Automated cardiac function assessment by echocardiography
AssigneeThe Regents of the University of California
InventorsRahul C. Deo, Jeffrey Zhang
Life of this patent
Estimating the term…
Abstract
A computer vision pipeline is provided for fully automated interpretation of cardiac function, using a combination of machine learning strategies to enable building a scalable analysis pipeline for echocardiogram interpretation. Videos from patients with heart failure can be analyzed and processed as follows: 1) preprocessing of echo studies; 2) convolutional neural networks (CNN) processing for view identification; 3) segmentation of chambers and delineation of cardiac boundaries using CNNs; 4); particle tracking to compute longitudinal strain; and 5) targeted disease detection.
Claims18
A method of analyzing images obtained from an echocardiogram, the method comprising performing by a computer system:
- receiving the images obtained from the echocardiogram;
- identifying a subset of images corresponding to a particular echo view of a heart using a first convolutional neural network, wherein each image of the subset of images includes a set of pixel values, and wherein the particular echo view is selected from a group consisting of apical 2-, 3-, and 4-chamber (A2c, A3c, and A4c), parasternal long axis (PLAX), parasternal short axis (PSAX), and inferior vena cava (IVC);
- feeding the sets of pixel values into a second convolutional neural network trained using sample echocardiograms with at least two different structures identified;
- for each pixel of each image of the subset of images, assigning a probability of the pixel being part of one of a plurality of structures using the second convolutional neural network, wherein at least one of the plurality of structures is part of the heart imaged in the echocardiogram;
- identifying pixels in each of the images that correspond to a first structure of the heart based on a comparison of the probabilities to a threshold;
- and
- providing an indication of the first structure in the subset of images.
A computer product comprising a non-transitory computer readable medium storing a plurality of instructions that when executed control a computer system to perform operations for analyzing images obtained from an echocardiogram, the operations comprising:
- receiving the images obtained from the echocardiogram;
- identifying a subset of images corresponding to a particular echo view of a heart using a first convolutional neural network, wherein each image of the subset of images includes a set of pixel values, and wherein the particular echo view is selected from a group consisting of apical 2-, 3-, and 4-chamber (A2c, A3c, and A4c), parasternal long axis (PLAX), parasternal short axis (PSAX), and inferior vena cava (IVC);
- feeding the sets of pixel values into a second convolutional neural network trained using sample echocardiograms with at least two different structures identified;
- for each pixel of each image of the subset of images, assigning a probability of the pixel being part of one of a plurality of structures using the second convolutional neural network, wherein at least one of the plurality of structures is part of the heart imaged in the echocardiogram;
- identifying pixels in each of the images that correspond to a first structure of the heart based on a comparison of the probabilities to a threshold;
- and
- providing an indication of the first structure in the subset of images.
Dependencies are inferred from the claim text.