Intelligent medical image landmark detection

Patent metadata

Field	Value
Publication number	US-9569736-B1
Application number	US-201615160699-A
Country	US
Kind code	B1
Filing date	May 20, 2016
Priority date	Sep 16, 2015
Publication date	Feb 14, 2017
Grant date	Feb 14, 2017

Abstract

Intelligent image parsing for anatomical landmarks and/or organs detection and/or segmentation is provided. A state space of an artificial agent is specified for discrete portions of a test image. A set of actions is determined, each specifying a possible change in a parametric space with respect to the test image. A reward system is established based on applying each action of the set of actions and based on at least one target state. The artificial agent learns an optimal action-value function approximator specifying the behavior of the artificial agent to maximize a cumulative future reward value of the reward system. The behavior of the artificial agent is a sequence of actions moving the agent towards at least one target state. The learned artificial agent is applied on a test image to automatically parse image content.

First claim

The invention claimed is:

1. A method for intelligent image parsing, the method comprising:
specifying a state space of an artificial agent for discrete portions of a training image;
determining a set of actions, each action specifying a possible change in a parametric space with respect to the test image;
establishing a reward system based on applying each action of the set of actions and based on at least one target state;
learning, by the artificial agent, an optimal action-value function approximator specifying the behavior of the artificial agent to maximize a cumulative future reward value of the reward system, wherein the behavior of the artificial agent is a sequence of actions moving the agent towards the at least one target state; and
applying the learned artificial agent on a test image to automatically parse image content.

2. The method of claim 1 , wherein applying the learned artificial agent on test images to automatically parse image content further comprises:
evaluating the optimal action-value function approximator for a current state space;
simultaneously obtaining the optimal action-value function approximator for all possible actions at each current state space; and
applying a reward policy of the optimal action-value function approximator.

3. The method of claim 2 , wherein applying the reward policy of the optimal action-value function approximator further comprises:
determining a next action of the artificial agent based on balancing maximization of the cumulative future reward value and completion of evaluation of each portion of each test image based on the state space.

4. The method of claim 1 , wherein the at least one target state is at least one anatomical landmark, and wherein the state space is defined by the position parameters of an anatomical landmark, and wherein the reward value is indicative of a proximity of a current state space to the at least one target state.

5. The method of claim 1 , wherein learning an optimal action-value function approximator further comprises:
generating an experience memory database including a predefined number of last evaluated states for a current training image;
sampling the experience memory database; and
updating parameters of the optimal action-value function approximator based on the experience memory.

6. The method of claim 1 , further comprising:
exploring an episodic trajectory for a training image based on a completed evaluation of each portion of the training image via the state space, wherein the episodic trajectory is indicative of the actions of the artificial agent as a sequence of visited states for the training image.

7. The method of claim 6 , wherein exploring the episodic trajectory for the training image further comprises:
storing episodic trajectories at pre-defined intervals of sequential completed evaluations of training images by the artificial agent; and
updating parameters of the optimal action-value function approximator based on the stored episodic trajectories.

8. The method of claim 1 , wherein the set of actions includes changing a position of the state space by at least one pixel of the image in an upwards, downwards, left, or right direction with respect to the image.

9. The method of claim 1 , wherein the target state is an anatomical object, and wherein the set of actions includes changing a position, an orientation, a scale or a shape of the current state.

10. The method of claim 7 , wherein the set of actions further includes an action in which the state space remains in the same state.

11. The method of claim 1 , wherein the target state is an anatomical object, and wherein at least one action of the set of actions is determined via input of a user.

12. The method of claim 1 , further comprising:
learning, by the artificial agent, example annotation actions received via user input or selected image optimization parameters; and
replicating the learned annotation actions or the selected image optimization parameters on the test image.

13. The method of claim 1 , wherein learning, by the artificial agent, an optimal action-value function approximator specifying the behavior of the artificial agent further comprising:
observing a plurality of user input to a processor associated with the artificial agent; and
replicating the observed plurality of user inputs; and
suggesting, by the artificial agent, a next action of the processor based on the replicated user inputs.

14. A system for artificial agent training for intelligent landmark identification in medical images, the system comprising:
at least one processor; and
at least one memory including computer program code for one or more programs; the at least one memory and the computer program code configured to, with the at least one processor, cause the system to:
evaluate a state space of discrete portions of each training image of a set of training images, wherein training images of the set of training images includes a landmark target, wherein the landmark target is marked on each training image;
change a position of the state space with respect to each training image via application of actions of a pre-defined set of actions;
determine a reward value of each position change of the state space based on a reward system, wherein the reward value is indicative of a proximity of a current state space to a pre-defined landmark target of each training image; and
optimize behavior of the artificial agent based on maximizing a cumulative future reward value based on the reward system, the set of actions, the state space, and the set of training images, wherein the behavior of the artificial agent is intelligent selection of next actions to achieve a position of the state space on the landmark target in each image.

15. The system of claim 14 , wherein configuration of the at least one memory and the computer program code to optimize behavior of the artificial agent, further causes the system to:
evaluate an optimal action-value function approximator for each current position of the state space;
simultaneously obtain the optimal action-value function approximator for all possible actions based on the current position of each state space; and
apply a reward policy of the optimal action-value function approximator.

16. The system of claim 15 , wherein configuration of the at least one memory and the computer program code to apply the reward policy of the optimal action-value function approximator, further causes the system to:
determine a next action of the artificial agent based on a balance of maximization of the cumulative future reward value based on the reward system and completion of evaluation of each portion of each training image based on the state space.

17. The system of claim 15 , the at least one memory and the computer program code configured to, with the at least one processor, cause the system to:
generate an experience memory database, wherein the experience memory includes a pre-defined number of last evaluated states for the current training image;
sample the experience memory; and
update parameters of the optimal action-value function approximator based on the sampled experience memory.

18. The system of claim 14 , the at least one memory and the computer program code configured to, with the at least one processor, cause the system to:
exploring an episodic trajectory for a training image based on a completed evaluation of each portion of the training image via the state space, wherein the episodic trajectory is indicative of the actions of the

Assignees

• Siemens Healthcare Gmbh

Inventors

• Ghesu Florin Cristian
• Georgescu Bogdan
• Neumann Dominik
• Mansi Tommaso
• Comaniciu Dorin
• Liu Wen
• Zhou Shaohua Kevin

Classifications

• A61B8/0883 Primary

  for diagnosis of the heart · CPC title

• G06V10/771

  Feature selection, e.g. selecting representative features from a multi-dimensional feature space · CPC title

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G06T7/0012

  Biomedical image inspection · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title