Plan optimization method, computing apparatus used for optimizing plan, and computer-readable medium

1 . A plan optimization method, comprising:

obtaining a plurality of medical variables at a current time, wherein at least one of the medical variables at the current time is caused by an implementation of medical variables at a previous time with a different value or variable type, and one of the medical variables at the current time corresponds to a state transition from a previous state at the previous time; and

determining plan information by inputting the medical variables at the current time into a machine learning model, wherein the plan information comprises at least one subplan information, each of the subplan information corresponds to at least one treatment manner record at a subsequent time, the previous time is earlier than the current time, and the subsequent time is later than the current time, the machine learning model is a causal graph model, and determining the plan information comprises:

determining causal relationships corresponding to the medical variables at the current time through the causal graph model, wherein

in response to the at least one of the medical variables at the current time being different from the medical variables at the previous time, generating causal relationships different from that corresponding to the previous time through the causal graph model, and accordingly changing a treatment manner record in at least one subplan information corresponding to the previous time; and

setting an intensity matrix, wherein the intensity matrix corresponds to a probability of a state transition of an effect node, an input of the previous state or an input of a new state at the current time corresponds to the state transition, and each of a plurality of elements of the intensity matrix corresponds to an instantaneous rate of the state transition of two states.

2 . The plan optimization method according to claim 1 , wherein determining the causal relationships corresponding to the medical variables at the current time through the causal graph model comprises:

setting a first period belonging to a first probability distribution, wherein the first period is a period between a precondition node and an action node, the precondition node corresponds to an input of the previous state, the action node is configured to trigger an intervention behavior of the causal relationships corresponding to the medical variables, and the intervention behavior is to change the causal relationships corresponding to the medical variables through the causal graph model;

setting a second period corresponding to the current time and belonging to a second probability distribution, wherein the second period is a delay of the intervention behavior and corresponds to a period from triggering the intervention behavior to actually executing the intervention behavior; and

setting a third period belonging to a third probability distribution, wherein the third period is a period between the action node and the effect node, and the effect node corresponds to an input of a new state at the current time.

3 . The plan optimization method according to claim 2 , wherein the first probability distribution is a delta distribution, the second probability distribution is a gamma distribution, and the third probability distribution is an exponential distribution.

4 . The plan optimization method according to claim 1 , wherein the instantaneous rate of at least one of the elements in the intensity matrix corresponding to the state transition in which the previous state is a first state corresponds to a period belonging to a probability distribution, the instantaneous rate of at least one of the elements in the intensity matrix corresponding to the state transition in which the previous state is a second state corresponds to transitioning to a new state without executing an intervention behavior, and the instantaneous rate of at least one of the elements in the intensity matrix corresponding to the state transition in which the previous state is a third state corresponds to not executing the intervention behavior while maintaining state.

5 . The plan optimization method according to claim 1 , wherein determining the causal relationships corresponding to the medical variables at the current time through the causal graph model comprises:

training the causal graph model through a continuous time structural equation model (CTSEM), and determining a causal relationship between a precondition node and an action node, or a causal relationship between the action node and the effect node, wherein the precondition node corresponds to an input of the previous state, the action node is configured to trigger an intervention behavior of causal relationships corresponding to the medical variables, and the effect node corresponds to an input of a new state at the current time.

6 . The plan optimization method according to claim 1 , wherein obtaining the medical variables at a current time comprises:

obtaining a medical image;

identifying a plurality of objects in the medical image; and

grouping the objects according to corresponding treatment effect of the objects to generate a grouping result, wherein the medical variables comprise the grouping result.

7 . The plan optimization method according to claim 6 , wherein grouping the objects according to the corresponding treatment effect of the objects comprises:

estimating the corresponding treatment effect of each of the objects; and

grouping the estimated treatment effects through a grouping algorithm.

8 . The plan optimization method according to claim 7 , wherein estimating the corresponding treatment effect of each of the objects comprises:

estimating a conditional average treatment effect (CATE) through a Bayesian structural equation model (BSEM), wherein the conditional average treatment effect is a treatment effect corresponding to one of the object.

