Abstract

In this talk, I will first cover general topics related to our long-term vision, experience, and existing infrastructure in supporting translational biomedical informatics research in a neurosurgical department. These topics will include our philosophy of team construction, the problems facing modern healthcare where amount of data to make sense is escalating, and in particular, the problems with current monitoring practice in an intensive care unit. I will then introduce a new direction we are pursuing to provide cognitive support for the medical decisions in ICU where continuous data are abundant. The second part of talk will cover three published papers from our group related to applying signal processing/modelling and machine learning techniques to intracranial pressure and cerebral hemodynamic signals. The first paper described an approach that combines an advanced intracranial pressure pulse morphology characterization method with pattern recognition techniques to predict acute intracranial pressure elevation. The second paper described a similar technique that can provide continuous monitoring of global cerebral hypoperfusion, as validated by Xeon133 CBF , using morphological features of intracranial pressure. The last paper utilized the concept of pulse wave propagation in the cerebral vasculature to derive a model of pulse onset latency relative to ECG QRS peak from an intracranial (for example, cerebral blood flow velocity at THE MCA ) and that of an extracranial pulse signal (for example, arterial blood pressure pulse at the radial artery). Using this model, one can reduce confounding extracranial influence on using intracranial pulse onset latency to characterize the cerebral vasculature as validated by a dataset from a cuff test experiment as presented in the paper.