Abstract

The management of patients having traumatic spinal cord injury (TSCI) would benefit from understanding and monitoring of spinal cord metabolic states. We hypothesized that the metabolism of the injured spinal cord could be visualized using Kohonen self-organizing maps. Sixteen patients with acute, severe spinal cord injuries were studied. Starting within 72 hours of the injury, and for up to a week, we monitored the injury site hourly for tissue glucose, lactate, pyruvate, glutamate and glycerol using microdialysis as well as intraspinal pressure and spinal cord perfusion pressure. A hexagonal Kohonen map, which is an unsupervised, self-organizing topology-preserving neural network, was used to analyze 3,366 hours of monitoring data. We first visualized the different spinal cord metabolic states. Our data show that the injured cord assumes one or more of four metabolic states. Based on their metabolite profiles, we termed these states near-normal, ischemic, hypermetabolic and distal. We then visualized how patients' intraspinal pressure and spinal cord perfusion pressure affect spinal cord metabolism. This revealed that for more than 60 % of the time, spinal cord metabolism is patient-specific; periods of high intraspinal pressure or low perfusion pressure are not associated with specific spinal cord metabolic patterns. Finally, we determined relationships between spinal cord metabolism and neurological status. Patients with complete deficits have shorter periods of near-normal spinal cord metabolic states (7±4% versus 58±12%, P<0.01, mean±standard error) and more variable metabolic responses (metabolism spread in 70±11 versus 40±6 hexagons, P<0.05), whereas patients with incomplete neurological deficits have longer and less variable metabolic responses. By visualizing the microdialysis data, Kohonen maps allow us to see these metabolic responses, and may thus aid us in treating patients with acute spinal cord injuries. This article is protected by copyright. All rights reserved.