⚡ electrical engineering

BIS-Based Model Reference Adaptive Control of Anesthesia Depth under Patient Variability and Time-Varying Delays

This paper proposes a Model Reference Adaptive Control (MRAC) strategy for automated, closed-loop anesthesia depth regulation using propofol and remifentanil that effectively handles inter-patient variability and time-varying delays without requiring patient-specific tuning, achieving stable Bispectral Index (BIS) maintenance within the clinically acceptable range of 40–60 in heterogeneous virtual patient simulations.

Original authors: Raha Rahimi, Marzieh Kamali, Fazaneh Shayegh, Zahra Baharlouei

Published 2026-08-03

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Raha Rahimi, Marzieh Kamali, Fazaneh Shayegh, Zahra Baharlouei

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are the conductor of a very delicate orchestra, but instead of violins and flutes, your instruments are the body's own chemistry. The music you are trying to create is a state of deep, painless sleep called general anesthesia. It's a high-stakes performance where the goal is to keep the patient completely unconscious and free from pain, but not so deep that their heart or breathing slows down dangerously. To do this, doctors use special drugs that act like volume knobs for the brain. However, every person's body is a different instrument; some are loud and react fast, others are quiet and slow. On top of that, the drugs don't work instantly—they take a little while to travel through the bloodstream and reach the brain, creating a confusing "lag" between giving the medicine and seeing the result. If a doctor tries to adjust the knobs manually while the music is playing, they might be reacting to yesterday's volume instead of today's, leading to a chaotic performance. This is the tricky puzzle scientists in the field of medical control systems have been trying to solve: how to automate this process so the "music" stays perfect, no matter who is on the operating table or how fast the drugs travel.

This paper introduces a clever new way to solve that puzzle using a "Model Reference Adaptive Control" (MRAC) system. Think of this system as a super-smart, self-tuning autopilot for anesthesia. Instead of needing to know the exact details of a specific patient's body before starting (like their weight, age, or how fast their liver works), this new controller learns on the fly. It has a built-in "ideal conductor" (the reference model) that knows exactly how the music should sound. As the real patient's body reacts, the controller constantly compares the actual music to the ideal one and instantly adjusts the drug knobs to match. The researchers tested this idea using computer simulations with a diverse group of 12 "virtual patients," each with different body types and drug sensitivities. They also threw in some curveballs, like sudden surgical pain (simulated as a "disturbance") and the tricky time delays that happen in real life.

The results from these simulations suggest that this new approach is highly effective. The system successfully guided the patients' brain activity, measured by a score called the Bispectral Index (BIS), into the safe "sweet spot" of 40 to 60. During the induction phase (when the patient is put to sleep), the system reached the target in about 4 to 6 minutes, with very little overshoot (the brain didn't get too sleepy too fast). Even when the virtual patients had different body chemistries or when the time it took for drugs to work changed, the controller kept the BIS score steady, hovering right around 50 with tiny deviations of just ±2. Crucially, the system handled the "lag" in the drugs without needing to be re-tuned for each person. The authors found that by using a special mathematical trick to reverse-engineer the complex relationship between the drugs and the brain signal, they could keep the system stable and safe. While these findings are currently limited to computer simulations and have not yet been tested on real humans, the study suggests that this adaptive, delay-compensating controller could be a promising solution for making automated anesthesia safer and more personalized in the future.

Technical Summary: BIS-Based Model Reference Adaptive Control of Anesthesia Depth under Patient Variability and Time-Varying Delays

Problem Statement
The administration of general anesthesia requires the precise regulation of hypnotic (propofol) and analgesic (remifentanil) drugs to maintain unconsciousness and analgesia. A primary challenge in automated closed-loop control is the significant inter-patient variability in pharmacokinetic/pharmacodynamic (PK/PD) parameters and the presence of physiological time-varying delays in drug distribution. Conventional control strategies, such as fixed-gain PID or Model Predictive Control (MPC), often rely on offline patient-specific parameter identification or assume instantaneous drug distribution. These limitations reduce their robustness in real clinical settings where patient physiology changes (e.g., cardiac output variations) and delays fluctuate. Furthermore, the nonlinear relationship between drug concentrations and the Bispectral Index (BIS) output, governed by a Hill function with drug interaction terms, complicates the design of linear feedback controllers.

