Ge Precision Healthcare LLC

Case Western Reserve University

General Electric Company

Soumya Ghose

Jhimli Mitra

Dattesh Dayanand Shanbhag

Chitresh Bhushan

Jed Douglas Pack

Thomas Kwok-Fah Foo

Desmond Teck Beng Yeo

Prem Venugopal

Peter Michael Edic

Anant Madabhushi

Cynthia Elizabeth Landberg Davis

Deepa Anand

Gopal Biligeri Avinash

Gireesha Chinthamani Rao

David Martin Mills

Tao Tan

Pál Tegzes

Lehel Ferenczi

László Ruskó

Radhika Madhavan

Khaled Salem Younis

Levente Imre Török

Lowell Scott Smith

André De Almeida Maximo

Arathi Sreekumari

Michael J MacDonald

Rakesh Shiradkar

Thomas Atta-Fosu

Amod Suhas Jog

Systems and methods are provided for an image processing system. In an example, a method includes acquiring a pathology dataset, acquiring a reference dataset, generating a deformation field by mapping points of a reference case of the reference dataset to points of a patient image of the pathology dataset, manipulating the deformation field, applying the deformation field to the reference case to generate a simulated pathology image including a simulated deformation pathology, and outputting the simulated pathology image.

Simulating pathology images based on anatomy data

A computer implemented method is provided. The method includes establishing, via multiple processors, a continuous federated learning framework including a global model at a global site and respective local models derived from the global model at respective local sites. The method also includes retraining or retuning, via the multiple processors, the global model and the respective local models without sharing actual datasets between the global site and the respective local sites but instead sharing synthetic datasets generated from the actual datasets.

System and method for deep learning techniques utilizing continuous federated learning with a distributed data generative model

Methods and systems are provided for automatic placement of at least one saturation band on a medical image, which may direct saturation pulses during a MRI scan. A method may include acquiring a localizer image of an imaging subject, determining a plane mask for the localizer image by entering the localizer image as input to a deep neural network trained to output the plane mask based on the localizer image, generating a saturation band based on the plane mask by positioning the saturation band at a position and an angulation of the plane mask, and outputting a graphical prescription for display on a display device, the graphical prescription including the saturation band overlaid on the medical image.

Methods and systems for automated saturation band placement

Systems and methods for workflow management for labeling the subject anatomy are provided. The method comprises obtaining at least one localizer image of a subject anatomy using a low-resolution medical imaging device. The method further comprises labeling at least one anatomical point within the at least one localizer image. The method further comprises extracting using a machine learning module a mask of the at least one localizer image comprising the at least one anatomical point label. The method further comprises using the mask to label at least one anatomical point on a high-resolution image of the subject anatomy based on the at least one anatomical point within the localizer image.

Workflow management for labeling the subject anatomy

A computer-implemented method includes obtaining, via a processor, clinical images including vessels and generating, via the processor, straightened-out images for each coronary tree path within respective clinical images, The method also includes extracting, via the processor, segmented 3D image patches, determining, via the processor, overlapping binary mask volumes for each segment, and predicting, via the processor, pressure drops across the segmented image patches using a trained deep neural network.

System and method for deep-learning based estimation of coronary artery pressure drop

A computer-implemented method includes obtaining, via a processor, segmented image patches of a vessel along a coronary tree path and associated coronary flow distribution for respective vessel segments in the segmented image patches. The method also includes determining, via the processor, a pressure drop distribution along an axial length of the vessel from the segmented image patches and the associated coronary flow distribution. The method further includes determining, via the processor, critical points in the pressure drop distribution. The method even further includes detecting, via the processor, a presence of a stenosis based on the critical points in the pressure drop distribution.

System and method for detecting stenosis

Techniques are described for optimizing deep learning model performance using image harmonization as a pre-processing step. According to an embodiment, a method comprises decomposing, by a system operatively coupled to a processor, an input image into sub-images. The method further comprises harmonizing the sub-images with corresponding reference sub-images of at least one reference image based on two or more different statistical values respectively calculated for the sub-images and the corresponding reference-sub images, resulting in transformation of the sub-images into modified sub-images images. In some implementations, the modified sub-images can be combined into a harmonized image having a more similar appearance to the at least one reference image relative to the input image. In other implementations, harmonized images and/or modified sub-images generated using these techniques can be used as ground-truth training samples for training one or more deep learning model to transform input images with appearance variations into harmonized images.

