Yale University

Koninklijke Philips Corporation N.V.

Siemens Medical Solutions Usa, Inc.

James S Duncan

Susanne Smolka

John Treilhard

Dorin Comaniciu

Puneet Sharma

Ali Kamen

Tommaso Mansi

Razvan Ionasec

Ming De Lin

Brian Letzen

Jean-François Geschwind

Clinton Wang

Johanna Van Breugel

Aaron Abajian

Julius Chapiro

MingDe Lin

Jingjing Kanik

One aspect of the invention provides a computer-implemented method of identifying one or more clinical factors associated with an artificial intelligence prediction. The computer implemented method includes: applying a previously trained deep neural network to one or more images for a subject to produce a prediction, the previously trained deep neural network comprising a plurality of nodes; comparing outputs of the nodes to previously identified patterns of node outputs for a plurality of individual clinical factors; and identifying which of the plurality of individual clinical factors are most correlated with the outputs of the nodes.

Interpretable deep machine learning for clinical radiology

In an image guidance method, a stream of cone-beam computed tomography (CBCT) images of an anatomical organ is acquired using a CBCT imaging device (). For each CBCT image (), the surface of the organ in the CBCT image is segmented to generate a segmented surface () of the anatomical organ in the CBCT image. Occluded shape reconstruction () using a statistical shape prior () may be used to reconstruct a missing or occluded surface portion. A pre-operative medical image () of the anatomical organ and the CBCT image are spatially registered by warping one of the images to the other of the images using point set registration of a segmented surface () of the organ in the preoperative medical image and the segmented surface of the organ in the CBCT image. A fused image () is displayed that combines the warped image and the other image.

CBCT to MR registration via occluded shape reconstruction and robust point matching

The described implementations relate to systems, methods, and apparatuses for generating regions of interest () from imaging data (). Specifically, the regions of interest are generated for tracking treatment efficacy in a more consistent and repeatable manner. The regions of interest can be generated from contrast medium and non-contrast medium enhanced scans () of a patient. Voxel data derived from the scans can be collected and distributed according to respective intensity values in order to identify mode voxels () for particular ranges () of intensities. Regions of interest () can then be generated for each identified mode voxel, and standard deviations for the regions of interest can be determined. One or more thresholds can be derived from the determined standard deviations in order to further filter the intensity values and identify filtered groups of voxels to be the resulting regions of interest.

Systems, methods, and apparatuses for generating regions of interest from voxel mode based thresholds

A mechanical property of anatomy is estimated from a patient in vivo, such as estimating a patient-specific material property of a valve. A morphological model is used to determine anatomy dynamics. A biomechanical model, using the anatomy dynamics, predicts the dynamics, based, at least in part, on one or more material properties. Using an inverse solution based on comparison of dynamics predicted by the biomechanical model and the dynamics determined from the morphological model, values for the material properties are determined.

Estimation of a mechanical property of anatomy from medical scan data

Growing community of inventors

Madison, CT, United States of America

James S Duncan