Optimizing patient treatment recommendations using reinforcement learning combined with recurrent neural network patient state simulation

A non-transitory computer readable medium comprising instructions that, when executed by at least one processor, configure the at least one processor to:

• train a Reinforcement Learning (RL) action generator based at least in part on the received patient data of a plurality of patients to generate recommended treatment actions based on at least one state of a patient;
• train a Recurrent Neural Network (RNN) state simulator based at least in part on the received patient data of the plurality of patients to simulate new states of a patient based on recommended treatment actions generated by the RL action generator;
• apply the RL action generator to a state of a patient in the received patient data to determine a next recommended treatment action for the patient;
• determine whether a convergence in the RL action generator has been reached;
• optimize the RL action generator until the convergence in the RL action generator is determined to be reached by iteratively repeating:
• applying the RNN state simulator to the determined next recommended treatment action to simulate a next state;
• applying the RL action generator to the simulated next state to determine the next recommended treatment action for a patient, the applying the RNN state simulator comprising:
• at an iteration, receiving, as inputs to a simulator node of said RNN, a prior simulated state from an immediate prior simulator iteration, an immediate prior recommended treatment action for the patient, and the determined next recommended treatment action from said RL action generator to generate the next simulated state;
• determining whether the convergence in the RL action generator has been reached;
• determine that the RL action generator has been optimized in response to determining that the convergence has been reached;
• receive a patient state as a query;
• apply the optimized RL action generator to the received patient state to determine an optimized recommended treatment action for the received patient state;
• and
• output the optimized recommended treatment action for the received patient state.