Methodology
The authors propose a Model Reference Adaptive Control (MRAC) framework designed to regulate anesthesia depth using a dual-drug infusion of propofol and remifentanil. The methodology integrates the following components:

Plant Modeling: The patient is modeled using standard three-compartment PK models (Schnider) for both drugs, extended to include time-varying delays ( $\tau(t)$ ) in the inter-compartmental transfer. The pharmacodynamic (PD) component links effect-site concentrations to the BIS output via a nonlinear Hill function that accounts for the synergistic interaction between propofol and remifentanil.
Inverse Hill Reconstruction: To address the nonlinearity of the Hill function and the ambiguity of multi-input control, the authors introduce a simplified inverse reconstruction strategy. By assuming the normalized remifentanil effect is proportional to the normalized propofol effect ( $Y_r \approx \mu Y_p$ ), the complex multi-drug interaction is reduced to a single-variable nonlinear relationship. An inverse Hill function ( $H^{-1}_\mu$ ) is then applied to the measured BIS signal to estimate the linearized propofol effect-site concentration ( $\hat{Y}_p$ ), which serves as the feedback variable.
Adaptive Control Law: The core controller is an MRAC scheme tailored for systems with relative degree two and unknown time-varying delays.
- Reference Model: A second-order reference model ( $W_m(s)$ ) defines the desired transient and steady-state response (target BIS of 50).
- Control Structure: The control input is generated using filtered regressors derived from the input and estimated output. To handle the relative degree and ensure stability, a first-order stable filter $Z(s) = s + p_0$ is introduced to render the error dynamics strictly positive real (SPR).
- Adaptation and Robustness: The controller parameters are updated online via an adaptive law derived from a Lyapunov-Krasovskii functional. A specific robustness term involving the sign of the tracking error and an integral term is included to compensate for unknown time-varying delays and external disturbances (e.g., surgical stimuli).
Stability Analysis: Theoretical stability is proven using a Lyapunov-Krasovskii functional, demonstrating that all closed-loop signals remain bounded and the tracking error asymptotically converges to zero, provided the delay rate of change is less than unity.

Key Contributions
The paper outlines five specific contributions:

Parameter Independence: The MRAC approach does not require exact patient-specific PK/PD parameter identification or prior tuning, unlike conventional PID or event-driven controllers.
Explicit Delay Handling: Unlike existing adaptive frameworks (e.g., Fractional-Order MRAC), this method explicitly incorporates physiological time-varying delays into the control design and stability analysis, rather than relying solely on fractional-order memory effects.
Nonlinear Compensation: The proposed simplified inverse Hill reconstruction allows for the estimation of a linear output from the nonlinear BIS signal, facilitating adaptive output-feedback control despite complex multi-drug interactions.
Robustness: The adaptive law continuously updates parameters online, ensuring stability against inter-patient variability and intraoperative disturbances without the need for controller recalibration.
Clinical Safety: The controller regulates infusion rates within clinically accepted standard ranges, preventing both overdosing and insufficient anesthesia.

Results
The proposed method was validated through simulations on a heterogeneous population of 12 virtual patients with varying biometric characteristics (age, weight, height, gender) and randomly generated PK/PD parameters.

Induction Phase: The controller successfully drove the BIS from baseline to the target of 50 across all patients. The Time to Target (TT) ranged from 104 to 193 seconds (mean ~145s), and the Settling Time (ST10) ranged from 141 to 360 seconds (mean ~263s). The BIS-NADIR (minimum BIS) remained above 40 in all cases, avoiding excessive depth.
Maintenance Phase: Under simulated surgical disturbances (double-step inputs), the controller rapidly restored the BIS to the target range. Recovery times from positive and negative disturbances were short (mean ~0.6 min), and the BIS remained within the 40–60 range with minimal overshoot (<5%) and small deviations (±2).
Drug Administration: Infusion profiles for propofol and remifentanil remained within typical clinical limits for both induction and maintenance phases.
Comparison: The proposed MRAC outperformed or matched existing strategies (Fractional-Order MRAC, Optimized PID, Nonlinear MPC) by achieving competitive tracking without requiring patient-specific model parameters or offline tuning.

Significance and Claims
The authors claim that this study provides a promising solution for robust and individualized closed-loop anesthesia management. The primary significance lies in the controller's ability to operate effectively without prior patient-specific parameter identification, thereby addressing a major barrier to the clinical deployment of automated anesthesia systems. By explicitly modeling and compensating for time-varying delays and nonlinear drug interactions, the proposed MRAC framework offers a theoretically grounded, scalable approach that maintains patient safety and stability across diverse physiological conditions. The work suggests that such a system could enhance patient safety, reduce drug consumption, and improve recovery times compared to manual or less adaptive automated methods.

More like this