Medical atlas registration

Patent metadata

Field	Value
Publication number	US-10692226-B2
Application number	US-201716097653-A
Country	US
Kind code	B2
Filing date	May 4, 2017
Priority date	May 4, 2016
Publication date	Jun 23, 2020
Grant date	Jun 23, 2020

Abstract

A system and method are provided for enabling atlas registration in medical imaging, said atlas registration comprising matching a medical atlas 300, 302 to a medical image 320 . The system and method may execute a Reinforcement Learning (RL) algorithm to learn a model for matching the medical atlas to the medical image, wherein said learning is on the basis of a reward function quantifying a degree of match between the medical atlas and the medical image. The state space of the RL algorithm may be determined on the basis of a set of features extracted from i) the atlas data and ii) the image data. As such, a model is obtained for medical atlas registration without the use, or with a reduced use, of heuristics. By using a machine learning based approach, the solution can easily be applied to different atlas matching problems, e.g., to different types of medical atlases and/or medical images.

First claim

The invention claimed is:

1. A system for enabling atlas registration in medical imaging, said atlas registration comprising matching a medical atlas to a medical image, the system comprising:
a first input interface for accessing atlas data defining the medical atlas;
a second input interface for accessing image data of the medical image;
a processor configured to:
execute a Reinforcement Learning algorithm to learn a model for matching the medical atlas to the medical image, wherein said learning is on a basis of a reward function quantifying a degree of match between the medical atlas and the medical image;
determine a state space for the Reinforcement Learning algorithm on a basis of a set of features extracted from i) the atlas data and ii) the image data;
apply a Convolutional Neural Network or Autoencoder to the atlas data and the image data to determine the set of features; and
determine an action space for the Reinforcement Learning algorithm on a basis of a predefined set of transformation actions which are available to be applied to the medical atlas, wherein the action space is structured into different levels;

wherein each of the different levels comprises a subset of the transformation actions, and wherein the different levels form a hierarchy of transformation actions in which selection of a sequence of transformation actions by the Reinforcement Learning algorithm is restricted to a downward progression in the hierarchy.

2. The system according to claim 1 , wherein the processor is configured to learn the set of features to be extracted from the atlas data and the image data using a machine learning algorithm.

3. The system according to claim 1 , wherein the action space is hierarchically structured to comprise less restrictive transformation actions upwards in the hierarchy and more restrictive transformation actions downwards in said hierarchy.

4. The system according to claim 3 , wherein the processor is configured to perform said hierarchically structuring of the action space using a machine learning algorithm or using pre-determined heuristics.

5. The system according to claim 3 wherein the processor is configured to determine the state space for the Reinforcement Learning algorithm further based on a current level in the hierarchy of transformation actions.

6. The system according to claim 1 , wherein the predefined set of transformation actions, which is available to be applied to the medical atlas, is defined by the atlas data.

7. The system according to claim 1 , wherein the processor is configured to determine the state space for the Reinforcement Learning algorithm further based on landmark data defining landmarks in the medical atlas.

8. The system according to claim 1 , wherein the Reinforcement Learning algorithm is a Deep Reinforcement Learning algorithm.

9. A workstation comprising the system according to claim 1 .

10. A non-transitory computer readable medium comprising data representing a model for atlas registration in medical imaging as generated by the system according to claim 1 .

11. A method of enabling atlas registration in medical imaging, said atlas registration comprising matching a medical atlas to a medical image, comprising:
accessing atlas data defining the medical atlas;
accessing image data of the medical image;
executing a Reinforcement Learning algorithm to learn a model for matching the medical atlas to the medical image, wherein said learning is on a basis of a reward function quantifying a degree of match between the medical atlas and the medical image;
determining a state space for the Reinforcement Learning algorithm on a basis of a set of features extracted from i) the atlas data and ii) the image data;
applying a Convolutional Neural Network or Autoencoder to the atlas data and the image data to determine the set of features; and
determining an action space for the Reinforcement Learning algorithm on a basis of a predefined set of transformation actions which are available to be applied to the medical atlas, wherein the action space is structured into different levels;
wherein each of the different levels comprises a subset of the transformation actions, and wherein the different levels form a hierarchy of transformation actions in which selection of a sequence of transformation actions by the Reinforcement Learning algorithm is restricted to a downward progression in the hierarchy.

12. A non-transitory computer readable medium comprising data representing instructions to cause a processor system to perform the method according to claim 11 .

Assignees

• Koninklijke Philips Nv

Inventors

• Bresch Erik
• Menkovski Vlado

Classifications

• G06T7/00

  Image analysis · CPC title

• G06T7/33 Primary

  using feature-based methods · CPC title

• G06T2207/20016

  Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title