Ultrasonic image analysis

Patent metadata

Field	Value
Publication number	US-10751029-B2
Application number	US-201916557261-A
Country	US
Kind code	B2
Filing date	Aug 30, 2019
Priority date	Aug 31, 2018
Publication date	Aug 25, 2020
Grant date	Aug 25, 2020

Abstract

Embodiments of a computer-implemented method of facilitating ultrasonic image analysis of a subject are disclosed. The method can involve receiving signals representing a set of ultrasound images of the subject, deriving one or more extracted feature representations from the set of ultrasound images, determining, based on the derived one or more extracted feature representations, a quality assessment value representing a quality assessment of the set of ultrasound images, determining, based on the derived one or more extracted feature representations, an image property associated with the set of ultrasound images, and producing signals representing the quality assessment value and the image property for causing the quality assessment value and the image property to be associated with the set of ultrasound images. Embodiments of a computer-implemented method of training one or more neural networks to facilitate ultrasonic image analysis is also disclosed. Other apparatuses, methods, systems, and computer-readable media are also disclosed.

First claim

The invention claimed is:

1. A computer-implemented method of facilitating ultrasonic image analysis of a subject, the method comprising:
receiving signals representing a set of ultrasound images of the subject;
deriving one or more extracted feature representations from the set of ultrasound images;
determining, based on the derived one or more extracted feature representations, a quality assessment value representing a quality assessment of the set of ultrasound images;
determining, based on the derived one or more extracted feature representations, an image property associated with the set of ultrasound images; and
producing signals representing the quality assessment value and the image property for causing the quality assessment value and the image property to be associated with the set of ultrasound images.

2. The method of claim 1 wherein the image property is a view category.

3. The method of claim 2 wherein deriving the one or more extracted feature representations from the ultrasound images comprises, for each of the ultrasound images, deriving a first feature representation associated with the ultrasound image.

4. The method of claim 3 wherein deriving the one or more extracted feature representations comprises, for each of the ultrasound images, inputting the ultrasound image into a commonly defined first feature extracting neural subnetwork to generate the first feature representation associated with the ultrasound image.

5. The method of claim 4 wherein deriving the one or more extracted feature representations comprises concurrently inputting each of a plurality of the ultrasound images into a respective implementation of the commonly defined first feature extracting neural network.

6. The method of claim 4 wherein the commonly defined first feature extracting neural network includes a convolutional neural network.

7. The method of claim 4 wherein deriving the one or more extracted feature representations comprises inputting the first feature representations into a second feature extracting neural network to generate respective second feature representations, each associated with one of the ultrasound images and wherein the one or more extracted feature representations include the second feature representations.

8. The method of claim 7 wherein the second feature extracting neural network is a recurrent neural network.

9. The method of claim 2 wherein determining the quality assessment value comprises inputting the one or more extracted feature representations into a quality assessment value specific neural network and wherein determining the image property comprises inputting the one or more extracted feature representations into an image property specific neural network.

10. The method of claim 9 wherein inputting the one or more extracted feature representations into the quality assessment value specific neural network comprises inputting each of the one or more extracted feature representations into an implementation of a commonly defined quality assessment value specific neural subnetwork and wherein inputting the one or more extracted feature representations into the image property determining neural network comprises inputting each of the one or more extracted feature representations into an implementation of a commonly defined image property specific neural network.

11. The method of claim 2 wherein producing signals representing the quality assessment value and the image property for causing the quality assessment value and the image property to be associated with the set of ultrasound images comprises producing signals for causing a representation of the quality assessment value and a representation of the image property to be displayed by at least one display in association with the set of ultrasound images.

12. A computer-implemented method of training one or more neural networks to facilitate ultrasonic image analysis, the method comprising:
receiving signals representing a plurality of sets of ultrasound training images;
receiving signals representing quality assessment values, each of the quality assessment values associated with one of the sets of ultrasound training images and representing a quality assessment of the associated set of ultrasound training images;
receiving signals representing image properties, each of the image properties associated with one of the sets of ultrasound training images; and
training a neural network, the training comprising, for each set of the plurality of sets of ultrasound training images, using the set of ultrasound training images as an input to the neural network and using the quality assessment values and the image properties associated with the set of ultrasound training images as desired outputs of the neural network.

13. The method of claim 12 wherein each of the image properties is a view category.

14. The method of claim 13 wherein the neural network includes a feature extracting neural network, an image property specific neural network, and a quality assessment value specific neural network and wherein:
the feature extracting neural network is configured to take an input set of the plurality of sets of ultrasound training images as an input and to output one or more extracted feature representations;
the image property specific neural network is configured to take the one or more extracted feature representations as an input and to output a representation of an image property associated with the input set of ultrasound training images; and
the quality assessment specific neural network is configured to take the one or more extracted feature representations as an input and to output a quality assessment value associated with the input set of ultrasound training images.

15. The method of claim 14 wherein the feature extracting neural network is configured to, for each of the ultrasound training images included in the input set of ultrasound training images, derive a first feature representation associated with the ultrasound image.

16. The method of claim 15 wherein the feature extracting neural network includes, for each of the ultrasound images included in the input set of ultrasound training images, a commonly defined first feature extracting neural network configured to take as an input the ultrasound training image and to output a respective one of the first feature representations.

17. The method of claim 16 wherein more than one implementation of the commonly defined first feature extracting neural networks are configured to concurrently generate the first feature representations.

18. The method of claim 16 wherein the commonly defined first feature extracting neural network is a convolutional neural network.

19. The method of claim 16 wherein the feature extracting neural network includes a second feature extracting neural network configured to take as an input the first feature representations and to output respective second feature representations, each associated with one of the ultrasound images included in the input set of ultrasound training images and wherein the one or more extracted feature representations include the second feature representations.

20. The method of claim 19 wherein the second feature extracting neural network is a recurrent neural network.

21. A system for facilitating ultrasonic image analysis comprising at least one processor configured to:
receive signals representing a set of ultrasound images of the subject;
derive one or more extracted feature representations from

Assignees

• Univ British Columbia

Inventors

• Abolmaesumi Purang
• Rohling Robert
• Tsang Teresa
• Liao Zhibin
• Abdi Amir

Classifications

• A61B8/466

  adapted to display 3D data · CPC title

• A61B8/5215

  involving processing of medical diagnostic data · CPC title

• A61B8/4405

  Device being mounted on a trolley · CPC title

• A61B8/483

  involving the acquisition of a 3D volume of data · CPC title

• A61B8/463

  characterised by displaying multiple images or images and diagnostic data on one display · CPC title