Kalman filter soft sensor to handle signal quality loss in closed-loop controlled anesthesia

Background and objective: This study aims to enhance the performance of a closed-loop anesthetic depth control system by fusing noise-corrupted clinical measurements with a non-perfect pharmacological model. Methods: We implement a Kalman filter to constitute a trade-off between model prediction and measurement signal dependence for depth of hypnosis (DoH) control using a previously evaluated PID controller. This trade-off is adjusted online, based on signal quality index (SQI) feedback, provided by the clinical DoH monitor, in this case assumed to be the bispectral index (BIS) monitor. Results: Our simulations show that the proposed solution leads to fundamental performance improvements over the traditional monitor feedback case, which fails to provide the required clinical performance when the SQI drops due to signal inference. In particular, the soft sensor approach increases the time of DoH within the recommended clinical range of 40–60 BIS from 71% to 99%, compared to simple feedback of the noisy monitor output. Conclusion: Our Kalman filter soft-sensor approach succeeds in importantly increasing system robustness to measurement signal disturbances by combining sensor measurements and model predictions.