Image harmonization for deep learning model optimization

A computer-implemented method of automatically labeling medical images is provided. The method includes clustering training images and training labels into clusters, each cluster including a representative template having a representative image and a representative label. The method also includes training a neural network model with a training dataset that includes the training images and the training labels, and target outputs of the neural network model are labels of the medical images. The method further includes generating a suboptimal label corresponding to an unlabeled test image using the trained neural network model, and generating an optimal label corresponding to the unlabeled test image using the suboptimal label and representative templates. In addition, the method includes updating the training dataset using the test image and the optimal label, retraining the neural network model, generating a label of an unlabeled image using the retrained neural network model, and outputting the generated label.

Systems and methods of automatic medical image labeling

A computer-implemented method includes generating, via a processor, synthetic vessels. The method also includes performing, via the processor, three-dimensional (3D) computational fluid dynamics (CFD) on the synthetic vessels for different flow rates to generate 3D CFD data. The method further includes extracting, via the processor, 3D image patches from the synthetic vessels. The method even further includes obtaining, via the processor, pressure drops across the 3D image patches from the 3D CFD data. The method yet further includes training, via the processor, a deep neural network utilizing the 3D image patches, the pressure drops, and associated flow rates to generate a trained deep neural network.

Systems and techniques that facilitate hybrid unsupervised and supervised image segmentation are provided. In various embodiments, a system can access a computed tomography (CT) image depicting an anatomical structure. In various aspects, the system can generate, via an unsupervised modeling technique, at least one class probability mask of the anatomical structure based on the CT image. In various instances, the system can generate, via a deep-learning model, an image segmentation based on the CT image and based on the at least one class probability mask.

Hybrid unsupervised and supervised image segmentation model

The subject matter discussed herein relates to multi-modal image alignment to facilitate biopsy procedures and post-biopsy procedures. In one such example, prostate structures (or other suitable anatomic features or structures) are automatically segmented in pre-biopsy MR and pre-biopsy ultrasound images. Thereafter, pre-biopsy MR and pre-biopsy ultrasound contours are aligned. To account for non-linear deformation of the imaged anatomic structure, a patient-specific transformation model is trained via deep learning based at least in part on the pre-biopsy ultrasound images. The pre-biopsy ultrasound images that are overlaid with the pre-biopsy MR contours and based off the deformable transformation model are then aligned with the biopsy ultrasound images. Such real-time alignment using multi-modality imaging techniques provides guidance during the biopsy and post-biopsy system.

Biopsy workflow using multimodal imaging

Embodiments access a set of radiological images acquired from a population of subjects, where a member of the set of radiological images includes a left atrium (LA) region; construct a statistical shape differential atlas from the images; generate a template LA model from the statistical shape differential atlas, where the template LA model includes a site of interest (SOI); acquire a pre-ablation radiological image of a region of tissue in a patient demonstrating atrial fibrillation (AF) pathology; generate a patient LA model from the pre-ablation image; compute a deformation field that registers the SOI to the patient LA model using deformable registration; compute a patient feature vector based on the deformation field; generate an AF probability score for the patient based on the feature vector; generate a classification of the patient based, at least in part, on the AF probability score; and display the classification or the AF probability score.

Differential atlas for identifying sites of recurrence (DISRN) in predicting atrial fibrillation recurrence

The subject matter discussed herein relates to the automatic, real-time registration of pre-operative magnetic resonance imaging (MRI) data to intra-operative ultrasound (US) data (e.g., reconstructed images or unreconstructed data), such as to facilitate surgical guidance or other interventional procedures. In one such example, brain structures (or other suitable anatomic features or structures) are automatically segmented in pre-operative and intra-operative ultrasound data. Thereafter, anatomic structure (e.g., brain structure) guided registration is applied between pre-operative and intra-operative ultrasound data to account for non-linear deformation of the imaged anatomic structure. MR images that are pre-registered to pre-operative ultrasound images are then given the same nonlinear spatial transformation to align the MR images with intra-operative ultrasound images to provide surgical guidance.

Deformable registration for multimodal images

Embodiments predict prostate cancer (PCa) biochemical recurrence (BCR) employing an image acquisition circuit that accesses a pre-treatment image of a region of tissue demonstrating PCa; a segmentation circuit that segments a prostate capsule represented in the image; a registration circuit that registers the segmented prostate with a BCR&#x2212; median template, and generates a registered surface of interest (SOI) mask by registering an SOI mask with the registered prostate; a mask circuit that generates a patient-specific SOI mask from the registered prostate and the registered SOI mask, and generates a patient-specific SOI mesh from the patient-specific SOI mask; a field effect induced organ distension (FOrge) circuit extracts a set of FOrge features from the patient-specific SOI mesh, and computes a probability that the region of tissue will experience BCR; and a classification circuit classifies the region of tissue as likely to experience BCR based on the probability.

Growing community of inventors

Niskayuna, NY, United States of America

Soumya Ghose