Abstract: Systems and methods for radiology image classification from noisy images in accordance with embodiments of the invention are illustrated. One embodiment includes noisy image classification device, including a processor, camera circuitry, and a memory containing a noisy image classification application, where the noisy image classification application directs the processor to obtain image data describing a first image taken of a second image using the camera circuitry, where the second image was produced by a medical imaging device, and where the first image is a noisy version of the second image, classify the image data using a neural network trained to be robust to noise, generate an investigation recommendation based on the classification, and provide the investigation recommendation via a display.

Abstract: Method for cutting a three-dimensional model of a dental scene, or “scene model.” The method includes acquiring a view of the scene model, called the “analysis view.” The method includes analyzing the analysis view by a neural network in order to identify, in the analysis view, at least one elementary zone representing an element of the dental scene, and assigning a value to at least one attribute of the elementary zone. The method includes identifying a region of the scene model represented by the elementary zone on the analysis view, and assigning, in the region, a value to an attribute of the scene model in accordance with the value of the attribute of the elementary zone.

Abstract: A system for generating a depth output for an image is described. The system receives input images that depict the same scene, each input image including one or more potential objects. The system generates, for each input image, a respective background image and processes the background images to generate a camera motion output that characterizes the motion of the camera between the input images. For each potential object, the system generates a respective object motion output for the potential object based on the input images and the camera motion output. The system processes a particular input image of the input images using a depth prediction neural network (NN) to generate a depth output for the particular input image, and updates the current values of parameters of the depth prediction NN based on the particular depth output, the camera motion output, and the object motion outputs for the potential objects.

Abstract: A medical image learning method of a medical image process apparatus includes preparing a plurality of body X-ray images for learning as an input of a learning data set, preparing internal diagnostic indicator information corresponding to each of the plurality of body X-ray images for learning as a label of the learning data set, and learning an artificial neural network model using the learning data set. The internal diagnostic indicator information includes information on a cardiovascular border. The information on the cardiovascular border includes information on at least one of an aortic knob, a pulmonary conus, a left atrial appendage, an upper right cardiac border, a lower right cardiac border, a lower left cardiac border, a descending aorta, a carina, an upper end point of a diaphragm, a right pulmonary artery, a posterior cardiac border, and an anterior spinal border.

Abstract: Systems and methods for estimating a gaze vector of an eye using a trained neural network. An input image of the eye may be received from a camera. The input image may be provided to the neural network. Network output data may be generated using the neural network. The network output data may include two-dimensional (2D) pupil data, eye segmentation data, and/or cornea center data. The gaze vector may be computed based on the network output data. The neural network may be previously trained by providing a training input image to the neural network, generating training network output data, receiving ground-truth (GT) data, computing error data based on a difference between the training network output data and the GT data, and modifying the neural network based on the error data.

Abstract: Apparatus and methods related to applying lighting models to images of objects are provided. A neural network can be trained to apply a lighting model to an input image. The training of the neural network can utilize confidence learning that is based on light predictions and prediction confidence values associated with lighting of the input image. A computing device can receive an input image of an object and data about a particular lighting model to be applied to the input image. The computing device can determine an output image of the object by using the trained neural network to apply the particular lighting model to the input image of the object.

Abstract: A deep-learning-based method for metal artifact reduction in CT images includes providing a dataset and a cGAN. The dataset includes CT image pairs, randomly partitioned into a training set, a validation set, and a testing set. Each Pre-CT and Post-CT image pairs is respectively acquired in a region before and after an implant is implanted. The Pre-CT and Post-CT images of each pair are artifact-free CT and artifact-affected CT images, respectively. The cGAN is conditioned on the Post-CT images, includes a generator and a discriminator that operably compete with each other, and is characterized with a training objective that is a sum of an adversarial loss and a reconstruction loss. The method also includes training the cGAN with the dataset; inputting the post-operatively acquired CT image to the trained cGAN; and generating an artifact-corrected image by the trained cGAN, where metal artifacts are removed in the artifact-corrected image.

Abstract: The present application relates to a method and system for generating multi-task learning-type generative adversarial network for low-dose PET reconstruction, and relates to the field of deep learning. The method includes connecting layers of the encoder with layers of the decoder by skip connection to provide a U-Net type picture generator; generating a group of generative adversarial networks by matching a plurality of picture generators with a plurality of discriminators in one-to-one manner; obtaining a first multi-task learning-type generative adversarial network; designing a joint loss function 1 for improving image quality; and training the first multi-task learning-type generative adversarial network according to the joint loss function 1 in combination with an optimizer to provide a second multi-task learning-type generative adversarial network.

Abstract: Computer-implemented method for determining nuclear medical image data sets in dynamic nuclear medical imaging. The method includes using a trained function to determine at least one further nuclear medical image data set for at least one frame if a basis raw data set is taken from one single frame, wherein input data of the trained function includes at least one of the nuclear medical raw data set of the respective frame or a preliminary reconstructed image reconstructed therefrom, and an already determined nuclear medical image data set.

Abstract: Embodiments of this application disclose a brain image segmentation method and apparatus, a network device, and a storage medium. The method includes obtaining, by a device, a to-be-segmented image group comprising a plurality of modal images of a brain. The device includes a memory storing instructions and a processor in communication with the memory. The method further includes performing, by the device, skull stripping according to the plurality of modal images to obtain a skull-stripped mask; separately performing, by the device, feature extraction on the plurality of modal images to obtain extracted features, and fusing the extracted features to obtain a fused feature; segmenting, by the device, encephalic tissues according to the fused feature to obtain an initial segmentation result; and fusing, by the device, the skull-stripped mask and the initial segmentation result to obtain a segmentation result corresponding to the to-be-segmented image group.

