Towards Safe Mechanical Ventilation Treatment Using Deep Offline Reinforcement Learning

Mechanical ventilation is a key form of life support for patients with pulmonary impairment. Healthcare workers are required to continuously adjust ventilator settings each patient, challenging and time consuming task. Hence, it would be beneficial develop an automated decision tool optimize treatment. We present DeepVent, Conservative Q-Learning (CQL) based offline Deep Reinforcement Learning (DRL) agent that learns predict the optimal parameters patient promote 90 day survival. design clinically relevant intermediate reward encourages continuous improvement vitals as well addresses challenge sparse in RL. find DeepVent recommends within safe ranges, outlined recent clinical trials. The CQL algorithm offers additional safety by mitigating overestimation value estimates out-of-distribution states/actions. evaluate our using Fitted Q Evaluation (FQE) demonstrate outperforms physicians from MIMIC-III dataset.