Koninklijke Philips Corporation N.v.

Christine Menking Swisher

Man Nguyen

Claudia Errico

Hua Xie

Jean-Luc Francois-Marie Robert

Oladimeji Feyisetan Farri

Sheikh Sadid Al Hasan

Rithesh Sreenivasan

Binyam Gebrekidan Gebre

Cristhian Mauricio Potes Blandon

Jonathan Rubin

Yuan Ling

Asif Rahman

Vijay Thakur Shamdasani

Bryan Conroy

Cornelis Conradus Adrianus Maria Van Zon

Yinhui Deng

Raghavendra Srinivasa Naidu

Ze He

Purnima Rajan

Homer Pien

Preetish Rath

The present disclosure describes ultrasound imaging systems and methods for ultrasonically inspecting biological tissue, such as liver and for automatically identifying and acquiring a view suitable for hepatic-renal echo-intensity ratio quantification, using one or more neural networks, which may be trained to perform image classification, segmentation, or a combination thereof to compute a confidence metric and to provide a recommendation for measurement ROIs placement on the image.

Ultrasound system with an artificial neural network for guided liver imaging

The present disclosure describes medical imaging systems and methods configured to generate medical images based on undersampled imaging data. The images may be generated by applying a neural network trained with samples of known fully sampled data and undersampled data derived from the known fully sampled data to a acquired sparsely sampled data. The training of the neural network may involve training adversarial generative network including a generator and a discriminator. The generator is trained with sets of known undersampled data until the generator is capable of generating estimated image data, which the classifier is incapable of differentiation as either real or fake, and the trained generator may then be applied to unknown undersampled data.

Ultrasound system with a neural network for producing images from undersampled ultrasound data

Techniques disclosed herein relate to identifying individuals in digital images. In some embodiments, a digital image may be acquired () that captures an environment containing at least a first subject. A first portion of the digital image depicting the first subject may be segmented () into a plurality of superpixels. For each superpixel of the plurality of superpixels: a semantic label may be assigned () to the superpixel; features of the superpixel may be extracted (); and a measure of similarity between the features extracted from the superpixel and features extracted from a reference superpixel identified in a reference digital image may be determined (), wherein the reference superpixel has a reference semantic label that matches the semantic label assigned to the superpixel. Based on the measures of similarity associated with the plurality of superpixels, it may be determined () that the first subject is depicted in the reference image.

Person identification systems and methods

A method () for generating a textual description of a medical image, comprising: receiving () a medical image of an anatomical region, the image comprising one or more abnormalities; segmenting () the anatomical region in the received medical image from a remainder of the image; identifying () at least one of the one or more abnormalities in the segmented anatomical region; extracting () one or more features from the identified abnormality; generating (), using the extracted features and a trained text generation model, a textual description of the identified abnormality; and reporting (), via a user interface of the system, the generated textual description of the identified abnormality.

Convolutional localization networks for intelligent captioning of medical images

Various embodiments of the present disclosure are directed to a salient medical imaging controller () employing an artificial intelligence engine () and a graphical user interface (). In operation, the artificial intelligence engine () includes one or more machine learning models () trained to render a feature assessment of a medical image. The graphical user interface () provides a user interaction with the artificial intelligence engine () to manipulate a salient visualization of the feature assessment of the medical image by the machine learning model(s) ().

Salient visual explanations of feature assessments by machine learning models

The present disclosure describes ultrasound imaging systems and methods configured to identify and remove image artefacts from ultrasound image frames by applying a neural network to the frames. Systems may include an ultrasound transducer configured to acquire echo signals responsive to ultrasound pulses transmitted toward a target region. One or more processors communicatively coupled with the ultrasound transducer may be configured to generate an image frame from the ultrasound echoes and apply a neural network to the image frame. The neural network determines whether an artefact is present in the image frame, and identifies the type of artefact detected. The processors can also generate an indicator conveying the presence of the artefact, which can be displayed on a user interface. The processors can further generate an instruction for adjusting the ultrasound transducer based on the presence and type of artefact present within the image frame.