Abstract: Disclosed techniques for image processing three-dimensional image data include: obtaining three-dimensional image data representing contiguous slices parallel to a plane, constructing training data from the image data by, for each of a plurality of angles: rotating the image data in the plane to produce rotated image data, blurring the rotated image data in a dimension parallel to the plane to produce low resolution rotated image data, and introducing aliasing into the low resolution rotated image data in the dimension parallel to the plane to produce aliased low resolution rotated image data, training an anti-aliasing neural network with the aliased low resolution image data and the low resolution image data, training a super-resolution neural network with the aliased low resolution image data and the rotated image data, and processing the image data using the trained anti-aliasing neural network and the trained super-resolution neural network to produce processed image data.

Abstract: Embodiments relate to a method for supporting X-ray image reading including receiving information associated with a reading target positioned in a reading space where X-rays pass through or are reflected off, acquiring a non X-RAY image of an item object based on the information associated with the reading target, and generating a fake X-RAY image of the item object by applying the non X-RAY image of the item object to the image transform model, and a system for performing the same.

Abstract: The present disclosure provides a method and apparatus of labeling a target in an image, and a computer recording medium. The method includes: acquiring a first neural network, the first neural network includes a multi-layer convolutional neural network and a fully connected layer, wherein each layer of the multi-layer convolutional neural network includes a convolutional layer, an activation function layer and a down-sampling layer arranged successively; processing the image by using the multi-layer convolutional neural network of the first neural network acquired so as to obtain a target position mask for the image; and labeling the target in the image based on the target position mask.

Abstract: An AI system may receive an image. The AI system may include a first AI model trained using labeled training images including images from prior mammograms to predict cancer and a second AI model trained using labeled training images including images from current mammograms to classify mammogram images. The second AI model may be initialized using the weights of the first AI model using transfer learning. The AI system may receive a classification output indicating a likely current breast cancer diagnosis or a likelihood of the user to develop breast cancer in the future.

Abstract: There is provided a method, comprising: accessing medical images of subjects, depicting contrast phases of contrast administered to the respective subject, accessing for a first subset of the medical images, metadata indicating a respective contrast phase, wherein a second subset of the medical images are unassociated with metadata, mapping each respective contrast phase of the contrast phases to a respective time interval indicating estimated amount of time from a start of contrast administration to time of capture of the respective medical image, creating a training dataset, by labelling images of the first subset with a label indicating the respective time interval, and including the second subset as non-labelled images, and training the ML model using the training dataset for generating an outcome of a target time interval indicating estimated amount of time from the start of contrast administration, in response to an input of a target medical image.

Abstract: Method of segmenting anatomical structures such as organs in 3D scans in an architecture that combines U-net, time-distributed convolutions and bidirectional convolutional LSTM.

Abstract: Techniques for enhancing image segmentation with the integration of deep learning are disclosed herein. An example method for atlas-based segmentation using deep learning includes: applying a deep learning model to a subject image to identify an anatomical feature, registering an atlas image to the subject image, using the deep learning segmentation data to improve a registration result, generating a mapped atlas, and identifying the feature in the subject image using the mapped atlas. Another example method for training and use of a trained machine learning classifier, in an atlas-based segmentation process using deep learning, includes: applying a deep learning model to an atlas image, training a machine learning model classifier using data from applying the deep learning model, estimating structure labels of areas of the subject image, and defining structure labels by combining the estimated structure labels with labels produced from atlas-based segmentation on the subject image.

Abstract: A machine-based learning method estimates a probability of bone fractures in a 3D image, more specifically vertebral fractures. The method and system utilizing such method utilize a data-driven computational model to learn 3D image features for classifying vertebra fractures. A three-dimensional medical image analysis system for predicting a presence of a vertebral fracture in a subject includes a 3D image processor for receiving and processing 3D image data of a 3D image of the subject, producing two or more sets of 3D voxels. Each of the sets of 3D voxels corresponds to an entirety of the 3D image and each of the sets of 3D voxels consists of equal 3D voxels of different dimensions. The system also includes a voxel classifier for assigning the 3D voxels one or more class probabilities each of the 3D voxels contains a fracture using a computational model, and a fracture probability estimator for estimating a probability of the presence of a vertebral fracture in the subject.

Abstract: A segmentation method for tumor regions in a pathological image of clear cell renal cell carcinoma based on deep learning includes data acquisition and pre-processing, building and training of a classification network SENet and prediction of tumor regions. The present invention studies clear cell renal cell carcinoma based on pathological images, yielding results with higher reliability than judgments made based on CT or MRI images. The present invention overcomes the drawback that the previous research on clear cell renal cell carcinoma is only limited to judgment on presence by being able to visually provide the position and size of tumor regions, which is convenient for the medical profession to better study the pathogenesis and directions to the treatment of clear cell renal cell carcinoma.

Abstract: To provide a technology of more accurately detecting spoofing in face authentication, without increasing a scale of a device configuration and a burden on a user. A spoofing detection device includes a facial image sequence acquisition unit, a line-of-sight change detection unit, a presentation information display unit, and a spoofing determination unit. The facial image sequence acquisition unit acquires a facial image sequence indicating the face of a user. The line-of-sight change detection unit detects information about a temporal change in the line-of-sight from the facial image sequence. The presentation information display unit displays presentation information presented to the user as part of an authentication process. The spoofing determination unit determines the likelihood of the face indicated by the facial image sequence being spoofing on the basis of the information about the temporal change in the line-of-sight with respect to the presentation information.