Intelligent multi-scale medical image landmark detection
Intelligent multi-scale image parsing determines the optimal size of each observation by an artificial agent at a given point in time while searching for the anatomical landmark. The artificial agent begins searching image data with a coarse field-of-view and iteratively decreases the field-of-view to locate the anatomical landmark. After searching at a coarse field-of view, the artificial agent increases resolution to a finer field-of-view to analyze context and appearance factors to converge on the anatomical landmark. The artificial agent determines applicable context and appearance factors at each effective scale.
1. A method for intelligent multi-scale image parsing of a medical image, the method comprising:
specifying a state space of an artificial agent for discrete portions of a training image, the state space comprising a parametric space and a scale space;
determining a set of actions comprising parametric actions specifying a possible change in the parametric space with respect to the training image and scale actions specifying a possible change in the scale space with respect to the training image;
establishing a reward system based on applying each action of the set of actions and based on at least one target location of the training image;
learning, by the artificial agent an optimal action-value function approximator specifying a behavior of the artificial agent by maximizing a cumulative future reward value of the reward system based on sequences of actions performed by the artificial agent, wherein parametric actions of the set of actions move the artificial agent towards the target location within a particular scale and scale actions of the set of actions increase a resolution of the artificial agent;
applying the learned artificial agent to the medical image to automatically parse image content of the medical image for a landmark location;
identifying the landmark target on the medical image when the cumulative reward value indicates a proximity of an adjacent state space within a pre-defined reward threshold distance value of the landmark target on the medical image; and
determining that the landmark target is not present in the medical image when the cumulative reward value is outside a pre-defined failure threshold distance value.
2. The method of claim 1 , wherein scale space actions comprise at least a zoom-in and zoom-out action.
3. The method of claim 1 , wherein the scale space is defined as:
L
(
x
;
t
)
=
∑
ξ
∈
z
3
T
(
ξ
;
t
)
I
(
x
-
ξ
)
,
where t∈ + denotes the continuous scale-level, x∈ 3 , L(x;0)=I(x) and T defines a one-parameter family of kernels.
4. The method of claim 1 , wherein applying the learned artificial agent comprises applying series of actions performed by the artificial agent that change the position and scale of a patch of the medical image in order to parse the medical image without performing an exhaustive search of an entirety of the medical image.
5. The method of claim 1 , wherein the set of actions is determined such that the artificial agent is configured to be able to search an entirety of the medical image.
6. The method of claim 1 , wherein a convergence point at a previous scale is used as a starting point at a subsequent scale for a subsequent action of the sequences of actions.
7. The method of claim 1 , wherein applying the learned artificial agent 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.
8. The method of claim 7 , wherein applying the reward policy comprises determining a next action of the artificial agent based on balancing maximization of the cumulative future reward value by actions changing the parametric space and by actions changing the scale space.
9. The method of claim 1 , wherein the at least one target location is an anatomical landmark location, and wherein the target location is defined by position parameters of the landmark anatomical landmark, and wherein a reward value is indicative of a proximity to at least one target state.
10. The method of claim 1 , wherein learning the optimal action-value function approximator further comprises:
generating an experience memory database including a predefined number of last evaluated parametric spaces and scale spaces for the medical image;
sampling the experience memory database; and
updating parameters of the artificial agent based on experience memory.
11. The method of claim 1 , wherein learning the optimal action-value function approximator further comprises:
parameterizing the behavior of the artificial agent using a deep neural network; and
optimizing the behavior of the artificial agent using an episodic trajectory for the training image based on discrete portions of the training image via the parametric space and scale space, wherein the episodic trajectory is indicative of a series actions of the set of actions of the artificial agent.
12. A system for intelligent multi-scale image parsing of a medical image, 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:
specify a state space of an artificial agent for discrete portions of a training image, the state space comprising a parametric space and a scale space;
determine a set of actions comprising parametric actions specifying a possible change in the parametric space with respect to the training image and scale actions specifying a possible change in the scale space with respect to the training image;
establish a reward system based on applying each action of the set of actions and based on at least one target location of the training image;
learn an optimal action-value function approximator specifying a behavior of the artificial agent by maximizing a cumulative future reward value of the reward system based on sequences of actions performed by the artificial agent, wherein parametric actions of the set of actions move the artificial agent towards the target location within a particular scale and scale actions of the set of actions increase a resolution of the artificial agent;
apply the artificial agent to the medical image to automatically parse image content of the medical image for a landmark location;
identify the landmark target on the medical image when the cumulative reward value indicates a proximity of an adjacent state space within a pre-defined reward threshold distance value of the landmark target on the medical image; and
determine, when the cumulative reward value is outside a pre-defined failure threshold distance value, that the landmark target is not present in the medical image.
13. The system of claim 12 , wherein learning the optimal action-value function approximator comprises encoding parameters of search strategy model in a multilayer data representation.
14. The system of claim 12 , wherein learning the optimal action-value function approximator comprises using experience memory from previously parsed patches at different scales.
15. The system of claim 12 , wherein the scale space is defined as:
L
(
x
;
t
)
=
∑
ξ
∈
z
3
T
(
ξ
;
t
)
I
(
x
-
ξ
)
,
where t∈ + denotes the continuous scale-level, x∈ 3 , L(x;0)=I(x) and T defines a one-parameter family of kernels.
16. The system of claim 12 , wherein applying the artificial agent comprises applying series of actions performed by the artificial agent that change the position and scale of a patch of the medical image in order to parse the medical image without performing an exhaustive search of an entirety of the medical image.
17. The system of claim 12 , wherein the set of actions is determined such that the artificial agent is configured to be able to search an entirety of the medical image.
18. The system of claim 12 , wherein a convergence point at a previous scale is used as a starting point at a subsequent scale for a subsequent action of the sequences of actions.
19. A method for intelligent multi-scale image parsing of a medical image, the method comprising:
evaluating discrete portions of the medical image defined by a position of a state space of an artificial agent relative to the medical image, the state space comprising a parametric space and a scale space, and determining a possible location of a target landmark using the artificial agent, the artificial agent trained to maximize a cumulative reward value of a reward system based on applying each action of a pre-defined set of actions comprising scale-space actions and parametric-space actions, wherein a behavior of the artificial agent comprises a sequence of actions moving the artificial agent towards the possible location of the target landmark of the medical image;
identifying the target landmark on the medical image when the cumulative reward value indicates a proximity of an adjacent state space within a pre-defined reward threshold distance value of the target landmark on the medical image; and
determining that the target landmark is not present in the medical image when the cumulative reward value is outside a pre-defined failure threshold distance value.
20. The method of claim 19 , wherein the sequence of actions comprises a path converging on the target landmark location by parsing less than an entirely of the medical image.