Pattern-recognition program for analysis of colon myoelectric and pressure data

Abstract

A pattern-recognition program was developed which emulates visual scoring of colonic myoelectric and pressure recordings. It smoothes digitized data with a moving average filter, computes difference scores between successive groups of three data points, and uses the signs of these difference scores to detect the beginning and end of waves. Adjacent waves are merged if their means are closer than 1.67 times the sum of their standard deviations, and amplitude and duration criteria are used to exclude nonsignificant waves. When compared to four experienced human scorers on randomly selected records, the program agreed as well with the human scorers as they agreed with each other, and it approached the level of agreement of these observers with themselves when they were asked to rescore the same records blindly four to six weeks later. Human scorers agreed with themselves on 36–71% of myoelectric slow waves and on 42–88% of pressure waves, compared to 100% test-retest reliability for the pattern-recognition program. Frequency histograms of the duration of waves detected by the pattern-recognition program differed from the spectra generated by the fast Fourier transform (FFT) method. This pattern-recognition program provides an alternative to spectral analysis for the reliable and objective quantification of colonic myoelectric slow waves and pressure waves.