Artificial neural networks for prediction and control in neurointensive care
The aim of the project is to improve the treatment of neurointensive care unit (NICU) patients with traumatic brain injury or subarachnaoidal haemhorrage. We intend to construct a decision support system, enabling the clinician to better handle the large amounts of data generated at a NICU.
Continuously monitored patient data are
analysed with a set of related techniques, most of which can be characterised
as artificial neural networks and/or Bayesian strategies. The analysis
is guided by an explicit modelling of the pathophysiological processes in
severe brain injury, building on existing knowledge of the involved mechanical,
hydromechanical, osmotic, metabolic and nervous mechanisms. These models will
also be tested using methods from control systems theory, partly using new
probing devices developed within the project. The results of these different
analyses will eventually be fed into the decision support system. Because
of the cross-scientific character of the study, it involves specialists in
neurosurgery, anesthesiology, statistics, signal analysis and systems engineering.
Severe brain injuries are a major health economic problem, and improvements in the care of these patients is an urgent aim not only from a humanitarian but also from an economic perspective. The data acquisition and modelling techniques used in this project also have general implications for the development of decision support systems, smart multiparameter alarms, and better pharmacokinetic models to improve our understanding of the effects of drug therapies.
Project managers: Bertil Rydenhag (Dept.
of Neurosurgery, The Sahlgrenska Academy, Göteborg); Helge Malmgren
(Dept. of Philosophy, Göteborg University).
Major partners: Richard Dybowski
(InferSpace, London); Lars Lindström (Dept.
of Clinical Physiology, Sahlgrenska University Hospital); Hans Sandholt (Mechatronics,
Chalmers School of Technology), Tim Howells (Department of Neuroscience,
Uppsala Akademiska Sjukhus).
References:
T. P. Howells, Modeling physiological time series data using Bayesian
neural networks. Edinburgh 2001
R. Dybowski & V. Grant (eds)., Clinical applications of artificial
neural networks. London 2001