9 . A computing apparatus used for optimizing a plan, comprising:

a storage, storing program code; and

a processor, coupled to the storage, loading the program code and executing:

obtaining a plurality of medical variables at a current time, wherein at least one of the medical variables at the current time is caused by an implementation of medical variables at a previous time with a different value or variable type, and one of the medical variables at the current time corresponds to a state transition from a previous state at the previous time; and

determining plan information by inputting the medical variables at the current time into a machine learning model, wherein the plan information comprises at least one subplan information, each of the subplan information corresponds to at least one treatment manner record at a subsequent time, the previous time is earlier than the current time, and the subsequent time is later than the current time, the machine learning model is a causal graph model, and determining the plan information comprises:

determining causal relationships corresponding to the medical variables at the current time through the causal graph model, wherein

in response to the at least one of the medical variables at the current time being different from the medical variables at the previous time, generating causal relationships different from that corresponding to the previous time through the causal graph model, and accordingly changing a treatment manner record in at least one subplan information corresponding to the previous time; and

setting an intensity matrix, wherein the intensity matrix corresponds to a probability of a state transition of the effect node, an input of the previous state or an input of a new state at the current time corresponds to the state transition, and each of a plurality of elements of the intensity matrix corresponds to an instantaneous rate of the state transition of two states.

10 . The computing apparatus used for optimizing the plan according to claim 9 , wherein the processor further executes:

setting a first period belonging to a first probability distribution, wherein the first period is a period between a precondition node and an action node, the precondition node corresponds to an input of the previous state, the action node is configured to trigger an intervention behavior of the causal relationships corresponding to the medical variables, and the intervention behavior is to change the causal relationships corresponding to the medical variables through the causal graph model;

setting a second period corresponding to the current time and belonging to a second probability distribution, wherein the second period is a delay of the intervention behavior and corresponds to a period from triggering the intervention behavior to actually executing the intervention behavior; and

setting a third period belonging to a third probability distribution, wherein the third period is a period between the action node and the effect node, and the effect node corresponds to an input of a new state at the current time.

11 . The computing apparatus used for optimizing the plan according to claim 10 , wherein the first probability distribution is a delta distribution, the second probability distribution is a gamma distribution, and the third probability distribution is an exponential distribution.

12 . The computing apparatus used for optimizing the plan according to claim 9 , wherein the instantaneous rate of at least one of the elements in the intensity matrix corresponding to the state transition in which the previous state is a first state corresponds to a period belonging to a probability distribution, the instantaneous rate of at least one of the elements in the intensity matrix corresponding to the state transition in which the previous state is a second state corresponds to transitioning to a new state without executing an intervention behavior, and the instantaneous rate of at least one of the elements in the intensity matrix corresponding to the state transition in which the previous state is a third state corresponds to not executing the intervention behavior while maintaining state.

13 . The computing apparatus used for optimizing the plan according to claim 9 , wherein the processor further executes:

training the causal graph model through a continuous time structural equation model, and determining a causal relationship between a precondition node and an action node, or a causal relationship between the action node and the effect node, wherein the precondition node corresponds to an input of the previous state, the action node is configured to trigger an intervention behavior of causal relationships corresponding to the medical variables, and the effect node corresponds to an input of a new state at the current time.

14 . The computing apparatus used for optimizing the plan according to claim 9 , wherein the processor further executes:

obtaining a medical image;

identifying a plurality of objects in the medical image; and

grouping the objects according to corresponding treatment effect of the objects to generate a grouping result, wherein the medical variables comprise the grouping result.

15 . The computing apparatus used for optimizing the plan according to claim 14 , wherein the processor further executes:

estimating the corresponding treatment effect of each of the objects; and

grouping the estimated treatment effects through a grouping algorithm.

16 . A non-transitory computer-readable medium, loading program code through a processor and executing following steps:

obtaining a plurality of medical variables at a current time, wherein at least one of the medical variables at the current time is caused by an implementation of medical variables at a previous time with a different value or variable type, and one of the medical variables at the current time corresponds to a state transition from a previous state at the previous time; and

determining plan information by inputting the medical variables at the current time into a machine learning model, wherein the plan information comprises at least one subplan information, each of the subplan information corresponds to at least one treatment manner record at a subsequent time, the previous time is earlier than the current time, and the subsequent time is later than the current time, the machine learning model is a causal graph model, and determining the plan information comprises:

determining causal relationships corresponding to the medical variables at the current time through the causal graph model, wherein

in response to the at least one of the medical variables at the current time being different from the medical variables at the previous time, generating causal relationships different from that corresponding to the previous time through the causal graph model, and accordingly changing a treatment manner record in at least one subplan information corresponding to the previous time; and

setting an intensity matrix, wherein the intensity matrix corresponds to a probability of a state transition of the effect node, an input of the previous state or an input of a new state at the current time corresponds to the state transition, and each of a plurality of elements of the intensity matrix corresponds to an instantaneous rate of the state transition of two states.