Thesis Defense for Antonio A. Albanese
Title: Physiology-based Mathematical Models for the Intensive Care Unit: Application to Mechanical Ventilation
Sponsor: Nicolas W. Chbat, PhD
Date & Time: Today, Tuesday, April 15, 2014, at 1:00 pm
Location: BME Conference Room
This work takes us a step closer to realizing personalized medicine, complementing empirical and heuristic ways in which clinicians typically work. This thesis presents mechanistic models of physiology. These models, given continuous signals from a patient, can be fine-tuned via parameter estimation methods so that the model’s outputs match the patient’s. We thus obtain a virtual patient mimicking the patient at hand. Therapeutic scenarios can then be applied and optimal diagnosis and therapy can thus be attained. As such, personalized medicine can then be achieved without resorting to costly genetics.
In particular, we have developed a novel comprehensive mathematical model of the cardiopulmonary system that includes cardiovascular circulation, respiratory mechanics, tissue and alveolar gas exchange, as well as short-term neural control. Validity of the model was proven by the excellent agreement with real patient data, under normo-physiological as well as hypercapnic and hypoxic conditions, taken from literature.
As a concrete example, a submodel of the lung mechanics was fine-tuned using real patient data and personalized respiratory parameters (resistance, R, and compliance, C) were estimated continually. This allows us to compute the patient’s effort (Work of Breathing), continuously and, more importantly, noninvasively.
Finally, the use of Bayesian estimation techniques, which allow incorporation of population studies and prior information about model’s parameters, was proposed in the context of patient-specific physiological models. A Bayesian Maximum a Posteriori Probability (MAP) estimator was implemented and applied to a case-study of respiratory mechanics. Its superiority against the classical Least Squares method was proven in data-poor conditions using both simulated and real animal data.
This thesis can serve as a platform for a plethora of applications for cardiopulmonary personalized medicine.