Abstract: A framework for visual explanation of classification. The framework trains (204) a generative model to generate new images that resemble input images but are classified by the classifier as belonging to one or more alternate classes. At least one explanation mask may then be generated (206) by performing optimization based on a current input image and a new image generated by the trained generative model from the current input image.

Abstract: A computer-implemented method for classifying a reconstruction includes receiving an uncategorized reconstruction and applying a trained classification function configured to classify the uncategorized reconstruction into one of a plurality of categories. The plurality of categories are based on a labeled data-set including a plurality of labeled reconstructions. The trained classification function uses reconstruction-invariant features for classification. The method further includes storing a label indicating a selected one of the plurality of categories for the uncategorized reconstruction.

Abstract: Systems and methods for detecting and classifying clinical features in medical images are disclosed. Natural language processes are applied to speech received from a dictation system to determine clinical and anatomical information for a medical image being viewed. In some examples, gaze location information identifying an eye position is received, as well as an image position for the medical image being viewed. Features of the medical image are detected and classified based on machine learning models. Anatomical associations are generated based on one or more of the classifications, the anatomical information, the gaze information, and the image position. The machine learning models can be trained based on the anatomical associations. In some examples, reports are generated based on the anatomical associations.

Abstract: A method for parametric image reconstruction and motion correction using whole-body motion fields includes receiving a nuclear imaging data set including a set of dynamic frames and generating at least one of a whole-body forward motion field and/or a whole-body inverse motion field for at least one frame in the set of frames. An iterative loop is applied to update at least one parameter used in a direct parametric reconstruction and at least one parametric image is generated based on the at least one parameter updated by the iterative loop. The iterative loop includes calculating a frame emission image for the at least one frame, generating a motion-corrected frame emission image based on the at least one whole-body forward motion field or a whole-body inverse motion field, and updating at least one parameter by applying a fit to the motion-corrected frame emission image.

Abstract: A method for parametric image reconstruction and motion correction using whole-body motion fields includes receiving a nuclear imaging data set including a set of dynamic frames and generating at least one of a whole-body forward motion field and/or a whole-body inverse motion field for at least one frame in the set of frames. An iterative loop is applied to update at least one parameter used in a direct parametric reconstruction and at least one parametric image is generated based on the at least one parameter updated by the iterative loop. The iterative loop includes calculating a frame emission image for the at least one frame, generating a motion-corrected frame emission image based on the at least one whole-body forward motion field or a whole-body inverse motion field, and updating at least one parameter by applying a fit to the motion-corrected frame emission image.

Abstract: Systems and methods for detecting and classifying clinical features in medical images are disclosed. Natural language processes are applied to speech received from a dictation system to determine clinical and anatomical information for a medical image being viewed. In some examples, gaze location information identifying an eye position is received, as well as an image position for the medical image being viewed. Features of the medical image are detected and classified based on machine learning models. Anatomical associations are generated based on one or more of the classifications, the anatomical information, the gaze information, and the image position. The machine learning models can be trained based on the anatomical associations. In some examples, reports are generated based on the anatomical associations.

Abstract: A computer-implemented method for classifying a reconstruction includes receiving an uncategorized reconstruction and applying a trained classification function configured to classify the uncategorized reconstruction into one of a plurality of categories. The plurality of categories are based on a labeled data-set including a plurality of labeled reconstructions. The trained classification function uses reconstruction-invariant features for classification. The method further includes storing a label indicating a selected one of the plurality of categories for the uncategorized reconstruction.

Abstract: A method for configuring a neural network comprises: accessing a plurality of three-dimensional (3D) emission image data sets collected by an emission scanner from respective brains of respective subjects; transforming each of the plurality of 3D emission image data sets to a respective two-dimensional (2D) image; cropping portions of each respective 2D image to remove image data corresponding to tissue outside of a striatum of each of the respective brains, to form respective cropped 2D striatum images; and training a neural network to detect a presence of a Parkinsonian syndrome using the cropped 2D striatum images.

Abstract: A method for configuring a neural network comprises: accessing a plurality of three-dimensional (3D) emission image data sets collected by an emission scanner from respective brains of respective subjects; transforming each of the plurality of 3D emission image data sets to a respective two-dimensional (2D) image; cropping portions of each respective 2D image to remove image data corresponding to tissue outside of a striatum of each of the respective brains, to form respective cropped 2D striatum images; and training a neural network to detect a presence of a Parkinsonian syndrome using the cropped 2D striatum images.

Abstract: In a method for retrieval of similar findings from a hybrid image dataset, a database of hotspots is prepared, wherein the hotspots are identified by binary strings encoding descriptors, and identify binary strings stored in the database are identified that resemble a new binary string.

Abstract: In a method for retrieval of similar findings from a hybrid image dataset, a database of hotspots is prepared, wherein the hotspots are identified by binary strings encoding descriptors, and identify binary strings stored in the database are identified that resemble a new binary string.

Abstract: A recognition pipeline automatically partitions a 3D image of the human body into regions of interest (head, rib cage, pelvis, and legs) and correctly labels each region. The 3D image is projected onto a 2D image using volume rendering. The 3D image may contain the whole body region or any subset. In a learning phase, training datasets are manually partitioned and labeled, and a training database is computed. In a recognition phase, images are partitioned and labeled based on the knowledge from the training database. The recognition phase is computationally efficient and may operate under 2 seconds in current conventional computer systems. The recognition is robust to image variations, and does not require the user to provide any knowledge about the contained regions of interest within the image.

Abstract: A recognition pipeline automatically partitions a 3D image of the human body into regions of interest (head, rib cage, pelvis, and legs) and correctly labels each region. The 3D image is projected onto a 2D image using volume rendering. The 3D image may contain the whole body region or any subset. In a learning phase, training datasets are manually partitioned and labeled, and a training database is computed. In a recognition phase, images are partitioned and labeled based on the knowledge from the training database. The recognition phase is computationally efficient and may operate under 2 seconds in current conventional computer systems. The recognition is robust to image variations, and does not require the user to provide any knowledge about the contained regions of interest within the image.