Time-frequency components of somatosensory evoked potentials in relation to the location of spinal cord injury

Abstract

Spinal cord injury usually causes permanent and devastating neurological deficits and disability. Surgical intervention can provide great help in SCI rehabilitation, but it requires accurate diagnosis of the exact location of neurological deficits caused by SCI. Current clinical level diagnosis methods depend largely on radiological examination which allows the anatomy of cervical spine to be inspected. These methods, however, do not provide accurate information regarding the deficits in neurological function of the spinal cord, because the correspondence between anatomical abnormalities and the existence of neurological impairment remains contentious. In clinical cases, such as multi-level cervical myelopathy, an accurate localization of the offending level with neurological deficits is still difficult. Previous studies showed that somatosensory evoked potentials (SEPs) consist of a series of components in the time-frequency domain that contain useful information about the physiological mechanism of the spinal cord. Changes of the characteristics of these components, therefore, are potential indicator of the location of neurological deficits in the spinal cord. In this study, relationship between SEP time-frequency components and the location of spinal cord injury is investigated. To validate that time-frequency features of SEPs are associated with the location of spinal cord injury, time-frequency distributions of SEPs following spinal cord injury at different levels were first investigated. Results showed that various injury locations could lead to distinct distribution patterns of SEPs, which verified this association. Next, to confirm whether there is location-specific information contained in the SEP time-frequency features, time-frequency patterns of SEP components after acute and chronic injury at the same level of the spinal cord were compared with each other. Findings inferred that SEP components are likely to possess information concerning the site of neurological deficits in the spinal cord while independent of the modality of injury. Findings in these two pilot studies suggest that the time-frequency components of SEPs may be useful in identifying the location of spinal cord injury. We proposed that this can be seen as a pattern recognition problem. A supervised classification method was then designed using SEP recordings following chronic spinal cord compression. Further, relationship between time-frequency distributions of SEP components following single- and multi-level spinal cord injury was explored. The distribution of multi-level injury was found to be strongly correlated with the distribution of the corresponding single-levels while weakly correlated with the distribution of other single-levels. This finding suggests that it is possible to simulate the SEP distribution of multi-level injury using single-level distributions by regression analysis. Finally, a hybrid dataset containing SEP recordings after acute and chronic spinal cord injury was used to train an optimized classifier. Its good performance demonstrated the usefulness of time-frequency components in detecting the location of spinal cord injury.