Intelligent ultrasound system for detecting image artefacts

Various embodiments of the present disclosure are directed to a raw diagnostic machine for a medical diagnosis of raw medical imaging data generated by a medical imaging machine as opposed to a medical diagnosis of a medical image conventionally reconstructed from the raw medical imaging data. In operation, the raw diagnostic engine includes a medical imaging diagnostic controller implementing a dimension reduction pre-processor for selecting or extracting one or more dimension reduced feature vectors from the raw medical imaging data, and further implementing a raw diagnostic artificial intelligence engine for rendering a diagnostic assessment of the raw medical imaging data as represented by the dimension reduced feature vector(s). The medical imaging diagnostic controller may further control a communication of the diagnostic assessment of the raw medical imaging data (e.g., a display, a printing, an emailing, a texting, etc.).

Machine learning on raw medical imaging data for clinical decision support

The present disclosure describes ultrasound imaging systems and methods configured to generate ultrasound images based on undersampled ultrasound data. The ultrasound images may be generated by applying a neural network trained with samples of known fully sampled data and undersampled data derived from the known fully sampled data to a acquired sparsely sampled data. The training of the neural network may involve training adversarial generative network including a generator and a discriminator. The generator is trained with sets of known undersampled data until the generator is capable of generating estimated image data, which the classifier is incapable of differentiation as either real or fake, and the trained generator may then be applied to unknown undersampled data.

Systems and computer implemented method of generating a simulated image of a baby based on an ultrasound image of a fetus. A method comprises acquiring an ultrasound image of the fetus, and using a machine learning model to generate the simulated image of the baby, based on the ultrasound image of the fetus, wherein the simulated image of the baby comprises a prediction of how the fetus will look as a baby.

Generating a simulated image of a baby

Clinical assessment devices, systems, and methods are provided. A clinical assessment system, comprising a processor in communication with an imaging device, wherein the processor is configured to receive, from the imaging device, a sequence of image frames representative of a contrast agent perfused subjects tissue across a time period; classify the sequence of image frames into a plurality of first tissue classes and a plurality of second tissue classes based on a spatiotemporal correlation among the sequence of image frames by applying a predictive network to the sequence of image frames to produce a probability distribution for the plurality of first tissue classes and the plurality of second tissue classes; and output, to a display in communication with the processor, the probability distribution for the plurality of first tissue classes and the plurality of second tissue classes.

Ultrasound lesion assessment and associated devices, systems, and methods

The invention discloses an apparatus for converting data and for assessing data. The apparatus comprises a processor. For converting data, the processor is configured to train a neural network arrangement to generate a first vector to represent input data, each element of the first vector representing a defined feature of the input data. For assessing data, the processor is configured to provide a neural network trained to generate a first vector representing input data; provide input data to the trained neural network; and generate, using the trained neural network, a first vector representing the input data, wherein each element of the generated first vector represents a defined feature of the input data. Methods and a computer program product are also disclosed.

Generating vectors from data

Methods and systems for detecting false alarms. Methods and systems described herein may receive data associated with an alarm signal using an interface, extract at least one artifact feature from the received data, and then receive a classification of the alarm signal as a true positive or false positive based on the at least one extracted artifact feature. The classifier may be configured to execute an ensemble of decision trees.

False alarm detection

A imaging system includes a camera, a display and a processor. The camera has color video acquisition capability, and is mounted to a distal end of an interventional instrument insertable within an object, the camera providing image frames for imaging vasculature of the object, each image frame including multiple pixels providing corresponding signals, respectively. The processor is programmed to receive the signals; amplify variations in at least one of color and motion of the signals corresponding to the multiple pixels; determine spatial phase variability, frequency and signal characteristics of at least some of the amplified signals corresponding to the multiple pixels, respectively; identify pixels indicative of abnormal vascularity based on the spatial phase variability, frequency and/or signal characteristics; create a vascular map corresponding to each, where each vascularity map includes a portion of the object having the abnormal vascularity; and operate the display to display each vascularity map.

System and method for detecting abnormal tissue using vascular features

Techniques disclosed herein relate to identifying individuals in digital images. In some embodiments, a digital image(s) that captures a scene containing one or more people may be acquired. The single digital image may be applied as input across a single machine learning model. In some implementations, the single machine learning model may be trained to perform a non-facial feature recognition task and a face-related recognition task. Output may be generated over the single machine learning model based on the input. The output may include first data indicative of non-facial features of a given person of the one or more people and second data indicative of at least a location of a face of the given person in the digital image relative to the non-facial features. In various embodiments, the given person may be identified based at least in part on the output.

Growing community of inventors

San Diego, CA, United States of America

Christine Menking Swisher