System and method for optimizing operations of a radiology service using ai powered gamification
Systems and methods for managing patient diagnostic and therapy workflows in a hospital and/or radiology centers. A radiology recommendation agent is trained using reinforcement learning and a simulation environment in which the agent takes actions and receives feedback from the simulation environment based on how its action affect the simulation environment over time.
1 . An AI powered gamification method for radiology center management, the method comprising:
generating a simulation environment of a radiology center, the simulation environment comprising radiology scheduling, planning, settings and scanning parameters for diagnostic imaging radiology procedures, diagnostics, and therapy workflows in the radiology center;
training an agent using the simulation environment in which the agent takes actions and receives feedback from the simulation environment based on how its action affect the simulation environment over time, wherein training the agent comprises training the agent in the simulation environment in the presence of at least one adversarial agent configured to perturb one or more operational parameters of the simulation environment, the adversarial agent receiving a reward when the agent fails to improve one or more key performance indicators, such that the agent is trained to generate recommendations that are robust to adverse and non-stationary operating conditions of the radiology center; and
implementing the agent to provide real time recommendations that optimize an operation of the radiology center.
2 . The method of claim 1 , wherein the feedback from the simulation environment comprises one or more reward values related to estimated key performance indicators.
3 . The method of claim 2 , wherein the estimated key performance indicators comprise at least one of patient wait times upon arrival, throughput of patients and the time to exam, staff utilization, and equipment utilization.
4 . The method of claim 2 , wherein the estimated key performance indicators are derived from real world results of actions used to generate the simulation environment.
5 . The method of claim 1 , further comprising:
continuously re-training the agent with new data from the physical world.
6 . The method of claim 1 , wherein the simulation environment comprises a Markov process defined over a state of physical entities and dynamics defined via state transition functions.
7 . The method of claim 6 , wherein generating the simulation environment comprises:
iteratively refining state representations and state transition functions until a reality gap between the forecasted observations based on a world model and observed data is statistically small.
8 . The method of claim 1 , wherein the simulation environment further comprises at least a patient specific model including an age, gender, and location of a patient.
9 . The method of claim 1 , wherein the agent is modeled as a single agent that estimates a state of the world, takes actions, and receives rewards based on whether the simulation environment evolves favorably or not.
10 . The method of claim 1 , wherein a plurality of agents are trained to operate and generate recommendations collaboratively.
11 . A system for radiology center management, the system comprising:
an agent configured to provide recommendations for radiology scheduling and operations, wherein training the agent comprises training the agent in a simulation environment in the presence of at least one adversarial agent configured to perturb one or more operational parameters of the simulation environment, the adversarial agent receiving a reward when the agent fails to improve one or more key performance indicators, such that the agent is trained to generate recommendations that are robust to adverse and non- stationary operating conditions of the radiology center, wherein the simulation environment comprises radiology scheduling, planning, options when performing a radiology procedure, diagnostics, and therapy workflows in a radiology center; and
a graphical user interface configured to output the recommendations to a radiology operator.
12 . The system of claim 11 , wherein the feedback from the simulation environment comprises one or more reward values related to estimated key performance indicators.
13 . The system of claim 12 , wherein the estimated key performance indicators comprise at least one of patient wait times upon arrival, throughput of patients and a time to exam, staff utilization, and equipment utilization.
14 . The system of claim 11 , wherein the agent is modeled as a single agent that estimates the state of the world, take actions, and receives rewards based on whether the simulation environment evolves favorably or not.
15 . The system of claim 11 , wherein a plurality of agents are trained to operate and generate recommendations collaboratively.
16 . A system for radiology center management, the system comprising:
a CT scanning system configured to scan a patient, the scan comprising one or more settings and scanning parameters;
a scheduling system configured to schedule the scan and radiology personnel for the scan; and
a radiology recommendation agent configured to recommend one or more actions for the scheduling system and the CT scanning system including recommended settings and scanning parameters, the radiology recommendation agent trained using reinforcement learning and a simulation environment in which the radiology recommendation agent takes actions and receives feedback from the simulation environment based on how its action affect the simulation environment over time, wherein training the agent comprises training the radiology recommendation agent in the simulation environment in the presence of at least one adversarial agent configured to perturb one or more operational parameters of the simulation environment, the adversarial agent receiving a reward when the radiology recommendation agent fails to improve one or more key performance indicators, such that the radiology recommendation agent is trained to generate recommendations that are robust to adverse and non-stationary operating conditions of the radiology center.
17 . The system of claim 16 , wherein the radiology recommendation agent is further configured to recommend follow up actions for the patient based on results of the scan.