Dynamic Prediction Model Of Somatosensory Evoked Potential Abnormal Warning For Intraoperative Spinal Cord Monitoring

Somatosensory-evoked potentials (SEPs) have been widely used for intraoperative spinal cord monitoring to minimize possible risks during spinal surgery. SEP signals recorded in the operating room are always embedded in considerable noise. In addition, there is variability in somatosensory evoked potentials resulting from the effect of multiple factors. Therefore, the detection and intraoperative variability of somatosensory evoked potential is very difficult to solve in clinical spinal cord monitoring.In this paper, for the diversity of factors influencing evoked potential variability and the complexity of the relationship between various factors, we made multiple factors analysis of the main non-surgical factors affecting the somatosensory evoked potential variability using grey theory; for the low SNR of somatosensory evoked potential and requirement of intraoperative real-time monitoring, constrained second-order blind source separation algorithm was applied to the signal-to-noise ratio; On this basis, the support vector regression was used to establish dynamic warning model for abnormal somatosensory evoked potential, which will improve the reliability of intraoperative monitoring. The main research contents and conclusions are as follows:(1) The variation rate of latencies and amplitudes were systematically analyzed to evaluate their correlations with anthropometry indicators and physiological parameters. The statistic analysis did not find any significant relationship between height, weight and age and SEP latency/amplitude changes. However, variations in latency and amplitude were correlated most closely with variations in temperature, PaCO2 and heart rate, followed by variations in diastolic and then systolic blood pressure. So specialists monitoring SEPs should take this into consideration when making decisions based on abnormal SEPs.(2) Established a spinal cord injury model on rat to determine the effectiveness of a proposed constrained second order blind identification method. According to the characteristics of the somatosensory evoked potential, we improved the traditional second-order blind identification algorithm, and performed single trial extraction of somatosensory-evoked potentials during contusion injury to the spinal cord. Compared with the traditional ensemble averaging method, results from constrained second order blind identification showed it can extract the clear single SEP waveform signal by a single trial and single trial somatosensory-evoked potential can predict spinal cord injury in the early stage, which not only reduce the extraction time, it will help to provide more information for early warning of spinal cord injury.(3) Proposed a support vector regression model for dynamic prediction of intraoperative somatosensory evoked potential changes associated with abnormal somatosensory evoked potential early warning. Considering non-surgical factors that resulting in the changes of somatosensory evoked potential and the nonlinear characteristics of the support vector regression, we established a somatosensory evoked potential dynamic prediction model, which can adjust monitoring baseline dynamically and gave a confidence interval under a certain confidence level. The predicted results showed that dynamic prediction model not only made a good fitting of the latency and amplitude, observed and predicted SEP has similar variation trend with different values, with acceptable errors. This not only increases the reliability of intraoperative monitoring, and reduces the monitoring error caused by the non-surgical factors. It will be helpful to develop an intelligent monitor model based expert system that can make a reliable decision for the potential spinal injury.This research systematically analyzed the non-surgical factors influencing SEP, mainly discussed the relationship between the main factors and variability of SEP, and used the animal experiments verified the dynamic tracking ability of single extraction technology for spinal cord injury. Finally, established dynamic early warning model of abnormal somatosensory evoked potential considering non-surgical factors influencing somatosensory evoked potential. This study can not only improve the reliability of somatosensory evoked potential monitoring, but also can understand better the state of spinal cord function at any time, which will help detect neurologic injury early and determine injury position to avoid irreversible neurological damage.