Objective: To analyze patterns of cerebral microdialysis in patients with traumatic brain injury and, with a neural network methodology, investigate pattern relationships to intracranial pressure and cerebral perfusion pressure.
Design: Retrospective.
Setting: University hospital, adult neurosurgical intensive care unit.
Patients: Twenty-six patients with severe traumatic brain injury. All consecutive traumatic brain injured patients (Glasgow Coma Scale < or =8) with microdialysis monitoring, analyzing glutamate, lactate, pyruvate, and glucose in both penumbral and nonpenumbral tissue.
Interventions: None; patients received the unit's standard neurointensive care procedure.
Measurements and main results: We used 2084 hrs of complete microdialysis data sets (eight markers) to train Kohonen self-organizing maps. The self-organizing map algorithm is a data-clustering method that reduces high-dimensional information to a two-dimensional representation on a grid (map), retaining local relationships in the data. Maps were colored (overlaid) for intracranial pressure, cerebral perfusion pressure, and outcome, to explore relationships with underlying microdialysis patterns. The maps exhibited a striking clustering of patients, with unique microdialysis patterns that were recognizable throughout the analysis period. This also held true for most microdialysis patterns characteristic of ischemia. These patients with ischemic patterns can have good outcomes, suggesting a disparity between microdialysis values and severity of traumatic brain injury.
Conclusion: Using an artificial neural network-like clustering technique, Kohonen self-organizing maps, we have shown that cerebral microdialysis, in traumatic brain injury, exhibits strikingly individualistic patterns that are identifiable throughout the analysis period. Because patients form their own clusters, microdialysis patterns, during periods of increased intracranial pressure or decreased cerebral perfusion pressure, will be found within these clusters. Consequently, no common pattern of microdialysis can be seen among patients within the range of our data. We suggest that these individualistic patterns reflect not only metabolic states of traumatic brain injury but also local gradients seen with small volume sampling. Future investigation should focus on relating these patterns, and movement within and from clusters, to metabolic states of the complex pathophysiology of traumatic brain injury.