Speaker
Description
Advances in computational methods are rapidly transforming modern medicine, particularly in safety-critical domains such as anesthesia. In these settings, clinicians must make complex and time-critical decisions, motivating the development of automated and data-driven support systems. A recent study demonstrated the feasibility of reinforcement learning (RL) based propofol infusion control using a clinical dataset \cite{YUN23}.
This work implements a learning framework based on the methodology proposed in the original study, providing an independent validation of the approach. To explore the generalizability of the method, the framework is subsequently evaluated on an additional clinical dataset, assessing its robustness beyond the original experimental constraints.
To provide further insight into the model, patient demographic data and administered drug doses are combined with pharmacokinetic–pharmacodynamic models for bispectral index determination. These data serve as the training input for the RL agent, which learns to control propofol infusion automatically, incorporating the effects of manual remifentanil administration by anesthetist. The framework is applied to an independent clinical dataset for evaluation.
The evaluations allow assessment of policy generalization, robustness across patients, and interactions with partially manual drug administration, while the study as a whole contributes to enhancing transparency and reproducibility in automated anesthesia systems.
Bibliography
@article{YUN23,
title = {Deep reinforcement learning-based propofol infusion control for anesthesia: {A} feasibility study with a 3000-subject dataset},
volume = {156},
issn = {00104825},
shorttitle = {Deep reinforcement learning-based propofol infusion control for anesthesia},
url = {https://linkinghub.elsevier.com/retrieve/pii/S0010482523002044},
doi = {10.1016/j.compbiomed.2023.106739},
language = {en},
urldate = {2026-03-14},
journal = {Computers in Biology and Medicine},
author = {Yun, Won Joon and Shin, MyungJae and Jung, Soyi and Ko, JeongGil and Lee, Hyung-Chul and Kim, Joongheon},
month = apr,
year = {2023},
pages = {106739},
}
