Study On The Role Of Time-Frequency Components Of Somatosensory Evoked Potentials In Accurate Diagnosis Of Spinal Cord Injury

BACKGROUND Iatrogenic spinal cord injury may occur in spinal surgery.If intraoperative spinal cord monitoring technology can prompt the precise location and pattern of spinal cord injury,it will help to detect and remove the source of injury as soon as possible,thus reducing or avoiding the spinal cord injury.Somatosensory evoked potentials are used in clinical practice to assist in the detection of spinal cord disease and the monitoring of intraoperative spinal cord function.However,in the past,intraoperative spinal cord monitoring was mainly based on the changes in the latency and amplitude of somatosensory evoked potentials to check the integrity of somatosensory conduction pathways.Many useful information in the evoked potentials was not fully utilized.The time-frequency analysis method can reveal many small components in the evoked potential.In the time-frequency domain,the detailed information of somatosensory evoked potentials can be extracted effectively.Studies have reported that the support vector machine method can use the time-frequency distribution pattern of somatosensory evoked potentials to accurately identify the location of cervical spinal cord injury at C4,C5,and C6.However,as the number of segments of spinal cord injury included in the study increases,there will be serious linear inseparability problems,making it is difficult to achieve accurate diagnosis of the location and pattern of spinal cord injury.Therefore,optimizing the time-frequency component analysis technology of somatosensory evoked potentials and establishing an accurate diagnosis method for the location and pattern of spinal cord injury to the whole spinal range is a research direction with a certain theoretical basis and clinical application value.METHODS In this paper,we used the hind limb somatosensory evoked potential data from spinal cord injury model to diagnose the location and pattern of the spinal cord injury.The spinal cord injury locations included C5 and C6 of the cervical spine,T1-T4 and T7-T13 of the thoracic spine,and L1-L6 of the thoracic spine.The spinal cord injury patterns were contusion and distraction.The somatosensory evoked potentials denoising procedure included 200 times averaging and 10-250 Hz band-pass filtering.Then the time-frequency decomposition was performed using a matched pursuit algorithm,and the obtained temporal frequency components could be described by latency,frequency,and energy.In this paper,a time-frequency feature extraction method based on k-centroid clustering was designed according to the distribution pattern of time-frequency components of somatosensory evoked potentials.This method divided the time-frequency domain into different sub-regions,and treated each sub-region as a different time-frequency feature.A noise component recognition scheme applicable to this feature extraction method was also designed:the features with less than 1.4%of the total number of components of current group were regarded as noise and deleted;the components corresponding to outlier points in each direction of latency,frequency and energy in each feature were also regarded as noise and deleted.A method of classifying the time-frequency components of somatosensory evoked potentials based on the Naive Bayesian principle was introduced,which could utilize the quantitative characteristics of time-frequency components in the features ignored by the previous study.The analysis method about time-frequency components was applied to the accurate diagnosis of spinal cord injury location in the cervical,thoracic,and lumbar spine,as well as the identification of contusions and distraction spinal cord injury patterns.Stable time-frequency features associated with the location and pattern information of spinal cord injury were identified.RESULTS In the experiment of detecting the integrity of the somatosensory conduction pathway,the new time-frequency analysis method obtained an accuracy of 90.5%,which was much higher than the 69.5%accuracy of the support vector machine method used in the existing research.The method was then applied to the hind limb somatosensory evoked potential data of the spinal cord injury model.The accuracy of identifying the position of C5 and C6 of the cervical spine was 86.1%;the accuracy of identifying the position of the upper thoracic,middle and lower thoracic spine was 79.2%,and the accuracy of identifying the position of the upper and lower lumbar spine was 81.8%.In the spinal cord injury pattern classification task,the average recognition accuracy of each individual segment data was 89.9%,and the average accuracy of the whole spine was 78.6%.The stable time-frequency features associated with SCI location and pattern were all concentrated in the time-frequency region of 20-35 ms of latency;in this region,changes in spinal cord injury location and pattern mainly affected the frequency and energy parameters of the components.In other regions,all three parameters of latency,frequency,and energy were influenced by the location or pattern of spinal cord injury.CONCLUSION In this study,a new method of time-frequency analysis of somatosensory evoked potentials based on the Naive Bayesian principle was introduced to identify spinal cord injuries with a higher capability,and its detection effect was significantly better than that of the existing support vector machine method.The new method has obtained a high accuracy rate for the accurate diagnosis of the location of spinal cord injury,demonstrating that spinal cord injury located in different positions of the spine can cause different time-frequency distribution patterns of somatosensory evoked potentials,and the distribution patterns can be used for accurate diagnosis of spinal cord injury location.In addition,the new method not only achieved a high accuracy in each classification task of spinal cord injury patterns in a single spinal segment,but also has the ability to diagnose spinal cord injury patterns in the whole spine.The distribution pattern of the temporal frequency components of somatosensory evoked potentials was demonstrated to be useful for the detection of spinal cord injury patterns.Although changes in the position of spinal cord injury in the whole spine would affect the distribution of time-frequency components,there are still some stable time-frequency features associated to the spinal cord injury pattern,which enables the accurate diagnosis of the spinal cord injury pattern in the whole spine.