Actively engaged in control systems and biomedical research, my interests are in developing a University-based Virtual Operating Room and Intensive Care Center. Such a center is motivated by the increasing requirement to develop fundamental new knowledge on drug delivery and patient monitoring advisory systems for the intensive care unit and the operating room. In this context an innovative research program that desires to combine experiments approved by the hospital ethics committee with mathematical modeling in a systems and control framework will enable the understanding and qualifying the conditions leading to a safe environment surrounding the patient. A few of the most relevant and innovative aspects of the current projects are:

The application of process monitoring in detecting changes of multiple physiological parameters, still largely unexplored, promises to make a positive impact in a health care setting. The healthcare professional’s memory of a patient’s physiological process is imperfect especially during long procedures when is difficult to maintain attention to the monitoring systems. Since current alarm systems do not consider trend changes, skilled clinicians must mentally combine multiple instantaneous values over time at the bedside. Automatic detection of significant changes in the physiological process, alternatively, can consistently identify subtle changes in a patient’s state and provide an early alert to the health care professional. The early alert, then, will trigger intervention and prevent adverse outcomes. Current monitor alarms are set to predefined limits unrelated to a specific patient. Thus, normal fluctuations in a patient’s physiological rhythms are often mistaken for significant abnormalities. Intelligent monitoring of physiological processes can differentiate a significant change in the trend of a physiological parameter from normal fluctuations. An intelligent alarm system integrates multivariate analysis to prevent redundant, correlated alarms. It is believed that providing this tool to the anesthetist will improve port operative recovery and lead to new knowledge in applying general anesthesia and neuromuscular block.
 
The objectives for this research project are: I) to collect a significant sample of clinical data to facilitate evaluation of performance of automated systems that detect ‘fault’ conditions in physiological process monitoring; II) to develop an intelligent monitoring decision support system for clinical anesthesiologists that integrates the steady stream of physiological data produced by clinical monitoring systems; III) to apply the ability of the intelligent monitoring system to recognize patterns and identify abnormalities to a control system engineered to reduce anesthesia-related adverse events.
 
A three-phase study to develop a decision support system that is able to represent changes in physiological parameters over time, automatically highlighting significant changes to the overall trend of a variable has been designed. Phase I is an observational study in which the ability of the process monitoring model to identify a process change will be assessed with reference to process changes observed by the anesthesiologist in real-time and post hoc visual inspection, by a selection of clinical experts. In Phase II, the ability of the process monitoring model to identify a process change will be assessed with reference to the anesthesiologist real-time observation of process changes and post hoc expert inspection in a statistically significant sample. In Phase III, anesthesiologists will perform a true assessment of the ability of the process monitoring model to identify a process change in real-time.
 

The third representative project to be undertaken reflects the concept of an advisory system as part of the general anesthesia procedures. Advances in modern anesthesia have been built mostly through the following three aspects of the practice: I) education, which has a key role in making anesthesia as safe and reliable as possible II) sophisticated equipment, that includes standard monitoring devices such as: mass spectrometers, capnographs, pulse oxymeters, heart rate and blood pressure III) the anesthesiologists themselves, who are accessing an extensive pharmacopoeia from which a combination of drugs is selected according to the patient status (i.e. medical records, allergies, age, etc.) and the type and duration of the operation. Similarly to the development of automated flight control in the aeronautic industry, automation in anesthesia is a natural evolutionary step. Previous attempts at closed loop anesthesia throughout the past 60 years were performed on a small number of patients during maintenance. Such achievements have shown that a using a controlled infusion pump (or vaporizer) tracking a given setpoint more accurately than an anesthesiologist is possible.
 
After reviewing the state of the art in this field it has been assessed that the performance obtained from automated anesthesia systems can be greatly improved by using modern control techniques. Anesthesia is characterized by a strong uncertainty in patients’ reactions to the administration of anesthetics and opioids. As a consequence close loop anesthesia systems can bring benefits to the current practice by providing a drug titration based on each patient individual response to the administration of drugs. By applying the proposed control systems is expected that the anesthetic state measured by hypnosis and analgesia will present less fluctuation. The controller will help compensate faster for surgical stimuli as well as detect the onset of the disturbance sooner than the anesthesiologist.
 
Ensuring nominal performance with respect to such uncertainties can be achieved using a robust control/advisory technique, together with an adaptation of the system gains. It is necessary to impose numerous constraints such as maximum allowed drug plasma concentrations and rates of infusion. As consequence the inclusion of a model based constrained predictive controller in cascade with the robust inner loop is recommended. It is also clear that close loop control of anesthesia should not be limited at the regulatory level which has been the main focus of the prior art. The existing synergism between opioids and anesthetics must be used as an advantage when optimizing the titration of the drugs. An appropriate solution will be to allow controller flexibility by which it can optimize drug usage with respect to criteria specified by the anesthesiologist.
 
Based on the developed models, the design of a supervisory/advisory system, integrated in the user accessible auxiliary screen of the requested anesthesia monitor is possible. The models used by the supervisory system will be based at start-up on models available in the literature. As the identification procedure progresses, a complete library of models can be built. From this library an online self-tuning procedure embedded in the advisory system will select the appropriate model. The supervisory/advisory system has to be tested during another experimental study. Feedback from anesthesiologists will be critical in developing the system’s user interface.
 
