Figure 2 shows the procedure when generating regressors for LEP in the TF-sMLR approach using the nonparametric approach based on PCA, while the same number of regressors has also been obtained for ERS and ERD. The first three principal component PCs obtained from the latency variability matrix explained 97.58 %, 95.30 %, and 95.36 % of the total variance for LEP, ERD, and ERS respectively, and the first three PCs obtained from the frequency variability matrix explained 96.60 %, 93.26 %, and 97.28 % of the total variance for LEP, ERD, and ERS respectively.

SEP is the most popular technique in spinal cord monitoring to avoid possible neurological complication because of its advantages of low cost, easy to use, non-invasive and higher successful rate. However, SEP monitoring is somewhat more technically challenging than other evoked potential methods because of its weaker signal and extremely low SNR, variable waveform and time-consuming acquisition. Ensemble averaging (EA) is the most widely used method, which unavoidably requires a longer measurement time. Significant irreversible neurological injury might have already occurred during the lengthy data collection time. In addition, there are a lot of features in single trial SEP which were ignored by averaging a relatively large number of consecutive trials. We developed a new method to extract SEP with rare channel by using one-unit second-order blind identification with reference (SOBI-R) algorithm (Fig. 3).