Abstract: Systems and methods are provided for correcting motion artifacts in magnetic resonance images. An image-to-image neural network is used to generate motion corrected magnetic resonance data given motion corrupted magnetic resonance data. The image-to-image neural network is coupled within an adversarial network to help refine the generated magnetic resonance data. The adversarial network includes a generator network (the image-to-image neural network) and a discriminator network. The generator network is trained to minimize a loss function based on a Wasserstein distance when generating MR data. The discriminator network is trained to differentiate the motion corrected MR data from motion artifact free MR data.

Abstract: A learning-based magnetic resonance fingerprinting (MRF) reconstruction method for reconstructing an MR image of a tissue space in an MR scan subject for a particular MR sequence is disclosed. The method involves using a machine-learning algorithm that has been trained to generate a set of tissue parameters from acquired MR signal evolution without using a dictionary or dictionary matching.

Abstract: Systems and methods are provided for generating a protocol independent image. A deep learning generative framework learns to recognize the boundaries and classification of tissues in an MRI image. The deep learning generative framework includes an encoder, a decoder, and a discriminator network. The encoder is trained using the discriminator network to generate a latent space that is invariant to protocol and the decoder is trained to generate the best output possible for brain and/or tissue extraction.

Abstract: A generative network is used for lung lobe segmentation or lung fissure localization, or for training a machine network for lobar segmentation or localization. For segmentation, deep learning is used to better deal with a sparse sampling of training data. To increase the amount of training data available, an image-to-image or generative network localizes fissures in at least some of the samples. The deep-learnt network, fissure localization, or other segmentation may benefit from generative localization of fissures.

Abstract: Deep reinforcement machine learning is used to control denoising (e.g., image regularizer) in iterative reconstruction for MRI compressed sensing. Rather than requiring different machine-learnt networks for different scan settings (e.g., acceleration of the MR compressed sensing), reinforcement learning creates a policy of actions to provide denoising and data fitting through iterations of the reconstruction given a range of different scan settings. This allows a user to scan as appropriate for the patient, the MR system, the application, and/or preferences while still providing an optimized reconstruction under sampling resulting from the MR compressed sensing.

Abstract: For classifying magnetic resonance image quality or training to classify magnetic resonance image quality, deep learning is used to learn features distinguishing between corrupt images base on simulation and measured similarity. The deep learning uses synthetic data without quality annotation, allowing a large set of training data. The deep-learned features are then used as input features for training a classifier using training data annotated with ground truth quality. A smaller training data set may be needed to train the classifier due to the use of features learned without the quality annotation.

Abstract: For determination of motion artifact in MR imaging, motion of the patient in three dimensions is used with a measurement k-space line order based on one or more actual imaging sequences to generate training data. The MR scan of the ground truth three-dimensional (3D) representation subjected to 3D motion is simulated using the realistic line order. The difference between the resulting reconstructed 3D representation and the ground truth 3D representation is used in machine-based deep learning to train a network to predict motion artifact or level given an input 3D representation from a scan of a patient. The architecture of the network may be defined to deal with anisotropic data from the MR scan.

Abstract: A method for processing medical image data comprises: inputting medical image data to a variational autoencoder configured to reduce a dimensionality of the medical image data to a latent space having one or more latent variables with latent variable values, such that the latent variable values corresponding to an image with no tissue of a target tissue type fit within one or more clusters; determining a probability that the latent variable values corresponding to the medical image data fit within the one or more clusters based on the latent variable values; and determining that a tissue of the target tissue type is present in response to a determination that the medical image data have less than a threshold probability of fitting within any of the one or more clusters based on the latent variable values.

Abstract: A method of reviewing neural scans includes receiving at least one landmark corresponding to an anatomical region. A plurality of images of tissue including the anatomical region is received and a neural network configured to differentiate between healthy tissue and unhealthy tissue within the anatomical region is generated. The neural network is generated by a machine learning process configured to receive the plurality of images of tissue and generate a plurality of weighting factors configured to differentiate between healthy tissue and unhealthy tissue. At least one patient image of tissue including the anatomical region is received and a determination is made by the neural network whether the at least one patient image of tissue includes healthy or unhealthy tissue.

Abstract: System and methods are provided for localizing a target object in a medical image. The medical image is discretized into a plurality of images having different resolutions. For each respective image of the plurality of images, starting from a first image and progressing to a last image with the progression increasing in resolution, a sequence of actions is performed for modifying parameters of a target object in the respective image. The parameters of the target object comprise nonlinear parameters of the target object. The sequence of actions is determined by an artificial intelligence agent trained for a resolution of the respective image to optimize a reward function. The target object is localized in the medical image based on the modified parameters of the target object in the last image.