Extensive simulation will allow the estimation of the maximum uncertainty that can be allowed in the system. A tradeoff between model accuracy and robust performance is required. Based on the simulation results, the focus will be on validating the single variable control system in a clinical environment. Once satisfactory results have been reached, we will adapt the single variable control systems to work in a full multivariable setting.

This research is focused on physiological modeling, circadian state estimation, and control system design to meet the tight performance requirements imposed by the circadian cycle control application. Since we advocate a model-based solution to circadian control, therefore a major focus of the work will be the development of complete circadian physiology models. Comprehensive models of circadian physiology exist for a number of the subject inputs and outputs. However, many of the relevant models are based on a qualitative approach to the data. Key investigations will be identified by an extensive literature search and the experimental data will be re-examined from a quantitative paradigm. Making use of system identification techniques, existing models will be adapted for model-based control.
 
Accurate physiological measurement is necessary for feedback control. As such, the integration and development of physiological sensors will follow the initial model investigation. Signal processing, for noise removal and pattern detection, will enhance the accuracy of the recorded measurements. Upon complete instrumentation of the subject, circadian state estimation techniques will be developed using available measurements and models. Finally, a controller based on complex non-linear physiological models, and capable of managing inherent subject constraints, will be developed. By applying state-of-the-art knowledge in system identification, estimation and observer design, adaptive control, hierarchical control schemes and model-based predictive control, in a collaborative environment the above goals can be reached.
 
A key factor for the success of the control system development is having accurate models of human circadian physiology. Accordingly, the first stage of this project involves further expansion of the existing physiological models. The focus will be on models linking the effects of actuation stimulus with sensed physiological variables of the human body.
 
The strongest actuator for circadian cycle control is light. Detailed models of its effects are available. Complete subject control will be achieved by augmenting this conventional actuation with additional inputs including: sleep schedules, physical exercise timing, intake of calories, and environmental temperature. The effects of these secondary inputs to human circadian physiology have been qualitatively examined. However, only few models suitable for the purposes of control have been developed.
 
The human body presents a large number of measurable outputs. Identification of the most significant circadian physiology outputs will focus the physiological sensor development into a number of primary areas. It is anticipated that these areas will include sensors to measure the heart rate, core body temperature, physical activity, and ambient light exposure. Availability of essential physiological measurements from the subject will allow validation of currently available research models and further modeling initiatives.
 
Remote and non-invasive sensors are required to monitor the subject without restricting independence. Existing non-invasive commercial sensors are available for a variety of physiological variables and some sensors will modified to meet specific custom requirements. The sensors will be combined into a portable integrated sensing system.
 
A circadian state estimation algorithm will be developed through the application of observer design theory such as model-based Kalman filtering. As a result a composite index based on available measurements of the endogenous circadian pacemaker phase and amplitude will be developed. A precise definition of the endogenous circadian pacemaker phase and amplitude will ensure the performance of the future control system by providing robust measurements.
 
Current models are focused on the primary effect of the circadian cycle on core body temperature and cognitive performance. A reassessment of these models in the light of new inputs and outputs is the object of this project milestone. Data required is grouped in a number of key areas: I) the relationship between various measurable body responses and the complex circadian cycle index; II) evoked potentials in the EEG signals used to establish levels of sleep; III) evoked potentials able to detect the muscle tonus; IV) maximum limits allowable in shifting the circadian cycle peak; V) maximum intake levels of sound, light, heat, exercise, calories over a determined period of time; VI) weight, sex, age, behavior of the subjects for the circadian cycle transition etc.
 
Following the development of the physiological models, sensors, and circadian state estimation techniques, the framework will be set for the application of closed-loop control. The SISO, model based controller previously developed will serve as foundation. This controller will be augmented with the inclusion of new models, and new functionality to meet tight operational requirements. To achieve this, research and development of advanced constraint-based optimization, non-linear multi-variable control, and adaptive learning will be emphasized. The existing SISO controller manipulates light input to achieve control over an individual’s circadian state. The inclusion of the new models will expand the controller into a non-linear MIMO system capable of manipulating light, sleep schedules, physical activity and eating schedules to achieve control over alertness and cognitive performance. One approach considered to deal with the non-linear aspects of the models is quasi-linear parameter variation.
 
The introduction of stimulus to an ambulatory human subject involves significant actuator constraints. Such constraints include I) maximum levels, II) rate at which inputs such as temperature or intake of food can be increased from the previous value III) the duration that actuation can be maintained, IV) scheduled activity of subject. Optimization techniques, such as constrained model-based prediction will provide an elegant solution to optimum control performance in the presence of complex constraints. An important consideration in systems interacting with human physiology is inter-subject variability. The feedback provided by the sensors and state-estimation techniques will provide a significant degree of compensation for this variability and will allow nominal subject models to be used with accuracy. This is a primary competitive advantage of closed-loop control over open-loop methods. An additional capability, which is on-line model identification and adaptation, will be developed for the controller. All control system development will be performed in Matlab, and following successful simulations, it will be ported to a “D-Space” rapid-prototyping system and to a personal-computer with sensor interface hardware capability. A graphical user interface will be developed to allow easy interaction during further refinement and testing stages. The use of the Matlab design environment and the rapid prototyping system will allow quick iterations through the testing and tuning cycle of the control algorithm development.