Modeling and estimation for non-invasive monitoring of intracranial pressure and cerebrovascular autoregulation

Brain tissue is highly vulnerable to unbalanced oxygen demand and supply. A few seconds of oxygen deficit may trigger neurological symptoms, and sustained oxygen deprivation over a few minutes may result in severe and often irreversible brain damage. The rapid dynamics coupled with the potential for severe injury necessitate continuous cerebrovascular monitoring in the populations at greatest risk for developing or exacerbating brain injury. Intracranial pressure (ICP), which is the pressure of the cerebrospinal fluid, is a vitally important variable to monitor in a wide spectrum of medical conditions involving the brain, such as traumatic brain injury, stroke, hydrocephalus, or brain tumors. However, clinical measurement of ICP is highly invasive, as it requires neurosurgical penetration of the skull and placement of a pressure sensor in the brain tissue or ventricular spaces. Measurement of ICP is thus currently limited to only those patient populations in which the benefits of obtaining the measurement outweigh the significant attendant risks, thus excluding a large pool of patients who could potentially benefit from ICP monitoring. The primary goal of our work is to address the non-invasive monitoring of ICP. A secondary aim of this work is to develop methods for the assessment of cerebrovascular autoregulation, which is the innate ability of the vasculature to maintain cerebral blood flow in the face of changes in cerebral perfusion pressure. Cerebrovascular autoregulation is often impaired in patients with brain trauma or stroke, and also in pre-term neonates, as their cerebrovascular system is not fully matured. We develop methods for non-invasive, continuous, calibration-free and patientspecific ICP monitoring. Specifically, we present a model-based approach to providing real-time estimates of ICP and cerebrovascular resistance and compliance, for each cardiac cycle, from nonor minimally-invasive time-synchronized measurements of arterial blood pressure and cerebral blood flow velocity in a major cerebral artery. In the first step, our approach exploits certain features of cerebrovascular physiology, along with model reduction ideas, to deduce a simple mathematical model of the cerebrovascular system. In the second step, we develop algorithms to compute robust estimates of model parameters by processing the measured waveforms through the constraints provided by the models dynamic equation. For validation, our non-invasive estimates of ICP were compared against invasive measurements from 45 comatose brain-injury patients, with a total of 35 hours of data (over 150,000 beats), providing more than 3,500 independent ICP estimates. Our estimates track measured ICP closely over a range of dynamic variations. Pooling all independent estimates resulted in a mean estimation error (bias) of less than 2 mmHg and a standard deviation of error of about 8 mmHg. We also suggest how variations in estimated cerebrovascular resistance and compliance in response to variations in cerebral perfusion pressure may be used to provide novel approaches for assessment of cerebrovascular autoregulation. Thesis Co-supervisor: Professor George C. Verghese Title: Professor of Electrical Engineering Thesis Co-supervisor: Dr. Thomas Heldt Title: Research Scientist