Abstract: A trained recurrent neural network having a set of control policies learned from application of a template dataset and one or more corresponding template deep network architectures may generate a deep network architecture for performing a task on an application dataset. The template deep network architectures may have an established level or performance in executing the task. A deep network based on the deep network architecture may trained to perform the task on the application dataset. The control policies of the recurrent neural network may be updated based on the performance of the trained deep network.

Abstract: Imaging from sequential scans is aligned based on patient information. A three-dimensional distribution of a patient-related object or objects, such as an outer surface of the patient or an organ in the patient, is stored with any results (e.g., images and/or measurements). Rather than the entire scan volume, the three-dimensional distributions from the different scans are used to align between the scans. The alignment allows diagnostically useful comparison between the scans, such as guiding an imaging technician to more rapidly determine the location of a same lesion for size comparison.

Abstract: Machine learning is used to train a network to predict the location of an internal body marker from surface data. A depth image or other image of the surface of the patient is used to determine the locations of anatomical landmarks. The training may use a loss function that includes a term to limit failure to predict a landmark and/or off-centering of the landmark. The landmarks may then be used to configure medical scanning and/or for diagnosis.

Abstract: A system includes: a movable X-ray tube scanner; a range sensor movable with the X-ray tube scanner; an X-ray detector positioned to detect X-rays from the X-ray tube passing through a standing subject between the X-ray tube and the X-ray detector; and a processor configured for automatically controlling the X-ray tube scanner to transmit X-rays to a region of interest of the patient while the subject is standing between the X-ray tube and the X-ray detector.

Abstract: A method and system for classification of endoscopic images is disclosed. An initial trained deep network classifier is used to classify endoscopic images and determine confidence scores for the endoscopic images. The confidence score for each endoscopic image classified by the initial trained deep network classifier is compared to a learned confidence threshold. For endoscopic images with confidence scores higher than the learned threshold value, the classification result from the initial trained deep network classifier is output. Endoscopic images with confidence scores lower than the learned confidence threshold are classified using a first specialized network classifier built on a feature space of the initial trained deep network classifier.

Abstract: A method and a system for automatically aligning a positionable X-ray source of an X-ray system in alignment with a mobile X-ray detector is disclosed where the X-ray system detects the position of the mobile X-ray detector using a 3D camera and then driving the positionable X-ray source to a position in alignment with the mobile X-ray detector.

Abstract: A system and method includes operation of a generation network to generate first generated computed tomography data based on a first instance of surface data, determination of a generation loss based on the first generated computed tomography data and on a first instance of computed tomography data which corresponds to the first instance of surface data, operation of a discriminator network to discriminate between the first generated computed tomography data and the first instance of computed tomography data, determination of a discriminator loss based on the discrimination between the first generated computed tomography data and the first instance of computed tomography data, determination of discriminator gradients of the discriminator network based on the discriminator loss, and updating of weights of the generation network based on the generation loss and the discriminator gradients.

Abstract: A method and system for anatomical landmark detection in medical images using deep neural networks is disclosed. For each of a plurality of image patches centered at a respective one of a plurality of voxels in the medical image, a subset of voxels within the image patch is input to a trained deep neural network based on a predetermined sampling pattern. A location of a target landmark in the medical image is detected using the trained deep neural network based on the subset of voxels input to the trained deep neural network from each of the plurality of image patches.

Abstract: A method of obtaining a medical image includes obtaining, via a camera, at least one surface image of a patient. A pose of the patient is determined from the at least one surface image of the patient using at least one spatial information module. The patient is positioned, via a moveable bed, to an imaging start position and a medical image of the patient is obtained using a medical imaging modality.

Abstract: A system and method includes generation of a first map of first descriptors based on pixels of a first two-dimensional depth image, where a location of each first descriptor in the first map corresponds to a location of a respective pixel of a first two-dimensional depth image, generation of a second map of second descriptors based on pixels of the second two-dimensional depth image, where a location of each second descriptor in the second map corresponds to a location of a respective pixel of the second two-dimensional depth image, upsampling of the first map of descriptors using a first upsampling technique to generate an upsampled first map of descriptors, upsampling of the second map of descriptors using a second upsampling technique to generate an upsampled second map of descriptors, generation of a descriptor difference map based on differences between descriptors of the upsampled first map of descriptors and descriptors of the upsampled second map of descriptors, generation of a geodesic preservation m

Abstract: A cloud security system and method designed to protect users' data in case of accidental leaks in a cloud computing environment. Secured hashing of the names of folders stored on the cloud data storage are generated and persisted using multiple iterations of cryptographic hash functions along with a concatenated random number for each of the folder names, thereby providing protection against vulnerability of the folder names. The proposed system is a dual-layer framework consisting of a control layer and a data layer. The control layer is responsible for cryptographic hashing and persistence of the folder name, hashed name, salt, and iterations in a database. The control layer communicates with the data layer and provides the hashed folder names to persist the user data cloud storage.

