Siemens Healthcare Gmbh

Siemens Medical Solutions Usa, Inc.

Siemens Aktiengesellschaft

Covera Health, Inc.

Siemens Healthineers Ag

The Johns Hopkins University

Siemens Heatlhcare Gmbh

Benjamin L Odry

Mariappan S Nadar

Atilla Peter Kiraly

Carol L Novak

Boris Mailhe

Xiao Chen

Hasan Ertan Cetingul

Pascal Ceccaldi

Li Zhang

Dorin Comaniciu

Bogdan Georgescu

Shaohua Kevin Zhou

Hong Shen

Sandro Braun

Puneet Sharma

Xiaoguang Lu

Daniel Robert Elgort

Shuping Qing

Silvia Bettina Arroyo Camejo

Bjoern Coenen

Ron Vianu

Richard J Herzog

James Robert Browning

Benjamin Sellman Suutari

Gregory Allen Dubbin

Yefeng Zheng

Sasa Grbic

Bernhard Geiger

David D Liu

Berthold Kiefer

Marie-Pierre Jolly

Christophe Chefd'hotel

Gregory G Slabaugh

Gozde Unal

Jens Gühring

Bin Lou

Qiu Wang

Robert Grimm

Jens Guehring

James G Reisman

Shang-Hong Lai

Christopher Vincent Alvino

Thomas Re

Li Pan

Francisco Pereira

Bryce Eakin

JinHyeong Park

Anasuya Das

Diran Guiliguian

Jeanne Verre

Mayalen Irene Catherine Etcheverry

Amer Abufadel

Bernhard Göbel

Shivam Kaushik

Tarmo Henrik Aijo

Xiaojin Dong

LuoLuo Liu

Jefferson Chen

W Nathaniel Brown

Jayashri Pawar

Ihab Kamel

Micha Kornreich

Arijit Sehanobish

Jiancong Wang

Aubrey Chow

Nathaniel W Brown

Described are techniques for processing text data. A process can include obtaining an input text data indicative of pathologies associated with a corresponding radiological image. A BERT-based machine learning network can be used to generate a plurality of location tags, each location tag associated with a sentence of input text data and indicative of an anatomical location in the corresponding radiological image. A plurality of sentence groups can be generated using the input text data and the plurality of location tags, each sentence group including sentences of input text data that are associated with a location tag indicative of the same anatomical location. A multi-task learning (MTL)-based machine learning network can be used to generate a plurality of sets of features, each set of features generated based on a particular sentence group and indicative of pathology severity predictions determined for the anatomical location associated with the particular sentence group.

Multi-task adapters and task similarity for efficient extraction of pathologies from medical reports

Described are techniques for uncertainty-aware anatomical landmark detection, using, for example, a deep reinforcement learning (DRL) anatomical landmark detection agent. For instance, a process can include generating one or more image features for an input medical image using a first sub-network of the anatomical landmark detection agent. A softmax layer of a second sub-network of the anatomical landmark detection agent can generate a plurality of discrete Q-value distributions for a set of allowable actions associated with movement of the agent within the medical image. An anatomical landmark location within the medical image can be predicted using the discrete Q-value distributions. An uncertainty can be determined for the predicted anatomical landmark location, based on an average full width half maximum (FWHM) calculated for the plurality of discrete Q-value distributions.

Uncertainty-aware deep reinforcement learning for anatomical landmark detection in medical images

Brain tumor or other tissue classification and/or segmentation is provided based on from multi-parametric MRI. MRI spectroscopy, such as in combination with structural and/or diffusion MRI measurements, are used to classify. A machine-learned model or classifier distinguishes between the types of tissue in response to input of the multi-parametric MRI. To deal with limited training data for tumors, a patch-based system may be used. To better assist physicians in interpreting results, a confidence map may be generated using the machine-learned classifier.

Tumor tissue characterization using multi-parametric magnetic resonance imaging

For detecting motion in MR imaging, a regression model, such as a convolutional neural network, is machine trained. To generalize to MR imagers, MR contrasts, or other differences in MR image generation, the regression model is trained adversarially. The discriminator for adversarial training discriminates between classes of the variation source (e.g., type of MR imager or type of contrast) based on values of features learned in the regression model for detecting motion. By adversarial training, the regression model learns features that are less susceptible or invariant to variation in image source.

