Abstract: Disclosed are apparatuses, systems, and techniques that enable compressed grid-based graph representations for efficient implementations of graph-mapped computing applications. The techniques include but are not limited to selecting a reference grid having a plurality of blocks, assigning nodes of the graph to blocks of the grid, and generating a graph representation that maps directions, relative to the reference grid, of nodal connections of the graph.

Abstract: PET/MR images are compensated with simplified adaptive algorithms for truncated parts of the body. The compensation adapts to a specific location of truncation of the body or organ in the MR image, and to attributes of the truncation in the truncated body part. Anatomical structures in a PET image that do not require any compensation are masked using a MR image with a smaller field of view. The organs that are not masked are then classified as types of anatomical structures, the orientation of the anatomical structures, and type of truncation. Structure specific algorithms are used to compensate for a truncated anatomical structure. The compensation is validated for correctness and the ROI is filled in where there is missing voxel data. Attenuation maps are generated from the compensated ROI.

Abstract: A non-transitory computer-readable medium stores instructions readable and executable by a workstation (18) including at least one electronic processor (20) to perform an image reconstruction method (100). The method includes: determining a weighting parameter (13) of an edge-preserving regularization or penalty of a regularized image reconstruction of an image acquisition device (12) for an imaging data set obtained by the image acquisition device; determining an edge sensitivity parameter (?) of the edge-preserving algorithm for the imaging data set obtained by the image acquisition device; and reconstructing the imaging data set obtained by the image acquisition device to generate a reconstructed image by applying the regularized image reconstruction including the edge-preserving regularization or penalty with the determined weighting and edge sensitivity parameters to the imaging data set obtained by the image acquisition device.

Abstract: Apparatuses, systems, and techniques to train and apply a first machine learning model to identify a plurality of regions of interest within an input image, and to train and apply a plurality of second machine learning models to identify one or more objects within each region of interest identified by the first machine learning model.

Abstract: Apparatuses, systems, and techniques are disclosed to generate a derived artificial intelligence (AI) model from a plurality of AI models. In at least one embodiment, at least one common feature shared among the plurality of AI models are identified, and the derived AI model is generated based on the at least one common feature shared among the plurality of AI models.

Abstract: Apparatuses, systems, and techniques to train and apply a first machine learning model to identify a plurality of regions of interest within an input image, and to train and apply a plurality of second machine learning models to identify one or more objects within each region of interest identified by the first machine learning model.

Abstract: Apparatuses, systems, and techniques to collect performance data for one or more computations tasks executed by a plurality of nodes of a computational pipeline and enable optimization of distribution of task execution among the plurality of nodes.

Abstract: Systems and methods for improving the degree to which programs utilize processor resources during execution. A number of different versions of a program are received, as is a set of performance metrics describing desired performance of the program versions. The programs are then analyzed to determine the amount of processor resources used on a particular processor when the programs are executed to meet the performance metrics. At runtime, a program version that meets its performance metrics without exceeding the available processor resources is selected for execution by the processor. Program versions may be versions written to utilize processors in differing manner, such as by adjusting the numerical precision at which operations are performed or stored. If no program version meets its performance metrics without exceeding the available processor resources, the performance metrics may be reduced and program selection may be based on these reduced performance metrics.

Abstract: Apparatuses, systems, and techniques to train and apply a first machine learning model to identify a plurality of regions of interest within an input image, and to train and apply a plurality of second machine learning models to identify one or more objects within each region of interest identified by the first machine learning model.

Abstract: A non-transitory computer-readable medium stores instructions readable and executable by at least one electronic processor (20) to perform an image reconstruction method (100). The method includes: performing iterative image reconstruction of imaging data acquired using an image acquisition device (12); selecting an update image from a plurality of update images produced by the iterative image reconstruction; processing the selected update image to generate a hot spot artifact map; and suppressing hot spots identified by the generated hot spot artifact map in a reconstructed image output by the iterative image reconstruction.

Abstract: A non-transitory computer readable medium storing instructions readable and executable by an imaging workstation (14) including at least one electronic processor (16) to perform a dataset generation method (100) operating on emission imaging data acquired of a patient for one or more axial frames at a corresponding one or more bed positions, the method comprising: (a) identifying a frame of interest from the one or more axial frames; (b) generating simulated lesion data by simulating emission imaging data for the frame of interest of at least one simulated lesion placed in the frame of interest; (c) generating simulated frame emission imaging data by simulating emission imaging data for the frame of interest of the patient; (d) determining a normalization factor comprising a ratio of the value of a quantitative metric for the simulated patient data and the value of the quantitative metric for the emission imaging data acquired of the same patient for the frame of interest; and (e) generating a hybrid data set

Abstract: Iterative reconstruction (20) of imaging data is performed to generate a sequence of update images (22) terminating at a reconstructed image. During the iterative reconstruction, at least one of an update image and a parameter of the iterative reconstruction is adjusted using an adjustment process separate from the iterative reconstruction. In some embodiments using an edge-preserving regularization prior (26), the adjustment process (30) adjusts an edge preservation threshold to reduce gradient steepness above which edge preservation applies for later iterations compared with earlier iterations. In some embodiments, the adjustment process includes determining (36, 38) for each pixel, voxel, or region of a current update image whether its evolution prior to the current update image 22) satisfies an artifact feature criterion. A local noise suppression operation (40) is performed on the pixel, voxel, or region if the evolution satisfies the artifact feature criterion and is not performed otherwise.