Abstract: For artifact reduction in a magnetic resonance imaging system, deep learning trains an image-to-image neural network to generate an image with reduced artifact from input, artifacted MR data. For application, the image-to-image network may be applied in real time with a lower computational burden than typical post-processing methods. To handle a range of different imaging situations, the image-to-image network may (a) use an auxiliary map as an input with the MR data from the patient, (b) use sequence metadata as a controller of the encoder of the image-to-image network, and/or (c) be trained to generate contrast invariant features in the encoder using a discriminator that receives encoder features.

Abstract: Automated sequence prediction is provided for a medical imaging session including a self-assessment mechanism. An initial scout sequence is performed of a patient or object. The initial scout sequence is validated. An abbreviated acquisition protocol is performed. The abbreviated acquisition protocol is validated. Additional sequences are performed. The sequences may also be configured based on the analysis of the previous scans using deep learning-based reasoning to select the next appropriate settings and procedures.

Abstract: Systems and methods are provided for generating segmented output from input regardless of the resolution of the input. A single trained network is used to provide segmentation for an input regardless of a resolution of the input. The network is recursively trained to learn over large variations in the input data including variations in resolution. During training, the network refines its prediction iteratively in order to produce a fast and accurate segmentation that is robust across resolution differences that are produced by MR protocol variations.

Abstract: For correction of an image from an imaging system, a deep-learnt generative model is used as a regularlizer in an inverse solution with a physics model of the degradation behavior of the imaging system. The prior model is based on the generative model, allowing for correction of an image without application specific balancing. The generative model is trained from good images, so difficulty gathering problem-specific training data may be avoided or reduced.

Abstract: A generative network is used for lung lobe segmentation or lung fissure localization, or for training a machine network for lobar segmentation or localization. For segmentation, deep learning is used to better deal with a sparse sampling of training data. To increase the amount of training data available, an image-to-image or generative network localizes fissures in at least some of the samples. The deep-learnt network, fissure localization, or other segmentation may benefit from generative localization of fissures.

Abstract: A method of reviewing neural scans includes receiving at least one landmark corresponding to an anatomical region. A plurality of images of tissue including the anatomical region is received and a neural network configured to differentiate between healthy tissue and unhealthy tissue within the anatomical region is generated. The neural network is generated by a machine learning process configured to receive the plurality of images of tissue and generate a plurality of weighting factors configured to differentiate between healthy tissue and unhealthy tissue. At least one patient image of tissue including the anatomical region is received and a determination is made by the neural network whether the at least one patient image of tissue includes healthy or unhealthy tissue.

Abstract: Deep reinforcement machine learning is used to control denoising (e.g., image regularizer) in iterative reconstruction for MRI compressed sensing. Rather than requiring different machine-learnt networks for different scan settings (e.g., acceleration of the MR compressed sensing), reinforcement learning creates a policy of actions to provide denoising and data fitting through iterations of the reconstruction given a range of different scan settings. This allows a user to scan as appropriate for the patient, the MR system, the application, and/or preferences while still providing an optimized reconstruction under sampling resulting from the MR compressed sensing.

Abstract: Systems and methods are provided for automatically designing RF pulses using a reinforcement machine-learnt classifier. Data representing an object and a selected outcome is accessed. A reinforcement learnt method identifies the RF pulse sequence that generates a result within a predefined value of the selected outcome. An MRI scanner images the object using the RF pulse sequence.

Abstract: Systems and methods are provided for correcting motion artifacts in magnetic resonance images. An image-to-image neural network is used to generate motion corrected magnetic resonance data given motion corrupted magnetic resonance data. The image-to-image neural network is coupled within an adversarial network to help refine the generated magnetic resonance data. The adversarial network includes a generator network (the image-to-image neural network) and a discriminator network. The generator network is trained to minimize a loss function based on a Wasserstein distance when generating MR data. The discriminator network is trained to differentiate the motion corrected MR data from motion artifact free MR data.

Abstract: A method and apparatus for automated boundary delineation of a tubular structure in a 3D medical image of a patient using an infinitely recurrent neural network (IRNN) is disclosed. An unraveled cross-section image corresponding to a portion of a tubular structure is extracted from 3D medical image. The unraveled cross-section image is divided into a plurality of image chunks. A boundary of the portion of the tubular structure is detected based on the plurality of image chunks using a trained IRNN. The trained IRNN repeatedly inputs a sequential data stream, including the plurality of image chunks of the unraveled cross-section image, for a plurality of iterations while preserving a memory state between iterations, and detects, for each image chunk of the unraveled cross-section image input to the trained IRNN in the sequential data stream, a corresponding section of the boundary of the portion of the tubular structure.