Contrast and/or system independent motion detection for magnetic resonance imaging

A natural language understanding method begins with a radiological report text containing clinical findings. Errors in the text are corrected by analyzing character-level optical transformation costs weighted by a frequency analysis over a corpus corresponding to the report text. For each word within the report text, a word embedding is obtained, character-level embeddings are determined, and the word and character-level embeddings are concatenated to a neural network which generates a plurality of NER tagged spans for the report text. A set of linked relationships are calculated for the NER tagged spans by generating masked text sequences based on the report text and determined pairs of potentially linked NER spans. A dense adjacency matrix is calculated based on attention weights obtained from providing the one or more masked text sequences to a Transformer deep learning network, and graph convolutions are then performed over the calculated dense adjacency matrix.

Computer-implemented natural language understanding of medical reports

For training data pairs comprising training text (a radiological report) and training images (radiological images associated with the radiological report), a first encoder network determines word embeddings for the training text. A concept is generated from the operation of layers of the first encoder network, which is regularized by a first loss between the generated concept and a labeled concept for the training text. A second encoder network determines features for the training image. A heatmap is generated from the operation of layers of the second encoder network, which is regularized by a second loss between the generated heatmap and a labeled heatmap for the training image. A categorical cross entropy loss is calculated between a diagnostic quality category (classified by an error encoder) and a labeled diagnostic quality category for the training data pair. A total loss function comprising the first, second, and categorical cross entropy losses is minimized.

Computer-implemented machine learning for detection and statistical analysis of errors by healthcare providers

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.

Motion determination for volumetric magnetic resonance imaging using a deep machine-learning model

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.

Target detection in latent space

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.

Magnetic resonance imaging quality classification based on deep machine-learning to account for less training data

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.

Artifact reduction by image-to-image network in magnetic resonance imaging

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.

Adaptive nonlinear optimization of shape parameters for object localization in 3D medical images

A method of evaluating airway wall density and inflammation including: segmenting a bronchial tree to create an airway wall map; for each branch, taking a set of locations that form the wall of each branch from the map and sampling the value in a virtual non-contrast image of the bronchial tree and, given a set of samples of pre-contrast densities, computing a value to yield a bronchial wall density for each branch to yield density measures; for each branch, taking the set of locations that form the wall of each branch from the map and sampling the value in a contrast agent map of the bronchial tree and, given the set of samples of contrast agent intake, computing a value to yield a bronchial wall uptake for each branch to yield inflammation measures; and using the density and inflammation measures to determine treatment or predict outcome for a patient.

Method and system for simultaneous evaluation of airway wall density and airway wall inflammation

A system and method including receiving magnetic resonance (MR) imaging data from a first MR scanner device, the MR imaging data including data for a plurality of MR scans of different structural or anatomical regions; generating, based on the MR imaging data, normalized reference data including statistical information for each MR scan; learning a transformation, based on the normalized reference data, to correlate a set of input MR imaging data to the normalized reference data; and storing a record of the transformed imaging data.

System and method for normalized reference database for MR images via autoencoders

Systems and methods are provided for synthesizing protocol independent magnetic resonance images. A patient is scanned by a magnetic resonance imaging system to acquire magnetic resonance data. The magnetic resonance data is input to a machine learnt generator network trained to extract features from input magnetic resonance data and synthesize protocol independent images using the extracted features. The machine learnt generator network generates a protocol independent segmented magnetic resonance image from the input magnetic resonance data. The protocol independent magnetic resonance image is displayed.

Image standardization using generative adversarial networks

A computer-implemented method for generating an assessment of traumatic brain injury (TBI) includes a TBI assessment computer receiving structural imaging data acquired by performing a structural imaging scan on an individual and generating a structural imaging score based on the structural imaging data. The TBI assessment computer receives functional imaging data acquired by performing a functional imaging scan on the individual and generates a functional imaging score based on the functional imaging data. The TBI assessment computer also receives diffusion imaging data acquired by performing a diffusion imaging scan on the individual and generates a diffusion imaging score based on the diffusion imaging data. Based on the structural imaging score, the functional imaging score, and the diffusion imaging score, the TBI assessment computer generates a TBI assessment score.

Modular automated scoring system for TBI assessment

Growing community of inventors

West New York, NJ, United States of America

Benjamin L Odry