Abstract: This disclosure introduces an approach that includes techniques for determining an optimal weighted execution sequence of available reconstruction algorithms using a multi-processor unit. The introduced approach includes executing a series of optimal weighted execution sequence candidates on a representative slice of the image data and comparing their results to select one of the candidates as the optimal weighted execution sequence.

Abstract: A non-transitory computer-readable medium storing instructions readable and executable by a workstation (18) including at least one electronic processor (20) to perform a quality control (QC) method (100). The method includes: receiving a current QC data set acquired by a pixelated detector (14) and one or more prior QC data sets acquired by the pixelated detector; determining stability levels of detector pixels (16) of the pixelated detector over time from the current QC data set and the one or more prior QC data sets; labeling a detector pixel of the pixelated detector as dead when the stability level determined for the detector pixel is outside of a stability threshold range; and displaying, on a display device (24) operatively connected with the workstation, an identification (28) of the detector pixels labelled as dead.

Abstract: A machine learning guided image segmentation process is performed by an electronic processor (10). Image segmentation (22) is performed to generate an initial segmented representation (50) of an anatomical structure in the medical image. Parameters of a geometric shape are fitted (52) to the anatomical structure in the medical image to produce initial fitted shape parameters (54). A classification is assigned for the anatomical structure in the medical image using at least one classifier (60) operating on the initial fitted shape parameters and the initial segmented representation of the anatomical structure. A final segmented representation (72) of the anatomical structure in the medical image is generated by operations including repeating (70) the image segmentation using the classification as prior knowledge. In illustrative embodiments, the anatomical structure is a heart and the geometric shape is an ellipsoid.

Abstract: An imaging data set (22) comprising detected counts along lines of response (LORs) is reconstructed (24) to generate a full-volume image at a standard resolution. A region selection graphical user interface (GUI) (26) is provided via which a user-chosen region of interest (ROI) is defined in the full-volume image, and this is automatically adjusted by identifying an anatomical feature corresponding to the user-chosen ROI and adjusting the user-chosen ROI to improve alignment with that feature. A sub-set (32) of the counts of the imaging data set is selected (30) for reconstructing the ROI, and only the selected sub-set is reconstructed (34) to generate a ROI image (36) representing the ROI at a higher resolution than the standard resolution. A fraction of the sub-set of counts may be reconstructed using different reconstruction algorithms (40) to generate corresponding sample ROI images, and a reconstruction algorithm selection graphical user interface (42) employs these sample ROI images.

Abstract: Apparatuses, systems, and techniques to transform input data for training neural networks. In at least one embodiment, one or more data transforms are identified in a sequence of data transforms and combined into one or more master data transforms to be performed by one or more parallel processing units in order to prepare data for training an untrained neural network.

Abstract: This disclosure introduces an approach that includes techniques for determining an optimal weighted execution sequence of available reconstruction algorithms using a multi-processor unit. The introduced approach includes executing a series of optimal weighted execution sequence candidates on a representative slice of the image data and comparing their results to select one of the candidates as the optimal weighted execution sequence.

Abstract: A diagnostic imaging system retrieves data (206) from a plurality of accessible data sources, the accessible data sources storing data including physiological data describing a subject to be imaged, a nature of a requested diagnostic image, image preferences of a clinician who requested the diagnostic image, and previously reconstructed images of the requested nature of the subject and/or other subjects, reconstruction parameters and/or sub-routines used to reconstruct the previously reconstructed images. The system analyzes (6, 12) the retrieved data to automatically generate reconstruction parameters and/or sub-steps specific to the nature of the requested diagnostic image, the subject, and the clinician image preferences. The system controls a display device (10, 216) to display the generated reconstruction parameters and/or sub-routines to the user for a user selection.

Abstract: A non-transitory computer-readable medium stores instructions readable and executable by at least one electronic processor (20) to perform an image reconstruction method (100). The method includes: performing iterative image reconstruction of imaging data acquired using an image acquisition device (12); selecting an update image from a plurality of update images produced by the iterative image reconstruction; processing the selected update image to generate a hot spot artifact map; and suppressing hot spots identified by the generated hot spot artifact map in a reconstructed image output by the iterative image reconstruction.