Methods of systems of generating virtual multi-dimensional models using image analysis
AssigneeGeneral Electric Company
InventorsAlberto Santamaria-Pang, Daniel Eugene Meyer, Michael Ernest Marino, Qing Li, Dmitry V. Dylov, Aritra Chowdhury
Life of this patent
Estimating the term…
Abstract
The present approach relates to the use of trained artificial neural networks, such as convolutional neural networks, to classify vascular structures, such as using a hierarchical classification scheme. In certain approaches, the artificial neural network is trained using training data that is all or partly derived from synthetic vascular representations.
Claims20
A method for generating a multi-dimensional synthetic vascular model, comprising:
- defining a set of control points in a three-dimensional coordinate system;
- interpolating between the control points to generate a skeleton of a synthetic blood vessel being generated;
- and
- generating a three-dimensional volume of the synthetic blood vessel about the skeleton.
A method for training and using an artificial neural network, comprising:
- generating a training data set of images, wherein at least a portion of the images are derived by taking slices through one or more synthetic vessels of a synthetic vascular model;
- providing the training data set of images to an artificial neural network to generate a trained neural network;
- providing an input set of clinical images to the trained neural network for classification of vascular features within the set of clinical images;
- and
- receiving an output from the trained neural network based on the classifications generated for the set of clinical images.
A vascular morphology classification neural network, comprising:
- an input layer configured to receive images depicting sectional views of one or more vascular vessels in a biological sample;
- two or more hidden layers trained using synthetic vessel images to classify morphological features present in the sectional views of the biological sample;
- and
- an output layer downstream from the hidden layers, wherein the output layer is configured to provide an output based on the classifications generated for the morphological features.
Dependencies are inferred from the claim text.