Abstract: A method for training a learning-based medical scanner including (a) obtaining training data from demonstrations of scanning sequences, and (b) learning the medical scanner's control policies using deep reinforcement learning framework based on the training data.

Abstract: A system and method includes operation of a generation network to generate first generated computed tomography data based on a first instance of surface data, determination of a generation loss based on the first generated computed tomography data and on a first instance of computed tomography data which corresponds to the first instance of surface data, operation of a discriminator network to discriminate between the first generated computed tomography data and the first instance of computed tomography data, determination of a discriminator loss based on the discrimination between the first generated computed tomography data and the first instance of computed tomography data, determination of discriminator gradients of the discriminator network based on the discriminator loss, and updating of weights of the generation network based on the generation loss and the discriminator gradients.

Abstract: A method and system for determining fractional flow reserve (FFR) for a coronary artery stenosis of a patient is disclosed. In one embodiment, medical image data of the patient including the stenosis is received, a set of features for the stenosis is extracted from the medical image data of the patient, and an FFR value for the stenosis is determined based on the extracted set of features using a trained machine-learning based mapping. In another embodiment, a medical image of the patient including the stenosis of interest is received, image patches corresponding to the stenosis of interest and a coronary tree of the patient are detected, an FFR value for the stenosis of interest is determined using a trained deep neural network regressor applied directly to the detected image patches.

Abstract: A method for estimating a body surface model of a patient includes: (a) segmenting, by a computer processor, three-dimensional sensor image data to isolate patient data from environmental data; (b) categorizing, by the computer processor, a body pose of the patient from the patient data using a first trained classifier; (c) parsing, by the computer processor, the patient data to an anatomical feature of the patient using a second trained classifier, wherein the parsing is based on a result of the categorizing; and (d) estimating, by the computer processor, the body surface model of the patient based on a result of the parsing. Systems for estimating a body surface model of a patient are described.

Abstract: A method for generating a nuclear image includes obtaining, via a camera, a surface image of a patient. A synthetic computed-tomography (CT) image of the patient is generated based on the surface image. First time-of-flight (TOF) data for the patient is obtained via a nuclear imaging modality. Attenuation correction is applied to the first TOF data. The synthetic image is applied as a density map during the attenuation correction. A nuclear image is generated from the attenuation corrected first TOF data.

Abstract: A correspondence between frames of a set of medical image data is determined where the set of medical image data includes at least one frame acquired without contrast medium and at least one frame acquired with contrast medium. First data representing a first image frame acquired without contrast medium is received. Second data representing a second image frame acquired with contrast medium is received. A position of a feature of a medical device in the second image frame is determined at least partly on the basis of a position of the feature determined from the first image frame.

Abstract: A method of deriving one or more medical scene model characteristics for use by one or more software applications is disclosed. The method includes receiving one or more sensor data streams. Each sensor data stream of the one or more sensor data steams includes position information relating to a medical scene. A medical scene model including a three-dimensional representation of a state of the medical scene is dynamically updated based on the one or more sensor data streams. Based on the medical scene model, the one or more medical scene model characteristics are derived.

Abstract: Systems, methods, and computer-readable media are disclosed for determining feature representations of 2.5D image data using deep learning techniques. The 2.5D image data may be synthetic image data generated from 3D simulated model data such as 3D CAD data. The 2.5D image data may be indicative of any number of pose estimations/camera poses representing virtual or actual viewing perspectives of an object modeled by the 3D CAD data. A neural network such as a convolution neural network (CNN) may be trained using the 2.5D image data as training data to obtain corresponding feature representations. The pose estimations/camera poses may be stored in a data repository in association with the corresponding feature representations. The learnt CNN may then be used to determine an input feature representation from an input 2.5D image and index the input feature representation against the data repository to determine matching pose estimation(s).

Abstract: Systems and methods are provided for automatic segmentation of a vessel. A sequence of image slices containing a vessel is acquired. Features maps are generated for each of the image slices using a trained fully convolutional neural network. A trained bi-directional recurrent neural network generates a segmented image based on the feature maps.