A DNN assisted parallel control architecture for automated anesthesia delivery addressing inter-patient variability

Area of Science:

• Anesthesiology
• Control Engineering
• Biomedical Engineering

Background:

• Accurate depth of hypnosis (DoH) control is vital in anesthesia for patient safety and surgical outcomes.
• Traditional PID controllers struggle with the nonlinear and patient-specific dynamics of anesthetic agents.
• Existing methods lack robustness across diverse patient profiles, necessitating advanced control strategies.

Purpose of the Study:

• To develop a robust and adaptive control framework for precise anesthesia depth management.
• To decouple setpoint tracking and disturbance rejection for improved controller performance.
• To enable personalized anesthetic control through patient-specific parameter optimization.

Main Methods:

• Implementation of a two-degree-of-freedom (2-DOF) parallel control scheme.
• Integration of an offline deep neural network (DNN)-based gain scheduling strategy for adaptability.
• Optimization of controller parameters using genetic algorithms for 13 patient categories, minimizing integral of absolute error (IAE).

Main Results:

• The proposed framework demonstrated significant improvements in maintaining the Bispectral Index (BIS) within the target range (40-60).
• Simulations confirmed the controller's ability to adapt to patient-specific characteristics for precise anesthetic delivery.
• The 2-DOF scheme effectively managed both setpoint tracking and disturbance rejection.

Conclusions:

• The robust and adaptive control framework offers a promising solution for personalized anesthesia management.
• DNN-based gain scheduling and genetic algorithm optimization enhance controller performance and patient safety.
• This approach represents a significant advancement in automated anesthesia control systems.