based on the indication, a plurality of second image slices from the plurality of second volumetric images, wherein the plurality of second image slices correspond to the particular image slice; and cause a display to display a display object based on the plurality of second image slices.

24. A system, comprising: at least one processor; and at least one computer-readable media storing computer-executable instructions that when executed by the at least one processor, cause the at least one processor to: receive, from a machine-learning model, a plurality of second volumetric images, wherein the plurality of second volumetric images are generated by the machine-learning model based on measurement data; receive an indication to display a particular image slice based on at least a portion of the measurement data; identify, based on the indication, a plurality of second image slices from the plurality of second volumetric images, wherein the plurality of second image slices correspond to the particular image slice; and cause a display to display a display object based on the plurality of second image slices.

25. The system of claim 24, wherein the measurement data is computer tomography (CT) data.

26. The system of claim 24, wherein each of the plurality of second image slices are different.

27. The system of claim 24, wherein to cause the display to display the display object based on the plurality of second image slices, the computer-executable instructions cause the at least one processor to display a sequence of the plurality of second image slices.

28. The system of claim 24, wherein to cause the display to display the display object based on the plurality of second image slices, the computer-executable instructions cause the at least one processor to generate a video based on the plurality of second image slices, and cause the display to display the video.

29. The system of claim 24, wherein the computer-executable instructions further cause the at least one processor to: determine a plurality of differences between the plurality of second image slices, wherein to cause the display to display the display object based on the plurality of second image slices, the computer-executable instructions cause the at least one processor to generate a composite image based on the plurality of second image slices, and cause the display to display the composite image and an indication of at least one difference of the plurality of differences on the composite image.

30. The system of claim 24, wherein to cause the display to display the display object based on the plurality of second image slices, the computer-executable instructions cause the at least one processor to generate a composite image based on the plurality of second image slices, and cause the display to display the composite image.

Images