Feature enhancement techniques for the detection of spinal cord injury

Evoked potentials (EPs) produced by median nerve stimulation and recorded from the somatosensory receiving area of cat cortex were obtained from anaesthetized cats and averaged before and after one of two treatments. Seven data transformations were applied to the average EPs with the idea of emphasizing features of the EP pertinent for classification. The transformations applied involved both time and frequency domain representation of the original average EPs. Statistical multivariate classification techniques were then used to obtain canonical plots of the original and transformed EPs. In every case, frequency domain analysis resulted in better separation between treatment categories. It was also found that the power spectrum which was normalized by its average power proved the best for classification purposes.